Photovoltaic power station power generation prediction method and system combined with dynamic update mechanism

Abstract

The present invention discloses a photovoltaic power station power generation prediction method and system incorporating a dynamic update mechanism, which relates to the field of photovoltaic power generation. The method comprises: collecting power generation data of a photovoltaic component array within a target photovoltaic power station within a preset historical window; performing iterative data fluctuation analysis on the power generation data array sequence in chronological order; determining a power generation prediction window based on the size of the data fluctuation factor, traversing and extracting power generation data of the photovoltaic component array within the power generation prediction window to obtain a prediction window power generation data array sequence; performing centralized power generation data screening on the prediction window power generation data array sequence to determine a target prediction window power generation data array; and calling a power generation prediction network layer to identify the target prediction window power generation data array to obtain a power generation prediction result. The method solves the technical problem of low prediction accuracy in existing photovoltaic power station power generation predictions, achieving the technical effect of improving the accuracy of photovoltaic power station power generation predictions.

Description

Technical Field

[0001] The present application relates to the field of photovoltaic power generation, and in particular to a method and system for predicting the power generation of a photovoltaic power station in combination with a dynamic update mechanism. Background Art

[0002] In the operation and management of photovoltaic power plants, accurate power generation forecasting is crucial for optimizing plant operations and improving energy efficiency. Currently, power generation forecasting for photovoltaic power plants primarily relies on statistical learning methods. These methods utilize historical power generation data and environmental factor data, using statistical methods such as regression analysis and time series analysis to predict power generation. Their predictive performance relies on extensive historical data and stable environmental factors. Because the power generation process of photovoltaic power plants is influenced by a variety of complex factors, these methods struggle to accurately capture the changing patterns of these factors when processing dynamically changing power generation data, resulting in low prediction accuracy.

[0003] In the current related technologies, there is a technical problem in the prediction of photovoltaic power generation, which is the low prediction accuracy. Summary of the Invention

[0004] The present application provides a photovoltaic power station power generation prediction method and system combined with a dynamic update mechanism, which collects power generation data of a photovoltaic component array in a target photovoltaic power station within a preset historical window to obtain a power generation data array sequence, performs iterative data fluctuation analysis on the power generation data array sequence in chronological order to obtain a data fluctuation factor, determines a power generation prediction window according to the size of the data fluctuation factor, traverses and extracts power generation data of the photovoltaic component array in the power generation prediction window to obtain a prediction window power generation data array sequence, performs centralized power generation data screening on the prediction window power generation data array sequence, determines a target prediction window power generation data array, calls a power generation prediction network layer to identify the target prediction window power generation data array, and obtains a power generation prediction result, thereby achieving the technical effect of improving the power generation prediction accuracy of the photovoltaic power station.

[0005] The present application provides a photovoltaic power station power generation prediction method combined with a dynamic update mechanism, comprising: collecting power generation data of a photovoltaic component array in a target photovoltaic power station within a preset historical window to obtain a power generation data array sequence; performing iterative data fluctuation analysis on the power generation data array sequence in chronological order to obtain a data fluctuation factor; determining a power generation prediction window based on the magnitude of the data fluctuation factor, traversing and extracting power generation data of the photovoltaic component array within the power generation prediction window to obtain a prediction window power generation data array sequence; performing centralized power generation data screening on the prediction window power generation data array sequence to determine a target prediction window power generation data array; and calling a power generation prediction network layer to identify the target prediction window power generation data array to obtain a power generation prediction result.

[0006] In a possible implementation, iterative data fluctuation analysis is performed on the power generation data array sequence in chronological order to obtain a data fluctuation factor, and the following processing is performed: a first power generation data array and a second power generation data array are extracted from the power generation data array sequence in chronological order; the first power generation data array and the second power generation data array are iteratively correlated to obtain a first interaction correlation matrix and a first interaction correlation power generation data array; the first interaction correlation power generation data array is iteratively correlated with a third power generation data array extracted in chronological order from the power generation data array sequence to obtain a second interaction correlation matrix and a second interaction correlation power generation data array; the N-2th interaction correlation power generation data array is iteratively correlated with the Nth power generation data array extracted in chronological order from the power generation data array sequence to obtain an N-1th interaction correlation matrix, where N is the number of power generation data arrays in the power generation data array sequence; and an iterative data fluctuation analysis network layer is called to perform fluctuation analysis on the first interaction correlation matrix, the second interaction correlation matrix, the N-2th interaction correlation matrix, and the N-1th interaction correlation matrix to obtain the data fluctuation factor.

[0007] In a possible implementation, the following processing is performed: one-to-one data similarity identification is performed on the first power generation data array and the second power generation data array to obtain a first similarity array; the first similarity array is normalized using a normalization function, and the processing result is matrixed to obtain a first interaction correlation matrix; the second power generation data array and the first interaction correlation matrix are calculated using a convolutional network to obtain a first interaction correlation power generation data array.

