A photovoltaic power prediction method and system based on multi-stage time series feature mining

Abstract

This invention discloses a photovoltaic power prediction method and system based on multi-stage temporal feature mining. The prediction method includes: extracting features from historical photovoltaic power sequences using an LSTM model to obtain power temporal features; extracting features from historical satellite cloud image sequences using a 3DCNN model to obtain cloud image spatiotemporal dynamic features; fusing the power temporal features and cloud image spatiotemporal dynamic features based on a residual cross-attention mechanism to obtain multimodal fused features; further mining the temporal characteristics in the fused features using a Transformer model to obtain deep multimodal temporal features; and outputting the prediction results through a multilayer perceptron to predict photovoltaic power for the next 1 to 4 hours. This invention can improve the accuracy of photovoltaic power prediction, providing support for the safe and stable operation of the power grid, and has certain engineering practical value.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic power generation technology, specifically relating to a photovoltaic power prediction method and system based on multi-stage time series feature mining. Background Technology

[0002] Solar energy, with its wide distribution, ease of extraction, high energy efficiency, and pollution-free characteristics, has gained global attention and become a key focus of renewable energy research. Photovoltaic (PV) power generation, with its advantages of quiet operation, cleanliness, high geographical flexibility, and short construction cycle, has become the optimal application of solar energy. However, PV power generation is affected by meteorological factors such as sunshine, cloud cover, and temperature, resulting in intermittent and fluctuating power output, posing a challenge to grid stability and limiting its widespread deployment. Therefore, accurate prediction of PV power generation is crucial.

[0003] In the field of photovoltaic (PV) power forecasting, satellite cloud imagery plays a crucial role. It provides real-time, wide-ranging weather conditions, including cloud cover, distribution, and movement, which is essential for predicting short-term changes in PV power. PV power sequences and satellite cloud image sequences exhibit significant temporal characteristics. However, existing forecasting methods suffer from insufficient extraction of temporal features and neglect the coupling and redundancy between power features and image features, thus limiting the improvement of PV power forecasting performance. How to fully utilize the temporal features of historical data and explore the coupling characteristics between power features and image features to further improve the performance of PV power forecasting models is a current challenge in enhancing solar energy utilization. Summary of the Invention

[0004] The purpose of this invention is to provide a photovoltaic power prediction method and system based on multi-stage time-series feature mining, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a photovoltaic power prediction method based on multi-stage time-series feature mining, comprising the following steps:

[0006] Step 1: Extract features from historical photovoltaic power sequences using an LSTM model to obtain power time-series features. This process can be represented as follows:

[0007]

[0008] Among them, F power For power timing characteristics, f LSTM For the LSTM model, m is the number of LSTM layers used, [p1, p2, ..., p n [ ] represents the historical photovoltaic power sequence, where n is the length of the time series;

[0009] Step 2: Extract features from historical satellite cloud image sequences using a 3DCNN model to obtain the spatiotemporal dynamic features of the cloud images. This process can be represented as: F image =f 3DCNN ([i1,i2,...,i n ]);

[0010] Among them, F image For the spatiotemporal dynamic characteristics of cloud maps, f 3DCNN For a 3DCNN model, [i1,i2,...,i n [This refers to] a sequence of historical satellite cloud images; 3DCNN can be expressed using the following formula:

[0011] h (l) =f (l) (w (l) *h (l-1) +b (l) );

[0012] Step 3: Based on the residual cross-attention mechanism, feature fusion is performed on the power time series features and the spatiotemporal dynamic features of the cloud map to obtain multimodal fusion features. The specific process is as follows:

[0013]

[0014] Step 4: Utilize the Transformer model to further mine the temporal characteristics in the fused features, obtaining deep multimodal temporal features. The specific process is as follows:

[0015]

[0016] Step 5: The prediction results are output through a multilayer perceptron to predict the photovoltaic power for the next 1 to 4 hours. Based on power time series characteristics, cloud map spatiotemporal dynamic characteristics, and deep multimodal time series characteristics, as well as the descriptions in Steps 2 and 3 above, power characteristics, image characteristics, and fused characteristics can be obtained from historical data. Finally, by relearning these three features, the predicted photovoltaic power result y can be obtained.

