CN116916195A - Passive optical network management method, device and readable storage medium - Google Patents

Abstract

The application discloses a passive optical network management method, equipment and a readable storage medium, belonging to the technical field of communication, comprising the following steps: integrating optical power data of ONU in PON and topology data registered by ONU in resource management system, and performing data preprocessing and feature extraction processing to obtain time slice sequence; sampling a sequence pair in the time slice sequence to obtain training data and test data; inputting training data into the first model group for training to obtain a second model group; inputting the test data into a second model group to obtain a sequence pair scoring; integrating time dimension and ONU layer of sequence pair dividing, and determining recommended secondary beam splitter corresponding to ONU; according to the recommended secondary optical splitter corresponding to the ONU and the original topological structure of the PON in the resource management system, carrying out noise reduction treatment on the original topological structure; the first model group comprises a plurality of regression models, and the second model group comprises a plurality of trained regression models.

Description

Passive optical network management method, equipment and readable storage medium

Technical field

This application belongs to the field of communication technology, and specifically relates to a passive optical network management method, equipment and readable storage medium.

Background technique

Personal home business areas. In recent years, with the continuous advancement of Passive Optical Network (PON) network communication technology, PON network applications have developed rapidly, and users have higher requirements for network performance. PON technology is a point-to-multipoint optical fiber access technology. It consists of an optical line terminal (Optical Line Terminal, OLT) on the office side, an Optical Network Unit (ONU) on the user side, and an optical distribution network (Optical Distribution Network, ODN). composition. There are four types of networking methods for PON systems: tree topology, ring topology, bus topology, and tree trunk redundant topology. The most common one is tree topology, as shown in Figure 1. The topological structure makes PON services more transparent, and in principle can be applied to signals of any standard and speed. At the same time, compared with point-to-point active optical networks, PON technology also has the characteristics of simple maintenance, low cost, and high transmission bandwidth, giving PON a strong competitive advantage and is regarded as the future access network. Direction of development. But at the same time, due to the "passive" nature of the PON network, the tree topology between the OLT and the ONU is not clear, and it is difficult to obtain in real time which secondary optical splitter in the ODN an ONU is connected to.

Contents of the invention

Embodiments of the present application provide a passive optical network management method, equipment and a readable storage medium, which can solve the current problem of difficulty in obtaining in real time which secondary optical splitter a certain ONU is connected to.

In the first aspect, a passive optical network management method is provided, including:

Integrate the optical power data of the ONU in the PON and the topology data registered by the ONU in the asset management system, and perform data preprocessing and feature extraction to obtain a time slice sequence;

Sample the sequence pairs in the time slice sequence to obtain training data and test data;

Input the training data into the first model group for training to obtain the second model group;

Input the test data into the second model group to obtain sequence pair scores;

Integrate the sequence pair scoring with the time dimension and the ONU level to determine the recommended secondary optical splitter corresponding to the ONU;

According to the recommended secondary optical splitter corresponding to the ONU and the original topology of the PON in the asset management system, perform noise reduction processing on the original topology;

Wherein, the first model group includes multiple regression models, and the second model group includes multiple trained regression models.

Optionally, sampling the sequence pairs in the time slice sequence to obtain training data and test data includes:

Traverse all sequence pairs located at the same first-level optical splitter and the same time in the time slice sequence;

The sequence pairs in which the two sequences belong to the same secondary spectrometer are marked as positive samples, and the sequence pairs in which the two sequences belong to different secondary spectrometers are marked as negative samples;

According to the positive sample and the negative sample, training data and test data are obtained.

Optionally, the training data is input into the first model group for training to obtain the second model group, including:

Input the training data into the first convolutional neural network CNN model for training to obtain the second CNN model;

Input the training data into the first recurrent neural network RNN model for training to obtain the second RNN model;

The training data is input into the first gradient boosting decision tree GBDT model for training to obtain the second GBDT model.

Optionally, inputting the test data into the second model group to obtain a sequence pair score includes:

Input the test data into the second CNN model to obtain the first score;

Input the test data into the second RNN model to obtain a second score;

Enter the test data into the second GBDT model to obtain a third score;

The first score, the second score and the third score are averaged to obtain the sequence pair score.

Optionally, the integration of the time dimension and the ONU level for the sequence pair scoring, and determining the recommended secondary optical splitter corresponding to the ONU includes:

For any ONU pair, select the highest sequence pair score as the sequence pair score of the ONU pair;

Determine multiple candidate ONU groups;

Calculate the average sequence pair score between the target ONU and the candidate ONU in each candidate ONU group;

Determine the recommended secondary spectrometer corresponding to the target ONU according to the candidate ONU group with the highest mean score in the sequence;

Each candidate ONU group includes multiple candidate ONUs under the same primary optical splitter, and each candidate ONU group corresponds to a different secondary optical splitter.

Optionally, the integration of the time dimension and the ONU level for the sequence pair scoring, and determining the recommended secondary optical splitter corresponding to the ONU includes:

Integrate and fill all the sequence pairs corresponding to the target ONU and the scores of all sequence pairs corresponding to the target ONU into a three-dimensional matrix. The three-dimensional matrix includes the time dimension, different secondary spectrometer dimensions and preset time period data existing data. Set the time period to other ONU dimensions that are different from the default secondary optical splitter;

Integrate the time dimension of the three-dimensional matrix and the different secondary spectrometer dimensions of the preset time period existing data to obtain a two-dimensional matrix;

Input the two-dimensional matrix into the Self-attention model to obtain the secondary spectrometer score of the target ONU and each secondary spectroscope;

According to the secondary optical splitter score, the recommended secondary optical splitter corresponding to the target ONU is determined.

In a second aspect, a passive optical network management device is provided, including:

The data preprocessing and feature extraction module is used to integrate the optical power data of the ONU in the PON and the topology data registered by the ONU in the asset management system, and perform data preprocessing and feature extraction to obtain a time slice sequence;

A sampling module, used to sample sequence pairs in the time slice sequence to obtain training data and test data;

A model training module, used to input the training data into the first model group for training to obtain the second model group;

A scoring module, used to input the test data into the second model group to obtain a sequence pair score;

Determining module, used to integrate the time dimension and ONU level for the sequence pair scoring, and determine the recommended secondary optical splitter corresponding to the ONU;

A processing module, configured to perform noise reduction processing on the original topology according to the recommended secondary optical splitter corresponding to the ONU and the original topology of the PON in the asset management system;

Wherein, the first model group includes multiple regression models, and the second model group includes multiple trained regression models.