[0008] In a possible implementation, the following processing is performed: obtaining a normalization function, wherein the normalization function is:

[0009]

[0010] Among them, SFT[lim(x i ,y i )] is the normalized value corresponding to the i-th first similarity in the first similarity array, e is the base of the natural logarithm, n is the total number of first similarities in the first similarity array, x i is the first power generation data of the first power generation data array, y i is the i-th second power generation data in the second power generation data array, lim(x i ,y i ) is the i-th first similarity in the first similarity array.

[0011] In a possible implementation, centralized power generation data screening is performed on the prediction window power generation data array sequence to determine a target prediction window power generation data array, and the following processing is performed: a first prediction window power generation data array is randomly extracted from the prediction window power generation data array sequence and used as a first screening center; a first screening neighborhood of the first screening center is constructed in the prediction window power generation data array sequence according to a preset similarity threshold; the first screening center is iteratively updated in the first screening neighborhood using a screening center iteration function to obtain an iterative screening center; an iterative screening neighborhood of the iterative screening center is again constructed in the prediction window power generation data array sequence according to the preset similarity threshold; the iterative screening center is iteratively updated in the iterative screening neighborhood using a screening center iteration function to obtain an updated iterative screening center; and so on, until the similarity of two iterative screening centers obtained in two adjacent iterations is greater than or equal to a preset iterative similarity, the iteration is stopped, and the prediction window power generation data array corresponding to the iterative screening center obtained in the last iteration is used as the target prediction window power generation data array.

[0012] In a possible implementation, the following processing is performed: the screening center iteration function is:

[0013]

[0014] Among them, Point(z) is the iterative screening center, N(z) is the first screening neighborhood, and z q is the power generation data array of the qth prediction window in the first screening neighborhood, z is the first screening center, K(z q -z) is a weight kernel function built based on the Gaussian function.

[0015] In a possible implementation, the power generation prediction network layer is called to identify the target prediction window power generation data array to obtain the power generation prediction result, and the following processing is performed: multiple sample prediction window power generation data arrays and multiple sample power generation prediction results are obtained as training data; the training data is used to supervise the training of a framework built based on a convolutional neural network to learn a one-to-one mapping relationship between the prediction window power generation data array and the sample power generation prediction results until the training converges, thereby obtaining the trained power generation prediction network layer.

[0016] The present application also provides a photovoltaic power station power generation prediction system combined with a dynamic update mechanism, including: a historical power generation data acquisition module, used to collect power generation data of the photovoltaic component array in the target photovoltaic power station within a preset historical window to obtain a power generation data array sequence; a data fluctuation analysis module, used to iteratively perform data fluctuation analysis on the power generation data array sequence in chronological order to obtain a data fluctuation factor; a predicted power generation data extraction module, used to determine a power generation prediction window according to the size of the data fluctuation factor, traverse and extract the power generation data of the photovoltaic component array in the power generation prediction window to obtain a prediction window power generation data array sequence; a centralized screening module, used to perform centralized power generation data screening on the prediction window power generation data array sequence to determine a target prediction window power generation data array; and a power generation prediction module, used to call a power generation prediction network layer to identify the target prediction window power generation data array to obtain a power generation prediction result.

[0017] The photovoltaic power station power generation prediction method and system proposed in this application in combination with a dynamic update mechanism first collects the power generation data of the photovoltaic component array in the target photovoltaic power station within a preset historical window to obtain a power generation data array sequence, then performs iterative data fluctuation analysis on the power generation data array sequence in chronological order to obtain a data fluctuation factor, then determines the power generation prediction window according to the size of the data fluctuation factor, traverses and extracts the power generation data of the photovoltaic component array in the power generation prediction window to obtain a prediction window power generation data array sequence, then performs centralized power generation data screening on the prediction window power generation data array sequence to determine the target prediction window power generation data array, and finally calls the power generation prediction network layer to identify the target prediction window power generation data array to obtain the power generation prediction result, thereby achieving the technical effect of improving the power generation prediction accuracy of the photovoltaic power station. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention are briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in precise order. Instead, various steps may be processed in reverse order or simultaneously as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0019] Figure 1 A flow chart of a photovoltaic power station power generation prediction method combined with a dynamic update mechanism provided in an embodiment of the present application.

[0020] Figure 2 A schematic diagram of the structure of a photovoltaic power station power generation prediction system combined with a dynamic update mechanism provided in an embodiment of the present application.

