Distributed photovoltaic power generation output prediction method and system, and related equipment

Abstract

The invention is suitable for the technical field of output prediction, and provides a distributed photovoltaic power generation output prediction method and system and related equipment, and the method comprises the steps: obtaining the historical data of distributed photovoltaic power generation output, and carrying out the data preprocessing of the historical data; constructing a spatio-temporal information graph based on the preprocessed data, wherein the spatio-temporal information graph takes a cellular node as a point and takes a relation between cells as an edge; determining a target space-time correlation coefficient matrix by calculating time correlation and space correlation between nodes of the space-time information graph; performing feature extraction on the target space-time correlation coefficient matrix by using a convolutional neural network to obtain a target feature vector; and inputting the feature vector into a time sequence convolutional network for prediction, and generating a distributed photovoltaic power generation output prediction result. According to the method, based on the spatial-temporal correlation, the learning ability of the model in the aspects of space and time is improved, better generalization ability is achieved, and the accuracy of model prediction is obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of output prediction, and in particular to a distributed photovoltaic power generation output prediction method, system and related equipment. Background Art

[0002] In recent years, as a new type of green energy production method, distributed photovoltaic power generation system advocates local consumption and reduces transmission losses, and therefore has broad development prospects. At present, there are two main data-driven photovoltaic output prediction methods: traditional methods based on time series modeling and intelligent algorithms based on machine learning. Intelligent algorithms are easy to obtain information, simple and flexible modeling, and have high accuracy. However, current machine learning only focuses on meteorological characteristics, ignores the important differentiated characteristics of each node in distributed photovoltaics, lacks the distinction of personalized information of power generators, and leads to poor accuracy of prediction results. In addition, due to the fragmented characteristics of distributed photovoltaics, the prediction model can only use numerical weather forecast data from nearby meteorological stations, resulting in a certain deviation between the actual weather values ​​and the collected meteorological data, resulting in errors in the prediction results.

[0003] In view of this, a distributed photovoltaic power generation output prediction method, system and related equipment are needed. Summary of the invention

[0004] The embodiments of the present application provide a distributed photovoltaic power generation output prediction method, system and related equipment to solve the problem of errors in output prediction results.

[0005] A first aspect of an embodiment of the present application provides a distributed photovoltaic power generation output prediction method, comprising:

[0006] Acquire historical data of distributed photovoltaic power generation output, and perform data preprocessing on the historical data;

[0007] Constructing a spatiotemporal information graph based on the preprocessed data, wherein the spatiotemporal information graph uses cell nodes as points and relationships between cells as edges;

[0008] Determine a target spatiotemporal correlation coefficient matrix by calculating the temporal correlation and spatial correlation between the nodes of the spatiotemporal information graph;

[0009] Using a convolutional neural network to extract features from the target spatiotemporal correlation coefficient matrix to obtain a target feature vector;

[0010] The feature vector is input into a time series convolutional network for prediction to generate a distributed photovoltaic power generation output prediction result.

[0011] Furthermore, the expression of the spatiotemporal information graph is as follows:

[0012] G SO=(V SO ,E SO );

[0013] Where: G SO V is a space-time information map. SO is the set of type I cell nodes, E SO is the set of type I cell edges.

[0014] Furthermore, determining the target spatiotemporal correlation coefficient matrix by calculating the temporal correlation and spatial correlation between the nodes of the spatiotemporal information graph includes:

[0015] The calculation formula of the target spatiotemporal correlation coefficient matrix is ​​as follows:

[0016]

[0017] Where: COV N is the time correlation matrix, COV S is the spatial correlation matrix, PV f and PV f+1 The covariance matrix of and are the corresponding standard deviations, and The corresponding correlation coefficients modeled as Pearson coefficients respectively.

[0018] Furthermore, the adjacency matrix is ​​established based on the correlation coefficient using a threshold Gaussian kernel method, and the expression is as follows:

[0019]

[0020] Where: i, j are the i-th and j-th cells respectively, θ is the standard deviation of the correlation coefficient, and γ is the threshold.

