A distributed photovoltaic prediction method based on long short-term memory neural network

Abstract

The application relates to a distributed photovoltaic prediction method and system based on a long short-term memory neural network. The spatial distribution and historical records of regional photovoltaic resources are collected through a photovoltaic sensor array, and a 2-D CNN-BLSTM deep learning network model is built by combining a deep convolution network technology and a deep recurrent network technology to predict the regional light conditions. The method improves the accuracy of photovoltaic prediction, can provide reliable information for new energy power generation scheduling and planning, and greatly enhances the safety and stability of the power grid.

Description

Technical Field

[0001] This application relates to the field of distributed generation, and more specifically, to a distributed photovoltaic prediction method and system based on convolutional and long short-term memory neural networks. Background Technology

[0002] Currently, photovoltaic (PV) power generation systems are widely used in power systems. PV power generation systems can be categorized into distributed and centralized types based on their distribution. Both centralized and distributed PV power generation systems exist depending on the application, but distributed systems, with their advantages of being economical, efficient, and small-scale, have a broader application prospect and have become an important form of PV power generation.

[0003] However, the power generation of photovoltaic (PV) systems is closely related to various environmental factors, such as climate and geographical location. Affected by external conditions, solar irradiance fluctuates constantly, resulting in discontinuous power generation and an inability to guarantee stable power supply. Therefore, research on PV power output prediction technology is crucial. For any type of PV power prediction, prediction research can be conducted using different methods, such as physical methods, statistical methods, and machine learning methods.

[0004] The physical method involves building a physical model using environmental information obtained from numerical weather prediction (NWP) and then directly calculating photovoltaic power. However, because the parameters provided by photovoltaic manufacturers often have limited accuracy, and empirical parameters vary between different regions, the physical model contains certain errors, exhibits poor local interference resistance, and weak robustness.

[0005] In addition, statistical forecasting methods typically employ time series analysis, regression analysis, grey theory, fuzzy theory, and spatiotemporal correlation methods. These methods analyze processed historical data on solar radiation and photovoltaic power generation to establish data models and predict photovoltaic power generation. They offer advantages such as simple modeling and strong applicability across different regions. However, implementing statistical methods faces challenges such as collecting large amounts of data and complex computational processing, making it difficult to meet the speed requirements of ultra-short-term forecasting.

[0006] Machine learning can effectively extract high-dimensional, complex, and nonlinear features and directly map them to the output. Machine learning-based prediction methods leverage this and have become one of the most commonly used methods for predicting time series data. Common deep neural networks (DNNs) mainly include convolutional neural networks (CNNs), deep belief networks (DBNs), stacked denoised autoencoders (SDAEs), and long short-term memory (LSTMs). Deep learning-based prediction models can consider the temporal and nonlinear characteristics of time series data and can find complex data correlations from large amounts of data, exhibiting better performance and robustness. Among them, hybrid models show even more satisfactory performance. Furthermore, the bidirectional long short-term memory network BiLSTM, an extension of the traditional unidirectional LSTM network, can learn bidirectional temporal features, further improving the accuracy of model predictions. Background Technical Literature: SIAMI-NAMINI S, TAVAKOLI N, NAMIN A. A comparative analysis of forecasting financial time series using ARIMA, LSTM, and BILSTM demonstrates that the BiLSTM model has superior predictive power compared to the single-direction LSTM model in time series analysis and can be used for power load forecasting. Background Technical Literature: Research on Power Load Forecasting Method Based on CNN-BiLSTM Algorithm, Wang Yuxin, Xi'an University of Technology. This paper proposes a combined CNN-BiLSTM model for power load forecasting, which shows superiority over single-structure LSTM models and CNN-LSTM combined models.

[0007] However, the most critical factor affecting the power generation of photovoltaic power generation systems, namely solar irradiance information, not only has very significant spatial characteristics but also significant temporal correlations. Existing algorithms typically struggle to comprehensively analyze historical irradiance information from multiple dimensions, including time and space, leading to inaccurate predictions.

[0008] To address the aforementioned issues, a new method and system for distributed photovoltaic forecasting is urgently needed. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention provides a distributed photovoltaic prediction method and system based on convolutional and long short-term memory neural networks. It generates a continuous time series of a 2D coefficient matrix from irradiance data, and uses this time series as input. A CNN-BLTSM model is then employed to simultaneously acquire the spatial and temporal features of the irradiance data, thereby achieving accurate prediction of the irradiance at the next moment.

