Photovoltaic power distribution network power control method and system

Abstract

The invention discloses a photovoltaic power distribution network power control method and system, and the method comprises the steps: carrying out the precise classification of historical data according to weather types through two times of clustering processing, building a matched photovoltaic power station power generation power prediction sub-model for different weather types, and improving the prediction accuracy of the power generation power of a photovoltaic power station; and the actual output of the active power and the reactive power of the photovoltaic power distribution network inverter is controlled based on the photovoltaic power station power generation power prediction value, so that the stability and the efficient output of the photovoltaic power distribution network are ensured.

Description

Technical Field

[0001] The present application belongs to the technical field of photovoltaic power station power generation, and in particular, relates to a photovoltaic distribution network power control method and system. Background Art

[0002] In recent years, the consumption of fossil energy has been increasing, and environmental pollution has become increasingly serious, which has become a common concern of the international community. Therefore, all countries are actively developing new energy sources. Solar energy is an inexhaustible clean energy, so photovoltaic power station power generation has broad application prospects. And with the implementation of "dual carbon", more and more photovoltaic power station power generation facilities will be put into the power grid. However, the random volatility of photovoltaic power station power generation is large, and the photovoltaic output power cannot be accurately estimated. This has an impact on the power system that cannot be underestimated and will bring a huge impact to the power grid. Therefore, it is of great significance to carry out photovoltaic power station output power prediction and control the actual output power of the photovoltaic distribution network based on the above prediction value to increase the photovoltaic grid-connected capacity, reduce the impact of photovoltaic power station power generation uncertainty on the power system, and reduce carbon emissions. Summary of the invention

[0003] In view of the defects in the prior art, the present invention provides a photovoltaic distribution network power control method and system, which improves the accuracy and reliability of photovoltaic power station power prediction, thereby realizing photovoltaic distribution network power control.

[0004] To achieve the above object, the present invention provides a photovoltaic power distribution network power control method, comprising the following steps:

[0005] S1, collecting the power generation of the photovoltaic power station in the historical time period and the corresponding historical meteorological data as historical sample data;

[0006] S2, clustering the historical sample data twice to obtain five weather type data sets: D1, D2, D3, D4 and D5, which respectively represent sunny, cloudy, haze, light rain and heavy rain weather type data sets; the five weather type data sets are used for subsequent training and prediction of the corresponding neural network power prediction sub-model;

[0007] S3, obtaining weather forecast data about the photovoltaic power station prediction point from the meteorological center, determining the weather forecast type of the prediction point, and according to the weather forecast type of the prediction point, selecting the neural network power prediction sub-model corresponding to the weather forecast type to output the predicted value of the photovoltaic power station power generation at the prediction point; wherein sunny weather, cloudy weather, haze weather, light rain weather and heavy rain weather correspond to the first BP neural network power prediction sub-model, the second BP neural network power prediction sub-model, the third BP neural network power prediction sub-model, the first LSTM neural network power prediction sub-model and the second LSTM neural network power prediction sub-model respectively;

[0008] S4, controlling the output active power and reactive power of the photovoltaic grid-connected inverter in the photovoltaic power station through the predicted value of the photovoltaic power station power generation at the prediction point.

[0009] Another object of the present invention is to provide a photovoltaic power distribution network power control system, wherein the photovoltaic power distribution network power control system is used to execute the above photovoltaic power distribution network power control method.

[0010] Another object of the present invention is to provide a computer-readable storage medium storing one or more programs, wherein the one or more programs enable a computer to execute the above-mentioned photovoltaic power distribution network power control method.

[0011] In combination with all the above technical solutions, the present invention has the following advantages compared with the prior art:

[0012] (1) By designing two clustering processes, the fuzzy C-means clustering algorithm is combined with the K-means clustering algorithm to achieve accurate division of weather type data, providing accurate basic data for subsequent photovoltaic power station power prediction.

[0013] (2) According to different weather types, the optimal power prediction sub-model is designed specifically to further improve the accuracy of power prediction of photovoltaic power stations, facilitate the subsequent control and adjustment of the actual power of the inverter and the stable and efficient output of the photovoltaic distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic diagram of executing the photovoltaic distribution network power control method of the present invention.