Optionally, the sampling module is specifically used for:

Traverse all sequence pairs located at the same first-level optical splitter and the same time in the time slice sequence;

The sequence pairs in which the two sequences belong to the same secondary spectrometer are marked as positive samples, and the sequence pairs in which the two sequences belong to different secondary spectrometers are marked as negative samples;

According to the positive sample and the negative sample, training data and test data are obtained.

Optionally, the model training module is specifically used for:

Input the training data into the first CNN model for training to obtain the second CNN model;

Input the training data into the first RNN model for training to obtain the second RNN model;

The training data is input into the first GBDT model for training to obtain the second GBDT model.

Optionally, the scoring module is specifically used for:

Input the test data into the second CNN model to obtain the first score;

Input the test data into the second RNN model to obtain a second score;

Enter the test data into the second GBDT model to obtain a third score;

The first score, the second score and the third score are averaged to obtain the sequence pair score.

Optionally, the determination module is specifically used for:

For any ONU pair, select the highest sequence pair score as the sequence pair score of the ONU pair;

Determine multiple candidate ONU groups;

Calculate the average sequence pair score between the target ONU and the candidate ONU in each candidate ONU group;

Determine the recommended secondary spectrometer corresponding to the target ONU according to the candidate ONU group with the highest mean score in the sequence;

Each candidate ONU group includes multiple candidate ONUs under the same primary optical splitter, and each candidate ONU group corresponds to a different secondary optical splitter.

Optionally, the determination module is specifically used for:

Integrate and fill all the sequence pairs corresponding to the target ONU and the scores of all sequence pairs corresponding to the target ONU into a three-dimensional matrix. The three-dimensional matrix includes the time dimension, different secondary spectrometer dimensions and preset time period data existing data. Set the time period to other ONU dimensions that are different from the default secondary optical splitter;

Integrate the time dimension of the three-dimensional matrix and the different secondary spectrometer dimensions of the preset time period existing data to obtain a two-dimensional matrix;

Input the two-dimensional matrix into the Self-attention model to obtain the secondary spectrometer score of the target ONU and each secondary spectroscope;

According to the secondary optical splitter score, the recommended secondary optical splitter corresponding to the target ONU is determined.

In a third aspect, an electronic device is provided, which is characterized in that it includes a processor and a memory. The memory stores programs or instructions that can be run on the processor. When the program or instructions are executed by the processor, The steps of implementing the passive optical network management method as described in the first aspect.

A fourth aspect provides a readable storage medium, characterized in that the readable storage medium stores programs or instructions, and when the programs or instructions are executed by a processor, the passive light source as described in the first aspect is implemented. Network management method steps.

In a fifth aspect, a chip is provided. The chip includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement wireless communication as described in the first aspect. Steps of source optical network management method.

In a sixth aspect, a computer program/program product is provided, the computer program/program product is stored in a storage medium, and the program/program product is executed by at least one processor to implement the wireless operation as described in the first aspect. Steps of source optical network management method.

In the embodiment of this application, data preprocessing and feature extraction are performed on the ONU's optical power data and the ONU's topology data registered in the asset management system to obtain a time slice sequence, and sample the sequence pairs in the time slice sequence. , realizes sampling of ONU levels under the same secondary optical splitter according to time slices, greatly increasing the amount of data that can be used for model learning; using a regression model to score sequence pairs, realizing a method based on the "regression" concept for "ONU pairs" Similarity scoring; using the technical solution of this application, the deep learning method is introduced into the field of PON topology information prediction and management, automatically obtaining the secondary optical splitter recommendation of any ONU, and registering errors in the PON network topology in real time. correct.

Description of the drawings

Figure 1 is an existing PON tree topology diagram;

Figure 2 is a schematic flowchart of a passive optical network management method provided by an embodiment of the present application;

Figure 3a is a schematic diagram of the overall principle of PON management provided by the embodiment of the present application;

Figure 3b is an overall operation flow chart of PON management provided by the embodiment of the present application;

Figure 3c is a schematic diagram of the PON data preprocessing and feature extraction principles provided by the embodiment of the present application;

Figure 3d is a schematic diagram of the PON data preprocessing and feature extraction operation flow provided by the embodiment of the present application;

Figure 3e is a flow chart of sampling and model training operations provided by the embodiment of the present application;

Figure 3f is a schematic diagram of the principles of PON sampling and model training CNN provided by the embodiment of this application;

Figure 3g is a schematic diagram of the PON sampling and model training RNN principle provided by the embodiment of this application;

Figure 3h is a flow chart of voting prediction operation provided by the embodiment of this application;

Figure 3i is one of the schematic diagrams of the PON voting prediction principle provided by the embodiment of this application;

Figure 3j is the second schematic diagram of the PON voting prediction principle provided by the embodiment of this application;

Figure 4 is a schematic structural diagram of a passive optical network management device provided by an embodiment of the present application;

FIG. 5 is a schematic structural diagram of an electronic device provided by an embodiment of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be clearly described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, but not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art fall within the scope of protection of this application.

The terms “first”, “second”, etc. used in this application are used to distinguish similar objects and are not used to describe a specific order or sequence. It is to be understood that the terms so used are interchangeable under appropriate circumstances so that the embodiments of the present application can be practiced in sequences other than those illustrated or described herein, and that "first" and "second" are distinguished objects It is usually one type, and the number of objects is not limited. For example, the first object can be one or multiple. In addition, "and/or" in this application means at least one of the connected objects. For example, "A or B" covers three options, namely, option one: including A and excluding B; option two: including B but excluding A; option three: including both A and B. The character "/" generally indicates that the related objects are in an "or" relationship.

During the research on technical issues, the inventor discovered:

In practice, the registered topology data may not be updated accurately due to actual changes such as incorrect insertion, omission, or machine relocation. In practice, the accuracy rate is only 50% to 60%. Network managers need to grasp the analysis in a timely manner and strengthen the management of various PON faults to solve network performance problems as soon as possible. Currently there are three solutions: manual implementation, hardware implementation and software implementation. Manual implementation requires arranging professional staff to inspect ONUs one by one, and introducing systematic inspection methods; hardware implementation requires installing a reflective chip on the optical splitter to achieve the "active" effect by reflecting light; software implementation uses The timing data information such as the transmit and receive power of each ONU obtained from the OLT. When ONUs under the same secondary optical splitter receive disturbances from the optical splitter, these timing data will show similar characteristics, so that clustering can be performed.

Related algorithms:

1. Clustering: The process of dividing a collection of physical or abstract objects into multiple classes consisting of similar objects is called clustering. A cluster generated by clustering is a collection of data objects that are similar to each other in the same cluster and different from objects in other clusters. "Birds of a feather flock together, and people divide into groups." There are a large number of classification problems in natural sciences and social sciences. Cluster analysis, also known as group analysis, is a statistical analysis method for studying (sample or indicator) classification problems. Cluster analysis originated from taxonomy, but clustering is not equal to classification. The difference between clustering and classification is that the classes required for clustering are unknown. Cluster analysis is very rich in content, including systematic clustering method, ordered sample clustering method, dynamic clustering method, fuzzy clustering method, graph theory clustering method, cluster prediction method, etc.