[0021] Furthermore, the use of a convolutional neural network to extract features from the target spatiotemporal correlation coefficient matrix to obtain a target feature vector includes:

[0022] The node feature expression is as follows:

[0023]

[0024] Where: is the feature description of the nodes in layer l, f relu is the activation function, W (l) is the weight matrix, d l is the characteristic dimension of the l-th layer node, H (l+1) is the node feature updated after layer l and input to the next layer, is the normalized matrix of the adjacency matrix A;

[0025] The output of the graph convolution obtained by summing all layers in the convolutional neural network is expressed as follows:

[0026]

[0027] Where: U∈R N×D , Z k is the probability transfer matrix of the lth step, W k ∈R N×D is the weight coefficient of the kth step, K b is the total number of diffusion steps.

[0028] Furthermore, the expression for inputting the feature vector into the temporal convolutional network for prediction is as follows:

[0029] X (h) =ReLU(F(X (h―1) )+X (j―1) );

[0030] Where: X (h) is the output, X (h―1) is the input and ReLU is the activation function.

[0031] Furthermore, the data preprocessing of the historical data includes data cleaning and normalization, and the data cleaning includes missing value processing, negative value processing and constant value processing.

[0032] Furthermore, the normalization process includes:

[0033] Find the maximum daily output based on the sample data of data cleaning;

[0034] In the daily maximum processing, find the maximum output value by month, recorded as P max ;

[0035] The daily force data is standardized and the expression is as follows:

[0036]

[0037] Where: P′ i The normalized daily output data.

[0038] A second aspect of the embodiment of the present application provides a distributed photovoltaic power generation output prediction system, including:

[0039] A historical data acquisition unit, used to acquire historical data of distributed photovoltaic power generation output and perform data preprocessing on the historical data;

[0040] A space-time information graph construction unit, used to construct a space-time information graph based on the preprocessed data, wherein the space-time information graph uses cell nodes as points and relationships between cells as edges;

[0041] A target spatiotemporal correlation coefficient matrix determination unit, used to determine the target spatiotemporal correlation coefficient matrix by calculating the time correlation and spatial correlation between the nodes of the spatiotemporal information graph;

[0042] A target feature vector determination unit, used for extracting features from the target spatiotemporal correlation coefficient matrix using a convolutional neural network to obtain a target feature vector;

[0043] The output prediction result generating unit is used to input the feature vector into the time series convolution network for prediction, so as to generate the distributed photovoltaic power generation output prediction result.

[0044] A third aspect of an embodiment of the present application provides a computer device, including a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, wherein when the processor executes the computer-readable instructions, the steps of the distributed photovoltaic power generation output prediction method as described in any one of the above items are implemented.

[0045] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0046] This application first obtains the historical data of distributed photovoltaic power generation output, and performs data preprocessing, and constructs a space-time information graph based on the preprocessed data, in which the space-time information graph uses cell nodes as points and the relationship between cells as edges; then the target space-time correlation coefficient matrix is ​​determined by calculating the time correlation and spatial correlation between the nodes of the space-time information graph; the graph convolutional neural network is used to extract features of the target space-time correlation coefficient matrix to obtain the target feature vector; finally, the feature vector is input into the time series convolutional network for prediction, and the distributed photovoltaic power generation output prediction result is generated. By deeply mining the space-time correlation of the output of the distributed photovoltaic system, the learning ability of the model in both space and time is effectively improved, and it has better generalization ability, which significantly improves the accuracy of the model prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of an embodiment of a distributed photovoltaic power generation output prediction method in the present invention;

[0048] Figure 2 It is the overall flow chart of the output prediction method in the present invention;

[0049] Figure 3 It is the spatiotemporal information graph in the present invention;

[0050] Figure 4 It is a structural diagram of the convolutional neural network in the present invention;

[0051] Figure 5 It is a structural diagram of the prediction model in the present invention;

[0052] Figure 6 This is a comparison chart of the prediction results in the present invention;

[0053] Figure 7 It is a structural block diagram of an embodiment of a distributed photovoltaic power generation output prediction system in the present invention;

[0054] Figure 8 The present invention provides a structural block diagram of a computer device. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0056] The distributed photovoltaic power generation output prediction method in this embodiment is used to improve the accuracy of distributed power generation output prediction. The implementation method in this embodiment can be implemented in the system, can be implemented in the server, and can also be implemented in the terminal, without specific limitation.

[0057] Embodiment 1

[0058] See also Figure 1 to Figure 2 , an embodiment of a distributed photovoltaic power generation output prediction method in the present invention comprises the following steps:

[0059] S11. Obtain historical data of distributed photovoltaic power generation output and perform data preprocessing on the historical data;

[0060] In this embodiment, the historical data obtained is the historical output data of each power station or each power generation system over a period of time, which may include historical power generation and data on weather factors that affect power generation. The amount of data is not limited here, but the acquisition of the data amount can make the model fit well.