[0010] The present invention adopts the following technical solution.

[0011] The first aspect of this invention relates to a distributed photovoltaic (PV) prediction method based on a long short-term memory (LSTM) neural network. The method includes the following steps: Step 1, collecting irradiance data of the location of the distributed PV system, and generating a continuous time series of a 2D sparse matrix based on the spatial and temporal dimension parameters of the irradiance data; Step 2, using a convolutional neural network to extract spatial features from the 2D sparse matrix, and simultaneously extracting single-point time series from the continuous time series of the 2D sparse matrix, and using an LSM neural network to extract temporal features from the single-point time series; Step 3, merging the temporal and spatial features to predict the irradiance at the location of the distributed PV system at the next moment, thereby achieving the prediction of distributed PV power generation.

[0012] Preferably, the irradiance data is collected based on an optical sensor array within the area of ​​the distributed photovoltaic system; and each irradiance data corresponds to the spatial location of an optical sensor and the irradiance collection time.

[0013] Preferably, based on the spatial location of the optical sensor, the irradiance data collected at the same time are used to form an original matrix; the spatial correlation between each irradiance data in the original matrix is ​​solved, and the original matrix is ​​simplified based on the spatial correlation to obtain a 2D sparse matrix.

[0014] Preferably, a CNN-BLSTM model is used to predict the irradiance at the next time step; and the CNN-BLSTM model includes a spatial feature extraction branch, a temporal feature extraction branch, and a prediction branch; wherein, the spatial feature extraction branch includes a multi-layer convolutional neural network; the temporal feature extraction branch includes a multi-layer long short-term memory neural network, a first fully connected layer, and a second fully connected layer; the prediction branch includes a feature connection layer, a third fully connected layer, and an output layer.

[0015] Preferably, the feature connection layer arranges the extracted spatial features into a one-dimensional array and then merges them with the temporal features.

[0016] Preferably, before merging the temporal and spatial features, the temporal features are pre-adjusted; and the pre-adjustment includes adjusting the time interval density and time span of the time series in the temporal features; at the same time, the size of the output layer convolution kernel in the prediction branch is adjusted.

[0017] Preferably, before merging the temporal and spatial features, the spatial features are pre-adjusted; and the pre-adjustment includes adjusting the weight of the spatial nodes according to the power generation contribution rate of the photovoltaic devices in the spatial nodes.

[0018] A second aspect of this invention relates to a distributed photovoltaic (PV) prediction system based on a long short-term memory (LSTM) neural network. The system includes an acquisition module, an extraction module, and a prediction module. The acquisition module acquires irradiance data of the location of the distributed PV system and generates a continuous time series of a 2D sparse matrix based on the spatial and temporal parameters of the irradiance data. The extraction module uses a convolutional neural network to extract spatial features from the 2D sparse matrix, extracts single-point time series from the continuous time series of the 2D sparse matrix, and uses an LSM to extract temporal features from the single-point time series. The prediction module combines the temporal and spatial features to predict the irradiance at the location of the distributed PV system at the next moment, thereby predicting the distributed PV power generation.

[0019] A third aspect of the present invention relates to a terminal, including a processor and a storage medium; the storage medium is used to store instructions; the processor is used to perform operations according to the instructions to execute the steps of the method in the first aspect of the present invention.

[0020] A fourth aspect of the present invention relates to a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the method of the first aspect of the present invention.

[0021] The beneficial effects of this invention are that, compared with the prior art, the distributed photovoltaic prediction method and system based on convolutional and long short-term memory neural networks in this invention generates a continuous time series of 2D coefficient matrices from irradiance data, and uses this time series as input. A CNN-BLTSM model is then employed to simultaneously acquire the spatial and temporal features of the irradiance data, thereby achieving accurate prediction of the irradiance at the next moment. This invention is effective and reliable; the comprehensive collection and combination of the temporal and spatial features of the irradiance data ensures the model's prediction efficiency and accuracy. Attached Figure Description

[0022] Figure 1 This is a schematic diagram illustrating the steps of a distributed photovoltaic prediction method based on a long short-term memory neural network according to the present invention.

[0023] Figure 2This is a schematic diagram illustrating the acquisition of irradiance data in a distributed photovoltaic prediction method based on a long short-term memory neural network according to the present invention.

[0024] Figure 3 This is a model block diagram of the CNN-BLSTM model in a distributed photovoltaic prediction method based on a long short-term memory neural network according to the present invention.