2.K-Means: k-means clustering algorithm is an iterative clustering analysis algorithm. The steps are to pre-divide the data into K groups, and then randomly select K objects as the initial cluster center, then calculate the distance between each object and each seed cluster center, and assign each object to the cluster center closest to it. Cluster centers and the objects assigned to them represent a cluster. Each time a sample is assigned, the cluster centers are recalculated based on the existing objects in the cluster. This process will be repeated until a certain termination condition is met. The termination condition can be that no (or a minimum number of) objects are reassigned to different clusters, no (or a minimum number of) cluster centers change anymore, and the sum of squared errors is locally minimized.

3.DBScan: Density-Based SpatialClustering of Applications with Noise (Density-Based SpatialClustering of Applications with Noise) is a relatively representative density-based clustering algorithm. Different from partitioning and hierarchical clustering methods, it defines clusters as the largest set of density-connected points, can divide regions with sufficiently high density into clusters, and can discover clusters of arbitrary shapes in noisy spatial databases.

4. DTW algorithm: There are many similarity or distance functions in time series data, the most prominent of which is DTW. In isolated word speech recognition, the simplest and most effective method is to use the Dynamic Time Warping (DTW) algorithm. This algorithm is based on the idea of dynamic programming (DP) and solves the template matching problem of different pronunciations. It is An earlier and more classic algorithm that appeared in speech recognition is used for isolated word recognition. The HMM algorithm requires a large amount of speech data during the training phase, and model parameters can be obtained through repeated calculations, while the DTW algorithm requires almost no additional calculations during training. Therefore, the DTW algorithm is still widely used in isolated word speech recognition.

The current solution has the following shortcomings:

The shortcomings of manual inspection are 1) cost: due to the large number of ONUs, the cost of inspecting one ONU is about 10 yuan, and within a province, the cost of inspecting all ONUs will be as high as tens of millions of yuan. In actual operation Even if about 20% of ONUs are inspected every year, the annual expenditure reaches millions of yuan; 2) Credibility: Workers’ inspections (especially in large-scale inspections) involve fraud and sabotage, and quality is difficult to guarantee; 3) Sustainability: Every year there are new installations, relocations and disassemblies, which lead to changes in the optical network structure. Manual troubleshooting cannot grasp these changes in real time.

The shortcomings of hardware troubleshooting are 1) cost: the cost of each reflective chip is about 20 yuan, even higher than the cost of a manual inspection; 2) maintenance: maintenance after chip damage is difficult and requires additional costs.

Under the premise of considering cost and sustainability, software troubleshooting has become the main research object. The main challenge is that 1) changes in external factors such as pressure on the optical fiber need to reach a certain level before they can be detected. Capture feature values and compare Difficulty; 2) The collection of signals is discontinuous. Each ONU sends power data (instantaneous data) every 5 minutes, and the time when different ONUs send power signals cannot be guaranteed to be the same, so a lot of information cannot be captured; 3) Due to Due to the diversity of specific locations where disturbances occur, ONUs under the same secondary optical splitter may not necessarily show similar waveforms. Due to these challenges, so far, known software troubleshooting methods such as directly clustering raw data still have the following shortcomings: 1) The clustering process relies too much on manually formulated rules and cannot fully utilize the large amount of optical power in the system. Data; 2) In the face of optical power data with large noise and missing information, accurate clustering cannot be achieved; 3) When the accuracy of the topological structure registered by the system is low, the algorithm will be greatly affected.

The passive optical network management method provided by the embodiments of the present application will be described in detail below with reference to the accompanying drawings through some embodiments and application scenarios.

Referring to Figure 2, an embodiment of the present application provides a passive optical network management method, including:

Step 201: Integrate the optical power data of the ONU in the PON and the topology data of the ONU registered in the asset management system, and perform data preprocessing and feature extraction to obtain a time slice sequence;

Step 202: Sample the sequence pairs in the time slice sequence to obtain training data and test data;

Step 203: Input the training data into the first model group for training to obtain the second model group;

Step 204: Input the test data into the second model group to obtain sequence pair scores;

Step 205: Integrate the sequence pair scoring in the time dimension and the ONU level, and determine the recommended secondary optical splitter corresponding to the ONU;

Step 206: Perform noise reduction processing on the original topology according to the recommended secondary optical splitter corresponding to the ONU and the original topology of the PON in the asset management system;

The first model group includes multiple regression models, and the second model group includes multiple trained regression models.

In the embodiment of this application, data preprocessing and feature extraction are performed on the ONU's optical power data and the ONU's topology data registered in the asset management system to obtain a time slice sequence, and sample the sequence pairs in the time slice sequence. , realizes sampling of ONU levels under the same secondary optical splitter according to time slices, greatly increasing the amount of data that can be used for model learning; using a regression model to score sequence pairs, realizing a method based on the "regression" concept for "ONU pairs" Similarity scoring; using the technical solution of this application, the deep learning method is introduced into the field of PON topology information prediction and management, automatically obtaining the secondary optical splitter recommendation of any ONU, and registering errors in the PON network topology in real time. correct.

It should be noted that the topology data registered in the above-mentioned asset management system is the recorded PON network topology structure data, but this data is not meant to be completely accurate. The data may be unupdated data, for example, due to incorrect insertion or leakage. Actual changes such as unplugging or relocating a machine will cause changes in the topology, and accurate updates cannot be guaranteed.

The overall principle of the technical solution of this application can be shown in Figure 3a;

The above step 201 can be summarized as a data preprocessing and feature extraction module, data preprocessing and feature engineering, which integrates the original ONU optical power data and the topology data registered in the asset management system, and transforms it into a module suitable for processing by the next module. sequence form, and uses a variety of feature engineering to extract effective information in the sequence;

The above steps 201 to 204 can be summarized as a sampling and model training module. According to the different belonging secondary spectrometers registered in the asset management system, positive and negative sequence samples are sampled from the sequence, and the deep learning model is used to capture any two The consistency of the sequence optical power fluctuations and a similarity score are given;

The above steps 205 to 206 can be summarized as a voting prediction module, which integrates the similarity of any sequence pair in the time dimension and the ONU level to obtain the ONU's scores for different secondary optical splitters, thereby matching the ONU to the corresponding secondary optical splitter. Optical splitter corrects errors in topological information in the asset management system and obtains a complete and correct topology.

