CN109002915B - Photovoltaic power station short-term power prediction method based on Kmeans-GRA-Elman model - Google Patents
Photovoltaic power station short-term power prediction method based on Kmeans-GRA-Elman model Download PDFInfo
- Publication number
- CN109002915B CN109002915B CN201810769372.9A CN201810769372A CN109002915B CN 109002915 B CN109002915 B CN 109002915B CN 201810769372 A CN201810769372 A CN 201810769372A CN 109002915 B CN109002915 B CN 109002915B
- Authority
- CN
- China
- Prior art keywords
- day
- sample
- predicted
- parameter
- power
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 25
- 238000013528 artificial neural network Methods 0.000 claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 18
- 230000002354 daily effect Effects 0.000 claims abstract description 11
- 230000003203 everyday effect Effects 0.000 claims abstract description 7
- 238000007781 pre-processing Methods 0.000 claims abstract description 5
- 210000002569 neuron Anatomy 0.000 claims description 14
- 230000005855 radiation Effects 0.000 claims description 14
- 238000010606 normalization Methods 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000007619 statistical method Methods 0.000 claims description 6
- 238000009792 diffusion process Methods 0.000 claims description 5
- 230000002159 abnormal effect Effects 0.000 claims description 4
- 238000010219 correlation analysis Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 150000001875 compounds Chemical class 0.000 claims description 2
- 230000006835 compression Effects 0.000 claims description 2
- 238000007906 compression Methods 0.000 claims description 2
- 230000007613 environmental effect Effects 0.000 claims description 2
- 238000010248 power generation Methods 0.000 description 17
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 10
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 10
- 238000012706 support-vector machine Methods 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Probability & Statistics with Applications (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Game Theory and Decision Science (AREA)
- Primary Health Care (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a short-term power prediction method of a photovoltaic power station based on a Kmeans-GRA-Elman model, which comprises the following steps: acquiring historical daily generated power of a photovoltaic power station and meteorological parameters of a weather station in a corresponding time period every day; preprocessing the data; clustering samples from the first day to the day before the day to be predicted in the historical days by using six statistical indexes and combining with an improved Kmeans algorithm, and determining the number of categories according to the profile coefficient; calculating the central point of each clustering meteorological characteristic value, and judging the category of the day to be predicted; determining a similar day and an optimal similar day of a day to be predicted; determining Elman neural network parameters; obtaining a training model; and inputting the parameter sample combination of the optimal similar day and the meteorological parameters of the day to be predicted into the training model to predict the generated power of the day to be predicted. The method and the device can improve the precision and accuracy of short-term power prediction of the photovoltaic power station under different weather conditions in different seasons.
Description
Technical Field
The invention belongs to a short-term power prediction technology of a photovoltaic power station, and particularly relates to a short-term power prediction method of the photovoltaic power station based on a Kmeans-GRA-Elman model.
Background
In recent years, with the development of socioeconomic, the problems of energy shortage and environmental pollution are highly regarded by various social circles, and the development and utilization of renewable energy have become important ways to solve the problems of energy and environment. In addition, with the increase of the power demand, the scale of the power grid is continuously enlarged, and the defects of high investment cost, high operation difficulty and the like of the traditional large-scale and high-concentration power generation are increasingly highlighted. Under the background, photovoltaic power generation is rapidly developed under the attention of countries in the world. For the technical level of photovoltaic power generation, the state puts higher demands, wherein the most difficult is the prediction of the photovoltaic power generation. Because the output of the photovoltaic power generation system is influenced by the solar irradiation intensity and weather factors, the output of the photovoltaic power generation system has randomness and volatility. Therefore, photovoltaic power generation is an uncontrolled source for large power grids, and randomness and volatility of photovoltaic power generation can impact the power grids. In order to meet the requirements of users and guarantee the safety of the power grid, the power grid can make corresponding scheduling strategies and plans, the photovoltaic power generation prediction can effectively help the power grid to make the plans, the scheduling of the power grid is facilitated, the relation between the photovoltaic power supply and the conventional power supply can be coordinated in real time, and the safe and stable operation of the power grid is promoted.
At present, the power prediction method of the photovoltaic power station can be mainly divided into an indirect prediction method and a direct prediction method. The indirect prediction method firstly needs to predict the solar radiation intensity, and then indirectly predicts the generating power of the photovoltaic power generation system according to the predicted value of the solar radiation intensity. The method needs accurate weather forecast information, and needs to be modeled for many times, and the prediction process is complex and tedious and is difficult to be practically applied. The direct prediction method is used for finding out the relation between the output power of the photovoltaic power generation system and the historical power and meteorological factors by carrying out statistical analysis on the historical output power of the photovoltaic power generation system and the relevant meteorological factors, and establishing a photovoltaic power station power prediction model, wherein the direct prediction method is mainly used for the current photovoltaic power prediction.
Common direct prediction methods include algorithms such as Artificial Neural Network (ANN), markov chain, time series method, Support Vector Machine (SVM), and the like. The most widely used artificial intelligence algorithm is based on ANN and SVM. The prediction algorithm mainly based on the SVM can better solve the situation of small samples and has higher precision, but the parameter optimization of the prediction algorithm needs long training time. Although the neural network is easy to fall into local minimum, the neural network is superior to a support vector machine in the aspect of prediction performance due to higher fitting and generalization capability and shorter training time, and is also successfully applied at present.