[0017] y = f MLP ([F power ,F image ,F final ]);

[0018] Among them, f MLP This is a multilayer perceptron model.

[0019] Preferably, in step one, LSTM can be expressed by the following formula:

[0020] i t =σ(W xi *x t +Whi *h t-1 +b i )

[0021] f t =σ(W xf *x t +W hf *h t-1 +b f )

[0022]

[0023] g t =σ(W xg *x t +W hg *h t-1 +b g )

[0024]

[0025] Where, x t h t c t These represent the input value, output value, and cell state of the LSTM at time t, respectively; h t-1 Let i be the output value of the LSTM at time t-1. t f t g t These are the input gate, forget gate, and output gate, used to control the flow of information; W xi W hi b i W xf W hf b f W xc W hc b c W xg W hg b g All represent learnable parameters, enabling the network to effectively process sequential data and retain long-term dependencies. tanh is the hyperbolic tangent function, σ(·) is the activation function, and "*" represents ordinary matrix multiplication. It represents the Hadamardi (or Hadama) stack.

[0026] Preferably, in step two, * represents the convolution operation, w (l) b represents the weights of the l-th convolutional layer. (l) f is the bias of the l-th convolutional layer. (l) h is the activation function of the l-th convolutional layer. (l) h (l-1)These are the outputs of the l-th and (l-1)-th convolutional layers, respectively. Furthermore, a three-dimensional max-pooling operation is used to downsample the feature map, reducing the computational cost and number of parameters while preventing overfitting. This process can be represented as follows:

[0027]

[0028] Where Y is the output feature map, (x, y, z) is the position index in the output feature map, g is the input feature map for the pooling operation, M, N, and O represent the dimensions of the pooling window in the width, height, and depth directions, respectively, and S d 、S h 、S w These represent the step size of the pooling window in the width, height, and depth directions, respectively, while m, n, and o are the index offsets of the pooling window at its current position.

[0029] Preferably, in step three, F fusion For multimodal fusion features, RS-Attention is a residual cross-attention mechanism. Q, K, and V are the query vector, key vector, and value vector, respectively. Q is obtained through a nonlinear mapping of power features, and K and V are obtained through a nonlinear mapping of spatiotemporal dynamic features of cloud maps. k Where is the dimension of the key vector, softmax is used to calculate the weights of the cross-attention mechanism.

[0030] Preferably, in step four, F final For multimodal time series features, f Transformer For the Transformer model, Let n represent the multimodal fusion features at each historical moment, where n is the dimension of the fusion features.

[0031] To achieve the above objectives, the present invention also provides the following technical solution:

[0032] A photovoltaic power prediction system based on multi-stage time-series feature mining includes a power feature extraction module, an image feature extraction module, a feature fusion module, a deep feature extraction module, and a time-series prediction module, which are used to execute the photovoltaic power prediction method based on multi-stage time-series feature mining described above.

[0033] The power feature extraction module extracts features from historical photovoltaic power sequences using an LSTM model to obtain power time-series features;

[0034] The image feature extraction module extracts features from historical satellite cloud image sequences using a 3DCNN model to obtain the spatiotemporal dynamic features of the cloud images;

[0035] The feature fusion module fuses power time-series features and cloud map spatiotemporal dynamic features based on the residual cross-attention mechanism to obtain multimodal fusion features;

[0036] The deep feature extraction module uses the Transformer model to further mine the temporal characteristics in the fused features to obtain deep multimodal temporal features;

[0037] The time-series prediction module takes power time-series characteristics, cloud map spatiotemporal dynamic characteristics, and deep multimodal time-series characteristics as inputs, and outputs photovoltaic power results for the next 1 to 4 hours through a multilayer perceptron.

[0038] To achieve the above objectives, the present invention also provides the following technical solution:

[0039] A computer device, comprising:

[0040] One or more processors;

[0041] Memory;

[0042] And one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, wherein when the programs are executed by the processors, they implement the steps of the photovoltaic power prediction method based on multi-stage time-series feature mining as described above.

[0043] Preferably, when the computer program is executed by the processor, it implements the steps of the photovoltaic power prediction method based on multi-stage time-series feature mining as described above.

