CN114548739A - Transformer substation flood prevention risk combination evaluation method - Google Patents

Abstract

The invention discloses a substation flood prevention risk combination assessment method, which relates to the field of power grid safety operation and data processing. A substation flood control risk combination assessment method, comprising: analyzing multi-dimensional substation flood control measurement data, dividing a flood control data set into a static data part and a dynamic monitoring data part; building a LightGBM-based static data evaluation sub-model; building an LSTM-based Dynamic monitoring data evaluation sub-model; using the entropy weight distribution method to effectively combine the static data evaluation sub-model and the dynamic monitoring data evaluation sub-model to construct a substation flood control risk combination evaluation algorithm model; use the substation flood control risk combination evaluation The algorithm model evaluates the flood control risk of the substation. The substation flood prevention risk combination evaluation method provided in the embodiment of the present invention constructs a substation flood prevention risk combination evaluation algorithm model, which can replace the subjective experience evaluation method in a data-driven manner, and is more efficient and accurate.

Description

A substation flood control risk combination assessment method

technical field

The present application relates to the field of safe operation of power grids and data processing, and in particular, to a combined assessment method for flood prevention risks in substations.

Background technique

At present, the power grid flood control risk assessment work mostly adopts the artificial experience method. At home and abroad, the flood control objects are often qualitatively assessed and the risk levels are divided according to red, orange, yellow and blue. The assessment results are subjective and fail to combine the advantages of big data. The efficiency of the method is not high, and the applicability to different flood control objects is not strong. At present, the quantitative risk assessment algorithms are roughly divided into: statistical analysis methods, machine learning algorithms, and deep learning algorithms.

Statistical analysis methods are mostly based on conceptual models of physical mechanisms, mathematical evaluation models established after qualitative analysis and physical process simulation of substation flood control risks. In the power grid flood control assessment work, comprehensive meteorological monitoring information, hydrological information, power grid operation information, combined with the experience of professional and technical personnel to evaluate the flood control capability and possible impact. Since qualitative methods are used for evaluation, the actual problems are often simplified mathematically in the modeling process, and it is difficult to take all influencing factors into account. Therefore, in practical applications, the evaluation will be inaccurate.

Machine learning algorithms build models driven by data, and can perform feature extraction and information mining on substation flood control data. Although machine learning methods can perform feature extraction and information mining, they cannot take into account the temporal relationship between data and ignore some features with temporal relationship.

The deep learning algorithm has a strong ability to mine implicit fluctuation laws, and has a stronger ability to mine meteorological monitoring information with a time series relationship in substation flood control data, and has stronger data adaptability. However, substation flood control data includes static data without time series relationship, such as hydrological information, power grid operation information, and meteorological monitoring information with time series relationship. Only a single evaluation algorithm cannot mine all the characteristic information.

The difficulty of the current substation flood control risk assessment task is that there are many influencing factors, and the data is usually mixed with time-series data and non-series data. Therefore, there is an urgent need for a combined assessment method for substation flood control risks that can combine time-series data and non-series data.

SUMMARY OF THE INVENTION

The purpose of the embodiments of the present invention is to provide a substation flood control risk combination evaluation method, build a substation flood control risk combination evaluation algorithm model, and replace the subjective experience evaluation method with a data-driven method, which is more efficient and accurate.

For solving the above-mentioned technical problems, the technical scheme adopted in the present invention is:

A substation flood control risk combination assessment method, comprising: analyzing multi-dimensional substation flood control measurement data, dividing a flood control data set into a static data part and a dynamic monitoring data part; building a LightGBM-based static data evaluation sub-model; building an LSTM-based Dynamic monitoring data evaluation sub-model; using the entropy weight distribution method to effectively combine the static data evaluation sub-model and the dynamic monitoring data evaluation sub-model to construct a substation flood control risk combination evaluation algorithm model; use the substation flood control risk combination evaluation The algorithm model evaluates the flood control risk of the substation.