[0021] Description of the accompanying drawings: historical power generation data collection module 10, data fluctuation analysis module 20, predicted power generation data extraction module 30, centralized screening module 40, power generation prediction module 50. DETAILED DESCRIPTION

[0022] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0023] In order to make the purpose, technical solutions and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting this application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0024] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict, and the terms “first\second” involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. The terms “including” and “having” and any variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or that are inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. The terms used herein are for the purpose of describing the embodiments of this application only.

[0025] The embodiment of the present application provides a photovoltaic power station power generation prediction method combined with a dynamic update mechanism, such as Figure 1 As shown, the method includes:

[0026] Step S100 : collecting power generation data of a photovoltaic module array in a target photovoltaic power station within a preset historical window to obtain a power generation data array sequence.

[0027] Specifically, the photovoltaic module array within the target photovoltaic power station is determined. The photovoltaic module array is a collection of multiple photovoltaic modules (such as solar panels) that are used to convert light energy into electrical energy. A preset historical window is set. This window is a time range that specifies the start and end time of collecting power generation data. The power generation data within the historical window is automatically collected through the data acquisition system of the photovoltaic power station (such as a SCADA system) to form a power generation data array sequence. This data includes but is not limited to the power generation of each photovoltaic module or array, power generation time, light intensity, temperature, etc.

[0028] Step S200 , performing iterative data fluctuation analysis on the power generation data array sequence in chronological order to obtain a data fluctuation factor.

[0029] Specifically, the acquired power generation data array sequence is iteratively analyzed in chronological order. Each iteration calculates the fluctuations between adjacent data points (such as rate of change, standard deviation, etc.), thereby obtaining a data fluctuation factor. The data fluctuation factor reflects the degree of fluctuation of the power generation data over time.

[0030] In one possible implementation, iterative data fluctuation analysis is performed on the power generation data array sequence in chronological order to obtain a data fluctuation factor. Step S200 further includes step S210 of extracting a first power generation data array and a second power generation data array from the power generation data array sequence in chronological order. Specifically, two consecutive power generation data arrays are first extracted from the power generation data array sequence in chronological order, and are named the first power generation data array and the second power generation data array, respectively. These two data arrays represent power generation data within two adjacent time periods in the time series.

[0031] Step S220, iteratively correlate the first power generation data array and the second power generation data array to obtain a first interactive correlation matrix and a first interactive correlation power generation data array. Specifically, an iterative correlation analysis is performed on the first power generation data array and the second power generation data array. This analysis can be performed based on correlation, covariance or other statistical metrics to quantify the degree of interactive correlation between the two data arrays. The result of the analysis is an interactive correlation matrix (first interactive correlation matrix), which reflects the strength of correlation between the two data arrays. The lower the data fluctuation, the deeper the degree of interactive correlation between the data. At the same time, based on this correlation analysis, a new power generation data array (first interactive correlation power generation data array) is generated, which integrates the information of the original data arrays and emphasizes the correlation between them.

[0032] Step S230, iteratively correlating the first interactively correlated power generation data array with the third power generation data array extracted in chronological order from the power generation data array sequence to obtain a second interactively correlated matrix and a second interactively correlated power generation data array; Step S240, iteratively correlating the N-2th interactively correlated power generation data array with the Nth power generation data array extracted in chronological order from the power generation data array sequence to obtain an N-1th interactively correlated matrix, where N is the number of power generation data arrays in the power generation data array sequence.

[0033] Specifically, steps S230 to S240 are an iterative extension of step S220. In S230, an iterative correlation analysis is performed on the first interactively correlated power generation data array and the third interactively correlated power generation data array extracted in chronological order, generating a second interactive correlation matrix and a second interactively correlated power generation data array. This process is repeated until the entire sequence of power generation data arrays has been processed. In S240, an iterative correlation is performed on the N-2 interactively correlated power generation data array and the N-th interactively correlated power generation data array to obtain the N-1 interactive correlation matrix. This results in a series of interactive correlation matrices that reflect the degree of correlation between the data in each time period within the entire sequence of power generation data arrays.

[0034] In step S250, an iterative data fluctuation analysis network layer is invoked to perform fluctuation analysis on the first interaction correlation matrix, the second interaction correlation matrix, the (N-2)th interaction correlation matrix, and the (N-1)th interaction correlation matrix to obtain the data fluctuation factor. Specifically, the iterative data fluctuation analysis network layer is invoked to process the sequence of interaction correlation matrices obtained in steps S220 to S240. This network layer is a deep learning model trained to identify fluctuation patterns within the matrices. The iterative data fluctuation analysis network layer analyzes the input interaction correlation matrix and outputs a data fluctuation factor, which quantifies the degree of fluctuation of the entire power generation data array sequence. This implementation method, through iterative correlation analysis, can capture the degree of interaction between data, which reflects the inherent connection and stability between the data. By processing these correlation matrices through the iterative data fluctuation analysis network layer, the degree of fluctuation of the entire data sequence is accurately quantified. This refined quantification allows the size of the prediction window to be adjusted based on the actual data fluctuations, thereby improving the accuracy and stability of the prediction.

