Photovoltaic power generation power prediction method based on spatio-temporal feature fusion

Abstract

The invention relates to the technical field of photovoltaic power generation, and discloses a photovoltaic power generation power prediction method based on spatial-temporal feature fusion, which comprises the following steps: acquiring an original signal, and optimizing the modal number K and bandwidth constraint alpha of VMD by adopting a CPO algorithm; taking the modal number K and the bandwidth constraint alpha as input parameters of a VMD algorithm, and decomposing the original signal into a plurality of modal components; for the plurality of modal components, extracting spatial features and time features in a manner of combining ST-GCN and LSTM, and fusing the spatial features and the time features to obtain spatial-temporal features; and mapping the fused spatial-temporal characteristics to a photovoltaic power prediction space through a full connection layer and a regression layer, and outputting a prediction value. According to the photovoltaic power generation power prediction method based on spatial-temporal feature fusion provided by the invention, photovoltaic power prediction is carried out based on combination of CPO optimized VMD, ST-GCN and LSTM, a basis can be provided for a power grid dispatching decision, and the power operation cost is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power generation, and in particular to a photovoltaic power generation power prediction method based on spatiotemporal feature fusion. Background Art

[0002] With the rapid growth of energy consumption, the deteriorating environment, and the increasing global demand for renewable energy, new energy sources such as solar and wind energy are gaining worldwide attention to promote sustainable development. The rapid growth of photovoltaic installed capacity has led to the rapid development of photovoltaic power generation technology, with enormous potential for further development. However, complex and unpredictable weather conditions affect the output of photovoltaic power stations, affecting factors such as solar radiation and ambient temperature. The output power of photovoltaic power generation systems is highly volatile and random, posing certain safety risks to the grid connection of photovoltaic systems. The large-scale and distributed connection of photovoltaic power stations to the grid poses significant challenges to the planning, operation, scheduling, and control of power systems. Summary of the Invention

[0003] The present invention provides a photovoltaic power prediction method based on spatiotemporal feature fusion. The photovoltaic power prediction is performed based on CPO optimization VMD, ST-GCN and LSTM, which can provide a basis for grid dispatching decision-making and reduce power operation costs.

[0004] The present invention provides a photovoltaic power generation power prediction method based on spatiotemporal feature fusion, comprising:

[0005] Obtaining the original signal, and optimizing the key parameters of VMD using the CPO algorithm; wherein the key parameters include the mode number K and the bandwidth constraint α;

[0006] Using the modal number K and the bandwidth constraint α as input parameters of the VMD algorithm, the original signal is decomposed into multiple modal components;

[0007] For multiple modal components, spatial features and temporal features are extracted by combining ST-GCN and LSTM, and the spatial features and temporal features are fused to obtain spatiotemporal features;

[0008] The fused spatiotemporal features are mapped to the photovoltaic power prediction space through the fully connected layer and the regression layer, and the predicted value is output.

[0009] Furthermore, the step of optimizing the key parameters of VMD using the CPO algorithm includes:

[0010] Initialize the particle swarm: Randomly initialize the individual particles in the CPO optimization algorithm, set the initial position and velocity of each particle, and represent the number of modes K to be optimized and the bandwidth constraint α;

[0011] VMD decomposition and reconstruction error calculation: For each particle, perform VMD decomposition on the original signal using the parameters it represents, and calculate the reconstruction error;

[0012] Particle update and global search: According to the optimization rules of the CPO algorithm, the particle speed and position are continuously updated, and the optimal solution is gradually approached through global search. Through iterative optimization, it finally converges to the optimal solution of the VMD decomposition parameters.