[0025] Figure 4 This is a schematic diagram of the radiance prediction curves for 17 photovoltaic deployment points in one embodiment of a distributed photovoltaic prediction method based on a long short-term memory neural network according to the present invention.

[0026] Figure 5 This is a schematic diagram illustrating the evaluation of radiance prediction results for 17 photovoltaic deployment sites in one embodiment of a distributed photovoltaic prediction method based on a long short-term memory neural network according to the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this invention are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments not described in this invention obtained by those skilled in the art based on the embodiments described in this invention without creative effort should fall within the protection scope of this invention.

[0028] Figure 1 This is a schematic diagram illustrating the steps of a distributed photovoltaic prediction method based on a long short-term memory neural network according to the present invention. Figure 1 As shown, the first aspect of the present invention relates to a distributed photovoltaic prediction method based on a long short-term memory neural network, the method comprising steps 1 to 3.

[0029] Step 1: Collect irradiance data of the location of the distributed photovoltaic system, and generate a continuous time series of 2D sparse matrix based on the spatial and temporal parameters of the irradiance data.

[0030] Understandably, in order to achieve real-time data acquisition of the power generation status of distributed photovoltaic systems, this invention primarily considers solar irradiance data, which affects the power generation capacity of distributed photovoltaic systems. Therefore, this invention can employ optical sensors to reasonably acquire the relevant irradiance data.

[0031] Preferably, the irradiance data is collected based on an optical sensor array within the area of ​​the distributed photovoltaic system; and each irradiance data corresponds to the spatial location of an optical sensor and the irradiance collection time.

[0032] Specifically, the irradiance data read by general photovoltaic prediction models is from a single point. When the geographical environment of distributed generation is more complex, the single-point photovoltaic prediction model cannot accurately predict photovoltaic fluctuations due to the lack of geospatial information. To solve this problem, we combine multiple sensors into a sensor array and centrally organize the data into matrix-style irradiance information.

[0033] Figure 2 This is a schematic diagram illustrating the acquisition of irradiance data in a distributed photovoltaic prediction method based on a long short-term memory neural network according to the present invention. Figure 2 As shown, the irradiance sensor array in this invention can be deployed within a specific area of ​​the photovoltaic system. The sensor array does not need to be distributed at equal intervals according to latitude and longitude; instead, it can be deployed at suitable locations based on the geographical environment information of the photovoltaic system's area. For example, multiple nearby areas with similar geographical environment information can reduce the number of sensors deployed. Conversely, areas with drastically changing geographical environment information can increase the number of sensors deployed. The deployment principle is to comprehensively collect irradiance information from all areas within the preset region.

[0034] By deploying irradiance sensors, this invention can subdivide specific locations within a preset area and generate two-dimensional irradiance data through this subdivision. It should be noted that since the sensors are not deployed entirely according to geographical location factors, while this method saves deployment costs, it increases the computational complexity. Therefore, this invention considers processing the data collected by the sensors to transform it into irradiance information for a uniform geographical location.

[0035] Since most similar locations do not have sensors, the irradiance data at these locations can be 0, so the resulting matrix can be a 2D sparse matrix.

[0036] Preferably, based on the spatial location of the optical sensor, the irradiance data collected at the same time are used to form an original matrix; the spatial correlation between each irradiance data in the original matrix is ​​solved, and the original matrix is ​​simplified based on the spatial correlation to obtain a 2D sparse matrix.

[0037] In another embodiment of the present invention, the original matrix can be simplified to obtain a 2D sparse matrix based on the actual spatial correlation between multiple irradiance data.

[0038] Specifically, the original matrix here is the unprocessed matrix containing the raw collected irradiance data. Since the locations of the sensors naturally exhibit spatial correlation—for example, two adjacent sensors in similar physical environments possess nearly identical physical characteristics, such as similar reflection and scattering of sunlight—this invention can calculate the correlation between the irradiance data collected by the two sensors. For instance, algorithms such as the Pearson correlation coefficient can be used to solve for the correlation between the irradiance data, and the original matrix can be simplified based on this correlation. This simplification can involve normalizing the data content in the matrix or setting some data values ​​to zero, thereby simplifying the original matrix into a sparse matrix to facilitate subsequent model calculations.

[0039] Step 2: Use a convolutional neural network to extract spatial features from the 2D sparse matrix, extract single-point time series from the continuous time series of the 2D sparse matrix, and use a long short-term memory neural network to extract time features from the single-point time series.