[0015] Figure 2 It is a schematic diagram of realizing the photovoltaic power distribution network power control system of the present invention. DETAILED DESCRIPTION

[0016] The following examples are combined with the attached Figure 1 Further description is given. Obviously, the described embodiments are partial embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the protection scope of the present invention.

[0017] The present invention provides a photovoltaic power distribution network power control method, referring to Figure 1 The photovoltaic power distribution network power control method is shown in the schematic diagram, and the photovoltaic power distribution network power control method includes the following steps:

[0018] S1. Collect the power generation of the photovoltaic power station in the historical time period and the corresponding historical meteorological data as historical sample data;

[0019] S2. The historical sample data is clustered twice to obtain five weather type data sets: D1, D2, D3, D4 and D5, which respectively represent sunny, cloudy, haze, light rain and heavy rain weather type data sets; the five weather type data sets are used for subsequent training and prediction of the corresponding neural network power prediction sub-model;

[0020] Preferably, the two clustering processes are specifically implemented as follows:

[0021] (2-1) The first clustering process uses the fuzzy C-means clustering algorithm to cluster the historical sample data into two categories {D1, D2, D3} and {D4, D5}, where {D1, D2, D3} is non-rainy weather and {D4, D5} is rainy weather;

[0022] (2-2) The second clustering process uses the K-means clustering algorithm to cluster the two types of results generated by the first clustering again, and obtains five weather type data sets D1, D2, D3, D4, and D5;

[0023] The classification of weather types is calculated by clustering historical data. The traditional method is to calculate various weather influencing factors in one cluster. However, the historical data has many feature dimensions. Calculating all features at the same time will cause data redundancy, and unnecessary data will reduce the classification accuracy during the calculation process. Therefore, the weather influencing factors will be analyzed here and divided into two clusters based on the characteristics of the data. Processing data of different weather types separately can not only use specific weather influencing factors to reduce the amount of data calculation, but also divide the weather more accurately.

[0024] In the first clustering process, sunny days, cloudy days and haze were classified as non-rainy days, and light rain and heavy rain were classified as rainy days. After data analysis, it was found that the scattered radiation and humidity on non-rainy days were significantly lower than those on rainy days, while the direct radiation was significantly higher than that on rainy days. Therefore, in the first clustering, humidity, scattered radiation and direct radiation were used as characteristic factors for clustering.

[0025] Preferably, the fuzzy C-means clustering algorithm (FCM) realizes clustering division of given elements by calculating the membership between spatial elements and cluster centers.

[0026] The specific implementation is as follows: Suppose there is a data set X = {x1, x2, ...x n}∈R d×n , where d is the dimension, n is the number of sample points in the data set, and x k ∈R d×n is the kth data point, dividing the data set into c fuzzy sub-data sets: {B i|i=1,2,...c}, 0≤c≤n, the objective function and constraints of the fuzzy C-means clustering algorithm are as follows:

[0027]

[0028] where U is the fuzzy C classification of X, V = {v1, v2, ..., v c} is the cluster center, m>1 is the fuzzy index, which is used to adjust the fuzziness, u ik is the matrix U∈R n×c The (i,k)th element of reflects the degree to which the kth data point belongs to the i-th cluster; when the fuzzy C-means clustering algorithm converges, each data point will be divided into the corresponding category.

[0029] In the second clustering process, the weather without rain is clustered again, and divided into sunny, cloudy and hazy; the rainy weather is also clustered again, and divided into light rain and heavy rain. In the clustering of sunny, cloudy and hazy, diffuse radiation and direct radiation are used as clustering characteristic factors. In the clustering of light rain and heavy rain, humidity, diffuse radiation, direct radiation and ambient temperature are used as clustering characteristic factors.