[0035] In one possible implementation, step S220 further includes step S221, performing one-to-one data similarity identification on the first power generation data array and the second power generation data array to obtain a first similarity array. Specifically, a similarity calculation method is used to perform one-to-one data similarity identification on the first power generation data array and the second power generation data array, that is, similarity calculation is performed on data points at corresponding positions in the two data arrays. Similarity can be calculated based on various distance metrics (such as Euclidean distance, Manhattan distance, cosine similarity, etc.) to quantify the degree of similarity between corresponding data points in the two data arrays, generating a similarity array, in which each element represents the similarity of data points at corresponding positions in the first power generation data array and the second power generation data array.

[0036] Step S222: Normalize the first similarity array using a normalization function and matrix the processing results to obtain a first interaction correlation matrix. Specifically, after obtaining the similarity array, it is processed using a normalization function to convert the similarity values ​​to a unified scale to facilitate subsequent calculations and analysis. The normalization function can be linear normalization, logarithmic normalization, min-max normalization, etc. The processed results are converted into a matrix form, namely a first interaction correlation matrix, which reflects the degree of interaction between the first power generation data array and the second power generation data array.

[0037] Step S223, using a convolutional network to calculate the second power generation data array and the first interactive correlation matrix to obtain a first interactive correlation power generation data array. Specifically, a convolutional neural network (CNN) is used to process the second power generation data array and the first interactive correlation matrix. CNN is a deep learning model that is applied to a two-dimensional data array. CNN extracts features, learns patterns, and makes predictions through structures such as convolutional layers, pooling layers, and fully connected layers. The second power generation data array and the first interactive correlation matrix are used as inputs to the CNN, and the trained network is used for calculation to output a first interactive correlation power generation data array. This new data array combines the information of the second power generation data array and the correlation information of the first interactive correlation matrix. This implementation method can capture the local similarity between data points through one-to-one data similarity recognition, and the normalization process ensures that the similarity values ​​are on a unified scale, which is convenient for subsequent analysis. By using a convolutional network for calculation, these similarities and correlation information can be extracted and utilized to generate a new data array that incorporates more useful features. This processing not only improves the richness of data representation, but also provides more accurate and comprehensive input data for subsequent power generation forecasts, ultimately helping to improve the accuracy and stability of power generation forecasts.

[0038] In a possible implementation, step S222 further includes step S2221, obtaining a normalization function, wherein the normalization function is:

[0039]

[0040] Among them, SFT[lim(x i ,y i )] is the normalized value corresponding to the i-th first similarity in the first similarity array, e is the base of the natural logarithm, n is the total number of first similarities in the first similarity array, x i is the first power generation data of the first power generation data array, y i is the i-th second power generation data in the second power generation data array, lim(x i ,y i ) is the i-th first similarity in the first similarity array.

[0041] Specifically, the normalization function is used to convert each similarity value in the first similarity array into a uniform scale between 0 and 1, lim(x i ,y i ) is an element in the first similarity array, representing two data points x i and y i The degree of similarity between them.

[0042] Step S300 , determining a power generation prediction window according to the magnitude of the data fluctuation factor, traversing and extracting power generation data of the photovoltaic module array in the power generation prediction window, and obtaining a power generation data array sequence of the prediction window.

[0043] Specifically, the size of the window for the next power generation forecast is dynamically determined based on the magnitude of the data fluctuation factor. A large fluctuation factor indicates unstable power generation data, so a shorter forecast window should be selected. Conversely, a small fluctuation factor indicates stable power generation data, so a longer forecast window can be selected. Within the determined forecast window, power generation data from the PV array is extracted traversally to form a power generation data array sequence for the forecast window.

[0044] Step S400 : performing centralized power generation data screening on the power generation data array sequence of the prediction window to determine the power generation data array of the target prediction window.

[0045] Specifically, data preprocessing technology is used to screen the obtained prediction window power generation data array sequence, remove outliers, missing values ​​and other data points that do not meet the requirements, ensure the quality of data used for prediction, and the screened data form the target prediction window power generation data array.

[0046] In one possible implementation, centralized power generation data screening is performed on the prediction window power generation data array sequence to determine a target prediction window power generation data array. Step S400 further includes step S410 of randomly extracting a first prediction window power generation data array from the prediction window power generation data array sequence and using it as a first screening center. Specifically, a random number generator is used to randomly select a subsequence from the prediction window power generation data array sequence as the initial screening center (the first screening center). The length of the subsequence is consistent with the size of the prediction window.