使用D中样本依次训练K棵回归树，且根据前树的评估效果建立树；待K棵回归树全部建成，将其评估值之和作为评估结果

进行输出。In some embodiments, the constructing a LightGBM-based static data evaluation sub-model includes: taking the historical value x _i ={x _i1 ,x _i2 ,..., _xiq } of the static data in the substation flood control data set as an input feature Matrix, the corresponding risk capability assessment probability value is used as the output

Use the samples in D to train K regression trees in turn, and build a tree according to the evaluation effect of the previous tree; when all K regression trees are built, the sum of their evaluation values is used as the evaluation result

to output.

In some embodiments, the construction of the LSTM-based dynamic monitoring data evaluation sub-model includes: taking the current dynamic data in the substation flood control data set as the x _(t) input model, and designing a time sliding window to process the data to calculate the dynamic cumulative value; using LSTM The forgetting gate, the input gate and the new memory unit in the combination extract the useful information output at the previous moment and the input at the current moment to obtain the long-term memory unit, and the evaluation result y2 is calculated by the long-term memory unit.

In some embodiments, the effective combination of the static data evaluation sub-model and the dynamic monitoring data evaluation sub-model by the entropy weight distribution method includes: using the entropy weight distribution method to measure the static data evaluation sub-model and the dynamic monitoring data evaluation sub-model. The dynamic monitoring data evaluates the degree of deviation of the sub-models, assigns a smaller weight to the sub-models with large evaluation deviation, and constructs a substation flood control risk combination evaluation algorithm model.

In some embodiments, the static data portion is derived from an equipment management system, including substation-specific information.

In some embodiments, the dynamic monitoring data is partly derived from micro-meteorological stations established by each substation, including meteorological data of rainfall, temperature, and wind speed.

The invention provides a substation flood control risk combination evaluation method, which uses the substation flood control risk combination evaluation algorithm model to evaluate the substation flood control risk. The algorithm model is driven by data to construct the LightGBM and LSTM combination evaluation algorithm model, which can maximize the use of There is information to evaluate. Firstly, the substation flood control data set is analyzed, and the static and dynamic data parts are divided according to the characteristics of the data. Secondly, LightGBM and LSTM are used to establish evaluation sub-models for the divided static and dynamic data parts, that is, for the non-sequential data in the data set. The LightGBM evaluation model, the time series data part adopts the LSTM evaluation model, and then realizes the effective combination between the two sub-models through the entropy weight distribution method, and constructs the substation flood control risk combination evaluation algorithm model.

The invention provides a substation flood control risk combination assessment method, which can analyze and divide the characteristics of the substation flood control data set, and build sub-models suitable for the characteristics of the divided data, which is more suitable for practical applications, and The factors considered are more comprehensive. At the same time, using the entropy weight distribution method to realize the effective combination between the sub-models can improve the accuracy of the combination evaluation algorithm model. Using the combination evaluation algorithm model of the present invention can better mine data features, and maximize the use of existing information for evaluation. . The substation flood control risk combination evaluation algorithm model of the present invention replaces the subjective experience evaluation method in a data-driven manner, and is more efficient and accurate.

Description of drawings

In order to illustrate the technical solutions in the present disclosure more clearly, the following briefly introduces the accompanying drawings that need to be used in some embodiments of the present disclosure. Obviously, the accompanying drawings in the following description are only the appendixes of some embodiments of the present disclosure. For those of ordinary skill in the art, other drawings can also be obtained from these drawings. In addition, the accompanying drawings in the following description may be regarded as schematic diagrams, and are not intended to limit the actual size of the product involved in the embodiments of the present disclosure, the actual flow of the method, and the like.

FIG. 1 is a substation flood control risk combination assessment algorithm framework according to some embodiments of the present disclosure;

2 is a sub-model for flood control risk assessment of LightGBM substations according to some embodiments of the present disclosure;

3 is a sub-model for flood control risk assessment of LSTM substations according to some embodiments of the present disclosure;

4 is a performance comparison diagram of common machine learning deep learning algorithms in substation flood control assessment tasks according to some embodiments of the present disclosure;

FIG. 5 is a comparison diagram of combined model evaluation in a substation flood control evaluation task according to some embodiments of the present disclosure.

Detailed ways

The technical solutions in some embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, but not all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments provided by the present disclosure fall within the protection scope of the present disclosure.