Compared with the traditional BP neural network, the Elman neural network has an additional receiving layer for receiving a feedback signal from the hidden layer and memorizing an output value of a neuron of the hidden layer at the previous moment, so that the network has sensitivity to historical data, and the capability of the network for processing dynamic information is improved. Hence a better Elman neural network than BP neural network is chosen herein as the prediction model. In order to enable the model to more accurately predict the power generation power of the photovoltaic power station under different weather conditions in different seasons, corresponding models need to be established according to different meteorological characteristics, and the historical date closest to the date to be predicted serves as the input of the model, so that the prediction precision can be greatly improved. Therefore, the similar day sample and the optimal similar day sample of the day to be predicted are searched by combining the Kmeans + + algorithm with the GRA algorithm and are respectively used as the training sample and the test sample of the model to be input. Therefore, the method can realize the rapid and accurate prediction of the generated power of the photovoltaic power station by adopting the hybrid improved Kmeans-GRA-Elman model.
At present, no research on the application of the Kmeans-GRA-Elman algorithm based on hybrid improvement to photovoltaic power station short-term power prediction is found in publicly published documents and patents.
Disclosure of Invention
In view of the above, the invention aims to provide a photovoltaic power plant short-term power prediction method based on a hybrid improved Kmeans-GRA-Elman model. And (3) clustering samples from the first day to the day before the day to be predicted in the historical days by using a Kmeans + + algorithm according to six statistical indexes (average power, standard deviation, variation coefficient, skewness coefficient, kurtosis coefficient and total power) in the normalized statistical analysis, and determining the most appropriate clustering category number according to the contour coefficient. And then calculating the Euclidean distance between the weather parameter characteristic value of the day to be predicted and the center point of the weather parameter characteristic value of each cluster sample set, and determining the category of the day to be predicted. And then selecting a similar day and an optimal similar day from the cluster sample set to which the day to be predicted belongs by utilizing a GRA algorithm according to the meteorological parameter characteristic value of the day to be predicted. And taking the generated power of each time of the similar day sample in the same day, meteorological parameters such as illumination, ambient temperature, humidity and wind speed in the same day and meteorological parameters in the next day as input, and taking the generated power of each time in the next day as output to train the prediction model based on the Elman neural network. And predicting the photovoltaic power generation power by taking the model and the generated power, the meteorological parameters and the meteorological parameters of the day to be predicted of the best similar day as the input of the model, and predicting the generated power of the day to be predicted at each moment.
The invention is realized by adopting the following scheme: a short-term power prediction method of a photovoltaic power station based on a Kmeans-GRA-Elman model comprises the following steps:
step S1: collecting historical daily generated power of a photovoltaic power station and meteorological parameters of a weather station in a corresponding time period every day, and combining the collected power and meteorological parameters to obtain a daily meteorological-power parameter sample combination;
step S2: preprocessing the weather-power parameter sample combination every day, removing abnormal data and carrying out normalization processing;
step S3: clustering samples from the first day to the day before the day to be predicted in the historical days by using six statistical indexes in the normalized statistical analysis in combination with an improved Kmeans algorithm, and determining the category number according to the contour coefficient;
step S4: according to the meteorological characteristic value of each clustering sample set; determining a cluster center position; judging the category of the day to be predicted by using the Euclidean distance;
step S5: determining a similar day and an optimal similar day of the day to be predicted in the same cluster sample set according to the meteorological characteristic value of the day to be predicted and combining a gray correlation analysis (GRA) algorithm;
step S6: determining parameters of the Elman neural network;
step S7: training an Elman neural network by using parameter sample combinations on similar days, and continuously modifying the number of neurons in a hidden layer to obtain a training model;
step S8: and inputting the parameter sample combination of the optimal similar day and the meteorological parameters of the day to be predicted into a training model to predict the generated power of the day to be predicted, and obtaining the output power value of each moment of the day to be predicted.
The method has the advantage that the generated power of the photovoltaic power station at each moment every hour in the next day can be predicted accurately in advance. Clustering the historical samples by using a Kmeans + + algorithm, determining a similar day and an optimal similar day in the clustering category to which the day to be predicted belongs by combining a GRA algorithm, and predicting by using an Elman neural network model, thereby further improving the accuracy and precision of short-term power generation power prediction of the photovoltaic power station under different weather conditions in different seasons.
Drawings
FIG. 1 is a block flow diagram of the present invention.
Fig. 2 shows power prediction results 1 of the experimental group and the control group according to the embodiment of the present invention.
Fig. 3 is a prediction error curve 1 at each time point of the experimental group and the control group according to the embodiment of the present invention.
Fig. 4 shows power prediction results 2 of the experimental group and the control group according to the embodiment of the present invention.
Fig. 5 is a prediction error curve 2 at each time point of the experimental group and the control group according to the embodiment of the present invention.
Fig. 6 shows power prediction results 3 of the experimental group and the control group according to the embodiment of the present invention.
Fig. 7 is a prediction error curve 3 at each time point of the experimental group and the control group according to the embodiment of the present invention.
Fig. 8 shows the power prediction results 4 of the experimental group and the control group according to the embodiment of the present invention.
Fig. 9 is a prediction error curve 4 at each time point of the experimental group and the control group according to the embodiment of the present invention.
