Photovoltaic power station mid-term output prediction method and system based on MCNN-BiGRU
By combining multi-channel convolutional neural network and bidirectional gated cycle unit, the problem of insufficient spatio-temporal feature capture of photovoltaic power generation in the prior art is solved, and the prediction accuracy and reliability are improved.
Patent Information
- Application Number
- CN202510339748.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-21
- Publication Date
- 2025-08-15
AI Technical Summary
Existing photovoltaic energy prediction methods cannot effectively capture the space-time coupling characteristics of photovoltaic power generation, resulting in low prediction accuracy, traditional methods are complex in calculations or rely on data quality, and deep learning algorithms cannot handle complex time-dependent and nonlinear relationships at the same time.
Combining multi-channel convolutional neural network (MCNN) and bidirectional gated cyclic unit (BiGRU), through data preprocessing, constructing multi-channel convolutional neural network model, designing bidirectional gate cyclic control unit, obtaining hyperparameters and weight bias combinations, and using test set data for testing.
It improves the prediction accuracy of mid-term output of photovoltaic power plants, reduces the impact of data noise, enhances the local feature extraction ability, captures the two-way dependence relationship of time series data, and ensures the reliability of the prediction results.
Smart Images
Figure CN120497868A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of photovoltaic output prediction, and in particular to a method and system for predicting the medium-term output of a photovoltaic power station based on MCNN-BiGRU. Background Art
[0002] Photovoltaic energy has attracted widespread attention as a clean and sustainable energy source. However, the intermittent and variable nature of photovoltaic power generation poses challenges to large-scale grid integration and effective utilization. Accurate photovoltaic energy output forecasting is crucial for optimizing grid scheduling and improving the operational efficiency and economic benefits of photovoltaic power plants.
[0003] Traditional photovoltaic energy forecasting methods, such as statistical and physical models, have significant shortcomings. Statistical models are computationally simple but limited in their ability to handle complex time-dependent and nonlinear relationships. While physical models can capture physical mechanisms, they are computationally expensive and highly dependent on the quality and accuracy of input data. Furthermore, single deep learning algorithms, such as convolutional neural networks and long-short-term memory neural networks, are unable to simultaneously capture the spatiotemporal coupling characteristics of photovoltaic power generation, resulting in low forecasting accuracy.
[0004] Multi-channel convolutional neural networks (MCNNs) can effectively extract local spatial features from photovoltaic energy data, helping the model capture information in the spatial dimension of the data, while bidirectional gated recurrent units (BiGRUs) can effectively process complex patterns in time series. Therefore, how to combine multi-channel convolutional neural networks and bidirectional gated recurrent units to design prediction methods with high-performance spatiotemporal feature capture is still under research. Summary of the Invention
[0005] In view of the above problems in the prior art, the present invention is proposed.
[0006] To solve the above technical problems, the present invention provides the following technical solution: a medium-term output prediction method for a photovoltaic power station based on MCNN-BiGRU, comprising: acquiring data, preprocessing and dividing the data; constructing a multi-channel convolutional neural network model and obtaining hyperparameters; designing a bidirectional gate recurrent control unit and obtaining the weight and bias combination of the control unit's loss function; and testing using test set data to obtain test results.
[0007] As a preferred solution of the MCNN-BiGRU-based photovoltaic power station medium-term output prediction method described in the present invention, the preprocessing and division include obtaining the photovoltaic power station medium-term output historical data, normalizing, denoising and labeling the data, and dividing the normalized data into a training set and a test set.
[0008] As a preferred solution of the photovoltaic power station medium-term output prediction method based on MCNN-BiGRU described in the present invention, wherein: the obtaining of hyperparameters includes constructing a convolutional neural network input layer based on a normalized photovoltaic power station medium-term output historical data training set; constructing a convolutional layer, a pooling layer and an output layer of the convolutional neural network based on the normalized photovoltaic power station medium-term output historical data training set and the convolutional neural network input layer; determining the hyperparameters that need to be optimized based on the constructed convolutional neural network, and using the Adam optimization algorithm to obtain the hyperparameters that minimize the loss function in the convolutional neural network.