[0040] In the method of the present invention, the prediction model is improved by taking into account both spatial and temporal information contained in the irradiance data.

[0041] In existing technologies, CNN-BiLSTM models have been used to predict the power generation of distributed photovoltaic systems. This invention can simultaneously employ CNN and BiLSTM models to solve for different dimensions of feature information in the irradiance data of the distributed photovoltaic system, and then, through feature connections in the connection layer, ultimately achieve irradiance prediction.

[0042] Figure 3 This is a model block diagram of the CNN-BLSTM model in a distributed photovoltaic prediction method based on a long short-term memory neural network according to the present invention. Figure 3 As shown, preferably, a CNN-BLSTM model is used to predict the irradiance at the next time step; and the CNN-BLSTM model includes a spatial feature extraction branch, a temporal feature extraction branch, and a prediction branch; wherein, the spatial feature extraction branch includes a multi-layer convolutional neural network; the temporal feature extraction branch includes a multi-layer long short-term memory neural network, a first fully connected layer, and a second fully connected layer; the prediction branch includes a feature connection layer, a third fully connected layer, and an output layer.

[0043] It is understandable that, such as Figure 3As shown, the method in this invention can input a matrix sequence as an image into a multi-layer convolutional neural network. By selecting parameters such as the size of the convolution window, similar features between various irradiance data in the matrix can be extracted. Since the features extracted in this way only consider the differences in irradiance collected by the sensor at different locations in a matrix, they can also be called spatial features.

[0044] Alternatively, another approach of this invention can employ a BiLSTM model to achieve accurate prediction of time series. Each time series here can be extracted from the time series of a 2D sparse matrix. Specifically, data from the same row and column of each sparse matrix can be collected and arranged chronologically to form a time series, thus allowing the use of a BiLSTM model for solving the problem. The BiLSTM model is the bidirectional LSTM model mentioned in the background section of this application; since it is already described in the prior art, it will not be elaborated upon here.

[0045] After extracting the temporal features from the time series using a multi-layer BiLSTM model, this invention connects the features from multiple time series using a first fully connected layer and a second fully connected layer.

[0046] Preferably, the feature connection layer arranges the extracted spatial features into a one-dimensional array and then merges them with the temporal features.

[0047] Understandably, after obtaining spatial and temporal features through the two branches mentioned above, this invention employs a feature connection layer to achieve a reasonable concatenation of the spatial and temporal features. This concatenation arranges all spatial features into a one-dimensional array, and then incorporates temporal information into each data point in the array. Following further processing by a fully connected layer and an output layer, this invention can predict the irradiance information for the next moment based on the input data for the predicted time.

[0048] In another embodiment of the present invention, the max pooling method can be used to extract spatial features in the CNN model.

[0049] Step 3: Combine the temporal and spatial features to predict the irradiance at the location of the distributed photovoltaic system at the next moment, thereby enabling the prediction of distributed photovoltaic power generation.

[0050] Specifically, in this invention, spatial feature values ​​are expanded into one-dimensional data and combined with temporal feature values. After processing by a fully connected layer, the final predicted value is obtained. This model can simultaneously process both spatial and temporal information of irradiance, and its efficiency and accuracy are far superior to traditional models that can only use temporal information.

[0051] Preferably, before merging the temporal and spatial features, the temporal features are pre-adjusted; and the pre-adjustment includes adjusting the time interval density and time span of the time series in the temporal features.

[0052] Understandably, to achieve a reasonable merging of temporal and spatial features, this invention can pre-adjust the temporal features generated by the temporal feature extraction branch to a certain extent before merging them with the spatial features. The pre-adjustment methods supported in this invention can include extracting a subset of features from all feature information in the time series at certain time intervals. Alternatively, it can involve collecting all feature data for a single time span in the time series. Therefore, this approach alters the time interval density and span to some extent, thus affecting the prediction results. Therefore, this invention can adjust the prediction results according to actual prediction needs, such as the time interval between the predicted period and the input data, and the temporal density of the data generated for the predicted period, to ensure the effectiveness and usability of the results.

[0053] In addition, the present invention can also adjust the size of the output layer convolution kernel in the prediction branch to achieve effective output of the prediction results.

[0054] Preferably, before merging the temporal and spatial features, the spatial features are pre-adjusted; and the pre-adjustment includes adjusting the weight of the spatial nodes according to the power generation contribution rate of the photovoltaic devices in the spatial nodes.