Detailed Description
The invention is further explained below with reference to the drawings and the embodiments.
The embodiment provides a photovoltaic power station short-term power prediction method based on a hybrid improved Kmeans-GRA-Elman model, and a flow chart is shown in FIG. 1. The method specifically comprises the following steps:
step S1: collecting historical daily generated power of a photovoltaic power station and meteorological parameters of a weather station in a corresponding time period every day, wherein the meteorological parameters comprise meteorological factors such as illumination, ambient temperature, humidity and wind speed, and combining the meteorological factors to obtain a daily meteorological-power parameter sample combination;
step S2: preprocessing the weather-power parameter sample combination every day, removing abnormal data and carrying out normalization processing;
step S3: clustering samples from the first day to the day before the day to be predicted in the historical days by using six statistical indexes in the normalized statistical analysis in combination with an improved Kmeans algorithm, and determining the category number according to the contour coefficient;
step S4: calculating the central point of each clustering meteorological characteristic value according to the meteorological characteristic values of each clustering sample set, and judging the category of the day to be predicted by using Euclidean distance;
step S5: determining a similar day and an optimal similar day of the day to be predicted in the same clustering sample set according to the meteorological characteristic value of the day to be predicted and combining a gray correlation analysis (GRA) algorithm;
step S6: determining the number m of input layer nodes, the number b of hidden layer nodes and the number p of output layer nodes of the Elman neural network, and initializing each weight and threshold of the Elman neural network;
step S7: the method comprises the steps of training an Elman neural network by using parameter sample combinations on similar days, and continuously modifying the number of neurons in a hidden layer by a trial and error method to obtain a training model;
step S8: and inputting the parameter sample combination of the optimal similar day and the meteorological parameters of the day to be predicted into the training model to predict the generated power of the day to be predicted, so as to obtain the output power value of each moment of the day to be predicted.
Preferably, the photovoltaic power station used for collecting data in this embodiment is an alice springs photovoltaic power station in australia, and the photovoltaic power station is composed of 22 photovoltaic panels with a rated value of 250W, and a photovoltaic array with a rated value of 5.5KW is used for grid-connected power generation through an inverter.
In this embodiment, the weather-power parameter sample combination in step S1 includes the historical daily generated power of the photovoltaic power station and the weather parameters of the weather station in the corresponding time period each day. The parameter sample combination is noted as (P)ki，Gki，Dki，Tki，Wki，Hki) (ii) a Wherein K is the serial number of the date of sample collection, represents the number of days, and is an integer from 1 to K, and K is an integer greater than 1; i is the time of sample collection in one day, represents the time number, and is an integer from 1 to I, wherein I is an integer greater than 1; pkiThe power parameter sample of the ith moment in the k-th day parameter sample combination is taken as the power parameter sample; gkiThe global horizontal radiation parameter sample of the ith moment in the kth day parameter sample combination is used as the global horizontal radiation parameter sample; dkiThe diffusion level radiation parameter sample of the ith time in the k day parameter sample combination is used as the sample; t iskiThe environmental temperature parameter sample at the ith moment in the kth day parameter sample combination is taken as the sample; wkiThe wind speed parameter sample at the ith moment in the kth day parameter sample combination is taken as the sample; hkiIs the relative humidity parameter sample at the ith moment in the parameter sample combination of the kth day.
In this embodiment, the preprocessing of the sample in step S2 mainly includes removing abnormal data and normalizing. Removing outlier data refers to removing days in which the historical data is negative or significantly erroneous. The specific method of normalization is as follows: when the same parameter sample is identical by adopting a proportional compression methodMapping to interval [0, 1]]In the power sample P ═ P (P)1i,P2i，…Pki，…PKi) For example, the specific mapping formula is:
wherein y' represents the data obtained after normalization, PimaxRepresenting the maximum value in the ith time of the data set P, PiminRepresenting the minimum value at the i-th instant of the data set P.
In this embodiment, the samples from the first day to the day before the day to be predicted in the historical day are clustered by using six statistical indexes in the normalized statistical analysis in combination with the improved Kmeans algorithm in step S3, and the number of categories is determined according to the profile coefficient. The six statistical indexes are (
σk，cvk，skk，kurk，Psumk) Wherein K is the number of the date of sample collection, represents the number of days, and is an integer from 1 to K.
Is the average power parameter sample of day k, σkSample of standard deviation parameters for day k, cvkIs a coefficient of variation parameter sample, sk, of day kkIs the skewness coefficient parameter sample at day k, kurkIs a sample of the kurtosis coefficient parameter at day k, PsumkTotal power parameter samples for day k. And (4) normalizing the six statistical indexes, then combining a Kmeans + + algorithm for clustering, and determining the category number according to the contour coefficient s. Selecting s>A clustering case of 0.45 is taken as a suitable clustering result. The specific formulas of calculation, normalization and calculation of the profile coefficient of each parameter sample are shown as follows. Take the prediction of 9/14 (spring in australia) in 2017, 26 (summer in australia) in 2018 in 2/26 (summer in australia), 30 (autumn in australia) in 2018 in 3/30 and 29 (winter in australia) in 2017 in 7/29 as examplesThe best calculated profile coefficients and the corresponding class numbers are shown in table 1.