[0013] Furthermore, the CPO algorithm accelerates the convergence speed through cyclic population reduction technology, and the algorithm mathematical model is:

[0014]

[0015] Where N represents the population size, T max Indicates the maximum number of cycles, N min represents the minimum number of individuals in the population of a new species, and t represents the current function evaluation;

[0016] The CPO algorithm includes two stages: global exploration and local development:

[0017] Global exploration phase: Crested porcupines use two defense strategies, visual and acoustic, to threaten predators. The mathematical model is:

[0018]

[0019] in, Indicates the position of the i-th individual iteration, τ1 represents a normally distributed random number, τ2 and τ3 are random numbers in the interval [0,1], is the optimal solution of the function at time t, is the position of the predator at time t, U1 is a binary vector consisting of 0 and 1, and r1 and r2 are random numbers between [1, N].

[0020] Local development stage: Crested porcupines attack predators through scent and physical attacks. The mathematical model is:

[0021]

[0022] Among them, γ t represents the defense factor, represents the odor diffusion factor, δ is the parameter that controls the search direction, α is the convergence speed factor, τ4 and τ5 are random numbers in the interval [0,1]. It is the inelastic collision force generated by an individual in the population attacking a predator.

[0023] Furthermore, the step of using the mode number K and the bandwidth constraint α as input parameters of the VMD algorithm to decompose the original signal into multiple modal components includes:

[0024] Signal decomposition: The original signal is decomposed into K modal components by VMD algorithm, and the IMF of each mode is obtained, including IMF1, IMF2, ...IMF k ; Each modal component contains information about the set frequency band of the signal;

[0025] Modal Image Generation: Generate a two-dimensional image for each modal component, and use visualization to further understand and analyze the frequency characteristics of the signal.

[0026] Furthermore, the original signal is decomposed into K modal components by the VMD algorithm to obtain the IMF of each mode, including IMF1, IMF2, ...IMF k In the step, VMD decomposes the input signal into K eigenmode components with limited bandwidth through alternating iterative updates through the variational constraint model; the variational constraint model of VMD is:

[0027]

[0028] Among them, {u i} are k eigenmodal components, {ω i} is the center frequency of k eigenmode components, δ(t) represents the Dirac function, u i (t) represents the value of the component at time t, Indicates u i (t) Signal after Hilbert transform.

[0029] Furthermore, the step of extracting spatial features and temporal features by combining ST-GCN with LSTM for multiple modal components, and fusing the spatial features and temporal features to obtain spatiotemporal features includes:

[0030] Spatial feature extraction: The spatial dependency information in the data matrix is ​​extracted through the graph convolution layer in ST-GCN;

[0031] Temporal feature extraction: Use LSTM networks to model time series data and capture long-term and short-term temporal dependencies;

[0032] Feature fusion: The time feature and the space feature are weighted and summed to obtain the fused space-time feature. The feature fusion method adopts weighted fusion, and the fusion formula is:

[0033] y=αI s +(1-α)I t

[0034] Among them, y represents the fusion result, I s Represents spatial features, I trepresents the time feature, and α is the weight.

[0035] Furthermore, in the step of extracting spatial dependency information from the data matrix through the graph convolution layer in ST-GCN, ST-GCN is composed of a stack of multiple spatiotemporal convolution modules, and the spatial features in the input are extracted through the graph convolution layer. The graph convolution is the product of the input signal x and the graph kernel Θ:

[0036] Θ*x=Θ(UΛU T )x=UΘ(Λ)U T x

[0037] Where “*” represents the convolution operator, U∈R n×n is the normalized graph Laplace matrix, Λ is the diagonal matrix composed of the eigenvalues ​​of the Laplace matrix, where the graph Laplace matrix L is:

[0038]

[0039] Among them, I is the identity matrix, A is the node adjacency matrix, and D is the weighted degree matrix of the graph, which represents the weight between nodes;

[0040] The convolution kernel Θ is approximately fitted by Chebyshev polynomials to reduce the complexity of calculation, then:

[0041]

[0042] Among them, K is the size of the graph convolution kernel, k represents the order of Chebyshev polynomial, θ k are the Chebyshev polynomial coefficients, is the eigenvector matrix after L scaling, and λ max is the maximum eigenvalue of L;

[0043] The recursive formula of Chebyshev polynomial is:

[0044] T k (i) = 2iT k-1 (i)-T k-2 (i)