[0047] Step S420 constructs a first screening neighborhood of the first screening center in the sequence of power generation data arrays within the prediction window according to a preset similarity threshold. Specifically, a similarity metric (e.g., Euclidean distance, cosine similarity, etc.) is used to calculate the similarity between each data array in the sequence of power generation data arrays within the prediction window and the first screening center. Data arrays with similarities greater than or equal to the preset threshold are selected according to a preset similarity threshold (a similarity criterion for determining whether a data array belongs to the screening neighborhood) to form the first screening neighborhood.

[0048] Step S430: Iteratively update the first screening center in the first screening neighborhood using a screening center iteration function to obtain an iterative screening center. Specifically, the first screening center is updated in the first screening neighborhood using the screening center iteration function (a function for updating the screening center). That is, a statistic of the data array in the first screening neighborhood is calculated according to the definition of the iteration function (such as a weighted average or median), and the data array corresponding to the statistic is used as the new iterative screening center.

[0049] In step S440, an iterative screening neighborhood of the iterative screening center is constructed in the power generation data array sequence of the prediction window again according to the preset similarity threshold. In step S450, the iterative screening center is iteratively updated in the iterative screening neighborhood using the screening center iteration function again to obtain an updated iterative screening center.

[0050] Specifically, steps S420 and S430 are repeated, but performed on the new iterative filtering center and its iterative filtering neighborhood, that is, the iterative filtering center obtained in the previous iteration is used as the new filtering center, the similarity is recalculated and a new filtering neighborhood is constructed, and then the iterative function is used again to update the filtering center.

[0051] Step S460, and so on, until the similarity of the two iterative screening centers obtained in two adjacent iterations is greater than or equal to the preset iterative similarity, the iteration is stopped, and the prediction window power generation data array corresponding to the iterative screening center obtained in the last iteration is used as the target prediction window power generation data array. Specifically, after each iteration, the similarity of the iterative screening centers of two adjacent iterations is calculated, and it is checked whether the condition for stopping the iteration is met. When the similarity of the iterative screening centers obtained in two adjacent iterations is greater than or equal to the preset iterative similarity (the similarity standard for determining when to stop the iteration), the iteration is stopped, and the data array corresponding to the iterative screening center obtained in the last iteration is used as the target prediction window power generation data array. This implementation method randomly initializes the screening center, constructs a screening neighborhood according to the similarity threshold in each iteration, and then uses the iterative function to update the screening center, so as to gradually approach a more representative and stable data array as the target prediction window power generation data array. This method reduces the impact of noise and outliers on the prediction results and improves the accuracy and robustness of the prediction.

[0052] In a possible implementation, step S430 further includes step S431, and the screening center iteration function is:

[0053]

[0054] Among them, Point(z) is the iterative screening center, N(z) is the first screening neighborhood, z q is the power generation data array of the qth prediction window in the first screening neighborhood, z is the first screening center, K(z q -z) is a weight kernel function built based on the Gaussian function.

[0055] Specifically, the weight kernel function constructed based on the Gaussian function calculates the weight according to a certain distance (such as Euclidean distance) between the data array and the first screening center. The larger the weight, the more similar the data array is to the first screening center, and the greater the contribution to updating the screening center. For each data array in the first screening neighborhood, its distance from the first screening center is calculated, and the corresponding weight is calculated using the weight kernel function. The product of each data array in the first screening neighborhood and its corresponding weight is summed to obtain a weighted sum. The weighted sum is divided by the sum of the weights (i.e., normalized) to obtain a new screening center. This new screening center is the iterative screening center. This implementation method uses a weight kernel function constructed based on a Gaussian function to update the screening center. The Gaussian function is smooth and can reduce the impact of noise and outliers on the screening center, making the updated screening center more stable, ensuring that the selection of the screening center is gradually optimized during the iterative process, thereby obtaining a more accurate and stable target prediction window power generation data array, thereby improving the accuracy and reliability of the photovoltaic power station power generation forecast.

[0056] Step S500: calling the power generation prediction network layer to identify the power generation data array of the target prediction window to obtain the power generation prediction result.

[0057] Specifically, using machine learning or deep learning technology, first construct a power generation prediction network layer (such as a neural network, support vector machine, etc.), which has been trained and can recognize power generation data and predict power generation. Then, the target prediction window power generation data array obtained is input into the network layer for recognition, and the power generation prediction result is output. The embodiment of the present application adopts the method of collecting the power generation data of the photovoltaic component array in the target photovoltaic power station within a preset historical window, obtaining a power generation data array sequence, iteratively analyzing the power generation data array sequence in chronological order, obtaining a data fluctuation factor, determining the power generation prediction window according to the size of the data fluctuation factor, traversing and extracting the power generation data of the photovoltaic component array in the power generation prediction window, obtaining a prediction window power generation data array sequence, performing centralized power generation data screening on the prediction window power generation data array sequence, determining the target prediction window power generation data array, calling the power generation prediction network layer to identify the target prediction window power generation data array, obtaining the power generation prediction result and other technical means, thereby achieving the technical effect of improving the power generation prediction accuracy of the photovoltaic power station.