The overall process of the technical solution of this application can be referred to as shown in Figure 3b. Each step is explained in detail below:

Referring to Figure 3c and Figure 3d, for step 201, integrating the optical power data of the ONU and the topology data registered in the asset management system, and performing data preprocessing and feature extraction processing may include the following processes:

(1) Integrate optical power data and topology data;

(2) Filter all empty accounts;

(3) Segment the data with a certain length to form numerous time slice sequences: for a given ONU, the length of time involved may reach several months or even half a year, and the duration of the disturbance is mostly at the hour level, so it is necessary to Segment the long-term optical power data of the ONU to form numerous time slice sequences. Through observation and analysis of the actual disturbance time span, we have clarified that the length of one day is used as the standard for dividing time slices;

(4) Eliminate time slice sequences with too high a proportion of null values: The probe located on the ONU will collect data every 10 minutes. However, in actual production and life, the probe cannot always be in working state, and there are considerable problems. In a large proportion of cases, the probe cannot collect data and returns a null value. If the proportion of null values in a slice sequence is too high, then truly meaningful perturbations may be skipped, while other error information will be captured by the model, so sequences with too high a proportion of null values need to be filtered out. Taking into account the number of remaining sequences, the number of ONUs, and the proportion of effective information in the sequence, we confirmed that sequences with a proportion of null values higher than 20% were deleted;

(5) Normalize the sequence: This model aims to predict the ONU topology of different cells. However, there are large differences in basic information such as the mean and variance of optical power in different cells, which affects the generalization performance of the model. There is a huge effect, so normalization to the first-order beamsplitter is required. In addition, disturbance information is the information that the model of this project needs to prioritize capturing, and basic information such as the mean and variance of different secondary spectrometers may also be quite different, which will affect the learning focus of the model. Based on various considerations, we confirmed that the sequence is normalized at the ONU level;

(6) Perform a sliding average on the sequence to capture features at different scales;

(7) Eliminate abnormal points.

It can be understood that the values of some parameters involved in the above process are only examples and can be flexibly adjusted according to actual needs, and do not constitute a limitation on the technical solution of the present application.

The sampling and model training operation process in the embodiment of this application can be referred to as shown in Figure 3e;

Among them, for step 202, sample the sequence pairs in the time slice sequence to obtain training data and test data, including:

(1) Traverse all sequence pairs in the time slice sequence that are located at the same first-level spectrometer and at the same time;

(2) Mark the sequence pairs in which the two sequences belong to the same secondary spectrometer as positive samples, and mark the sequence pairs in which the two sequences belong to different secondary spectrometers as negative samples;

(3) Obtain training data and test data based on positive samples and negative samples.

In the embodiment of this application, for the data set, we traverse all sequence pairs located in the same primary spectrometer and at the same time. If the two sequences belong to the same secondary spectrometer, they are marked as positive samples, otherwise they are negative samples.

Optionally, for the test data set, we traverse all pairs of sequences located at the same primary beam splitter and at the same time. If the two sequences belong to the same secondary beam splitter, they are marked as positive samples, otherwise they are marked as negative samples. For the training data set, we can appropriately relax the sampling requirements for negative samples, allowing them not to be located in the same first-level optical splitter or at the same time, so as to reduce the impact of errors in the topological information registered by the asset management system on training.

For step 203, input the training data into the first model group for training, and obtain the second model group, including:

(1) Input the training data into the first Convolutional Neural Networks (CNN) model for training to obtain the second CNN model;

(2) Input the training data into the first Recurrent Neural Network (RNN) model for training to obtain the second RNN model;

(3) Input the training data into the first Gradient Boosting Decision Tree (GBDT) model for training to obtain the second GBDT model.

It should be noted that the above input of training data into three models is a parallel process, that is, the above (1) to (3) can be executed simultaneously.

The above CNN and RNN correspond to the deep neural network in Figure 3e.

In the embodiment of this application, three regression models, CNN, RNN, and GBDT, are used for similarity training. Different from the current conventional DTW+DBSCANE method, the conventional method is a 'clustering' concept, which classifies multiple ONUs at the same time. What we use is a 'regression' concept, and we also score the similarity of each two onu pairs.

For step 204, input the test data into the second model group to obtain sequence pair scores, including:

(1) Enter the test data into the second CNN model to obtain the first score;

(2) Enter the test data into the second RNN model to obtain the second score;

(3) Enter the test data into the second GBDT model to obtain the third score;

(4) Take the average of the first score, the second score and the third score to obtain the sequence pair score.

It should be noted that the above input of test data into three models is a parallel process, that is, the above (1) to (3) can be executed synchronously.

The above step (4) can be understood as an "integrated learning" link, that is, as shown in Figure 3e, after training three models with training data, integrated learning can be logically understood as unifying them into one model. After the test data passes through the model, the sequence pairs are scored to finally check the prediction effect. Specifically, the data will flow into three models at the same time, a score will be calculated respectively, and then the three scores will be averaged to summarize a final score.

The specific training and calculation processes of the three models are described below:

Deep neural network model: We use an integrated model of CNN and RNN for recommendation. We randomly initialize the prediction model parameters $\theta$, use the stochastic gradient descent algorithm, and use the hyperparameters $\phi$ to optimize the AUC index of the prediction model on the training set, and search for the optimal training hyperparameters $\ based on the results on the test set. hat{\phi}$, trains the prediction model with optimal hyperparameters on all training data, and gives a score for any sequence pair on the test data.

Noisy learning: Considering that the topological structure in the system is inaccurate, and the labeling of positive and negative samples is not necessarily correct, we adopt a noisy learning method based on peer loss (peer loss), and randomly select from the data set during the learning process. Sample a noisy label for training to improve the accuracy of the model learning in an inaccurate label environment.

CNN: Our convolutional neural network uses one-dimensional convolutions:

in:

S _uv represents the u-th element of the v-th channel of the output sequence S.

Represents the element in the i-th row and j-th column of the weight matrix of the v-th convolution kernel. Here, i corresponds to the one-dimensional index of the convolution kernel (ranging from 1 to R), and j corresponds to the channel index of the input sequence h (ranging from 1 to C).

h _t(u-1)-p+i,j represents the element of the (t(u-1)-p+i)th row and jth column of the input sequence h. Here, t(u-1) is the step size multiplied by (u-1), which represents the distance the convolution kernel slides on the input sequence. Subtracting p takes into account the effect of padding, and adding i is the local index of the convolution kernel when it slides over the input sequence.

b ^v represents the bias term of the v-th convolution kernel.

i represents the row index of the convolution kernel weight matrix w ^v (corresponding to the one-dimensional index of the convolution kernel), ranging from 1 to R.

j represents the column index of the convolution kernel weight matrix w ^v (corresponding to the channel index of the input sequence h), ranging from 1 to C.