[0030] Preferably, the K-means clustering algorithm is specifically implemented as follows: manually select K data objects as initial centers from the data set output by the first clustering process, calculate the distance from each data object to each center, and according to the nearest neighbor principle, all data objects will be divided into the cluster represented by the center closest to it, calculate the mean of the data objects in each newly generated cluster and update them as the new center of each cluster, compare the centers within adjacent iterations, and if the change does not exceed the specified threshold, the algorithm converges and outputs the result; otherwise, all data objects are re-divided according to the new center until the convergence condition of the algorithm is met.

[0031] The step of manually selecting K data objects from the data set output by the first clustering process as initial centers includes manually selecting three data objects, namely, one sunny day, one cloudy day, and one haze day, from the data set {D1, D2, D3} as initial centers;

[0032] The manually selecting K data objects from the data set output by the first clustering process as the initial centers includes manually selecting two data objects, one day of light rain and one day of heavy rain, as the initial centers from the data set {D4, D5}.

[0033] The traditional K-Means algorithm generally randomly selects K data objects from the data set as the initial centers, and its clustering results fluctuate with different initial clustering centers, and may fall into a local optimal situation. The present invention optimizes the traditional K-Means algorithm to meet the actual needs of weather type classification. First, one sunny day, one cloudy day and one haze day are manually selected as clustering centers, or one light rain day and one heavy rain day are manually selected as clustering centers, thereby reducing the dependence of the K-Means algorithm on the initial value and improving the efficiency and stability of the algorithm.

[0034] S3, obtaining weather forecast data about the photovoltaic power station prediction point from the meteorological center, determining the weather forecast type of the prediction point, and according to the weather forecast type of the prediction point, selecting the neural network power prediction sub-model corresponding to the weather forecast type to output the predicted value of the photovoltaic power station power generation at the prediction point; wherein sunny weather, cloudy weather, haze weather, light rain weather and heavy rain weather correspond to the first BP neural network power prediction sub-model, the second BP neural network power prediction sub-model, the third BP neural network power prediction sub-model, the first LSTM neural network power prediction sub-model and the second LSTM neural network power prediction sub-model respectively;

[0035] Preferably, the specific implementation is as follows:

[0036] (3-1) When the weather forecast type of the prediction point is sunny weather, the power generation of the photovoltaic power station is predicted based on the first BP neural network power prediction submodel that has been pre-trained. The first BP neural network power prediction submodel comprises an input layer, a hidden layer and an output layer. The input layer has 6 neurons, 4 of which represent the historical data of the power generation of the four adjacent photovoltaic power stations before the prediction point, and the remaining 2 are the solar irradiance and ambient temperature of the prediction point provided by the meteorological center. The number of neurons in the hidden layer is M1, M1>6, and in order to prevent overfitting in the training and learning process, M1 is preferably 7-12, wherein the hidden layer excitation function of the first BP neural network power prediction submodel adopts a sigmoid function.

[0037] (3-2) When the weather forecast type of the prediction point is cloudy weather, the power generation of the photovoltaic power station is predicted based on the pre-trained second BP neural network power prediction sub-model, the second BP neural network power prediction sub-model comprises an input layer, a hidden layer and an output layer, the input layer has 6 neurons, 4 of which represent the historical data of the power generation of the four adjacent photovoltaic power stations before the prediction point, and the remaining 2 are the solar irradiance and ambient temperature of the prediction point provided by the meteorological center; the number of neurons in the hidden layer is M2, M2>6, in order to prevent overfitting in the training and learning process, M2 is preferably 7-12, wherein the hidden layer excitation function of the second BP neural network power prediction sub-model adopts a softmax function;

[0038] (3-3) When the weather forecast type of the prediction point is haze weather, the power generation of the photovoltaic power station is predicted based on the pre-trained third BP neural network power prediction sub-model, the third BP neural network power prediction sub-model comprises an input layer, a hidden layer and an output layer, the input layer has 6 neurons, 4 of which represent the historical data of the power generation of the four adjacent photovoltaic power stations before the prediction point, and the remaining 2 are the solar irradiance correction value and the ambient temperature of the prediction point provided by the meteorological center; the number of hidden layer neurons is M3, M3>6, in order to prevent overfitting in the training and learning process, M3 is preferably 7-12, wherein the hidden layer excitation function of the third BP neural network power prediction sub-model adopts the tanh function; wherein the solar irradiance correction value is predicted and realized based on the support vector machine regression SVR model, the input of the support vector machine regression SVR model is the solar irradiance, ambient temperature and PM2.5 data of the prediction point provided by the meteorological center, and the output is the solar irradiance correction value of the prediction point;