[0009] As a preferred solution of the photovoltaic power station medium-term output prediction method based on MCNN-BiGRU described in the present invention, the convolutional layer, pooling layer and output layer of the convolutional neural network are constructed, including constructing the convolutional layer of the convolutional neural network through convolution and bias operations based on the normalized photovoltaic power station medium-term output historical data training set; performing a maximum pooling operation on the activated convolution result to construct the convolutional neural network pooling layer; and constructing the convolutional neural network output layer based on the output after maximum pooling and the photovoltaic power station medium-term output training set.
[0010] As a preferred solution of the medium-term output prediction method of a photovoltaic power station based on MCNN-BiGRU described in the present invention, wherein: the convolution layer of the convolutional neural network is constructed through convolution and bias operations, including obtaining the convolution output result according to the number of input channels and the number of output channels; adding a bias to the convolution output result of the output channel to obtain the final convolution output; applying an activation function to the final convolution output to obtain the activated convolution result.
[0011] As a preferred solution of the MCNN-BiGRU-based photovoltaic power station medium-term output prediction method described in the present invention, the bidirectional gate recurrent control unit is designed according to the time series characteristics of the normalized photovoltaic power station medium-term output historical data training set; the Adam optimization algorithm is used to calculate the gradient of the bidirectional gate recurrent control unit, adaptively adjust the learning rate according to the first-order moment estimate and second-order moment estimate of the gradient, and update the weight and bias; and minimize the loss function during the iterative process to obtain the weight and bias combination that minimizes the loss function.
[0012] As a preferred solution of the medium-term output prediction method of a photovoltaic power station based on MCNN-BiGRU described in the present invention, the testing using test set data includes testing the MCNN-BiGRU using the test set data to obtain the mean absolute error and root mean square error of the MCNN-BiGRU.
[0013] As a preferred solution of the photovoltaic power station medium-term output prediction system based on MCNN-BiGRU described in the present invention, it includes: a data acquisition module, a construction model module, a design control unit module and a test model module;
[0014] The data acquisition module acquires data and performs pre-processing and division on the data;
[0015] The model construction module constructs a multi-channel convolutional neural network model and obtains hyperparameters;
[0016] The control unit module is designed to design a bidirectional gate loop control unit and obtain a weight and bias combination of the control unit loss function;
[0017] The test model module uses the test set data to perform testing and obtain test results.
[0018] A computer device includes a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, it implements the steps of any one of the methods for predicting the medium-term output of a photovoltaic power station based on MCNN-BiGRU.
[0019] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of any one of the methods for predicting the medium-term output of a photovoltaic power station based on MCNN-BiGRU are implemented.
[0020] The beneficial effects of the present invention are as follows: by acquiring and preprocessing the original data, operations such as noise removal, missing value filling, and standardization are realized, thereby reducing the deviation caused by data problems and ensuring the reliability of the prediction results; by constructing a multi-channel convolutional neural network (MCNN), it is realized that different feature channels are processed in parallel through multiple convolution kernels, thereby enhancing the ability to extract local features; by designing a bidirectional gated recurrent unit (BiGRU), it is realized that bidirectional dependencies in time series data can be captured, that is, not only information from the forward data flow can be learned, but also reverse information can be learned, thereby more comprehensively understanding the dynamic changes of the data; by testing the deep learning model, the performance of the model is evaluated to ensure its prediction ability on actual data. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0022] Figure 1 A schematic flow chart of a method for predicting the medium-term output of a photovoltaic power station based on MCNN-BiGRU is provided as an embodiment of the present invention.
[0023] Figure 2 A multi-channel convolutional neural network structure diagram of a photovoltaic power station medium-term output prediction method based on MCNN-BiGRU provided in one embodiment of the present invention.
[0024] Figure 3 A diagram of the gate cycle unit structure of a photovoltaic power station medium-term output prediction method based on MCNN-BiGRU provided in one embodiment of the present invention.
[0025] Figure 4 An embodiment of the present invention provides a photovoltaic power station medium-term output prediction method based on MCNN-BiGRU and a photovoltaic power station medium-term output prediction result based on a back propagation neural network.
[0026] Figure 5 An embodiment of the present invention provides a method for predicting the medium-term output of a photovoltaic power station based on MCNN-BiGRU and a medium-term output prediction result of a photovoltaic power station based on MCNN-BiGRU. DETAILED DESCRIPTION
[0027] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.
[0028] Example 1
[0029] Reference Figure 1-Figure 5 , which is the first embodiment of the present invention, provides a method for predicting the medium-term output of a photovoltaic power station based on MCNN-BiGRU, including:
[0030] S1: Acquire data, preprocess and divide the data.