R represents the size of the convolution kernel (kernel_size).

C represents the number of input channels (in_channel).

u represents the row index of the output sequence S.

v represents the channel index of the output sequence S.

p represents the padding size (padding).

t represents the stride.

We first aggregate the two sequences in the sequence pair, convert them to specified dimensions using a linear model, and then process them through a multi-layer convolutional neural network. The schematic diagram is shown in Figure 3f. The output of the convolutional neural network

$output=[x_1, x_2, ldots, x_h]$ The final result is obtained after linear layer processing:

{\rm score}=\varphi(\theta_0+\theta_1x_1+\cdots+\theta_hx_h)\quad{

\rm where}\quad\varphi(x)＝\dfrac{1}{1+e^{-x}}

Score is the score given by our convolutional neural network model on the similarity of this sequence to the sample.

RNN: Our recurrent neural network uses the GRU standard model, where t is our sequence data time point:

h ^t-1′ ＝h ^t-1 ⊙r ^t

h ^t = (1-z ^t )⊙h ^t-1 +z ^t ⊙h ^t′

Among them, r ^t is the reset gate at time t. The reset gate is used to control the influence of the hidden state information of the previous time step on the candidate hidden state. z ^t is the update gate at time t. The update gate is used to control the degree of retention of the hidden state information of the previous time step in the current time step. and/> are the weight matrices of the reset gate and the update gate at time t respectively, x ^t is the input at time t, h ^t-1 is the hidden state at time t-1, h ^t′ represents the candidate hidden state at time t, h ^t represents t The new hidden state at time, sigmoid is the activation function, and tanh represents the hyperbolic tangent activation function.

First, a separate GRU is used to process the two sequences in the sequence pair separately, and then a unified GRU is used for processing after aggregation. The schematic diagram is shown in Figure 3g. The output of the recurrent neural network is processed similarly to the convolutional neural network to obtain the final score of sequence pair similarity.

Gradient boosting decision tree: We directly compare the sequence to the sample or extract features, and then use the gradient boosting iterative tree to process. We can first construct a decision tree with only the root node based on the existing data:

f ₀ (x)＝argmin _y ∑L(y _i γ)

Among them, L is the loss function; f ₀ (x) represents the calculation initialization prediction value; y represents the real value, i=1, 2,..., N, N is the number of training samples; γ represents the current model prediction value. Then iteratively do the following:

A. For each sample, calculate the residual of the current model:

r _mi =y _i -f _m-1 (x _i ), i = 1, 2...N

Among them, r _mi is the residual of this round, _yi is the true value of the i-th sample, and f _m-1 ( _xi ) is the predicted value of the first m-1 decision tree combination.

B. Fit the residual r _mi to learn a regression tree and obtain T _m (x).

C. Combine the regression tree and the model of the previous iteration to update the model:

f _m (x)＝f _m-1 (x)+T _m (x)

f _m (x) is the updated model, f _m-1 (x) is the previous round model, and T _m (x) is the newly learned model.

In this way, the gradient boosting iterative tree learns high-order nonlinear feature combinations through many decision trees to improve the final effect. The final formula is as follows, where M is the number of trees and F _M (x) is the final model:

Ensemble learning: Considering that different models can capture different information in the data, in order to improve the accuracy of the final model, we average the output of the above-trained models as the final sequence pair score and output it.

The running process of the voting prediction module program in the embodiment of this application can be referred to as shown in Figure 3h;

Regarding step 205, in an optional implementation, the sequence pair scoring is integrated in the time dimension and the ONU level to determine the recommended secondary optical splitter corresponding to the ONU, including:

(1) For any ONU pair, select the highest sequence pair score as the sequence pair score of the ONU pair;

(2) Determine multiple candidate ONU groups;

(3) Calculate the mean sequence pair score between the target ONU and the candidate ONU in each candidate ONU group;

(4) Based on the candidate ONU group with the highest mean score in the sequence, determine the recommended secondary spectrometer corresponding to the target ONU;

Each candidate ONU group includes multiple candidate ONUs under the same primary optical splitter, and each candidate ONU group corresponds to a different secondary optical splitter.

In the embodiment of this application, the time dimension and ONU level are integrated successively;

In the time dimension, since disturbances are random events, there are a large number of sequence pairs that do not capture the consistency of the disturbance. Therefore, for any ONU pair, we select the highest sequence pair score as the final score of the ONU pair. At the ONU level, we average the scores of all ONUs under the secondary optical splitter to obtain the ONU score for any secondary optical splitter. The principle is shown in Figure 3i. For each ONU, its True Label (the secondary optical splitter to which it belongs) objectively exists, but it is unknown to us. We can only obtain and use the Noisy Label on the registered topology information. First, we need to select a target ONU and select all ONUs under the same first-level optical splitter as candidate ONUs. These candidate ONUs can be divided into different clusters (representing different secondary optical splitters) according to their Noisy Label. When the error rate is less than 50%, it can generally be guaranteed that candidate ONUs with True Label account for the majority in the cluster with Noisy Label. . Then, the average score of the clusters with the same True Label as the target ONU and the target ONU will be higher. Calculate the average score of the target ONU and these clusters, and take the highest one to get the secondary spectrometer predicted by the model.

The above error rate refers to the error rate of registered topology information in the asset management system. Here we mentioned the two concepts of TrueLabel and NoisyLabel. True Label represents the grouping of physical authenticity. For example, the three ONUs [A, D, E] are really under the same group of secondary optical splitters in reality. The Noisy Label represents the record information in the asset management system. This information may be accurate or inaccurate. For example, it may be recorded in the asset management system that three ONUs [A, D, G] are the same group. Secondary beam splitter. This deviates from the real situation. The error rate we are talking about here refers to the deviation between the information recorded in the asset management system and the real physical world. We believe that if the error rate in the overall asset management system is less than 50%, the overall algorithm logic is controllable.

Regarding step 205, in another optional implementation, the sequence pair scoring is integrated with the time dimension and the ONU level to determine the recommended secondary optical splitter corresponding to the ONU, including:

(1) Integrate and fill all the sequence pairs corresponding to the target ONU and the scores of all sequence pairs corresponding to the target ONU into a three-dimensional matrix. The three-dimensional matrix includes the time dimension, different secondary spectrometer dimensions and preset data for the preset time period. The time period is different from other ONU dimensions under the default secondary optical splitter;

(2) Integrate the time dimension of the three-dimensional matrix and the different secondary spectrometer dimensions of the data existing in the preset time period to obtain a two-dimensional matrix;

(3) Input the two-dimensional matrix into the self-attention mechanism (Self-attention) model to obtain the secondary spectrometer score of the target ONU and each secondary spectroscope;

(4) Based on the secondary optical splitter score, determine the recommended secondary optical splitter corresponding to the target ONU.