[0061] There may be many abnormal problems in the data generated during the photovoltaic output process. Therefore, it is necessary to preprocess the data to eliminate some unreasonable data and screen the qualified data. Specifically, data preprocessing includes data cleaning and normalization, among which data cleaning includes missing value processing, negative value processing and constant value processing.

[0062] The missing data mainly refers to the missing of the entire time period and the partial missing of the field. The interruption of communication during the photovoltaic output process, the failure of the collection equipment and the failure of the storage cutoff will all lead to data missing, as follows:

[0063] Let X = {x1, x2, ... x n} is the time series of photovoltaic output, if there is x i satisfy:

[0064]

[0065] Based on the above-set conditions, it is determined that there is data missing. For long-term data missing, the data record of this photovoltaic user is directly deleted. If the missing values ​​are small, no processing is performed and it is regarded as photovoltaic output fluctuation.

[0066] Negative values ​​and constant values ​​are typical anomalies in output data, which may be caused by reverse connection of the acquisition terminal or abnormal monitoring data storage, and manifested as:

[0067] x i ={0│x i =A,i=1,2,..,n};

[0068] According to the above negative value and constant value exceptions, the processing method here is to directly delete the data record.

[0069] Data normalization processing includes the following:

[0070] 1. Find the maximum daily output based on the sample data of data cleaning;

[0071] 2. Find the maximum output value by month in the daily maximum processing, record it as P max ;

[0072] 3. Standardization of daily output data, the expression is as follows:

[0073]

[0074] Where: P′ i The normalized daily output data.

[0075] The above normalization is to map the daily output time series to [0,1] under the same weather type. The larger the ratio of the own output to the historical maximum output, the better the weather environment on that day.

[0076] S12. construct a spatiotemporal information graph based on the preprocessed data, wherein the spatiotemporal information graph uses cell nodes as points and relationships between cells as edges;

[0077] In this embodiment, Figure 3 As shown in Figure 1, the spatiotemporal information graph can help discover the spatiotemporal correlation between nodes in a distributed system. By superimposing and comparing the spatiotemporal information of nodes, the mutual influence and dependency between nodes can be discovered. This is conducive to understanding the spatial interaction between nodes, the information transmission path, and the spatiotemporal diffusion law of the system.

[0078] The expression of the space-time information graph is as follows:

[0079] G SO =(V SO ,E SO );

[0080] Where: G SO V is a space-time information map. SO is the set of type I cell nodes, E SO is the set of type I cell edges.

[0081] S13. Determine the target spatiotemporal correlation coefficient matrix by calculating the temporal correlation and spatial correlation between the spatiotemporal information graph nodes;

[0082] In this embodiment, the spatial correlation between distributed systems is calculated using the covariance-based spatiotemporal correlation, the spatial and temporal correlations in the historical PV output data between paired sites are modeled, and these correlations are used to construct a complete spatiotemporal correlation coefficient matrix.

[0083] The calculation formula of the target spatiotemporal correlation coefficient matrix is ​​as follows:

[0084]

[0085]

[0086] Where: COV N is the time correlation matrix, COV S is the spatial correlation matrix, PV f and PV f+1 The covariance matrix of and are the corresponding standard deviations, and The corresponding correlation coefficients modeled as Pearson coefficients respectively.

[0087] Based on the correlation coefficient, the adjacency matrix is ​​established using the threshold Gaussian kernel method. The expression is as follows:

[0088]

[0089] Where: i, j are the i-th and j-th cells respectively, θ is the standard deviation of the correlation coefficient, and γ is the threshold.

[0090] The above adjacency matrix reflects the connection relationship between nodes in the graph. By integrating the adjacency matrix of spatial correlation, GCN can accurately capture the proximity relationship between nodes, thereby effectively learning the information of the spatiotemporal graph.

[0091] S14. Using a graph convolutional neural network to extract features from the target spatiotemporal correlation coefficient matrix to obtain a target feature vector;

[0092] like Figure 4 As shown in Figure 1, the graph convolutional neural network (GCN) learns the representation of nodes by utilizing the connection relationship between nodes and local neighboring information. The expression of multi-layer GCN updating node features according to the following propagation rule is as follows:

[0093]

[0094] Where: is the feature description of the nodes in layer l, f relu is the activation function, W (l) is the weight matrix, d l is the characteristic dimension of the l-th layer node, H (l+1) is the node feature updated after layer l and input to the next layer, is the normalized matrix of the adjacency matrix A.