[0031] It should be noted that if Figure 1 As shown in S100 , preprocessing and dividing the data includes obtaining the historical mid-term output data of the photovoltaic power station, normalizing, denoising and labeling the data, and dividing the normalized data into a training set and a test set.
[0032] Furthermore, the module temperature (℃), ambient temperature (℃), atmospheric pressure (hPa), humidity (%), total radiation (W / m2 ), direct radiation (W / m 2 ) and diffuse radiation (W / m 2 ), the mid-term output historical data of the photovoltaic power station with actual output (W), and perform data normalization, denoising and labeling based on the obtained mid-term output historical data of the photovoltaic power station;
[0033] The normalization process is expressed as:
[0034]
[0035] Among them, z i is the data after maximum and minimum normalization processing, y i is the original data, y min is the minimum value of the original data feature, y max is the maximum value of the original data feature;
[0036] The normalized data is divided into training set and test set.
[0037] S2: Construct a multi-channel convolutional neural network model and obtain hyperparameters.
[0038] It should be noted that if Figure 1 As shown in S200, obtaining the hyperparameters includes constructing a convolutional neural network input layer based on the normalized training set of historical data on the mid-term output of the photovoltaic power station; constructing the convolutional layer, pooling layer and output layer of the convolutional neural network based on the normalized training set of historical data on the mid-term output of the photovoltaic power station and the input layer of the convolutional neural network; determining the hyperparameters that need to be optimized based on the constructed convolutional neural network, and using the Adam optimization algorithm to obtain the hyperparameters that minimize the loss function in the convolutional neural network.
[0039] Construct a multi-channel convolutional neural network (MCNN) with convolutional layers and pooling layers, and use the Adam optimization algorithm to obtain the hyperparameters that minimize the loss function in the convolutional neural network;
[0040] like Figure 2 As shown in the figure, based on the normalized training set of historical data of mid-term output of photovoltaic power stations, the input layer of MCNN is constructed, including the batch size N of samples and the number of input channels C. in , input height H in and the input width W in ; Based on the normalized photovoltaic power station mid-term output historical data training set and the input layer structure of MCNN, the convolution layer, pooling layer and output layer of MCNN are constructed.
[0041] Furthermore, based on the normalized training set of historical data of the photovoltaic power station’s mid-term output, the convolution layer of the MCNN is constructed through convolution and bias operations; according to the number of input channels Cin and the number of output channels C out , press the formula to get the nth sample, the Cth out The convolution output Z of the hth row and wth column of the output channel n,cout,h,w , expressed as:
[0042]
[0043] Among them, k h Refers to the height of the convolution kernel, k w Refers to the width of the convolution kernel;
[0044] For c out The convolution output Z of the hth row and wth column of the output channel n,cout,h,w Adding the bias b, we get the final convolution output A, which is expressed as:
[0045]
[0046] Apply activation function f to the final convolution output A active , get the convolution result after activation O n,cout,h,w , expressed as:
[0047]
[0048] The convolution result after activation O n,cout,h,ww Perform the maximum pooling operation to construct the pooling layer of MCNN, which is expressed as:
[0049]
[0050] Among them, (p h ;p w ) is the pooling kernel size, (s h ;s w ) is the pooling step size; according to the output P after the maximum pooling n,cout,h,w And the photovoltaic power station mid-term output training set, construct the output layer of MCNN, which is expressed as:
[0051]
[0052] Among them, W f,c,i is the weight of the output layer, b f,c is the bias of the output layer;
[0053] According to the constructed MCNN, the hyperparameters that need to be optimized are determined, such as learning rate, batch size, convolution kernel size, pooling size, etc. The Adam optimization algorithm is used to obtain the hyperparameters that minimize the loss function in the convolutional neural network.
[0054] S3: Design a bidirectional gate recurrent control unit and obtain the weight and bias combination of the control unit's loss function.
[0055] It should be noted that if Figure 1 As shown in S300, designing a bidirectional gate recurrent control unit includes designing it based on the time series characteristics of the normalized photovoltaic power station mid-term output historical data training set; using the Adam optimization algorithm, by calculating the gradient of the bidirectional gate recurrent control unit, adaptively adjusting the learning rate according to the first-order moment estimate and the second-order moment estimate of the gradient, and updating the weight and bias; minimizing the loss function during the iterative process, and obtaining the weight and bias combination that minimizes the loss function.