In the embodiment of this application, the time dimension is unified with the ONU level and the learnable model: for a given ONU, we integrate all the information related to it (the similarity scores of all sequences under it and sequence pairs formed by other sequences) and Filled with a three-dimensional matrix M _a*b*d , as shown in Figure 3j, where a represents the time dimension, b represents the different secondary spectrometers with data in a given period, and d represents the given period and the given secondary spectrometer. Different other ONUs under the controller. Integrate the two dimensions a and b into one dimension and mark the time and secondary spectrometer information through Positional Encoding to obtain the two-dimensional matrix N _g*d , and use the Self-attention mechanism to encode the matrix:

Among them, Q, K, and V are the three matrices required in the self-attention mechanism (Q: Query matrix; K: Key matrix, V: Value matrix). in:

Query (Q) matrix: represents the information of the current element and is used to match the Key matrix of other elements to determine which elements to focus on.

Key (K) matrix: represents information about other elements and is used to match the Query matrix of the current element to determine which elements to focus on.

Value(V) matrix: represents the actual information of other elements and is used to generate output based on the calculated weights.

The basic calculation process of the self-attention mechanism is expressed in the above formula:

1.QK ^T calculates the dot product between Query and Key vectors. The dot product can measure the similarity between the Query and Key vectors. A larger dot product value indicates that the two are more similar, and a smaller dot product value indicates that the two are dissimilar.

2. Divide the dot product by a scaling factor, where the scaling factor is the square root of the Key vector dimension The purpose of this operation is to avoid gradient instability caused by excessive dot product values.

3. Perform softmax normalization on the calculated dot product. The softmax function can convert any real number into a probability distribution, ensuring that the sum of all weights is 1. In this way, we have a weight matrix that measures how much attention each element in the sequence deserves relative to other elements.

4. Use the normalized weight matrix to perform a weighted sum of the Value vectors. The purpose of this step is to linearly combine the Value vectors based on the calculated weights to generate a new representation. This new representation contains relationship information to other elements.

Through the process described above, the deep learning method is introduced into the field of PON topology information prediction and management, and combined with a variety of time series data processing methods, it is possible to automatically obtain the secondary optical splitter recommendation of any ONU and perform real-time analysis of the PON network topology structure. Correcting the error registration can significantly improve the accuracy and coverage rate.

Optionally, for the noise reduction processing of the original topology in step 206, the verification can be performed based on the artificially generated partially mislabeled topology and optical power data sets as shown in Table 1;

Table 1

The three rows of data in Table 1 indicate that after checking the prediction effect, we use the score of a certain type of secondary optical splitter that the target ONU in the test data belongs to as its confidence level. For example, ONU A belongs to the third group of secondary optical splitters, and its score is 0.7 points. Then 0.7 is used as its confidence score. We took the top 10%, the top 20% and up to 90% to calculate the accuracy and true negative rate indicators respectively.

It should be noted that in the specific solution process described above, the specific model algorithm formulas used and the parameter settings involved are examples, which can be flexibly adjusted according to actual needs, and do not constitute a limitation on the technical solution of this application.

For the passive optical network management method provided by the embodiments of the present application, the execution subject may be a virtual device. In the embodiments of the present application, the passive optical network management device executed by the passive optical network management method is used as an example to illustrate the passive optical network management device provided by the embodiments of the present application.

Referring to Figure 4, an embodiment of the present application provides a passive optical network management device 400, which includes:

The data preprocessing and feature extraction module 401 is used to integrate the optical power data of the ONU in the PON and the topology data of the ONU registered in the asset management system, and perform data preprocessing and feature extraction processing to obtain a time slice sequence;

The sampling module 402 is used to sample sequence pairs in the time slice sequence to obtain training data and test data;

The model training module 403 is used to input training data into the first model group for training to obtain the second model group;

The scoring module 404 is used to input the test data into the second model group to obtain the sequence pair score;

The determination module 405 is used to integrate the sequence pair scoring in the time dimension and the ONU level, and determine the recommended secondary optical splitter corresponding to the ONU;

The processing module 406 is used to perform noise reduction processing on the original topology according to the recommended secondary optical splitter corresponding to the ONU and the original topology of the PON in the asset management system;

The first model group includes multiple regression models, and the second model group includes multiple trained regression models.

Optionally, the sampling module is specifically used for:

Traverse all sequence pairs located at the same first-level optical splitter and the same time in the time slice sequence;

The sequence pairs in which the two sequences belong to the same secondary spectrometer are marked as positive samples, and the sequence pairs in which the two sequences belong to different secondary spectrometers are marked as negative samples;

Based on the positive samples and negative samples, training data and test data are obtained.

Optionally, the model training module is specifically used for:

Input the training data into the first CNN model for training to obtain the second CNN model;

Input the training data into the first RNN model for training to obtain the second RNN model;

Input the training data into the first GBDT model for training, and obtain the second GBDT model.

Optionally, the scoring module is specifically used for:

Input the test data into the second CNN model to get the first score;

Input the test data into the second RNN model to get the second score;

Input the test data into the second GBDT model to obtain the third score;

The first score, the second score, and the third score are averaged to obtain the sequence pair score.

Optionally, determine the module specifically used for:

For any ONU pair, select the highest sequence pair score as the sequence pair score of the ONU pair;

Determine multiple candidate ONU groups;

Calculate the average sequence pair score between the target ONU and the candidate ONU in each candidate ONU group;

Based on the candidate ONU group with the highest average score in the sequence, determine the recommended secondary spectrometer corresponding to the target ONU;

Each candidate ONU group includes multiple candidate ONUs under the same primary optical splitter, and each candidate ONU group corresponds to a different secondary optical splitter.

Optionally, determine the module specifically used for:

Integrate and fill all the sequence pairs corresponding to the target ONU and the scores of all sequence pairs corresponding to the target ONU into a three-dimensional matrix. The three-dimensional matrix includes the time dimension, different secondary spectrometer dimensions with data existing in the preset time period, and the preset time period and Default different other ONU dimensions under the secondary optical splitter;

Integrate the time dimension of the three-dimensional matrix and the different secondary spectrometer dimensions of the data existing in the preset time period to obtain a two-dimensional matrix;

Input the two-dimensional matrix into the Self-attention model to obtain the secondary spectrometer score of the target ONU and each secondary spectroscope;

Based on the secondary optical splitter score, determine the recommended secondary optical splitter corresponding to the target ONU.