[0044] Compared with the prior art, the beneficial effects of the present invention are:

[0045] (1) Power temporal features and image spatiotemporal features were extracted using LSTM and 3DCNN respectively. Based on the Transformer model, the deep temporal characteristics of the fusion features were further explored. By capturing the temporal characteristics of historical data in multiple stages, the model's ability to represent the temporal features of multi-source data was successfully improved.

[0046] (2) The residual cross-attention mechanism is applied to fuse power features and image features, explore the coupling characteristics between the two features, reduce the redundancy of the fused features, and improve the availability of feature information. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the photovoltaic power prediction method of the present invention;

[0048] Figure 2 This is a schematic diagram of the long short-term memory neural network used in this invention;

[0049] Figure 3This is a schematic diagram of the Transformer model used in this invention;

[0050] Figure 4 This is a schematic diagram of the residual cross-attention mechanism proposed in this invention;

[0051] Figure 5 This is a schematic diagram of the prediction result fitting using the proposed method in an embodiment of the present invention. Detailed Implementation

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0053] This invention provides, for example Figure 1 One technical solution shown is a photovoltaic power prediction method based on multi-stage time-series feature mining, which includes the following steps:

[0054] Step 1: Extract features from historical photovoltaic power sequences using an LSTM model to obtain power time-series features;

[0055] Step 2: Extract features from historical satellite cloud image sequences using a 3DCNN model to obtain the spatiotemporal dynamic features of the cloud images;

[0056] Step 3: Based on the residual cross-attention mechanism, feature fusion is performed on the power time series features and the spatiotemporal dynamic features of the cloud map to obtain multimodal fusion features;

[0057] Step 4: Utilize the Transformer model to further mine the temporal characteristics in the fused features to obtain deep multimodal temporal features;

[0058] Step 5: The prediction results are output by the multilayer sensor to predict the photovoltaic power in the next 1 to 4 hours.

[0059] Specifically, in this embodiment of the invention, power timing features are extracted in step one using the following method;

[0060] Feature extraction of historical photovoltaic power sequences using an LSTM model yields power time-series features, the process of which can be represented as follows:

[0061]

[0062] Among them, F power The power timing characteristics are given, where m is the number of layers in the LSTM used, [p1, p2, ..., p...]. n[ ] represents the historical photovoltaic power sequence, where n is the length of the time series; for example... Figure 2 As shown, LSTM can be expressed by the following formula:

[0063] i t =σ(W xi *x t +W hi *h t-1 +b i )

[0064] f t =σ(W xf *x t +W hf *h t-1 +b f )

[0065]

[0066] g t =σ(W xg *x t +W hg *h t-1 +b g )

[0067]

[0068] Where, x t h t c t These represent the input value, output value, and cell state of the LSTM at time t, respectively; h t-1 Let i be the output value of the LSTM at time t-1. t f t g t These are the input gate, forget gate, and output gate, used to control the flow of information; W xi W hi b i W xf W hf b f W xc W hc b c W xg W hg b g All represent learnable parameters, enabling the network to effectively process sequence data and possess the ability to remember long-term dependencies; σ(·) is the activation function. It represents the Hadamardi (or Hadama) stack.

[0069] In step two of this invention, a 3DCNN model is used to extract features from historical satellite cloud image sequences to obtain the spatiotemporal dynamic features of the cloud images. The process can be represented as follows:

[0070] F image =f 3DCNN ([i1,i2,...,i n ])

[0071] Among them, F image For the spatiotemporal dynamic characteristics of cloud maps, f 3DCNN For a 3DCNN model, [i1,i2,...,i n [This refers to] a sequence of historical satellite cloud images; 3DCNN can be expressed using the following formula:

[0072] h (l) =f (l) (w (l) *h (l-1) +b (l) );

[0073] Where * represents the convolution operation, w (l) b represents the weights of the l-th convolutional layer. (l) f is the bias of the l-th convolutional layer. (l) h is the activation function of the l-th convolutional layer. (l) h (l-1) These are the outputs of the l-th and (l-1)-th convolutional layers, respectively. Furthermore, a three-dimensional max-pooling operation is used to downsample the feature map, reducing the computational cost and number of parameters while preventing overfitting. This process can be represented as:

[0074]

[0075] Where Y is the output feature map, (x, y, z) is the position index in the output feature map, g is the input feature map for the pooling operation, M, N, and O represent the dimensions of the pooling window in the width, height, and depth directions, respectively, and S d 、S h 、S w , respectively, represent the step size of the pooling window in the width, height, and depth directions, and m, n, and o are the index offsets of the pooling window at the current position; at the same time, the Batch Normalization operation is applied to reprocess the output feature map to speed up the model convergence and improve the model stability.