[0045] Among them, T0(i)=1, T1(i)=i, then the graph convolution formula can be expressed as:

[0046] Furthermore, the step of mapping the fused spatiotemporal features to the photovoltaic power prediction space through the fully connected layer and the regression layer and outputting the predicted value includes:

[0047] The fused spatiotemporal features are processed through the fully connected layer to extract deeper features and further enhance the expressive power of the model;

[0048] The extracted features are mapped to the photovoltaic power prediction result space through the regression layer to generate the final photovoltaic power prediction value. At the same time, the prediction results of each modal component are weighted and superimposed to obtain a more accurate prediction result.

[0049] The present invention also provides a photovoltaic power generation prediction device based on spatiotemporal feature fusion, comprising:

[0050] The optimization module is used to obtain the original signal and optimize the key parameters of VMD using the CPO algorithm; wherein the key parameters include the mode number K and the bandwidth constraint α

[0051] A decomposition module, configured to use the mode number K and the bandwidth constraint α as input parameters of a VMD algorithm to decompose the original signal into multiple modal components;

[0052] The extraction module is used to extract spatial features and temporal features from multiple modal components by combining ST-GCN and LSTM, and fuse the spatial features and temporal features to obtain spatiotemporal features;

[0053] The prediction module is used to map the fused spatiotemporal features to the photovoltaic power prediction space through a fully connected layer and a regression layer, and output a predicted value.

[0054] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0055] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0056] The beneficial effects of the present invention are:

[0057] The present invention uses the CPO algorithm to optimize the VMD modal number K and bandwidth constraint α, and uses these as input parameters of the VMD algorithm to decompose the original signal into multiple modal components. The spatiotemporal features are then extracted and fused by combining ST-GCN with LSTM. Finally, the fused spatiotemporal features are mapped to the photovoltaic power prediction space through a fully connected layer and a regression layer, and the predicted value is output. Through multiple stages of feature extraction, data processing, and fusion, the prediction accuracy and model robustness are significantly improved, providing a basis for grid dispatching decisions, reducing power operating costs, and providing strong technical support for the spatiotemporal complementary coordinated control of multiple energy sources. This is of great significance for ensuring the safety and stability of the power system and promoting the optimized operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1Schematic diagram of a method flow according to an embodiment of the present invention.

[0059] Figure 2 FIG. 1 is a schematic diagram of the device structure according to an embodiment of the present invention.

[0060] Figure 3 Schematic diagram of the internal structure of a computer device according to an embodiment of the present invention.

[0061] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0062] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0063] like Figure 1 As shown, the present invention provides a photovoltaic power generation prediction method based on spatiotemporal feature fusion. Through multiple stages of feature extraction, data processing and fusion, the prediction accuracy and model robustness are significantly improved. The specific steps are as follows:

[0064] S1. Obtain the original signal and use the CPO algorithm to optimize the key parameters of VMD; wherein the key parameters include the mode number K and the bandwidth constraint α.

[0065] S101. Initialize the particle swarm: Randomly initialize the individual particles in the CPO optimization algorithm, set the initial position and velocity of each particle, and represent the number of modes K to be optimized and the bandwidth constraint α;

[0066] S102. VMD decomposition and reconstruction error calculation: For each particle, perform VMD decomposition on the original signal using the parameters it represents, and calculate the reconstruction error; the reconstruction error reflects the effect of VMD decomposition.

[0067] S103, particle update and global search: According to the optimization rules of the CPO algorithm, the speed and position of the particles are continuously updated, and the optimal solution is gradually approached through global search. Through iterative optimization, it finally converges to the optimal solution of the VMD decomposition parameters, reducing the noise component and enhancing the characteristic expression of the signal.