The passive optical network management device in the embodiment of the present application may be an electronic device, such as an electronic device with an operating system, or may be a component in the electronic device, such as an integrated circuit or chip. The electronic device may be a terminal or other devices other than the terminal. For example, terminals may include but are not limited to the types of terminals 11 listed above, and other devices may be servers, network attached storage (Network Attached Storage, NAS), etc., which are not specifically limited in the embodiment of this application.

The passive optical network management device provided by the embodiments of the present application can implement each process implemented by the method embodiments of Figures 2 to 3j, and achieve the same technical effect. To avoid duplication, the details will not be described here.

Referring to Figure 5, an embodiment of the present invention provides an electronic device 500, including: at least one processor 501, a memory 502, a user interface 503, and at least one network interface 504. The various components in electronic device 500 are coupled together by bus system 505 .

It can be understood that the bus system 505 is used to implement connection communication between these components. In addition to the data bus, the bus system 505 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, various buses are labeled as bus system 505 in FIG. 5 .

The user interface 503 may include a display, a keyboard or a clicking device (for example, a mouse, a trackball, a touch pad or a touch screen, etc.).

It can be understood that the memory 502 in the embodiment of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory can be read-only memory (Read-Only Memory, ROM), programmable read-only memory (Programmable ROM, PROM), erasable programmable read-only memory (Erasable PROM, EPROM), electrically removable memory. Erase programmable read-only memory (Electrically EPROM, EEPROM) or flash memory. The volatile memory may be random access memory (RAM), which is used as an external cache. By way of illustration, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (Dynamic RAM, DRAM), synchronous dynamic random access memory (Synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (Double DataRate SDRAM, DDRSDRAM), enhanced synchronous dynamic random access memory (Enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (Synch Link DRAM, SLDRAM) and Direct Rambus RAM (DRRAM). The memory 502 described in embodiments of the invention is intended to include, but is not limited to, these and any other suitable types of memory.

In some embodiments, memory 502 stores the following elements, executable modules or data structures, or subsets thereof, or extensions thereof: operating system 5021 and application programs 5022.

Among them, the operating system 5021 includes various system programs, such as framework layer, core library layer, driver layer, etc., which are used to implement various basic services and process hardware-based tasks. Application program 5022 includes various application programs, such as media players, browsers, etc., and is used to implement various application services. The program that implements the method of the embodiment of the present invention may be included in the application program 5022.

In the embodiment of the present invention, the electronic device 500 may also include: a program stored on the memory 502 and executable on the processor 501. When the program is executed by the processor 501, the steps of the method provided by the embodiment of the present invention are implemented.

The methods disclosed in the above embodiments of the present invention can be applied to the processor 501 or implemented by the processor 501 . The processor 501 may be an integrated circuit chip with signal processing capabilities. During the implementation process, each step of the above method can be completed by instructions in the form of hardware integrated logic circuits or software in the processor 501 . The above-mentioned processor 501 can be a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), an off-the-shelf programmable gate array (Field Programmable Gate Array, FPGA), or other available processors. Programmed logic devices, discrete gate or transistor logic devices, discrete hardware components. Each method, step and logical block diagram disclosed in the embodiment of the present invention can be implemented or executed. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc. The steps of the method disclosed in conjunction with the embodiments of the present invention can be directly implemented by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software module can be located in random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, registers and other computer-readable storage media that are mature in this field. The computer-readable storage medium is located in the memory 502. The processor 501 reads the information in the memory 502 and completes the steps of the above method in combination with its hardware. Specifically, the computer program is stored on the computer-readable storage medium.

It can be understood that the embodiments described in the embodiments of the present invention can be implemented using hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more ASICs, DSPs, digital signal processing devices (DSP Devices, DSPDs), programmable logic devices (Programmable Logic Devices, PLDs), FPGAs, general-purpose processors, controllers, microcontrollers, etc. Controller, microprocessor, other electronic unit for performing the functions described in this application, or a combination thereof.

Embodiments of the present application also provide a readable storage medium, with programs or instructions stored on the readable storage medium. When the program or instructions are executed by a processor, each process of the above passive optical network management method embodiment is implemented, and can achieve the same technical effect, so to avoid repetition, we will not repeat them here.

Wherein, the processor is the processor in the terminal described in the above embodiment. The readable storage medium includes computer readable storage media, such as computer read-only memory ROM, random access memory RAM, magnetic disk or optical disk, etc. In some examples, the readable storage medium may be a non-transitory readable storage medium.

An embodiment of the present application further provides a chip. The chip includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the above passive optical network management method. Each process of the embodiment can achieve the same technical effect, so to avoid repetition, it will not be described again here.

It should be understood that the chips mentioned in the embodiments of this application may also be called system-on-chip, system-on-a-chip, system-on-chip or system-on-chip, etc.

Embodiments of the present application further provide a computer program/program product. The computer program/program product is stored in a storage medium. The computer program/program product is executed by at least one processor to implement the above passive optical network management. Each process of the method embodiment can achieve the same technical effect, so to avoid repetition, it will not be described again here.

It should be noted that, in this document, the terms "comprising", "comprising" or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article or device that includes a series of elements not only includes those elements, It also includes other elements not expressly listed or inherent in the process, method, article or apparatus. Without further limitation, an element defined by the statement "comprises a..." does not exclude the presence of additional identical elements in a process, method, article, or device that includes that element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, but may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved. Functions may be performed, for example, the methods described may be performed in an order different from that described, and various steps may also be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

Through the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of computer software products and necessary general hardware platforms, and of course can also be implemented by hardware. The computer software product is stored in a storage medium (such as ROM, RAM, magnetic disk, optical disk, etc.) and includes a number of instructions to cause the terminal or network side device to execute the methods described in various embodiments of the present application.

The embodiments of the present application have been described above in conjunction with the accompanying drawings. However, the present application is not limited to the above-mentioned specific implementations. The above-mentioned specific implementations are only illustrative and not restrictive. Those of ordinary skill in the art will Inspired by this application, many forms of implementations can be made without departing from the purpose of this application and the scope protected by the claims, and these implementations all fall within the protection of this application.

Claims (14)

1. A method for passive optical network management, comprising:

integrating optical power data of an Optical Network Unit (ONU) in a Passive Optical Network (PON) and topology data registered by the ONU in a resource management system, and performing data preprocessing and feature extraction processing to obtain a time slice sequence;

Sampling the sequence pairs in the time slice sequence to obtain training data and test data;

inputting the training data into a first model group for training to obtain a second model group;

inputting the test data into the second model group to obtain a sequence pair scoring;

integrating time dimension and ONU layer of the sequence pairs, and determining recommended secondary beam splitters corresponding to the ONU;

according to the recommended secondary optical splitter corresponding to the ONU and the original topological structure of the PON in the resource management system, performing noise reduction treatment on the original topological structure;

the first model group comprises a plurality of regression models, and the second model group comprises a plurality of trained regression models.