[0039] Preferably, the first BP neural network power prediction sub-model, the second BP neural network power prediction sub-model and the third BP neural network power prediction sub-model are trained based on the following steps:

[0040] Step 3-1-1: According to the system input and output sequence (X', Y'), determine the number of network input layer nodes to be 6, the number of hidden layer nodes to be 1, and initialize the connection weights ω between the input layer, hidden layer, and output layer neurons i'j ,ω jk' ; Initialize the hidden layer threshold a, output layer threshold b, and give the learning rate and hidden layer activation function;

[0041] Step 3-1-2: Hidden layer output calculation; according to the input vector X', the connection weight ω between the input layer and the hidden layer i'j And the hidden layer threshold a, calculate the hidden layer output H;

[0042]

[0043] In the above formula, l is the number of hidden layer nodes; f is the hidden layer activation function;

[0044] Step 3-1-3: Output layer output calculation; according to the hidden layer output H, the connection weight ω jk' and threshold b, calculate the BP neural network prediction output O;

[0045]

[0046] Step 3-1-4: Calculate the network prediction error e based on the network prediction output O and the expected output Y', and update the network connection weight ω based on the network prediction error e i'j ,ω jk' , hidden layer threshold a and output layer threshold b.

[0047] Step 3-1-5: Determine whether the algorithm iteration has ended. If not, return to step 3-1-2.

[0048] Preferably, since the support vector machine has the advantage of processing small sample nonlinear problems, a support vector machine regression SVR model can be established for the case where the amount of haze weather data is small to correct the solar irradiance of the prediction point. The specific implementation is as follows:

[0049] Construct a support vector machine regression SVR model, which is expressed as follows:

[0050] y=g(x)=wφ(x)+b1; w, b1 are model parameters, φ is nonlinear mapping; x is the input parameter, including solar irradiance, ambient temperature and PM2.5 data in the weather forecast data of the forecast point; y is the output solar irradiance correction value;

[0051] In SVR, a variable ε is used to prevent overfitting. ε>0 can also be understood as the tolerance of deviation. The optimization objective form of SVR can be written as the formula:

[0052] C' is the regularization constant. The above equation is solved by Lagrange multipliers. When |g(x)-y|≥ε, L(g(x)-y)=|g(x)-y|-ε, and in other cases, L(g(x)-y)=0.

[0053] (3-4) When the weather forecast type of the forecast point is light rain, the power generation of the photovoltaic power station is predicted based on the first LSTM neural network power prediction sub-model that has been trained in advance, wherein the first LSTM neural network power prediction sub-model consists of three parts: an input gate, an output gate, and a forget gate;

[0054] (3-5) When the weather forecast type of the forecast point is heavy rain, the power generation of the photovoltaic power station is predicted based on the pre-trained second LSTM neural network power prediction sub-model, where the second LSTM neural network power prediction sub-model consists of three parts: an input gate, an output gate, and a forget gate;

[0055] Preferably, the related processing update formulas of the first LSTM neural network power prediction sub-model and the second LSTM neural network power prediction sub-model when working are as follows:

[0056] f t =sigmoid(W f ·[h t-1 ,x t ]+b f ),

[0057] i t =sigmoid(W i ·[h t-1 ,x t ]+b i ),

[0058]

[0059] o t =sigmoid(W o ·[h t-1 ,x t ]+b o ),

[0060] h t =o t *tanh(C t ),

[0061] where x t are input parameters, including the solar irradiance and ambient temperature predicted by the meteorological center at the prediction point and the sequence representation of the historical output power of photovoltaic power plants at multiple neighboring prediction points, f t It is the forget gate, t is the input gate, o t is the output gate, represents the candidate state created by the tanh layer to be added to the unit at the current moment, C t Indicates the current state of the unit, C t-1 represents the state of the unit at the previous moment, h t represents the output of the current unit, and h t-1 represents the output of the unit at the previous moment, sigmoid and tanh are activation functions, W f , W i , W C , Wo is the weight vector, b f , b i , b C , b o is the bias vector.