[0055] In the method of this invention, spatial features also need to be pre-adjusted before merging. This adjustment allows the prediction results of each irradiance to be directly calculated into the final predicted power generation result. Specifically, the method can weight the spatial irradiance features of each spatial node, and the weights can be multiplied according to the power generation contribution rate of each node. For example, at a certain node, although the spatial irradiance value is low, if the capacity of the photovoltaic equipment at that node is large, the scale of light energy to electrical energy conversion that it can achieve per unit irradiance is larger. This scale can be used as a weight to pre-calculate the actual power generation. After calculation, the adjusted spatial features are merged with the time series, and then the predicted power generation result is directly output.

[0056] Figure 4 This is a schematic diagram of radiance prediction curves for 17 photovoltaic deployment points in one embodiment of a distributed photovoltaic prediction method based on a long short-term memory neural network according to the present invention. Figure 4 As shown, the irradiance data collected by the optical sensor disclosed in the Measurement and Instrumentation Data Center of this invention was used as the data source to simulate a photovoltaic system with 17 photovoltaic device deployment points.

[0057] exist Figure 4 In this study, a two-dimensional spatiotemporal solar irradiance sequence of the location of distributed photovoltaic power plants is used as input to predict the changes at the next moment. The illustration shows the model's prediction of irradiance changes over a whole day, demonstrating that the proposed hybrid model can accurately track peaks even under rapidly changing irradiance conditions.

[0058] Figure 5 This is a schematic diagram illustrating the evaluation of radiance prediction results for 17 photovoltaic deployment sites in one embodiment of a distributed photovoltaic prediction method based on a long short-term memory neural network according to the present invention. Figure 5 As shown, a regression analysis was performed on the output power (i.e., the irradiance in this invention, which can be calculated based on irradiance data) response of a distributed photovoltaic system with 17 photovoltaic deployment sites. Figure 5 As can be seen, the predicted values ​​are randomly distributed around the regression line, without any overall overestimation or underestimation, indicating that the prediction model does not have significant inherent errors. Our proposed method has shown acceptable performance in 17 photovoltaic subsystems, helping us to obtain more accurate prediction results.

[0059] It is understood that, in order to implement the various functions in the methods provided in the embodiments of this application, the terminal includes corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, based on the algorithm steps of the examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0060] This application embodiment can divide the terminal into functional modules according to the above method example. For example, each function can be divided into its own functional module, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware or as a software functional module. It should be noted that the module division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0061] The terminal device includes at least one processor, a bus system, and at least one communication interface. The processor may be a central processing unit (CPU), or it may be replaced by a field-programmable gate array (FPGA), an application-specific integrated circuit (ASIC), or other hardware. Alternatively, the FPGA or other hardware may work together with the CPU as a processor.

[0062] The memory can be read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions, random access memory (RAM) or other types of dynamic storage devices capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed discs, laser discs, optical discs, universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited to these. The memory can exist independently and be connected to the processor via a bus. The memory can also be integrated with the processor.

[0063] The hard drive can be a mechanical hard drive or a solid-state drive (SSD), etc. The interface card can be a host bus adapter (HBA), a redundant array of independent disks (RID), an expander card, or a network interface controller (NIC), etc., and this embodiment of the invention is not limited to any particular type. The interface card in the hard drive module communicates with the hard drive. The storage node communicates with the interface card of the hard drive module to access the hard drive in the hard drive module.

[0064] The hard drive interface can be Serial Attached Small Computer System Interface (SAS), Serial Advanced Technology Attachment (SATA), or Peripheral Component Interconnect Express (PCIe), etc.

[0065] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software programs, implementation can be, in whole or in part, in the form of a computer program product. This computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device containing one or more servers, data centers, etc., that can be integrated with the medium. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks (SSDs)).

[0066] A fourth aspect of the present invention relates to a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the method of the first aspect of the present invention.

[0067] The computer program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions. This electronic circuitry can execute the computer-readable program instructions to implement various aspects of the invention.