In the formula (I), the compound is shown in the specification,
σ, cv, sk, kur, Psum respectively represent the average power, standard deviation, coefficient of variation, skewness coefficient, kurtosis coefficient, and total power per day. I denotes the respective time of day at which the sample is collected and I denotes the total time of day.
Wherein x' represents the data obtained after normalization, xminAnd xmaxMinimum and maximum values, y, representing an array of samplesminTake-1, ymax1 is taken.
Where s (i) represents a contour coefficient, i represents a sample in each cluster sample set, a (i) represents an average value of dissimilarities of the sample i to other points in the same cluster, and b (i) represents a minimum value of the average dissimilarities of the sample i to other clusters.
TABLE 1 Profile factor and class number of daily clusters to be predicted
| 2017-9-14 (spring) | 2018-2-26 (summer) | 2018-3-30 (autumn) | 2017-7-29 (winter) | |
| Coefficient of contour | 0.4868 | 0.4925 | 0.4960 | 0.4768 |
| Number of categories | 3 | 3 | 3 | 3 |
In this embodiment, in step S4, it is necessary to calculate a central point of each clustering weather feature value according to the weather feature value of each clustering sample set, and determine the category to which the day to be predicted belongs by using the euclidean distance. The meteorological characteristic value is recorded as (G)kmax，Gkmin，Dkmax,Dkmin,Tkmax,Tkmin,Wkmax,Wkmin,Hkmax,Hkmin) Wherein k is the number of the date of sample collection, represents the number of days, and is an integer from 1 to N. GkmaxAnd GkminFor maximum and minimum global level radiation parameter samples, DkmaxAnd DkminFor maximum and minimum diffusion level radiation parameter samples, TkmaxAnd TkminFor maximum and minimum ambient temperature parameter samples, WkmaxAnd WkminFor maximum and minimum wind speed parameter samples, HkmaxAnd HkminAre maximum and minimum relative humidity parameter samples. And calculating the average value of the characteristic values of each cluster, namely determining the cluster center position. And calculating the distance between the day to be predicted and each cluster central point by using a Euclidean distance formula, and attributing the day to be predicted to the cluster with the minimum distance. The Euclidean distance formula is as follows:
in the formula (d)0cRepresenting the Euclidean distance, x, of the day to be predicted from each cluster0A weather characteristic value representing a day to be predicted,
representing the c-th cluster center point.
In this embodiment, in step S5, it is necessary to determine the similar day and the best similar day of the day to be predicted in the same cluster sample set according to the weather characteristic value of the day to be predicted (the same as in step S4) in combination with a gray correlation analysis (GRA) algorithm. Calculating the association degree of the day to be predicted and each sample in the same cluster sample set, determining the date with the association degree larger than a certain threshold value as a similar day, and determining the day with the maximum association degree in the last 10 days of the cluster sample set as the best similar day. Taking the prediction of 14 days in 2017 (spring of australia), 26 days in 2018 and 2 months (summer of australia), 30 days in 2018 and 3 months (autumn of australia) and 29 days in 2017 and 7 months (winter of australia) as examples, the meteorological parameters of the 4 days obtained according to the meteorological forecast are shown in table 2. According to table 2, the association degrees of the 4 days and the respective samples of the same category are respectively calculated, the date with the association degree greater than a certain threshold is determined as the similar day, the day with the maximum association degree in the last 10 days of the cluster sample set is determined as the best similar day, and the threshold value and the best similar day of the 4 days and the corresponding best association degree are shown in table 3. The correlation degree is calculated as shown in the following equation. As can be seen from table 3, in this example, the best similar day of the day to be predicted, which is the 9 th and 14 th day in 2017 (spring in australia), is the 9 th and 7 th day in 2017, and the best correlation value is 0.8788. The best similarity day of 26 in 2018 (summer in australia) to be predicted is 25 in 2018 in 2 and 25, and the best relevance value is 0.8311. The best similar day of the predicted day of 3 and 30 months in 2018 (fall in australia) was day 23 and 3 months in 2018, and the best relevance value was 0.9354. The best similarity day of 29 th 7 th 2017 (winter in australia) was 26 th 7 th 2017, with a best relevance value of 0.9113.
In the formula, riRepresenting the degree of association, wherein n represents the number of characteristic values; xiiA correlation coefficient representing the sample and the day to be predicted; the formula is as follows:
wherein y (n) represents the weather characteristic value normalized by the day to be predicted, and xiAnd (n) represents the meteorological characteristic value after the historical day normalization, rho represents a resolution coefficient, rho can be 0.5, and n represents the number of the characteristic values.
TABLE 2 solar weather parameters to be predicted
TABLE 3 threshold and best similarity day for determining similarity day and corresponding best association for the day to be predicted
| Date | 2017-9-14 | 2018-2-26 | 2018-3-30 | 2017-7-29 |
| Threshold value | 0.8 | 0.75 | 0.9 | 0.8 |
| Best similar day | 2017-9-7 | 2018-2-25 | 2018-3-23 | 2017-7-26 |
| Best degree of association | 0.8788 | 0.8311 | 0.9354 | 0.9113 |
In this embodiment, the Elman neural network described in step S6 is specifically set as: the number of nodes of an input layer is 23 neurons, the number of nodes of a hidden layer is 10 neurons, the number of nodes of an output layer is 11 neurons, the number of iterations is 2000, and other parameters are set by default. Wherein, the formula of the number of the hidden layer neurons is as follows:
in the formula, b represents the number of nodes of the hidden layer, m represents the number of nodes of the input layer, and p represents the number of nodes of the output layer. a is 1-10.