[0062] S4, controlling the output active power and reactive power of the photovoltaic grid-connected inverter in the photovoltaic power station through the predicted value of the photovoltaic power station power generation at the prediction point.

[0063] The control of the output active power and reactive power of photovoltaic grid-connected inverters is crucial to the stability of the power grid and the balance of power supply and demand.

[0064] In addition, the present invention provides a photovoltaic distribution network power control system, which is used to execute the photovoltaic distribution network power control method. It should be noted that those skilled in the art should understand that the implementation functions of the system in the implementation of the photovoltaic distribution network power control system can be understood with reference to the relevant description of the photovoltaic distribution network power control method. The functions of the system shown in the implementation of the photovoltaic distribution network power control system can be implemented by a program (executable instruction) running on a processor, or by a specific logic circuit.

[0065] Preferably, refer to the attached Figure 2 As shown, the photovoltaic distribution network power control system includes a data acquisition module, a clustering processing module, a photovoltaic power prediction module and a photovoltaic power control module; wherein the data acquisition module is used to collect the power generation of the photovoltaic power station in the historical time period and the corresponding historical meteorological data as historical sample data; the clustering processing module is used to use two clustering processes on the historical sample data to obtain a data set of 5 weather types; the photovoltaic power prediction module is used to obtain the weather forecast data of the photovoltaic power station prediction point from the meteorological center, determine the weather forecast type of the prediction point, and select the neural network power prediction sub-model corresponding to the weather forecast type according to the weather forecast type of the prediction point to output the power generation prediction value of the photovoltaic power station at the prediction point; the photovoltaic power control module is used to control the output active power and reactive power of the photovoltaic grid-connected inverter in the photovoltaic power station through the power generation prediction value of the photovoltaic power station at the prediction point.

[0066] Accordingly, the present invention also provides a computer-readable storage medium, in which computer executable instructions are stored, and when the computer executable instructions are executed by a processor, the various method implementations of the present application are implemented. Computer-readable storage media include permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be a computer-readable instruction, a data structure, a module of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0067] In addition, it should be understood that the above description is only a preferred embodiment of this specification and is not intended to limit the scope of protection of this specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification should be included in the scope of protection of one or more embodiments of this specification.

Claims

1. A photovoltaic power distribution network power control method, characterized in that: The following steps are involved: S1, collecting the power generation of the photovoltaic power station in the historical time period and the corresponding historical meteorological data as historical sample data; S2, clustering the historical sample data twice to obtain five weather type data sets: D1, D2, D3, D4 and D5, which respectively represent sunny, cloudy, haze, light rain and heavy rain weather type data sets; the five weather type data sets are used for subsequent training and prediction of the corresponding neural network power prediction sub-model; S3, obtaining weather forecast data about the photovoltaic power station prediction point from the meteorological center, determining the weather forecast type of the prediction point, and according to the weather forecast type of the prediction point, selecting the neural network power prediction sub-model corresponding to the weather forecast type to output the predicted value of the photovoltaic power station power generation at the prediction point; wherein sunny weather, cloudy weather, haze weather, light rain weather and heavy rain weather correspond to the first BP neural network power prediction sub-model, the second BP neural network power prediction sub-model, the third BP neural network power prediction sub-model, the first LSTM neural network power prediction sub-model and the second LSTM neural network power prediction sub-model respectively; S4, controlling the output active power and reactive power of the photovoltaic grid-connected inverter in the photovoltaic power station through the predicted value of the photovoltaic power station power generation at the prediction point.

2. The photovoltaic power distribution network power control method according to claim 1, characterized in that: The two clustering processes in step S2 are specifically implemented as follows: (2-1) The first clustering process uses the fuzzy C-means clustering algorithm to cluster the historical sample data into two categories {D1, D2, D3} and {D4, D5}, where {D1, D2, D3} is non-rainy weather and {D4, D5} is rainy weather; (2-2) The second clustering process uses the K-means clustering algorithm to cluster the two types of results generated by the first clustering again, and obtains 5 weather type data sets D1, D2, D3, D4, and D5.