[0056] Furthermore, the design of the bidirectional gate recurrent control unit includes designing a forward GRU layer according to the forward time series characteristics of the training set of the photovoltaic power station's mid-term output historical data; and constructing a reverse GRU layer by imitating the design structure of the forward GRU according to the reverse time series characteristics of the training set of the photovoltaic power station's mid-term output historical data, and obtaining the reverse hidden state h b,t ; According to the forward hidden state h f,t and the reverse hidden state h b,t , get the output y of the bidirectional gate loop control unit t ;
[0057] According to the time series characteristics of the normalized photovoltaic power station mid-term output historical data training set, a bidirectional gate recurrent control unit (BiGRU) is designed, such as Figure 3 As shown in Figure 1, the forward GRU layer is designed based on the forward time series characteristics of the training set of historical data of the photovoltaic power station's mid-term output. The update gate of the forward GRU is designed based on the retention amount of the hidden state of the photovoltaic power station's mid-term output at the previous moment, which is expressed as:
[0058] z t =σ(W z x t +U z h f,t-1 +b z )
[0059] Among them, z t is the output of the update gate; x t is the input vector of the input sequence at time step t; W z For input x t To the weight matrix of the update gate; h f,t-1 is the hidden state of the forward GRU at time step t-1; U z is the hidden state h at the previous moment f,t-1 To the weight matrix of the update gate; b z is the bias vector of the update gate; σ is the activation function;
[0060] According to the reset amount of the hidden state at the last moment of the mid-term output of the photovoltaic power station, the reset gate of the forward GRU is designed, which is expressed as:
[0061] r t =σ(W r x t +U r h f,t-1 +b r )
[0062] Among them, r t is the output of the reset gate, W r For input x t To the weight matrix of the reset gate, U r is the hidden state h at the previous moment f,t-1 To the weight matrix of the reset gate; b r is the bias vector for resetting the gate;
[0063] According to the update gate and reset gate of the forward GRU, the candidate hidden state h is obtained c,f,t and the forward hidden state h f,t Update h f,t , expressed as,
[0064] h c,f,t =tanh(W h x t +r t ⊙U h h f,t-1 +b h )
[0065] h f,t =(1-z t )⊙h f,t-1 +z t ⊙h c,f,t
[0066] Among them, W h For input x t To the candidate hidden state h c,f,t The weight matrix of U h is the hidden state h at the previous moment f,t-1 To the hidden state h c,f,t The weight matrix of b h is the hidden state h c,f,t The bias vector of
[0067] According to the reverse time series characteristics of the training set of the photovoltaic power station mid-term output history data, the reverse GRU layer is constructed by imitating the design structure of the forward GRU to obtain the reverse hidden state h b,t ; According to the forward hidden state h f,t and the reverse hidden state h b,tGet the final output y of BiGRU t , expressed as:
[0068] y t =[h f,t ,h b,t ]
[0069] The Adam optimization algorithm is used to calculate the gradient of the bidirectional gated recurrent control unit (BiGRU), adaptively adjust the learning rate according to the first-order moment estimate and second-order moment estimate of the gradient, and update the weights and biases. The loss function is minimized in the iterative process, and finally the weight and bias combination that minimizes the loss function is obtained.
[0070] S4: Use the test set data to perform testing and obtain test results.
[0071] It should be noted that if Figure 1 As shown in S400 , using the test set data for testing includes testing the MCNN-BiGRU using the test set data to obtain the mean absolute error and the root mean square error of the MCNN-BiGRU.
[0072] like Figure 4 As shown in Figure 2, the prediction results of the back propagation (BP) neural network for the mid-term output of the photovoltaic power station are as follows: MAE is 16.005 and RMSE is 27.044; Figure 5 As shown in the figure, the prediction results of MCNN-BiGRU for the medium-term output of photovoltaic power stations have a MAE of 15.130 and an RMSE of 22.365. Compared with the prediction results of BP neural network, the MAE of MCNN-BiGRU is reduced by 0.875 and the RMSE is reduced by 4.679.
[0073] Example 2
[0074] The second embodiment of the present invention is different from the previous embodiment in that:
[0075] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.