[0068] The CPO algorithm mainly accelerates the convergence speed through cyclic population reduction technology. The mathematical model of the algorithm is:

[0069]

[0070] Where N represents the population size, T max Indicates the maximum number of cycles, N min represents the minimum number of individuals in the population of a new species, and t represents the current function evaluation;

[0071] The CPO algorithm simulates the unique defense method adopted by crested porcupines in the face of predators. It mainly consists of two stages: global exploration and local exploitation:

[0072] 1) Global exploration phase: Crested porcupines use two defense strategies, visual and acoustic, to threaten predators. The mathematical model is:

[0073]

[0074] in, Indicates the position of the i-th individual iteration, τ1 represents a normally distributed random number, τ2 and τ3 are random numbers in the interval [0,1], is the optimal solution of the function at time t, is the position of the predator at time t, U1 is a binary vector consisting of 0 and 1, and r1 and r2 are random numbers between [1, N].

[0075] 2) Local development stage: Crested porcupines attack predators through scent and physical attacks. The mathematical model is:

[0076]

[0077] Among them, γ t represents the defense factor, represents the odor diffusion factor, δ is the parameter that controls the search direction, α is the convergence speed factor, τ4 and τ5 are random numbers in the interval [0,1]. It is the inelastic collision force generated by an individual in the population attacking a predator.

[0078] S2. Use the optimal modal number K and bandwidth constraint α obtained by CPO optimization as the input parameters of the VMD algorithm to decompose the original signal into multiple modal components. These modal components represent the components of the original signal in different frequency domains. The specific steps are as follows:

[0079] S201, signal decomposition: Decompose the original signal into K modal components using the VMD algorithm, and obtain the IMF (intrinsic mode function) of each mode, such as IMF1, IMF2, ...IMF k , each modal component contains the information of a certain frequency band of the signal.

[0080] VMD decomposition uses a variational constraint model to decompose the input signal into K eigenmode components with limited bandwidth through alternating iterative updates. The variational constraint model of VMD is:

[0081]

[0082] Among them, {u i} are k eigenmodal components, {ωi} is the center frequency of k eigenmode components, δ(t) represents the Dirac function, u i (t) represents the value of the component at time t, Indicates u i (t) Signal after Hilbert transform.

[0083] In solving the variational problem, the Lagrange operator λ and the penalty factor α are introduced to transform the constrained variational problem into an unconstrained problem. The Lagrange augmented function is:

[0084]

[0085] By alternating multiplication method, the component u k (t), center frequency ω k And the Lagrange factor is continuously updated iteratively:

[0086]

[0087] S202, Modal Image Generation: Generate a two-dimensional image for each modal component, using visualization to further understand and analyze the frequency characteristics of the signal. These modal components can intuitively display the characteristics of each frequency band of the signal, facilitating subsequent feature extraction and modeling.

[0088] S3. To accurately capture the spatiotemporal variations in photovoltaic power generation, we combined the ST-GCN (Spatiotemporal Graph Convolutional Network) with the LSTM to extract spatiotemporal features. The ST-GCN effectively extracts spatial dependencies, while the LSTM captures temporal features and weightedly integrates spatiotemporal information. The specific steps are as follows:

[0089] S301, Spatial Feature Extraction (ST-GCN): The graph convolutional layer in ST-GCN extracts spatial dependency information from the data matrix. During photovoltaic power generation, there is spatial correlation between PV plants in different locations. ST-GCN effectively captures this spatial interaction and enhances spatial feature representation.

[0090] ST-GCN is composed of multiple spatiotemporal convolution modules stacked together, and the spatial features in the input are extracted through the graph convolution layer. The graph convolution is the product of the input signal x and the graph kernel Θ:

[0091] Θ*x=Θ(UΛU T )x=UΘ(Λ)U T x

[0092] Where “*” represents the convolution operator, U∈R n×n is the normalized graph Laplace matrix, Λ is the diagonal matrix composed of the eigenvalues ​​of the Laplace matrix, where the graph Laplace matrix L is:

[0093]

[0094] Among them, I is the identity matrix, A is the node adjacency matrix, and D is the weighted degree matrix of the graph, which represents the weights between nodes.