3. The photovoltaic power distribution network power control method according to claim 2, characterized in that: The fuzzy C-means clustering algorithm achieves clustering of given elements by calculating the membership between spatial elements and cluster centers. Its specific implementation is as follows: Suppose there is a data set X = {x1, x2, ...x n }∈R d×n , where d is the dimension, n is the number of sample points in the data set, and x k ∈R d is the kth data point, dividing the data set into c fuzzy sub-data sets: {B i |i=1,2,...c}, 0≤c≤n, the objective function and constraints of the fuzzy C-means clustering algorithm are as follows: where U is the fuzzy C classification of X, V = {v1, v2, ..., v c } is the cluster center, m>1 is the fuzzy index, which is used to adjust the fuzziness, u ik is the matrix U∈R n×c The (i, k)th element of reflects the degree to which the kth data point belongs to the i-th cluster; when the fuzzy C-means clustering algorithm converges, each data point will be divided into the corresponding category.

4. The photovoltaic power distribution network power control method according to claim 2 or 3, characterized in that: The K-means clustering algorithm is specifically implemented as follows: manually select K data objects as initial centers from the data set output by the first clustering process, calculate the distance from each data object to each center, and according to the nearest neighbor principle, all data objects will be divided into clusters represented by the center closest to it, respectively calculate the mean of the data objects in each newly generated cluster and update it as the new center of each cluster, compare the centers within adjacent iterations, and if the change does not exceed the specified threshold, the algorithm converges and outputs the result; otherwise, all data objects are re-divided according to the new center until the convergence condition of the algorithm is met; The method of manually selecting K data objects from the data set output by the first clustering process as the initial centers includes manually selecting 3 data objects, namely, 1 sunny day, 1 cloudy day and 1 haze day, from the data set {D1, D2, D3} as the initial centers; and manually selecting 2 data objects, namely, 1 light rain day and 1 heavy rain day, from the data set {D4, D5} as the initial centers.

5. The photovoltaic power distribution network power control method according to claim 4, characterized in that: Step S3 includes: (3-1) When the weather forecast type of the prediction point is sunny weather, the power generation of the photovoltaic power station is predicted based on the first BP neural network power prediction sub-model that has been trained in advance. The first BP neural network power prediction sub-model comprises an input layer, a hidden layer and an output layer. The input layer has 6 neurons, 4 of which represent the historical data of the power generation of the four adjacent photovoltaic power stations before the prediction point, and the remaining 2 are the solar irradiance and ambient temperature of the prediction point provided by the meteorological center; the number of neurons in the hidden layer is M1, M1>6, and the hidden layer excitation function of the first BP neural network power prediction sub-model adopts a sigmoid function; (3-2) When the weather forecast type of the prediction point is cloudy weather, the power generation of the photovoltaic power station is predicted based on the pre-trained second BP neural network power prediction sub-model, the second BP neural network power prediction sub-model comprises an input layer, a hidden layer and an output layer, the input layer has 6 neurons, 4 of which represent the historical data of the power generation of the four adjacent photovoltaic power stations before the prediction point, and the remaining 2 are the solar irradiance and ambient temperature of the prediction point provided by the meteorological center; the number of neurons in the hidden layer is M2, M2>6, wherein the hidden layer excitation function of the second BP neural network power prediction sub-model adopts a softmax function; (3-3) When the weather forecast type of the prediction point is haze weather, the power generation of the photovoltaic power station is predicted based on the pre-trained third BP neural network power prediction sub-model, the third BP neural network power prediction sub-model comprises an input layer, a hidden layer and an output layer, the input layer has 6 neurons, 4 of which represent the historical data of the power generation of the four adjacent photovoltaic power stations before the prediction point, and the remaining 2 are the solar irradiance correction value and ambient temperature of the prediction point provided by the meteorological center; the number of hidden layer neurons is M3, M3>6, wherein the hidden layer excitation function of the third BP neural network power prediction sub-model adopts the tanh function; wherein the solar irradiance correction value is predicted and realized based on the support vector machine regression SVR model, the input of the support vector machine regression SVR model is the solar irradiance, ambient temperature and PM2.5 data of the prediction point provided by the meteorological center, and the output is the solar irradiance correction value of the prediction point; (3-4) When the weather forecast type of the forecast point is light rain, the power generation of the photovoltaic power station is predicted based on the first LSTM neural network power prediction sub-model that has been trained in advance, wherein the first LSTM neural network power prediction sub-model consists of three parts: an input gate, an output gate, and a forget gate; (3-5) When the weather forecast type of the forecast point is heavy rain, the power generation power of the photovoltaic power station is predicted based on the pre-trained second LSTM neural network power prediction sub-model, and the second LSTM neural network power prediction sub-model consists of three parts: input gate, output gate, and forget gate.