Throughout the specification and claims, the term "comprising" is to be interpreted in an open, inclusive sense, ie, "including, but not limited to," unless the context requires otherwise. In the description of the specification, the terms "one embodiment", "some embodiments", "exemplary embodiment", "example" or "some examples" etc. are intended to indicate specific features, structures related to the embodiment or example , materials or properties are included in at least one embodiment or example of the present disclosure. The schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials or characteristics described may be included in any suitable manner in any one or more embodiments or examples.

An embodiment of the present invention provides a substation flood control risk combination assessment method, as shown in FIG. 1 , including: S100-S500.

S100, analyze the multi-dimensional substation flood control metering data, and divide the flood control data set into a static data part and a dynamic monitoring data part.

S200, construct a static data evaluation sub-model based on LightGBM.

The static data part of the above substation flood control data has the characteristics of many dimensions and small capacity. Therefore, a static data evaluation sub-model is constructed based on LightGBM.

S300, construct an LSTM-based dynamic monitoring data evaluation sub-model.

The dynamic monitoring data part of the above substation flood control data is generally recorded at fixed time intervals, and there is a time series relationship between adjacent data. Therefore, a dynamic monitoring data evaluation sub-model is constructed based on LSTM.

S400, using the entropy weight distribution method to effectively combine the static data evaluation sub-model and the dynamic monitoring data evaluation sub-model, and construct a substation flood prevention risk combination evaluation algorithm model.

S500, using the substation flood prevention risk combination evaluation algorithm model to evaluate the substation flood prevention risk.

The invention provides a substation flood control risk combination evaluation method, which uses the substation flood control risk combination evaluation algorithm model to evaluate the substation flood control risk. The algorithm model is driven by data to construct the LightGBM and LSTM combination evaluation algorithm model, which can maximize the use of There is information to evaluate. Firstly, the substation flood control data set is analyzed, and the static and dynamic data parts are divided according to the characteristics of the data. Secondly, LightGBM and LSTM are used to establish evaluation sub-models for the divided static and dynamic data parts, that is, for the non-sequential data in the data set. The LightGBM evaluation model, the time series data part adopts the LSTM evaluation model, and then realizes the effective combination between the two sub-models through the entropy weight distribution method, and constructs the substation flood control risk combination evaluation algorithm model.

In some embodiments, the static data portion is derived from a facility management system, including substation-specific information.

The above-mentioned inherent information of the substation has remained unchanged for a long time, and the static engineering method was used to evaluate the flood prevention risk capability of the substation in the past. The details are shown in Table 1:

Table 1 Flood control static data of substations

In some embodiments, the dynamic monitoring data is partly derived from micro-meteorological stations established by each substation, including meteorological data of rainfall, temperature, and wind speed.

The above dynamic monitoring data part is generally sampled at an interval of 1 hour, which belongs to time series data and can affect the real-time flood control capability of the substation.

In some embodiments, as shown in FIG. 2 , the construction of a LightGBM-based static data evaluation sub-model includes: S210-S230.

由此变电站防汛静态数据可以表示为：S210, the historical value x _i ={x _i1 ,x _i2 ,...,x _iq } of the static data in the flood control data set of the substation is used as the input feature matrix, and the corresponding risk capability evaluation probability value is used as the output quantity

From this, the flood control static data of the substation can be expressed as:

D={(x _i ,y),i=1,2,...,n} (1)

S220, use the samples in D to train K regression trees in sequence, and build a tree according to the evaluation effect of the previous tree.

Among them, the use of histogram-based feature discretization reduces memory consumption and speeds up operation.

进行输出。即S230, when all K regression trees are built, the sum of their evaluation values is used as the evaluation result (that is, the above-mentioned risk ability evaluation probability value)

to output. which is

In some embodiments, as shown in FIG. 3 , the construction of an LSTM-based dynamic monitoring data evaluation sub-model includes: S310-S320.

S310, the current dynamic data in the flood control data set of the substation is used as the x _(t) input model, and the design time sliding window processes the data to calculate the dynamic cumulative value.