[0095] The expression of the normalized matrix is ​​as follows:

[0096]

[0097] Where: for is the degree matrix of , and I is the identity matrix.

[0098] The output of graph convolution can be obtained by summing up all layers, and the expression is as follows:

[0099]

[0100] Where: U∈R N×D , Z k is the probability transfer matrix of the kth step, W k ∈R N×D is the weight coefficient of the kth step, K b is the total number of diffusion steps.

[0101] S15. Input the feature vector into the time series convolutional network for prediction to generate the distributed photovoltaic power generation output prediction result.

[0102] In this embodiment, the information processed by the graph convolutional neural network is input into the time series convolutional network for prediction. The prediction model is as follows: Figure 5 As shown in Figure 2, TCN introduces residual network and dilated causal convolution on the basis of CNN. The residual network contains two layers of convolution and nonlinear mapping. WeightNorm and Dropout are added to each layer to regularize the network. For the input X of the residual network (h―1)After the residual network, the output is X (h) , the expression is as follows:

[0103] X (h) =ReLU(F(X (h―1) )+X (h―1) );

[0104] Where: ReLU is the activation function, and its expression is as follows:

[0105]

[0106] The following examples verify the above method, as follows:

[0107] The power generation data of a distributed photovoltaic system in a certain city in my country in March 2023 is used, with a sampling granularity of 15 minutes, and the distributed photovoltaic system is used as the node of Class I cell. The historical data of the past week is used to predict the data of the next day. In order to verify the effectiveness of the output prediction method in the present invention, the method proposed in the present invention is experimentally compared with the GNN-TCN model, GCN-LSTM model, TCN model, and CNN-LSTM model. The prediction result of cell 1 is as follows Figure 6 As shown in the figure, the prediction accuracy of the method proposed in the present invention is higher than that of several other models, and the prediction curve is closer to the real output curve than that of other models, and has higher accuracy and stability than other methods.

[0108] The above embodiment takes into account the characteristics of the distributed power generation system, mines its features from both time and space, and considers the correlation of time and space. At the same time, the prediction model has a stronger learning ability for the time and space information graph than other networks. Combining these two, the accuracy of distributed photovoltaic power generation output prediction is greatly improved.

[0109] Embodiment 2

[0110] See also Figure 7 , an embodiment of a distributed photovoltaic power generation output prediction system in the present invention comprises the following steps:

[0111] The historical data acquisition unit 101 is used to acquire the historical data of the distributed photovoltaic power generation output and perform data preprocessing on the historical data;

[0112] A spatiotemporal information graph construction unit 102 is used to construct a spatiotemporal information graph based on the preprocessed data, wherein the spatiotemporal information graph uses cell nodes as points and relationships between cells as edges;

[0113] A target spatiotemporal correlation coefficient matrix determining unit 103 is used to determine a target spatiotemporal correlation coefficient matrix by calculating the temporal correlation and spatial correlation between the spatiotemporal information graph nodes;

[0114] A target feature vector determination unit 104 is used to extract features from a target spatiotemporal correlation coefficient matrix using a convolutional neural network to obtain a target feature vector;

[0115] The output prediction result generating unit 105 is used to input the feature vector into the time series convolution network for prediction, and generate the distributed photovoltaic power generation output prediction result.

[0116] For the specific definition of the prediction system, please refer to the definition of the prediction method above, which will not be repeated here. Each module in the above prediction system can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0117] Embodiment 3

[0118] See also Figure 8 The present invention provides a computer device, including a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor. When the processor executes the computer-readable instructions, the steps of the above method are implemented.

[0119] Those of ordinary skill in the art will appreciate that the units of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition of each example has been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0120] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored. In addition, each functional unit in each embodiment of the present invention can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units.

[0121] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-0nlyMemory), random access memory (RAM, RandomAccessMemory), mobile hard disk, magnetic disk or optical disk, etc., which can store program code.