6. The photovoltaic power distribution network power control method according to claim 5, characterized in that: The first BP neural network power prediction sub-model, the second BP neural network power prediction sub-model and the third BP neural network power prediction sub-model are trained based on the following steps: Step 3-1-1: According to the system input and output sequence (X', Y'), determine the number of network input layer nodes to be 6, the number of hidden layer nodes to be 1, and initialize the connection weights ω between the input layer, hidden layer, and output layer neurons i'j ,ω jk' ; Initialize the hidden layer threshold a, output layer threshold b, and give the learning rate and hidden layer activation function; Step 3-1-2: Hidden layer output calculation; according to the input vector X', the connection weight ω' between the input layer and the hidden layer ij And the hidden layer threshold a, calculate the hidden layer output H; In the above formula, l is the number of hidden layer nodes; f is the hidden layer activation function; Step 3-1-3: Output layer output calculation; according to the hidden layer output H, the connection weight ω jk' and threshold b, calculate the BP neural network prediction output O; m' is a natural number; Step 3-1-4: Calculate the network prediction error e based on the network prediction output O and the expected output Y', and update the network connection weight ω based on the network prediction error e i'j ,ω jk' , hidden layer threshold a and output layer threshold b; Step 3-1-5: Determine whether the algorithm iteration has ended. If not, return to step 3-1-2.

7. The photovoltaic power distribution network power control method according to claim 5, characterized in that: In step (3-3), a support vector machine regression SVR model is established for haze weather to correct the solar irradiance at the prediction point. The specific implementation is as follows: construct a support vector machine regression SVR model, The model is expressed as follows: y=g(x)=wφ(x)+b1; w, b1 are model parameters, φ is a nonlinear mapping; x is the input parameter, including solar irradiance, ambient temperature and PM2.5 data in the weather forecast data of the prediction point; y is the output solar irradiance correction value.

8. The photovoltaic power distribution network power control method according to claim 5, characterized in that: The related processing update formulas of the first LSTM neural network power prediction sub-model and the second LSTM neural network power prediction sub-model during operation are as follows: f t =sigmoid(W f ·[h t-1 ,x t ]+b f ), i t =sigmoid(W i ·[h t-1 ,x t ]+b i ), o t =sigmoid(W o ·[h t-1 ,x t ]+b o ), h t =o t *tanh(C t ), where x t are input parameters, including the solar irradiance and ambient temperature predicted by the meteorological center at the prediction point and the sequence representation of the historical output power of photovoltaic power plants at multiple neighboring prediction points, f t It is the forget gate, t is the input gate, o t is the output gate, represents the candidate state created by the tanh layer to be added to the unit at the current moment, C t Indicates the current state of the unit, C t-1 represents the state of the unit at the previous moment, h t represents the output of the current unit, and h t-1 represents the output of the unit at the previous moment, sigmoid and tanh are activation functions, W f , W i , W C , W o is the weight vector, b f 、b i 、b C 、b o is the bias vector.

9. A photovoltaic power distribution network power control system, characterized in that: The photovoltaic power distribution network power control system is used to execute the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: One or more programs are stored, and the one or more programs enable a computer to execute the method according to any one of claims 1 to 8.

Images