In this embodiment, in step S7, according to the selected date to be predicted, the parameter sample combination of the current day and the meteorological parameter of the next day in the similar day sample set of the date to be predicted are used as inputs, and the output power value of each hour of the next day is used as an output to train the Elman neural network model. The input-output combination is expressed as (P)k，
Tkmax,Tkmin,
T(k+1)max,T(k+1)min,
Pk+1) Wherein K is the number of the date of sample collection, represents the number of days, and is an integer from 1 to K. The first 13 variables represent input variables in the input-output combination and the last variable represents output variables. In which P in the input variablekFor the power parameter samples at various times of day k,
and
the samples of the global level radiation parameters were taken on day k and day k +1,
and
the day-averaged diffusion horizontal radiance parameter samples, T, for day k and day k +1kmaxAnd T(k+1)maxMaximum ambient temperature parameter samples, T, for day k and day k +1kminAnd T(k+1)minThe minimum ambient temperature parameter samples for day k and day k +1,
and
are the day average wind speed parameter samples of the k day and the k +1 day,
and
outputting a variable P for the day average relative humidity parameter samples of the k day and the k +1 dayk+1Samples of the output power parameter per hour at various times on day k + 1. The number of the neurons of the hidden layer can be continuously modified through a trial and error method to obtain a training model.
In this embodiment, in step S8, it is necessary to predict the generated power every hour on the day to be predicted by using a trained model, with the parameter sample combination on the best similar day and the meteorological parameters on the day to be predicted as inputs. The input sample setting is the same as the step S7 setting.
Correspondingly, all data samples are used as training samples, the GRA algorithm is adopted to determine the optimal similar day, and the experiment input by taking the optimal similar day sample as a test sample is used as a control group; and (3) clustering by adopting a Kmeans + + algorithm, determining a similar day and an optimal similar day by combining a GRA algorithm, taking the similar day sample as a training sample, and taking an experiment input by taking the optimal similar day sample as a test sample as an experiment group. The experiment is carried out by arbitrarily selecting 1 day of each of spring, summer, autumn and winter as the day to be predicted, the operation is carried out for 10 times, the prediction results of the experimental group and the control group and the prediction error curves at each moment are shown in figures 2-9, and the model error index values are shown in table 4. Wherein RMSE is the root mean square error, MAPE is the mean absolute percent error, R2To determine the coefficients, t is the program run time. RMSE, MAPE, R2The calculation formula of (c) is as follows. RMSE at 9, 14 (Australian spring) 3.5098kW, MAPE 2.4556%, R on 2017 to be predicted20.9964, t is 58.5124 s. Day to be predictedRMSE 7.5822kW, MAPE 4.1643%, R on 26.2.2018 (summer in Australia)20.9883, t is 56.4642 s. RMSE of 4.3614kW, MAPE of 2.7878%, R of 3, 30 and 2018 (autumn of Australia) to be predicted20.9953, t is 27.5976 s. RMSE 2.4634kW, MAPE 2.0861%, R on 7/29/2017 (winter in Australia) to be predicted20.9988, t is 52.7167 s. MAPE error in four seasons is within 4.5%, R2Reaches over 0.988.
Preferably, the MAPE of the experimental group has an average value of 2.8735% and an improvement of about 3 percentage points over 5.5105% of the control group, the RMSE of the experimental group has an average value of 4.4792kW and an improvement of 3.2456kW over 7.7248kW of the control group, and the R of the experimental group2The mean value was 0.9947, which is 0.0086 higher than 0.9861 in the control group, the mean value t in the experimental group was 48.8227s, and the value was 23.9355s lower than 72.7582s in the control group. From the mean square error, the MAPE mean square error of the experimental group was 0.9070%, which was about 1 percentage point less than 1.9158% of the control group, the RMSE mean square error of the experimental group was 2.2095kW, which was 0.9073kW less than 3.1168kW of the control group, and the R of the experimental group2The mean square error was 0.0045, which is 0.0054 less than 0.0099 for the control group. The MAPE mean, RMSE mean and 3 mean variance values were all lower in the experimental group than in the control group, while R was lower2The mean value is greater than the control group. Therefore, the photovoltaic power station short-term power prediction method based on the hybrid improved Kmeans-GRA-Elman model is higher in accuracy, higher in speed and better in prediction effect.
In the formula, Pf，iRepresenting the predicted value of the output power of the photovoltaic power station, Pm,iThe measured value of the output power of the photovoltaic power station is shown,
and the actual daily total output power value of the photovoltaic power station is represented, and N represents the sampling number of the power generation time period of the photovoltaic power station.
TABLE 4 model error index
The above description is only a preferred embodiment of the present invention, and all equivalent changes and modifications made in accordance with the claims of the present invention should be covered by the present invention.