[0095] The convolution kernel Θ is approximately fitted by Chebyshev polynomials to reduce the complexity of calculation, then:

[0096]

[0097] Among them, K is the size of the graph convolution kernel, k represents the order of Chebyshev polynomial, θ k are the Chebyshev polynomial coefficients, is the eigenvector matrix after L scaling, and λ max is the largest eigenvalue of L.

[0098] The recursive formula of Chebyshev polynomial is:

[0099] T k (i) = 2iT k-1 (i)-T k-2 (i)

[0100] Among them, T0(i)=1, T1(i)=i, then the graph convolution formula can be expressed as:

[0101]

[0102] S302, Temporal Feature Extraction (LSTM): Use the LSTM network to model time series data and capture both long-term and short-term temporal dependencies. LSTM has a strong memory capacity in time series modeling and can effectively process the characteristics of different time nodes in the photovoltaic power generation process.

[0103] S303, feature fusion: The time feature and the space feature are weighted and summed to obtain the fused space-time feature; the feature fusion method adopts weighted fusion, and the fusion formula is:

[0104] y=αI s +(1-α)I t

[0105] Among them, y represents the fusion result, I s Represents spatial features, I t represents the time feature, and α is the weight.

[0106] S4, map the fused spatiotemporal features to the photovoltaic power prediction space through the fully connected layer and the regression layer, and output the predicted value. The specific steps are as follows:

[0107] S401, fully connected layer: The fused spatiotemporal features are processed through the fully connected layer to extract deeper features and further enhance the expressive power of the model.

[0108] S402, Regression Layer: The regression layer maps the extracted features to the PV power prediction result space to generate the final PV power prediction value. At the same time, the prediction results of each modal component are weighted and superimposed to obtain a more accurate prediction result.

[0109] The present invention uses the CPO algorithm to optimize the VMD modal number K and bandwidth constraint α, and uses these as input parameters of the VMD algorithm to decompose the original signal into multiple modal components. The spatiotemporal features are then extracted and fused by combining ST-GCN with LSTM. Finally, the fused spatiotemporal features are mapped to the photovoltaic power prediction space through a fully connected layer and a regression layer, and the predicted value is output. Through multiple stages of feature extraction, data processing, and fusion, the prediction accuracy and model robustness are significantly improved, providing a basis for grid dispatching decisions, reducing power operating costs, and providing strong technical support for the spatiotemporal complementary coordinated control of multiple energy sources. This is of great significance for ensuring the safety and stability of the power system and promoting the optimized operation of the power grid.

[0110] like Figure 2 As shown, the present invention also provides a photovoltaic power prediction device based on spatiotemporal feature fusion, comprising:

[0111] Optimization module 1 is used to obtain the original signal and optimize the key parameters of VMD using the CPO algorithm; wherein the key parameters include the mode number K and the bandwidth constraint α

[0112] Decomposition module 2, configured to use the mode number K and bandwidth constraint α as input parameters of a VMD algorithm to decompose the original signal into multiple modal components;

[0113] Extraction module 3 is used to extract spatial features and temporal features from multiple modal components by combining ST-GCN and LSTM, and fuse the spatial features and temporal features to obtain spatiotemporal features;

[0114] The prediction module 4 is used to map the fused spatiotemporal features to the photovoltaic power prediction space through a fully connected layer and a regression layer, and output a predicted value.

[0115] In one embodiment, the optimization module 1 includes:

[0116] Initialization unit, used to initialize the particle swarm: randomly initialize the individual particles in the CPO optimization algorithm, set the initial position and velocity of each particle, representing the number of modes K to be optimized and the bandwidth constraint α;

[0117] The calculation unit is used for VMD decomposition and reconstruction error calculation: for each particle, the original signal is decomposed by VMD using the parameters it represents, and the reconstruction error is calculated;

[0118] The update unit is used for particle update and global search: according to the optimization rules of the CPO algorithm, the particle speed and position are continuously updated, and the optimal solution is gradually approached through global search. Through iterative optimization, it finally converges to the optimal solution of the VMD decomposition parameters.