[0068] The beneficial effects of this invention are that, compared with the prior art, the distributed photovoltaic prediction method and system based on convolutional and long short-term memory neural networks in this invention generates a continuous time series of 2D coefficient matrices from irradiance data, and uses this time series as input. A CNN-BLTSM model is then employed to simultaneously acquire the spatial and temporal features of the irradiance data, thereby achieving accurate prediction of the irradiance at the next moment. This invention is effective and reliable; the comprehensive collection and combination of the temporal and spatial features of the irradiance data ensures the model's prediction efficiency and accuracy.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A distributed photovoltaic prediction method based on a long short-term memory neural network, characterized in that, The method includes the following steps: Step 1: Collect irradiance data of the location of the distributed photovoltaic system, and generate a continuous time series of a 2D sparse matrix based on the spatial dimension parameters and time dimension parameters of the irradiance data; The irradiance data is collected based on an optical sensor array within the area of ​​the distributed photovoltaic system. Each irradiance data corresponds to the spatial location of an optical sensor and the irradiance collection time. Based on the spatial location of the optical sensor, the irradiance data collected at the same time are combined into an original matrix. The spatial correlation between each irradiance data in the original matrix is ​​solved, and the original matrix is ​​simplified based on the spatial correlation to obtain a 2D sparse matrix. Step 2: Use a convolutional neural network to extract spatial features from the 2D sparse matrix, extract single-point time series from the continuous time series of the 2D sparse matrix, and use a long short-term memory neural network to extract time features from the single-point time series. Step 3: Combine the temporal and spatial features to predict the irradiance at the location of the distributed photovoltaic system at the next moment, thereby achieving the prediction of distributed photovoltaic power generation. The CNN-BiLSTM model is used to predict the irradiance at the next time step. The CNN-BiLSTM model includes a spatial feature extraction branch, a temporal feature extraction branch, and a prediction branch. The spatial feature extraction branch includes a multi-layer convolutional neural network. The temporal feature extraction branch includes a multi-layer long short-term memory neural network, a first fully connected layer, and a second fully connected layer. The prediction branch includes a feature connection layer, a third fully connected layer, and an output layer. The feature connection layer arranges the extracted spatial features into a one-dimensional array and then merges it with the temporal features.

2. The distributed photovoltaic prediction method based on a long short-term memory neural network as described in claim 1, characterized in that: Before merging the temporal and spatial features, the temporal features are pre-adjusted; and... The pre-adjustment includes adjusting the time interval density and time span of the time series in the time features; Simultaneously, the size of the output layer convolution kernel in the prediction branch is adjusted.

3. The distributed photovoltaic prediction method based on a long short-term memory neural network as described in claim 2, characterized in that: Before merging the temporal and spatial features, the spatial features are pre-adjusted; and... The pre-adjustment includes adjusting the weights of the space nodes based on the power generation contribution rate of the photovoltaic devices in the space nodes.

4. A distributed photovoltaic prediction system based on a long short-term memory neural network, running the distributed photovoltaic prediction method based on a long short-term memory neural network according to any one of claims 1-3, characterized in that: The system includes a data acquisition module, an extraction module, and a prediction module; wherein, The acquisition module is used to acquire irradiance data at the location of the distributed photovoltaic system, and generate a continuous time series of a 2D sparse matrix based on the spatial dimension parameters and time dimension parameters of the irradiance data. The irradiance data is collected based on an optical sensor array within the area of ​​the distributed photovoltaic system. Each irradiance data corresponds to the spatial location of an optical sensor and the irradiance collection time. Based on the spatial location of the optical sensor, the irradiance data collected at the same time are combined into an original matrix. The spatial correlation between each irradiance data in the original matrix is ​​solved, and the original matrix is ​​simplified based on the spatial correlation to obtain a 2D sparse matrix. The extraction module is used to extract spatial features from the 2D sparse matrix using a convolutional neural network, extract single-point time series from the continuous time series of the 2D sparse matrix, and extract time features from the single-point time series using a long short-term memory neural network. The prediction module is used to combine time features and spatial features to predict the irradiance of the location of the distributed photovoltaic system at the next moment, thereby realizing the prediction of the distributed photovoltaic power generation. The CNN-BiLSTM model is used to predict the irradiance at the next time step. The CNN-BiLSTM model includes a spatial feature extraction branch, a temporal feature extraction branch, and a prediction branch. The spatial feature extraction branch includes a multi-layer convolutional neural network. The temporal feature extraction branch includes a multi-layer long short-term memory neural network, a first fully connected layer, and a second fully connected layer. The prediction branch includes a feature connection layer, a third fully connected layer, and an output layer. The feature connection layer arranges the extracted spatial features into a one-dimensional array and then merges it with the temporal features.

5. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-3.

Images