Claims (9)
1. A photovoltaic power station short-term power prediction method based on a Kmeans-GRA-Elman model is characterized by comprising the following steps: the method comprises the following steps:
step S1: collecting historical daily generated power of a photovoltaic power station and meteorological parameters of a weather station in a corresponding time period every day, and combining the collected power and meteorological parameters to obtain a daily meteorological-power parameter sample combination;
step S2: preprocessing the weather-power parameter sample combination every day, removing abnormal data and carrying out normalization processing;
step S3: clustering samples from the first day to the day before the day to be predicted in the historical days by using six statistical indexes in the normalized statistical analysis in combination with an improved Kmeans algorithm, and determining the category number according to the contour coefficient;
the six statistical indicators in the step S3 are labeled
Wherein K is the serial number of the date of sample collection, represents the number of days, and is an integer from 1 to K, and K is an integer greater than 1;
is the average power parameter sample of day k, σkSample of standard deviation parameters for day k, cvkIs a coefficient of variation parameter sample, sk, of day kkIs the skewness coefficient parameter sample at day k, kurkIs a sample of the kurtosis coefficient parameter at day k, PsumkTotal power parameter samples for day k;
step S4: according to the meteorological characteristic value of each clustering sample set; determining a cluster center position; judging the category of the day to be predicted by using the Euclidean distance;
step S5: determining a similar day and an optimal similar day of the day to be predicted in the same cluster sample set according to the meteorological characteristic value of the day to be predicted and combining a gray correlation analysis (GRA) algorithm;
step S6: determining parameters of the Elman neural network;
step S7: training an Elman neural network by using parameter sample combinations on similar days, and continuously modifying the number of neurons in a hidden layer to obtain a training model;
step S8: and inputting the parameter sample combination of the optimal similar day and the meteorological parameters of the day to be predicted into a training model to predict the generated power of the day to be predicted, and obtaining the output power value of each moment of the day to be predicted.
2. The method for photovoltaic power plant short-term power prediction based on a Kmeans-GRA-Elman model as claimed in claim 1, wherein: the meteorological-power parameter sample combination in the step S1 is denoted as (P)ki，Gki，Dki，Tki，Wki，Hki) (ii) a Wherein K is the serial number of the date of sample collection, represents the number of days, and is an integer from 1 to K, and K is an integer greater than 1; i is the time of sample collection in one day, represents the time number, and is an integer from 1 to I, wherein I is an integer greater than 1; pki is a power parameter sample at the ith moment in the parameter sample combination of the kth day; gkiThe global horizontal radiation parameter sample of the ith moment in the kth day parameter sample combination is used as the global horizontal radiation parameter sample; dkiFor the kth day in the combination of the day k parameter samplesDiffusion level radiation parameter samples at i moments; t iskiThe environmental temperature parameter sample at the ith moment in the kth day parameter sample combination is taken as the sample; wkiThe wind speed parameter sample at the ith moment in the kth day parameter sample combination is taken as the sample; hkiIs the relative humidity parameter sample at the ith moment in the parameter sample combination of the kth day.
3. The method for photovoltaic power plant short-term power prediction based on a Kmeans-GRA-Elman model as claimed in claim 1, wherein: the specific method of normalization in step S2 is: and adopting a proportional compression method to map the same parameter sample into the interval [0, 1] at the same time.
4. The method for photovoltaic power plant short-term power prediction based on a Kmeans-GRA-Elman model as claimed in claim 1, wherein: the step S3
After the six statistical indexes are normalized, clustering is carried out by combining a Kmeans + + algorithm, and the category number is determined according to the contour coefficient s;
selecting a clustering condition with s larger than 0.45 as a proper clustering result, wherein the specific formulas of calculation, normalization and contour coefficient calculation of each parameter sample are as follows:
in the formula (I), the compound is shown in the specification,
sigma, cv, sk, kur and Psum respectively represent the average power, standard deviation, variation coefficient, skewness coefficient, kurtosis coefficient and total power of each day, I represents each time of sample collection in one day, and I represents the total time of one day;
wherein x' represents the data obtained after normalization, xminAnd xmaxMinimum and maximum values, y, representing an array of samplesminTake-1, ymaxTaking 1;
where s (i) represents the sample contour coefficient, a (i) represents the average of the dissimilarities of the sample to other points in the same cluster, and b (i) represents the minimum of the average dissimilarities of the sample to other clusters.
5. The method for photovoltaic power plant short-term power prediction based on a Kmeans-GRA-Elman model as claimed in claim 1, wherein: step S4 includes the following specific steps:
step S41: the meteorological characteristic value is recorded as
(Gkmax，Gkmin，Dkmax，Dkmin，Tkmax，Tkmin，Wkmax，Wkmin，Hkmax，Hkmin) Wherein K is the serial number of the date of sample collection, represents the number of days, and is an integer from 1 to K, and K is an integer greater than 1; gkmaxAnd GkminFor maximum and minimum global level radiation parameter samples, DkmaxAnd DkminFor maximum and minimum diffusion level radiation parameter samples, TkmaxAnd TkminFor maximum and minimum ambient temperature parameter samples, WkmaxAnd WkminFor maximum and minimum wind speed parameter samples, HkmaxAnd HkminMaximum and minimum relative humidity parameter samples;
step S42: calculating the average value of each characteristic value of each cluster, namely determining the cluster center position;
step S43: calculating the distance between the day to be predicted and each cluster central point by using an Euclidean distance formula, and attributing the day to be predicted to the cluster with the minimum distance, wherein the Euclidean distance formula is as follows:
in the formula (d)0cRepresenting the Euclidean distance, x, of the day to be predicted from each cluster0A weather characteristic value representing a day to be predicted,
representing the c-th cluster center point.