[0119] In one embodiment, in the update unit, the CPO algorithm accelerates convergence by using a cyclic population reduction technique. The mathematical model of the algorithm is:

[0120]

[0121] Where N represents the population size, T max Indicates the maximum number of cycles, N min represents the minimum number of individuals in the population of a new species, and t represents the current function evaluation;

[0122] The CPO algorithm includes two stages: global exploration and local development:

[0123] Global exploration phase: Crested porcupines use two defense strategies, visual and acoustic, to threaten predators. The mathematical model is:

[0124]

[0125] in, Indicates the position of the i-th individual iteration, τ1 represents a normally distributed random number, τ2 and τ3 are random numbers in the interval [0,1], is the optimal solution of the function at time t, is the position of the predator at time t, U1 is a binary vector consisting of 0 and 1, and r1 and r2 are random numbers between [1, N].

[0126] Local development stage: Crested porcupines attack predators through scent and physical attacks. The mathematical model is:

[0127]

[0128] Among them, γ t represents the defense factor, represents the odor diffusion factor, δ is the parameter that controls the search direction, α is the convergence speed factor, τ4 and τ5 are random numbers in the interval [0,1]. It is the inelastic collision force generated by an individual in the population attacking a predator.

[0129] In one embodiment, the decomposition module 2 includes:

[0130] Decomposition unit, used for signal decomposition: decompose the original signal into K modal components through the VMD algorithm, and obtain the IMF of each mode, including IMF1, IMF2, ...IMF k ; Each modal component contains information about the set frequency band of the signal;

[0131] The generation unit is used for modal image generation: a two-dimensional image is generated for each modal component, and the frequency characteristics of the signal can be further understood and analyzed through visualization.

[0132] In one embodiment, in the decomposition unit, VMD decomposition decomposes the input signal into K eigenmode components with limited bandwidth through alternating iterative updates using a variational constraint model; the variational constraint model of VMD is:

[0133]

[0134] Among them, {u i} are k eigenmodal components, {ω i} is the center frequency of k eigenmode components, δ(t) represents the Dirac function, u i (t) represents the value of the component at time t, Indicates u i (t) Signal after Hilbert transform.

[0135] In one embodiment, the extraction module 3 includes:

[0136] The first extraction unit is used for spatial feature extraction: the spatial dependency information in the data matrix is ​​extracted through the graph convolution layer in ST-GCN;

[0137] The second extraction unit is used for temporal feature extraction: LSTM networks are used to model time series data and capture long-term and short-term temporal dependencies.

[0138] The fusion unit is used for feature fusion: the time feature and the spatial feature are weighted and summed to obtain the fused spatiotemporal feature. The feature fusion method adopts weighted fusion, and the fusion formula is:

[0139] y=αI s +(1-α)I t

[0140] Among them, y represents the fusion result, I s Represents spatial features, I t represents the time feature, and α is the weight.

[0141] In one embodiment, in the first extraction unit, ST-GCN is composed of a stack of multiple spatiotemporal convolution modules, and the spatial features in the input are extracted through the graph convolution layer. The graph convolution is the product of the input signal x and the graph kernel Θ:

[0142] Θ*x=Θ(UΛU T )x=UΘ(Λ)U T x

[0143] Where “*” represents the convolution operator, U∈R n×n is the normalized graph Laplace matrix, Λ is the diagonal matrix composed of the eigenvalues ​​of the Laplace matrix, where the graph Laplace matrix L is:

[0144]

[0145] Among them, I is the identity matrix, A is the node adjacency matrix, and D is the weighted degree matrix of the graph, which represents the weight between nodes;

[0146] The convolution kernel Θ is approximately fitted by Chebyshev polynomials to reduce the complexity of calculation, then:

[0147]

[0148] Among them, K is the size of the graph convolution kernel, k represents the order of Chebyshev polynomial, θ k are the Chebyshev polynomial coefficients, is the eigenvector matrix after L scaling, and λ max is the maximum eigenvalue of L;

[0149] The recursive formula of Chebyshev polynomial is:

[0150] T k (i) = 2iT k-1 (i)-T k-2 (i)

[0151] Among them, T0(i)=1, T1(i)=i, then the graph convolution formula can be expressed as:

[0152] In one embodiment, the prediction module 4 includes:

[0153] The processing unit is used to process the fused spatiotemporal features through the fully connected layer to extract deeper features and further enhance the expressiveness of the model;

[0154] The mapping unit is used to map the extracted features to the photovoltaic power prediction result space through the regression layer to generate the final photovoltaic power prediction value, and at the same time perform weighted superposition on the prediction results of each modal component to obtain a more accurate prediction result.