[0058] In one possible implementation, the power generation prediction network layer is called to identify the target prediction window power generation data array to obtain a power generation prediction result. Step S500 further includes step S510, which obtains multiple sample prediction window power generation data arrays and multiple sample power generation prediction results as training data. Specifically, power generation data is extracted from historical power generation data according to the power generation prediction window to form multiple sample prediction window power generation data arrays. Each sample prediction window power generation data array contains the power generation data of the photovoltaic module array within a specific time period. The sample power generation prediction result is the actual power generation data corresponding to the sample prediction window power generation data array. The sample prediction window power generation data array and the corresponding sample power generation prediction result are used to train the power generation prediction network layer.

[0059] Step S520, supervise the training of the framework built based on the convolutional neural network using the training data, learn the one-to-one mapping relationship between the prediction window power generation data array and the sample power generation prediction results, until the training converges, and obtain the trained power generation prediction network layer. Specifically, a CNN model is designed, which includes an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer. The input layer receives the sample prediction window power generation data array, and the output layer outputs the predicted power generation. The loss function mean square error (MSE) is used to measure the difference between the model prediction results and the actual sample power generation prediction results, and stochastic gradient descent SGD is used to minimize the loss function, thereby updating the weight parameters of the CNN model. The training data is input into the CNN model in batches, the prediction results are calculated by forward propagation, and the gradient is calculated and the model weights are updated by back propagation. Repeat this process until the loss function converges or the preset number of training rounds is reached. This implementation uses a supervised training method based on CNN. CNN has powerful feature extraction capabilities and can automatically learn useful features from power generation data. Through supervised training with a large amount of sample data, the model can learn the complex mapping relationship between power generation data and power generation, thereby making accurate predictions on new data.

[0060] In the above, refer to Figure 1 The photovoltaic power generation prediction method combined with the dynamic update mechanism according to the embodiment of the present invention is described in detail. Figure 2 A photovoltaic power generation prediction system incorporating a dynamic update mechanism according to an embodiment of the present invention is described.

[0061] A photovoltaic power station power generation forecasting system incorporating a dynamic update mechanism according to an embodiment of the present invention is designed to address the low prediction accuracy issues of existing technologies, thereby improving the accuracy of photovoltaic power generation forecasts. The photovoltaic power station power generation forecasting system incorporating a dynamic update mechanism includes a historical power generation data acquisition module 10, a data fluctuation analysis module 20, a predicted power generation data extraction module 30, a centralized screening module 40, and a power generation forecasting module 50.

[0062] The historical power generation data acquisition module 10 is used to collect the power generation data of the photovoltaic component array in the target photovoltaic power station within a preset historical window to obtain a power generation data array sequence; the data fluctuation analysis module 20 is used to perform iterative data fluctuation analysis on the power generation data array sequence in chronological order to obtain a data fluctuation factor; the predicted power generation data extraction module 30 is used to determine the power generation prediction window according to the size of the data fluctuation factor, traverse and extract the power generation data of the photovoltaic component array in the power generation prediction window to obtain a prediction window power generation data array sequence; the centralized screening module 40 is used to perform centralized power generation data screening on the prediction window power generation data array sequence to determine the target prediction window power generation data array; the power generation prediction module 50 is used to call the power generation prediction network layer to identify the target prediction window power generation data array to obtain a power generation prediction result.

[0063] Next, the specific configuration of the data fluctuation analysis module 20 will be described in detail. As described above, the data fluctuation analysis is performed iteratively on the power generation data array sequence in chronological order to obtain a data fluctuation factor. The data fluctuation analysis module 20 may further include: a power generation data array extraction unit for extracting a first power generation data array and a second power generation data array from the power generation data array sequence in chronological order; an iterative correlation unit for iteratively correlating the first power generation data array and the second power generation data array to obtain a first interaction correlation matrix and a first interaction correlation power generation data array; iteratively correlating the first interaction correlation power generation data array with a third power generation data array extracted in chronological order from the power generation data array sequence to obtain a second interaction correlation matrix and a second interaction correlation power generation data array; iteratively correlating the (N-2)th interaction correlation power generation data array with the (N)th power generation data array extracted in chronological order from the power generation data array sequence to obtain an (N-1)th interaction correlation matrix, where N is the number of power generation data arrays in the power generation data array sequence; and a fluctuation analysis unit for calling an iterative data fluctuation analysis network layer to perform fluctuation analysis on the first interaction correlation matrix, the second interaction correlation matrix, the (N-2)th interaction correlation matrix, and the (N-1)th interaction correlation matrix to obtain the data fluctuation factor.