6. The method for photovoltaic power plant short-term power prediction based on a Kmeans-GRA-Elman model as claimed in claim 1, wherein: step S5 includes the following specific steps:
step S1: calculating the association degree of the day to be predicted and each sample in the same cluster sample set, wherein the calculation method of the association degree is shown as the following formula:
in the formula, riRepresenting the degree of association, wherein n represents the number of characteristic values; xiiA correlation coefficient representing the sample and the day to be predicted; the formula is as follows:
wherein y (n) represents the weather characteristic value normalized by the day to be predicted, and xiAnd (n) represents the meteorological characteristic value after historical day normalization, rho represents a resolution coefficient, and n represents the number of characteristic values.
7. The method for photovoltaic power plant short-term power prediction based on a Kmeans-GRA-Elman model as claimed in claim 1, wherein: the Elman neural network described in step S6 is specifically set as: the number of nodes of an input layer is 23 neurons, the number of nodes of a hidden layer is 10 neurons, the number of nodes of an output layer is 11 neurons, the number of iterations is 2000, and other parameters are set by default; wherein, the formula of the number of the hidden layer neurons is as follows:
in the formula, b represents the number of nodes of the hidden layer, m represents the number of nodes of the input layer, and p represents the number of nodes of the output layer; a is a constant and is 1-10.
8. The method for photovoltaic power plant short-term power prediction based on a Kmeans-GRA-Elman model as claimed in claim 1, wherein: step S7 includes the following specific steps:
step S71: the input-output combination is described as
Wherein K is the serial number of the date of sample collection, represents the number of days, and is an integer from 1 to K, and K is an integer greater than 1; the first 13 variables represent input variables in the input-output combination, and the last variable represents an output variable; in which P in the input variablekFor the power parameter samples at various times of day k,
and
the samples of the global level radiation parameters were taken on day k and day k +1,
and
the day-averaged diffusion horizontal radiance parameter samples, T, for day k and day k +1kmaxAnd T(k+1)maxMaximum ambient temperature parameter samples, T, for day k and day k +1kminAnd T(k+1)minMinimum ambient temperature parameter samples for day k and day k +1, WkAnd Wk+1Is a daily average wind speed parameter sample, H, for day k and day k +1kAnd
outputting a variable P for the day average relative humidity parameter samples of the k day and the k +1 dayk+1Samples of the hourly output power parameters at various times on day k + 1;
step S72: and continuously modifying the number of the neurons of the hidden layer by a trial and error method to obtain a training model.
9. The method for photovoltaic power plant short-term power prediction based on a Kmeans-GRA-Elman model as claimed in claim 8, wherein: in step S8, the parameter sample combination of the best similar day and the meteorological parameters of the day to be predicted are used as input, and the trained model is used to predict the generated power every hour of the day to be predicted; the input sample setting is the same as the step S7 setting.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810769372.9A CN109002915B (en) | 2018-07-13 | 2018-07-13 | Photovoltaic power station short-term power prediction method based on Kmeans-GRA-Elman model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810769372.9A CN109002915B (en) | 2018-07-13 | 2018-07-13 | Photovoltaic power station short-term power prediction method based on Kmeans-GRA-Elman model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109002915A CN109002915A (en) | 2018-12-14 |
| CN109002915B true CN109002915B (en) | 2022-03-18 |
Family
ID=64599682
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810769372.9A Active CN109002915B (en) | 2018-07-13 | 2018-07-13 | Photovoltaic power station short-term power prediction method based on Kmeans-GRA-Elman model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109002915B (en) |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543911B (en) * | 2018-11-29 | 2020-09-18 | 中国农业科学院农业信息研究所 | Sunlight radiation prediction method and system |
| CN109711609B (en) * | 2018-12-15 | 2022-08-12 | 福州大学 | Photovoltaic power station output power prediction method based on wavelet transformation and extreme learning machine |
| CN110009037B (en) * | 2019-04-03 | 2020-10-27 | 中南大学 | Short-term engineering wind speed prediction method and system based on physical information coupling |
| CN110163437B (en) * | 2019-05-23 | 2022-05-13 | 太原理工大学 | Day-ahead photovoltaic power generation power prediction method based on DPK-means |
| CN110148068A (en) * | 2019-05-23 | 2019-08-20 | 福州大学 | One kind being based on meteorological data similarity analysis and LSTM neural fusion photovoltaic plant ultra-short term power forecasting method |
| CN110348648A (en) * | 2019-08-02 | 2019-10-18 | 国网电子商务有限公司 | A kind of predicting power of photovoltaic plant method and device |
| CN111091139B (en) * | 2019-11-18 | 2024-04-19 | 特变电工西安电气科技有限公司 | Photovoltaic prediction method, device and equipment for similar day clustering and readable storage medium |
| CN112508246B (en) * | 2020-11-26 | 2022-09-16 | 国电南瑞科技股份有限公司 | Photovoltaic power generation power prediction method based on similar days |