[0076] like Figure 3As shown, step three uses a residual cross-attention mechanism to fuse power time-series features and cloud image spatiotemporal dynamic features to obtain multimodal fused features. First, by calculating the correlation between power time-series features and cloud image spatiotemporal dynamic features, key feature information in the cloud image spatiotemporal dynamic features is selected. Second, the key cloud image spatiotemporal dynamic features and power time-series features are merged through a concatenation operation to obtain the fused features. The specific process of the residual cross-attention mechanism is as follows:

[0077]

[0078] Among them, F fusion For multimodal fusion features, RS-Attention is a residual cross-attention mechanism. Q, K, and V are the query vector, key vector, and value vector, respectively. Q is obtained through a nonlinear mapping of power features, and K and V are obtained through a nonlinear mapping of spatiotemporal dynamic features of cloud maps. k Given the dimension of the key vector, softmax is used to calculate the weights of the cross-attention mechanism. Through the above attention mechanism, the prediction model can fully exploit the coupling characteristics between power features and image features, and reduce the redundancy of feature information.

[0079] In step four of this invention, the Transformer model is used to further mine the temporal characteristics in the fused features to obtain deep multimodal temporal features. The specific process can be represented as follows:

[0080]

[0081] Among them, F final For multimodal time series features, f Transformer For the Transformer model, For each historical moment, the multimodal fusion feature is represented, where n is the dimension of the fusion feature; for example... Figure 4 As shown, multi-head attention, as the core of the Transformer model, enables the model to simultaneously focus on different parts of the input sequence, thereby improving its attentional ability and capturing complex dependencies between input and output. The mathematical representation of the multi-head attention mechanism is as follows:

[0082] MultiHead(Q,K,V)=Concat(head1,...,head h W O ,head i =Attention(QW i Q ,KW i K VW i V )

[0083] In this context, MultiHead represents multi-head attention mechanism, Q, K, and V are the query vector, key vector, and value vector, respectively, Concat represents the concatenation operation, and head... i —The weight result generated by the i-th attention operation, where h is the number of times the attention mechanism is implemented, and W l Q 、W l K 、W l V 、W O These are the weighting coefficients of the linear mapping.

[0084] In step five of this invention, based on power time-series characteristics, spatiotemporal dynamic characteristics of cloud maps, and deep multimodal time-series characteristics, the predicted photovoltaic power y is obtained through a multilayer perceptron.

[0085] y = f MLP ([F power ,F image ,F final ]);

[0086] Among them, f MLP This is a multilayer perceptron model.

[0087] To verify the performance of the proposed method, the following experiments were conducted in this embodiment. The dataset used was [database name missing], and the model performance was evaluated using quantitative error metrics. The satellite cloud images and photovoltaic power data used in this study came from a photovoltaic power station in China (2017-2018). 80% of this dataset was used for training and 20% for testing. Three error analysis metrics were used in this embodiment: Mean Error (MAE), Root Mean Square Error (RMSE), and Coefficient of Determination (R²). 2 ), defined as follows:

[0088]

[0089]

[0090] Where N represents the total number of test samples, y i and Let these represent the predicted value and the actual value of the i-th sample, respectively. This represents the mean of the actual values.

[0091] To visually reflect the prediction results, Figure 5The photovoltaic power generation prediction curve obtained one hour in advance using the method of this invention is shown, demonstrating that the method can accurately predict photovoltaic power. To further evaluate the prediction performance of the prediction model, four comparison models were set up, denoted as S1, S2, S3, and S4. S1 uses 2DCNN to extract spatial features from cloud images, S2 uses ConvLSTM to extract spatiotemporal features from cloud images, and S3 uses MLP and 3DCNN to extract feature information from historical power and satellite cloud images, respectively. Furthermore, Table 1 shows the prediction error of each model for photovoltaic power prediction 10 minutes in advance. It can be seen that compared to the comparison models, the model constructed based on this invention has smaller prediction errors and superior performance.