[0064] Among them, the iterative association unit may further include: a data similarity identification subunit for performing one-to-one data similarity identification on the first power generation data array and the second power generation data array to obtain a first similarity array; a normalization processing subunit for performing normalization processing on the first similarity array using a normalization function, and matrixing the processing result to obtain a first interactive correlation matrix; a calculation subunit for using a convolutional network to calculate the second power generation data array and the first interactive correlation matrix to obtain a first interactively correlated power generation data array.

[0065] The normalization processing subunit may further include: a normalization function acquisition component for acquiring a normalization function, wherein the normalization function is:

[0066]

[0067] Among them, SFT[lim(x i ,y i )] is the normalized value corresponding to the i-th first similarity in the first similarity array, e is the base of the natural logarithm, n is the total number of first similarities in the first similarity array, x i is the first power generation data of the first power generation data array, y i is the i-th second power generation data in the second power generation data array, lim(x i ,y i ) is the i-th first similarity in the first similarity array.

[0068] The specific configuration of the centralized screening module 40 will be described in detail below. As described above, centralized screening of the prediction window power generation data array sequence is performed to determine a target prediction window power generation data array. The centralized screening module 40 may further include: a first screening center acquisition unit for randomly extracting a first prediction window power generation data array from the prediction window power generation data array sequence and using it as a first screening center; a first screening neighborhood construction unit for constructing a first screening neighborhood of the first screening center in the prediction window power generation data array sequence according to a preset similarity threshold; and a screening center iteration unit for iteratively updating the first screening center in the first screening neighborhood using a screening center iteration function to obtain an iterative screening center, then again constructing an iterative screening neighborhood of the iterative screening center in the prediction window power generation data array sequence according to the preset similarity threshold, and again iteratively updating the iterative screening center in the iterative screening neighborhood using the screening center iteration function to obtain an updated iterative screening center, and so on until the similarity between two iterative screening centers obtained in two adjacent iterations is greater than or equal to a preset iterative similarity, at which point the iteration is terminated, and the prediction window power generation data array corresponding to the iterative screening center obtained in the last iteration is used as the target prediction window power generation data array.

[0069] The screening center iteration unit may further include: a screening center iteration function acquisition subunit for acquiring a screening center iteration function, wherein the screening center iteration function is:

[0070]

[0071] Among them, Point(z) is the iterative screening center, N(z) is the first screening neighborhood, z qis the power generation data array of the qth prediction window in the first screening neighborhood, z is the first screening center, K(z q -z) is a weight kernel function built based on the Gaussian function.

[0072] The specific configuration of the power generation prediction module 50 will be described in detail below. As described above, the power generation prediction network layer is called to identify the target prediction window power generation data array to obtain a power generation prediction result. The power generation prediction module 50 may further include: a training data acquisition unit for acquiring multiple sample prediction window power generation data arrays and multiple sample power generation prediction results as training data; and a power generation prediction network layer acquisition unit for using the training data to perform supervised training on a framework constructed based on a convolutional neural network, learning a one-to-one mapping relationship between the prediction window power generation data array and the sample power generation prediction results until the training converges, thereby obtaining the trained power generation prediction network layer.

[0073] The photovoltaic power station power generation prediction system combined with a dynamic update mechanism provided in an embodiment of the present invention can execute the photovoltaic power station power generation prediction method combined with a dynamic update mechanism provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0074] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other and are not used to limit the scope of protection of the present invention.

[0075] The above specific embodiments do not constitute a limitation to the scope of protection of this application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of this application should be included in the scope of protection of this application. In some cases, the actions or steps recorded in this application can be performed in an order different from that in the embodiments and can still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A photovoltaic power station power generation prediction method combined with a dynamic update mechanism is characterized by: The method comprises: Collecting power generation data of the photovoltaic module array in the target photovoltaic power station within a preset historical window to obtain a power generation data array sequence; Performing iterative data fluctuation analysis on the power generation data array sequence in chronological order to obtain a data fluctuation factor; Determining a power generation prediction window according to the magnitude of the data fluctuation factor, traversing and extracting power generation data of the photovoltaic assembly array within the power generation prediction window to obtain a power generation data array sequence within the prediction window; Performing centralized power generation data screening on the prediction window power generation data array sequence to determine a target prediction window power generation data array; Calling the power generation prediction network layer to identify the power generation data array of the target prediction window to obtain a power generation prediction result; The iterative data fluctuation analysis is performed on the power generation data array sequence in chronological order to obtain the data fluctuation factor, including: Extracting a first power generation data array and a second power generation data array from the power generation data array sequence in chronological order; Iteratively correlating the first power generation data array and the second power generation data array to obtain a first interactive correlation matrix and a first interactively correlated power generation data array; Iteratively correlating the first interactively correlated power generation data array with a third power generation data array extracted in chronological order from the power generation data array sequence to obtain a second interactively correlated matrix and a second interactively correlated power generation data array; Iteratively correlating the N-2th interactively correlated power generation data array with the Nth power generation data array extracted in chronological order from the power generation data array sequence to obtain an N-1th interactively correlated matrix, where N is the number of power generation data arrays in the power generation data array sequence; The iterative data fluctuation analysis network layer is called to perform fluctuation analysis on the first interaction correlation matrix, the second interaction correlation matrix, the N-2th interaction correlation matrix and the N-1th interaction correlation matrix to obtain the data fluctuation factor.