[0076] Logic and / or steps otherwise described herein, which may be considered, for example, as an ordered list of executable instructions for implementing the logical functions, may be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" may be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.
[0077] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.
[0078] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0079] Example 3
[0080] The third embodiment of the present invention provides a photovoltaic power station medium-term output prediction system based on MCNN-BiGRU, which is characterized by comprising a data acquisition module, a construction model module, a design control unit module and a test model module;
[0081] The data acquisition module acquires data and performs pre-processing and division on the data;
[0082] The model construction module constructs a multi-channel convolutional neural network model and obtains hyperparameters;
[0083] The control unit module is designed to design a bidirectional gate loop control unit and obtain a weight and bias combination of the control unit loss function;
[0084] The test model module uses the test set data to perform testing and obtain test results.
[0085] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.
Claims
1. A method for predicting the mid-term output of a photovoltaic power station based on MCNN-BiGRU, characterized by: include: Acquire data, preprocess and divide the data; Construct a multi-channel convolutional neural network model and obtain hyperparameters; Design a bidirectional gate recurrent control unit and obtain the weight and bias combination of the control unit's loss function; Use the test set data to test and get the test results.
2. The method for predicting the medium-term output of a photovoltaic power station based on MCNN-BiGRU according to claim 1, characterized in that: The preprocessing and division include obtaining the mid-term output history data of the photovoltaic power station, normalizing, denoising and labeling the data, and dividing the normalized data into a training set and a test set.
3. The method for predicting the medium-term output of a photovoltaic power station based on MCNN-BiGRU according to claim 2, characterized in that: The obtaining of hyperparameters includes constructing a convolutional neural network input layer based on a normalized training set of historical data on the mid-term output of a photovoltaic power station; constructing a convolutional layer, a pooling layer, and an output layer of the convolutional neural network based on the normalized training set of historical data on the mid-term output of a photovoltaic power station and the input layer of the convolutional neural network; determining hyperparameters that need to be optimized based on the constructed convolutional neural network, and using the Adam optimization algorithm to obtain hyperparameters that minimize the loss function in the convolutional neural network.
4. The method for predicting the medium-term output of a photovoltaic power station based on MCNN-BiGRU according to claim 3, characterized in that: The convolutional layer, pooling layer and output layer of the convolutional neural network are constructed, including constructing the convolutional layer of the convolutional neural network through convolution and bias operations based on the normalized photovoltaic power station medium-term output historical data training set; performing a maximum pooling operation on the activated convolution result to construct the convolutional neural network pooling layer; and constructing the convolutional neural network output layer based on the output after maximum pooling and the photovoltaic power station medium-term output training set.
5. The method for predicting the medium-term output of a photovoltaic power station based on MCNN-BiGRU according to claim 4, characterized in that: The convolution layer of the convolutional neural network is constructed through convolution and bias operations, including obtaining a convolution output result according to the number of input channels and the number of output channels; adding a bias to the convolution output result of the output channel to obtain a final convolution output; and applying an activation function to the final convolution output to obtain an activated convolution result.
6. The method for predicting the medium-term output of a photovoltaic power station based on MCNN-BiGRU according to claim 5, characterized in that: The design of the bidirectional gate cycle control unit includes designing it according to the time series characteristics of the normalized photovoltaic power station mid-term output historical data training set; Using the Adam optimization algorithm, the gradient of the bidirectional gate recurrent control unit is calculated, and the learning rate is adaptively adjusted according to the first-order moment estimate and second-order moment estimate of the gradient to update the weights and biases; Minimize the loss function during the iteration process and obtain the weight and bias combination that minimizes the loss function.
7. The method for predicting the medium-term output of a photovoltaic power station based on MCNN-BiGRU according to claim 6, characterized in that: The testing using the test set data includes testing the MCNN-BiGRU using the test set data to obtain the mean absolute error and root mean square error of the MCNN-BiGRU.
8. A system for predicting mid-term output of a photovoltaic power station based on an MCNN-BiGRU according to any one of claims 1 to 7, characterized in that: Including data acquisition module, construction model module, design control unit module and test model module; The data acquisition module acquires data and performs pre-processing and division on the data; The model construction module constructs a multi-channel convolutional neural network model and obtains hyperparameters; The control unit module is designed to design a bidirectional gate loop control unit and obtain a weight and bias combination of the control unit loss function; The test model module uses the test set data to perform testing and obtain test results.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.