[0092] Table 1 Comparison of prediction results from different models

[0093]

[0094] In summary, the photovoltaic power prediction method based on multi-stage temporal feature mining designed in this invention first extracts features from historical photovoltaic power sequences using an LSTM model to obtain power temporal features; then, it extracts features from historical satellite cloud image sequences using a 3DCNN model to obtain cloud image spatiotemporal dynamic features; finally, it fuses the power temporal features and cloud image spatiotemporal dynamic features based on a residual cross-attention mechanism, reducing the redundancy of the fused features and improving the usability of feature information to obtain multimodal fused features; and then, it further mines the temporal characteristics in the fused features using a Transformer model to obtain deep multimodal temporal features. Experimental results show that this invention can improve the accuracy of photovoltaic power prediction, provide support for the safe and stable operation of the power grid, and has certain engineering practical value.

[0095] Based on the same technical concept as the method embodiments, the present invention also provides a photovoltaic power prediction system based on multi-stage time-series feature mining, including a power feature extraction module, an image feature extraction module, a feature fusion module, a deep feature extraction module, and a time-series prediction module, for executing the above-described photovoltaic power prediction method based on multi-stage time-series feature mining.

[0096] The power feature extraction module extracts features from historical photovoltaic power sequences using an LSTM model to obtain power time-series features;

[0097] The image feature extraction module extracts features from historical satellite cloud image sequences using a 3DCNN model to obtain the spatiotemporal dynamic features of the cloud images;

[0098] The feature fusion module fuses power time-series features and cloud map spatiotemporal dynamic features based on the residual cross-attention mechanism to obtain multimodal fusion features;

[0099] The deep feature extraction module uses the Transformer model to further mine the temporal characteristics in the fused features to obtain deep multimodal temporal features;

[0100] The time-series prediction module takes power time-series characteristics, cloud map spatiotemporal dynamic characteristics, and deep multimodal time-series characteristics as inputs, and outputs photovoltaic power results for the next 1 to 4 hours through a multilayer perceptron.

[0101] It should be understood that the photovoltaic power prediction system based on multi-stage time-series feature mining in the embodiments of the present invention can realize all the technical solutions in the above method embodiments. The functions of each functional module can be specifically implemented according to the methods in the above method embodiments. The specific implementation process can be referred to the relevant descriptions in the above embodiments, which will not be repeated here.

[0102] The present invention also provides a computer device, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, wherein when the programs are executed by the processors, they implement the steps of the photovoltaic power prediction method based on multi-stage time-series feature mining as described above.

[0103] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the photovoltaic power prediction method based on multi-stage time-series feature mining as described above.

[0104] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus (systems), computer devices, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0105] This invention is described with reference to a flowchart of a method according to embodiments of the invention. It should be understood that each step in the flowchart and combinations thereof can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the process. Figure 1 A device for a function specified in one or more processes.

[0106] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 The function specified in one or more processes.

[0107] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 Steps of a specified function in one or more processes.