2. The method of claim 1, wherein sampling pairs of sequences in the time slice sequence to obtain training data and test data comprises:

traversing all sequence pairs which are positioned at the same level of the spectroscope and have the same time in the time slice sequence;

marking the same sequence pair of the two-level optical splitters of the two sequences as positive samples, and marking the different sequence pair of the two-level optical splitters of the two sequences as negative samples;

And obtaining training data and test data according to the positive sample and the negative sample.

3. The method of claim 1, wherein the inputting the training data into the first model set for training results in a second model set, comprising:

inputting the training data into a first convolutional neural network CNN model for training to obtain a second CNN model;

inputting the training data into a first cyclic neural network (RNN) model for training to obtain a second RNN model;

and inputting the training data into a GBDT model of the first gradient lifting decision tree for training to obtain a second GBDT model.

4. A method according to claim 3, wherein said inputting said test data into said second model set yields a sequence pair score, comprising:

inputting the test data into the second CNN model to obtain a first score;

inputting the test data into the second RNN model to obtain a second score;

inputting the test data into the second GBDT model to obtain a third score;

and averaging the first scoring, the second scoring and the third scoring to obtain the sequence pair scoring.

5. The method of claim 1, wherein the integrating the time dimension of the sorting of the sequence pairs with the ONU level determines a recommended secondary splitter corresponding to the ONU, comprising:

Selecting the highest sequence pair scoring for any ONU pair as the sequence pair scoring of the ONU pair;

determining a plurality of candidate ONU groups;

calculating a sequence pair scoring mean value between the target ONU and the candidate ONU in each candidate ONU group;

determining a recommended secondary beam splitter corresponding to the target ONU according to the candidate ONU group with the highest grading average value of the sequence pairs;

each candidate ONU group comprises a plurality of candidate ONUs under the same primary optical splitter, and each candidate ONU group corresponds to a different secondary optical splitter.

6. The method of claim 1, wherein the integrating the time dimension of the sorting of the sequence pairs with the ONU level determines a recommended secondary splitter corresponding to the ONU, comprising:

dividing, integrating and filling all sequence pairs corresponding to a target ONU and all sequence pairs corresponding to the target ONU into a three-dimensional matrix, wherein the three-dimensional matrix comprises a time dimension, different two-level beam splitters dimensions of data in a preset time period and other ONU dimensions of different preset time periods and preset two-level beam splitters;

integrating the time dimension of the three-dimensional matrix and the dimensions of different secondary beamsplitters of the data in the preset time period to obtain a two-dimensional matrix;

Inputting the two-dimensional matrix into a Self-attention mechanism Self-attention model to obtain a secondary beam splitter score of the target ONU and each secondary beam splitter;

and determining a recommended secondary optical splitter corresponding to the target ONU according to the secondary optical splitter scoring.

7. A passive optical network management device, comprising:

the data preprocessing and feature extraction module is used for integrating optical power data of the ONU in the PON and topology data registered by the ONU in a resource management system, and performing data preprocessing and feature extraction processing to obtain a time slice sequence;

the sampling module is used for sampling samples of the sequence pairs in the time slice sequence to obtain training data and test data;

the model training module is used for inputting the training data into the first model group for training to obtain a second model group;

the scoring module is used for inputting the test data into the second model group to obtain a sequence pair score;

the determining module is used for integrating the time dimension and the ONU layer for dividing the sequence pairs and determining a recommended secondary beam splitter corresponding to the ONU;

the processing module is used for carrying out noise reduction processing on the original topological structure according to the recommended secondary optical splitter corresponding to the ONU and the original topological structure of the PON in the resource management system;

The first model group comprises a plurality of regression models, and the second model group comprises a plurality of trained regression models.

8. The apparatus of claim 7, wherein the sampling module is specifically configured to:

traversing all sequence pairs which are positioned at the same level of the spectroscope and have the same time in the time slice sequence;

marking the same sequence pair of the two-level optical splitters of the two sequences as positive samples, and marking the different sequence pair of the two-level optical splitters of the two sequences as negative samples;

and obtaining training data and test data according to the positive sample and the negative sample.

9. The apparatus of claim 7, wherein the model training module is specifically configured to:

inputting the training data into a first CNN model for training to obtain a second CNN model;

inputting the training data into a first RNN model for training to obtain a second RNN model;

and inputting the training data into a first GBDT model for training to obtain a second GBDT model.

10. The apparatus according to claim 9, wherein the scoring module is specifically configured to:

inputting the test data into the second CNN model to obtain a first score;

Inputting the test data into the second RNN model to obtain a second score;

inputting the test data into the second GBDT model to obtain a third score;

and averaging the first scoring, the second scoring and the third scoring to obtain the sequence pair scoring.

11. The apparatus of claim 7, wherein the determining module is specifically configured to:

selecting the highest sequence pair scoring for any ONU pair as the sequence pair scoring of the ONU pair;

determining a plurality of candidate ONU groups;

calculating a sequence pair scoring mean value between the target ONU and the candidate ONU in each candidate ONU group;

determining a recommended secondary beam splitter corresponding to the target ONU according to the candidate ONU group with the highest grading average value of the sequence pairs;

each candidate ONU group comprises a plurality of candidate ONUs under the same primary optical splitter, and each candidate ONU group corresponds to a different secondary optical splitter.

12. The apparatus of claim 7, wherein the determining module is specifically configured to:

dividing, integrating and filling all sequence pairs corresponding to a target ONU and all sequence pairs corresponding to the target ONU into a three-dimensional matrix, wherein the three-dimensional matrix comprises a time dimension, different two-level beam splitters dimensions of data in a preset time period and other ONU dimensions of different preset time periods and preset two-level beam splitters;

Integrating the time dimension of the three-dimensional matrix and the dimensions of different secondary beamsplitters of the data in the preset time period to obtain a two-dimensional matrix;

inputting the two-dimensional matrix into a Self-saturation model to obtain a secondary beam splitter score of the target ONU and each secondary beam splitter;

and determining a recommended secondary optical splitter corresponding to the target ONU according to the secondary optical splitter scoring.

13. An electronic device comprising a processor and a memory storing a program or instructions executable on the processor, which when executed by the processor, implement the steps of the passive optical network management method of any one of claims 1 to 6.

14. A readable storage medium, characterized in that it stores thereon a program or instructions, which when executed by a processor, implement the steps of the passive optical network management method according to any of claims 1 to 6.