[0122] It can be understood that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A method for predicting the output of distributed photovoltaic power generation. It is characterized in that include: Acquire historical data of distributed photovoltaic power generation output, and perform data preprocessing on the historical data; Constructing a spatiotemporal information graph based on the preprocessed data, wherein the spatiotemporal information graph uses cell nodes as points and relationships between cells as edges; Determine a target spatiotemporal correlation coefficient matrix by calculating the temporal correlation and spatial correlation between nodes of the spatiotemporal information graph; Using a convolutional neural network to extract features from the target spatiotemporal correlation coefficient matrix to obtain a target feature vector; The feature vector is input into a time series convolutional network for prediction to generate a distributed photovoltaic power generation output prediction result.

2. The distributed photovoltaic power generation output prediction method according to claim 1, It is characterized in that The expression of the space-time information graph is as follows: G SO =(V SO ,E SO ); Where: E SO V is a space-time information map. SO is the set of type I cell nodes, E SO is the set of type I cell edges.

3. The distributed photovoltaic power generation output prediction method according to claim 1, It is characterized in that The step of determining a target spatiotemporal correlation coefficient matrix by calculating the temporal correlation and spatial correlation between the nodes of the spatiotemporal information graph includes: The calculation formula of the target spatiotemporal correlation coefficient matrix is ​​as follows: Where: COV N is the time correlation matrix, COV S is the spatial correlation matrix, PV f and PV f+1 The covariance matrix of and are the corresponding standard deviations, and The corresponding correlation coefficients modeled as Pearson coefficients respectively.

4. The distributed photovoltaic power generation output prediction method according to claim 3, It is characterized in that The adjacency matrix is ​​established based on the correlation coefficient using the threshold Gaussian kernel method, and the expression is as follows: Where: i, j are the i-th and j-th cells respectively, is the standard deviation of the correlation coefficient, and γ is the threshold.

5. The distributed photovoltaic power generation output prediction method according to claim 1, It is characterized in that The convolutional neural network is used to extract features from the target spatiotemporal correlation coefficient matrix to obtain a target feature vector. include: The node feature expression is as follows: Where: is the feature description of the nodes in layer l, f relu is the activation function, W (l) is the weight matrix, d l is the characteristic dimension of the l-th layer node, H (l+1) is the node feature updated after layer l and input to the next layer, is the normalized matrix of the adjacency matrix A; The output of the graph convolution obtained by summing all layers in the convolutional neural network is expressed as follows: Where: U∈R N×D , Z k is the probability transfer matrix of the kth step, W k ∈R N×D is the weight coefficient of the kth step, K b is the total number of diffusion steps.

6. The distributed photovoltaic power generation output prediction method according to claim 1, It is characterized in that The expression for inputting the feature vector into the temporal convolutional network for prediction is as follows: X (h) =ReLU(F(X (h―1) )+X (h―1) ): Where: X (h) is the output, X (h―1) is the input and ReLU is the activation function.

7. The distributed photovoltaic power generation output prediction method according to claim 1, It is characterized in that The data preprocessing of the historical data includes data cleaning and normalization, and the data cleaning includes missing value processing, negative value processing and constant value processing.

8. The distributed photovoltaic power generation output prediction method according to claim 7, It is characterized in that The normalization process includes: Find the maximum daily output based on the sample data of data cleaning; In the daily maximum processing, find the maximum output value by month, recorded as P max ; The daily force data is standardized and the expression is as follows: Where: P′ i The normalized daily output data.

9. A distributed photovoltaic power generation output prediction system, It is characterized in that include: A historical data acquisition unit, used to acquire historical data of distributed photovoltaic power generation output and perform data preprocessing on the historical data; A space-time information graph construction unit, used to construct a space-time information graph based on the preprocessed data, wherein the space-time information graph uses cell nodes as points and relationships between cells as edges; A target spatiotemporal correlation coefficient matrix determination unit, configured to determine a target spatiotemporal correlation coefficient matrix by calculating the temporal correlation and spatial correlation between nodes of the spatiotemporal information graph; A target feature vector determination unit, used for extracting features from the target spatiotemporal correlation coefficient matrix using a convolutional neural network to obtain a target feature vector; The output prediction result generating unit is used to input the feature vector into the time series convolution network for prediction, so as to generate the distributed photovoltaic power generation output prediction result.

10. A computer device comprising a memory, a processor, and computer readable instructions stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer-readable instructions, the steps of the distributed photovoltaic power generation output prediction method according to any one of claims 1 to 8 are implemented.

Images