[0108] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A photovoltaic power prediction method based on multi-stage time-series feature mining, characterized in that, Includes the following steps: Step 1: Extract features from historical photovoltaic power sequences using an LSTM model to obtain power time-series features. The process is as follows: Among them, F power For power timing characteristics, f LSTM For the LSTM model, m is the number of LSTM layers used, [p1, p2, ..., p n [ ] represents the historical photovoltaic power sequence, where n is the length of the time series; Step 2: Extract features from historical satellite cloud image sequences using a 3DCNN model to obtain the spatiotemporal dynamic features of the cloud images. This process is represented as: F image =f 3DCNN ([i1,i2,...,i n ]); Among them, F image For the spatiotemporal dynamic characteristics of cloud maps, f 3DCNN For a 3DCNN model, [i1,i2,...,i n [This refers to] a sequence of historical satellite cloud images; 3DCNN is expressed using the following formula: h (l) =f (l) (w (l) *h (l-1) +b (l) ); * indicates a convolution operation, w (l) b represents the weights of the l-th convolutional layer. (l) f is the bias of the l-th convolutional layer. (l) h is the activation function of the l-th convolutional layer. (l) h (l-1) These are the outputs of the l-th and (l-1)-th convolutional layers, respectively; the process is represented as follows: Where Y is the output feature map, [x,y,z] is the position index in the output feature map, g is the input feature map for the pooling operation, M, N, and O represent the dimensions of the pooling window in the width, height, and depth directions, respectively, and S d S h S w These are the movement steps of the pooling window in the width, height, and depth directions, respectively, and m, n, and o are the index offsets of the pooling window at the current position. Step 3: Based on the residual cross-attention mechanism, feature fusion is performed on the power time series features and the spatiotemporal dynamic features of the cloud map to obtain multimodal fusion features. The specific process is as follows: Among them, F fusion For multimodal fusion features, RS-Attention is a residual cross-attention mechanism. Q, K, and V are the query vector, key vector, and value vector, respectively. Q is obtained through nonlinear mapping of power temporal features, and K and V are obtained through nonlinear mapping of cloud map spatiotemporal dynamic features. k Given the dimension of the key vector, softmax is used to calculate the weights of the cross-attention mechanism; Step 4: Utilize the Transformer model to further mine the temporal characteristics in the fused features, obtaining deep multimodal temporal features. The specific process is as follows: Among them, F final For deep multimodal temporal features, f Transformer For the Transformer model, For each historical moment, n represents the multimodal fusion feature, where n is the dimension of the fusion feature. Step 5: The prediction results are output by a multilayer perceptron to predict the photovoltaic power for the next 1 to 4 hours. Based on the power time series characteristics, the spatiotemporal dynamic characteristics of the cloud map, and the deep multimodal time series characteristics, relearning is performed to obtain the predicted photovoltaic power result y: y=f MLP ([F power ,F image ,F final ]); Among them, f MLP This is a multilayer perceptron model.

2. The photovoltaic power prediction method based on multi-stage time-series feature mining according to claim 1, characterized in that: In step one, LSTM is expressed by the following formula: i t =σ(W xi *x t +W hi *h t-1 +b i ) f t =σ(W xf *x t +W hf *h t-1 +b f ) g t =σ(W xg *x t +W hg *h t-1 +b g ) Where, x t h t c t These represent the input value, output value, and cell state of the LSTM at time t, respectively; h t-1 Let i be the output value of the LSTM at time t-1. t f t g t These are the input gate, forget gate, and output gate, used to control the flow of information; W xi W hi b i W xf W hf b f W xc W hc b c W xg W hg b g All represent learnable parameters, tanh is the hyperbolic tangent function, σ(·) is the activation function, and "*" represents ordinary matrix multiplication. It represents the Hadamardi (or Hadama) stack.

3. A photovoltaic power prediction system based on multi-stage time-series feature mining, characterized in that, It includes a power feature extraction module, an image feature extraction module, a feature fusion module, a deep feature extraction module, and a time series prediction module, and is used to execute the photovoltaic power prediction method based on multi-stage time series feature mining as described in any one of claims 1-2.

4. The photovoltaic power prediction system based on multi-stage time-series feature mining according to claim 3, characterized in that: Power feature extraction module: Extracts features from historical photovoltaic power sequences using an LSTM model to obtain power time-series features; The image feature extraction module extracts features from historical satellite cloud image sequences using a 3DCNN model to obtain the spatiotemporal dynamic features of the cloud images; The feature fusion module fuses power time-series features and cloud map spatiotemporal dynamic features based on the residual cross-attention mechanism to obtain multimodal fusion features; The deep feature extraction module uses the Transformer model to further mine the temporal characteristics in the fused features to obtain deep multimodal temporal features; The time-series prediction module takes power time-series characteristics, cloud map spatiotemporal dynamic characteristics, and deep multimodal time-series characteristics as inputs, and outputs photovoltaic power prediction results for the next 1 to 4 hours through a multilayer perceptron.

5. A computer device, characterized in that, include: One or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, wherein when the programs are executed by the processors, they implement the steps of the photovoltaic power prediction method based on multi-stage time-series feature mining as described in any one of claims 1-2.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the photovoltaic power prediction method based on multi-stage time-series feature mining as described in any one of claims 1-2.

Images