2. The photovoltaic power generation prediction method combined with a dynamic update mechanism according to claim 1, characterized in that: include: performing one-to-one data similarity identification on the first power generation data array and the second power generation data array to obtain a first similarity array; Normalizing the first similarity array using a normalization function, and matrixing the processing result to obtain a first interaction correlation matrix; The second power generation data array and the first interactive correlation matrix are calculated using a convolutional network to obtain a first interactive correlation power generation data array.

3. The photovoltaic power generation prediction method combined with a dynamic update mechanism according to claim 2, characterized in that: include: Obtain a normalization function, wherein the normalization function is: ； in, is the normalized value corresponding to the i-th first similarity in the first similarity array, is the base of natural logarithms, is the total number of first similarities in the first similarity array, is the i-th first power generation data in the first power generation data array, is the i-th second power generation data in the second power generation data array, is the i-th first similarity in the first similarity array.

4. The photovoltaic power generation prediction method combined with a dynamic update mechanism according to claim 1, characterized in that: Centrally screening the power generation data of the prediction window power generation data array sequence to determine the target prediction window power generation data array includes: Randomly extracting a first prediction window power generation data array from the prediction window power generation data array sequence and using it as a first screening center; constructing a first screening neighborhood of the first screening center in the prediction window power generation data array sequence according to a preset similarity threshold; Iteratively updating the first screening center in the first screening neighborhood using a screening center iteration function to obtain an iterative screening center; again constructing an iterative screening neighborhood of the iterative screening center in the prediction window power generation data array sequence according to the preset similarity threshold; Iteratively updating the iterative screening center in the iterative screening neighborhood using the screening center iteration function again to obtain an updated iterative screening center; The same process is repeated until the similarity between the two iterative screening centers obtained in two adjacent iterations is greater than or equal to the preset iterative similarity, and the iteration is stopped. The prediction window power generation data array corresponding to the iterative screening center obtained in the last iteration is used as the target prediction window power generation data array.

5. The photovoltaic power generation prediction method combined with a dynamic update mechanism according to claim 4, characterized in that: The screening center iteration function is: ； in, For iterative screening center, is the first screening neighborhood, is the power generation data array of the qth prediction window in the first screening neighborhood, For the first screening center, is a weight kernel function built based on the Gaussian function.

6. The photovoltaic power generation prediction method combined with a dynamic update mechanism according to claim 1, characterized in that: The power generation prediction network layer is called to identify the power generation data array of the target prediction window to obtain the power generation prediction result, including: Acquire multiple sample prediction window power generation data arrays and multiple sample power generation prediction results as training data; The framework based on the convolutional neural network is supervised and trained using the training data to learn the one-to-one mapping relationship between the power generation data array of the prediction window and the sample power generation prediction results until the training converges, thereby obtaining the trained power generation prediction network layer.

7. The photovoltaic power generation prediction system combined with the dynamic update mechanism is characterized by: The system is used to implement the photovoltaic power station power generation prediction method combined with a dynamic update mechanism according to any one of claims 1 to 6, and the system includes: A historical power generation data acquisition module is used to collect power generation data of the photovoltaic module array in the target photovoltaic power station within a preset historical window to obtain a power generation data array sequence; a data fluctuation analysis module, configured to perform iterative data fluctuation analysis on the power generation data array sequence in chronological order to obtain a data fluctuation factor; A predicted power generation data extraction module is used to determine a power generation prediction window according to the size of the data fluctuation factor, traverse and extract the power generation data of the photovoltaic module array in the power generation prediction window, and obtain a power generation data array sequence of the prediction window; A centralized screening module, configured to perform centralized power generation data screening on the prediction window power generation data array sequence to determine a target prediction window power generation data array; The power generation prediction module is used to call the power generation prediction network layer to identify the power generation data array of the target prediction window to obtain the power generation prediction result.

Images