S320, using the combination of forget gate, input gate and new memory unit in LSTM to extract useful information output at the previous moment and input at the current moment, obtain a long-term memory unit, and calculate the evaluation result y ₂ by the long-term memory unit.

o_(t)分别表示遗忘门、输入门、新记忆单元和输出门。c _(t-1) and c _(t) represent the long-term memory units of the previous moment and the current moment, respectively; f _(t) , i _(t) ,

o _(t) denote the forget gate, input gate, new memory unit, and output gate, respectively.

Among them, the forgetting gate, the input gate and the new memory unit can extract the useful information in the output h _(t-1) of the previous moment and the input x _(t) of the current moment, and obtain the long-term memory unit c _(t) :

Then, the flood control risk assessment value is calculated by the output gate and the long-term memory unit, and the calculation method is as follows:

h _(t) = o _(t) tanh(c _(t) ) (4)

The final h _(t) at time t is the evaluation result generated by the LSTM-based dynamic monitoring data evaluation sub-model

Using the memory structure of LSTM, it can consider the effect of the pre-order data on the evaluation results of the post-order data, meet the time series characteristics of the dynamic monitoring data part of the flood control data, improve the reliability of the flood control risk assessment, and be more in line with the actual situation.

In some embodiments, the effective combination of the static data evaluation sub-model and the dynamic monitoring data evaluation sub-model by the entropy weight distribution method includes:

The entropy weight distribution method is used to measure the deviation degree of the static data evaluation sub-model and the dynamic monitoring data evaluation sub-model, and the sub-model with large evaluation deviation is assigned a smaller weight to construct the substation flood control risk combination evaluation algorithm model.

Aiming at the practical significance of the weight distribution of the substation flood control risk portfolio assessment model, the specified constraints are:

The entropy weight distribution method is used to solve the weights of the above two sub-models and realize the model combination at the same time. The main steps are as follows:

In order to verify the effectiveness of the substation flood control risk combination assessment method provided by the present invention, the following is a comparison experiment of a single algorithm and a comparison experiment of a combined model to verify the performance of the model provided by the present invention, and to verify that the present invention is relative to mainstream machines. Outstanding benefits of learning, deep learning, and combinatorial models.

Aiming at the characteristic dimension difference of substation flood control data, through Min-MaxNormalization processing, the data is mapped to the [0,1] interval, that is,

The processed data is used as a dataset, and is divided into a training set and a test set according to the ratio of 7:3. In order to adapt to the substation flood control data set, the goodness of fit R ² is selected to measure the fitting degree of the test set prediction and the observed value. The closer R ² is to 1, the better the fitting degree is, and the more realistic the model evaluation results are.

In order to verify the accuracy of the model, the mean absolute error (MAE) is selected as an index to measure the accuracy of the model. The average error can well reflect the actual situation of the evaluation value error. The smaller the MAE value, the more accurate the model.

In addition, the mean square error (MSE) is used to evaluate the degree of change of the data. The smaller the MSE value, the better the accuracy of the model evaluation.

In order to verify the validity of the model proposed in this example, based on the same data, compare the effects of the method proposed in this example with LightGBM, LSTM algorithms, and common machine learning and deep learning algorithms. The comparison results are shown in Table 2 and intuitively shown in Figure 4.

Table 2 Comparative experimental results of flood control assessment algorithms for a single substation

Observing Table 2 and Figure 4, it can be seen that after dividing the dynamic and static data, the model provided by the present invention is used to conduct experiments. Compared with the single LightGBM algorithm, the MAE index is reduced by 31.57%, and compared with the LSTM algorithm, it is reduced by 48.2%. It is proved that the model provided by the present invention performs better in the evaluation of the flood control capability of the substation.

Compared with the machine learning algorithm, the performance of the LightGBM algorithm is slightly higher than that of the XGBoost algorithm, and the running speed is faster. Compared with the deep learning algorithm, the LSTM algorithm has a better effect, because it can not only extract the flood control data information of substations like a convolutional neural network (CNN), but also has a stronger ability to mine time series information than a recurrent neural network (RNN). And the machine learning algorithm is generally better than the deep learning method. This is because the dimension of the static data part of the flood control data in the substation is high, and the dynamic monitoring data part has a greater impact on the final evaluation result is the rainfall data. When the rainfall data does not change much, flood control The data will be static, and the unchanged data is equivalent to reducing the amount of data. At this time, the effect of machine learning on small data will be better.