| CN112668611B (en) * | 2020-12-08 | 2024-02-02 | 湖南工业大学 | Kmeans and CEEMD-PE-LSTM-based short-term photovoltaic power generation power prediction method |
| CN112927097A (en) * | 2021-01-29 | 2021-06-08 | 国网辽宁省电力有限公司阜新供电公司 | Photovoltaic power generation short-term prediction method based on GRA-ABC-Elman model |
| CN113313298A (en) * | 2021-05-21 | 2021-08-27 | 上海电力大学 | Feature selection-based hybrid Kmeans-GRA-SVR photovoltaic power generation power prediction method |
| CN115660132B (en) * | 2022-08-05 | 2024-01-30 | 科大数字(上海)能源科技有限公司 | Photovoltaic power generation power prediction method and system |
| CN116432874A (en) * | 2023-06-14 | 2023-07-14 | 青岛鼎信通讯科技有限公司 | Distributed photovoltaic power prediction method based on characteristic power |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106447098A (en) * | 2016-09-22 | 2017-02-22 | 许昌许继软件技术有限公司 | Photovoltaic ultra-short period power predicting method and device |
| CN106779129A (en) * | 2015-11-19 | 2017-05-31 | 华北电力大学（保定） | A kind of Short-Term Load Forecasting Method for considering meteorologic factor |
-
2018
- 2018-07-13 CN CN201810769372.9A patent/CN109002915B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106779129A (en) * | 2015-11-19 | 2017-05-31 | 华北电力大学（保定） | A kind of Short-Term Load Forecasting Method for considering meteorologic factor |
| CN106447098A (en) * | 2016-09-22 | 2017-02-22 | 许昌许继软件技术有限公司 | Photovoltaic ultra-short period power predicting method and device |
Non-Patent Citations (4)
| Title |
|---|
| Day-ahead forecasting of photovoltaic output power with similar cloud space fusion based on incomplete historical data mining;Xiuyuan Yang 等;《Applied Energy》;20170908;第683–696页 * |
| Hourly Solar Irradiance Prediction Based on Support Vector Machine and Its Error Analysis;Kuk Yeol Bae;《IEEE TRANSAC TIONS ON POWER SYSTEMS》;20170331;第32卷(第2期);第935-945页 * |
| Short-term forecasting of wind power generation Based on the Similar Day and Elman Neural Network;Xiaoyu Zhang 等;《2015 IEEE Symposium Series on Computational Intelligence》;20160111;第647-650页 * |
| 基于气象因子权重相似日的短期光伏功率预测;李伟 等;《广东电力》;20180430;第31卷(第4期);第59-64页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109002915A (en) | 2018-12-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109002915B (en) | Photovoltaic power station short-term power prediction method based on Kmeans-GRA-Elman model | |
| CN109086928B (en) | Photovoltaic power station real-time power prediction method based on SAGA-FCM-LSSVM model | |
| CN107766990B (en) | Method for predicting power generation power of photovoltaic power station | |
| CN112257941B (en) | Photovoltaic power station short-term power prediction method based on improved Bi-LSTM | |
| CN110909919A (en) | Photovoltaic power prediction method of depth neural network model with attention mechanism fused | |
| CN108205717A (en) | A kind of photovoltaic generation power Multiple Time Scales Forecasting Methodology | |
| CN110705789A (en) | Photovoltaic power station short-term power prediction method | |
| WO2024051524A1 (en) | Joint prediction method and apparatus for hydraulic, wind and photovoltaic generation power | |
| Yang et al. | Day-ahead forecasting of photovoltaic output power with similar cloud space fusion based on incomplete historical data mining | |
| Melzi et al. | Hourly solar irradiance forecasting based on machine learning models | |
| CN114792156B (en) | Photovoltaic output power prediction method and system based on curve characteristic index clustering | |
| CN105373857A (en) | Photovoltaic power station irradiance prediction method | |
| CN112990558B (en) | Meteorological temperature and illumination prediction method based on deep migration learning | |
| CN111027775A (en) | Step hydropower station generating capacity prediction method based on long-term and short-term memory network | |
| CN112215428B (en) | Photovoltaic power generation power prediction method and system based on error correction and fuzzy logic | |
| CN115796004A (en) | Photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models | |
| CN111369045A (en) | Method for predicting short-term photovoltaic power generation power | |
| CN111626473A (en) | Two-stage photovoltaic power prediction method considering error correction | |
| CN112927097A (en) | Photovoltaic power generation short-term prediction method based on GRA-ABC-Elman model | |
| CN116187540B (en) | Wind power station ultra-short-term power prediction method based on space-time deviation correction | |
| CN109902340A (en) | It is a kind of meter and complicated meteorology coupled characteristic multi-source-lotus combine scene generating method | |
| CN115759389A (en) | Day-ahead photovoltaic power prediction method based on weather type similar day combination strategy | |
| CN116470486A (en) | Ultra-short-term photovoltaic power generation power prediction method, device and storage medium | |
| CN116611702A (en) | Integrated learning photovoltaic power generation prediction method for building integrated energy management | |
| CN113112085A (en) | New energy station power generation load prediction method based on BP neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20230330 Address after: 327, Building 1, No.1 Xinzhen Road, Xinwei Town, Xiang'an District, Xiamen City, Fujian Province, 361000 Patentee after: Hengchao Construction Engineering Group Co.,Ltd. Address before: 350108 No. 2 Xueyuan Road, New District of Fuzhou University, Minhou County, Fuzhou City, Fujian Province Patentee before: FUZHOU University |
|
| TR01 | Transfer of patent right |