The model provided by the present invention can divide dynamic and static data according to the characteristics of the data to construct evaluation sub-models adapted to the characteristics of the data respectively, and then distribute the weight model according to the entropy weight distribution method, and the error index is smaller than a single mainstream machine learning and deep learning evaluation algorithm .

For algorithms with better performance comparison results of a single algorithm, a comparison experiment is conducted according to the data set division strategy and the combination strategy of machine learning and deep learning described in this example. The performance comparison results of the combined model are shown in Table 3. Some results in the test set are displayed, and the comparison curve between the evaluation results of each confounding model and the actual data is drawn as shown in Figure 5.

Table 3 Comparative experimental results of flood control assessment algorithms for combined substations

It can be seen from Table 3 that after dividing the dynamic and static data sets, the evaluation effect of the hybrid model is better than that of the single model, and the error index of the model in this example is smaller. It can be seen from Figure 5 that the black solid line is the actual data value, and the red dotted line is the evaluation of this example model. Compared with the other two hybrid models, the evaluation result of this example model is closer to the actual data.

To sum up, in the substation flood control risk assessment scenario with high precision requirements, the model provided by the present invention can obtain the highest accuracy on the premise of satisfying the speed, and is most suitable for substation flood control risk assessment.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Other modifications or equivalent replacements made by those of ordinary skill in the art to the technical solutions of the present invention, as long as they do not depart from the spirit and the technical solutions of the present invention. The scope should be included in the scope of the claims of the present invention.

Claims (6)

1. A flood prevention risk combination assessment method for a transformer substation is characterized by comprising the following steps:

analyzing flood prevention metering data of the multi-dimensional transformer substation, and dividing a flood prevention data set into a static data part and a dynamic monitoring data part;

constructing a static data evaluation sub-model based on the LightGBM;

constructing a dynamic monitoring data evaluation sub-model based on the LSTM;

effectively combining the static data evaluation submodel and the dynamic monitoring data evaluation submodel by adopting an entropy weight distribution method to construct a transformer substation flood prevention risk combination evaluation algorithm model;

and evaluating the flood prevention risk of the transformer substation by using the transformer substation flood prevention risk combined evaluation algorithm model.

2. The transformer substation flood prevention risk combination assessment method according to claim 1,

the method for constructing the static data evaluation submodel based on the LightGBM comprises the following steps:

concentrating the historical value x of static data in the flood prevention data of the transformer substation_i＝{x_i1,x_i2,...,x_iqUsing the risk assessment probability value as an output quantity

Sequentially training K regression trees by using the samples in the step D, and establishing the trees according to the evaluation effect of the former trees; when all K regression trees are built, taking the sum of the evaluation values as an evaluation result

And outputting the data.

3. The transformer substation flood prevention risk combination assessment method according to claim 2,

the method for constructing the LSTM-based dynamic monitoring data evaluation sub-model comprises the following steps:

taking the current dynamic data in the flood prevention data set of the transformer substation as x_(t)Inputting a model, designing a time sliding window to process data and calculating a dynamic accumulated value;

extracting useful information output at the last moment and input at the current moment by using a forgetting gate, an input gate and a new memory unit combination in the LSTM to obtain a long-term memory unit, and calculating to obtain an evaluation result y by using the long-term memory unit₂。

4. The transformer substation flood prevention risk combination assessment method according to claim 3,

the effective combination of the static data evaluation submodel and the dynamic monitoring data evaluation submodel by adopting an entropy weight distribution method comprises the following steps:

and measuring the deviation degrees of the static data evaluation submodel and the dynamic monitoring data evaluation submodel by adopting an entropy weight distribution method, distributing smaller weight to the submodel with large evaluation deviation, and constructing a transformer substation flood prevention risk combination evaluation algorithm model.

5. The transformer substation flood prevention risk combination assessment method according to claim 1,

the static data part is derived from the equipment management system and comprises inherent substation information.

6. The transformer substation flood prevention risk combination assessment method according to claim 1,

the dynamic monitoring data part is from a microclimate station established by each transformer substation and comprises meteorological data of rainfall, temperature and wind speed.

Images