[0155] The above modules and units are used to execute the corresponding steps in the above photovoltaic power generation prediction method based on spatiotemporal feature fusion. The specific implementation method thereof is described in the above method embodiment and will not be repeated here.

[0156] like Figure 3 As shown, the present invention also provides a computer device, which can be a server, and its internal structure can be as follows Figure 3 As shown. The computer device includes a processor, memory, network interface and database connected via a system bus. The processor of the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store all data required for the process of the photovoltaic power generation prediction method based on the fusion of spatiotemporal features. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the photovoltaic power generation prediction method based on the fusion of spatiotemporal features is implemented.

[0157] Those skilled in the art will understand that Figure 3 The structure shown in is merely a block diagram of a portion of the structure related to the present application solution and does not constitute a limitation on the computer device to which the present application solution is applied.

[0158] An embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, any one of the above-mentioned photovoltaic power generation prediction methods based on spatiotemporal feature fusion is implemented.

[0159] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM).

[0160] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0161] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A photovoltaic power generation prediction method based on spatiotemporal feature fusion, characterized in that: include: Obtaining the original signal, and optimizing the key parameters of VMD using the CPO algorithm; wherein the key parameters include the mode number K and the bandwidth constraint α; Using the modal number K and the bandwidth constraint α as input parameters of the VMD algorithm, the original signal is decomposed into multiple modal components; For multiple modal components, spatial features and temporal features are extracted by combining ST-GCN and LSTM, and the spatial features and temporal features are fused to obtain spatiotemporal features; The fused spatiotemporal features are mapped to the photovoltaic power prediction space through the fully connected layer and the regression layer, and the predicted value is output.

2. The photovoltaic power generation prediction method based on spatiotemporal feature fusion according to claim 1 is characterized in that: The step of optimizing the key parameters of VMD using the CPO algorithm includes: Initialize the particle swarm: Randomly initialize the individual particles in the CPO optimization algorithm, set the initial position and velocity of each particle, and represent the number of modes K to be optimized and the bandwidth constraint α; VMD decomposition and reconstruction error calculation: For each particle, perform VMD decomposition on the original signal using the parameters it represents, and calculate the reconstruction error; Particle update and global search: According to the optimization rules of the CPO algorithm, the particle speed and position are continuously updated, and the optimal solution is gradually approached through global search. Through iterative optimization, it finally converges to the optimal solution of the VMD decomposition parameters.

3. The photovoltaic power generation prediction method based on spatiotemporal feature fusion according to claim 2 is characterized in that: The CPO algorithm accelerates the convergence speed by using cyclic population reduction technology. The mathematical model of the algorithm is: Where N represents the population size, T max Indicates the maximum number of cycles, N min represents the minimum number of individuals in the population of a new species, and t represents the current function evaluation; The CPO algorithm includes two stages: global exploration and local development: Global exploration phase: Crested porcupines use two defense strategies, visual and acoustic, to threaten predators. The mathematical model is: in, Indicates the position of the i-th individual iteration, τ1 represents a normally distributed random number, τ2 and τ3 are random numbers in the interval [0,1], is the optimal solution of the function at time t, is the position of the predator at time t, U1 is a binary vector consisting of 0 and 1, and r1 and r2 are random numbers between [1, N]. Local development stage: Crested porcupines attack predators through scent and physical attacks. The mathematical model is: Among them, γ t represents the defense factor, represents the odor diffusion factor, δ is the parameter that controls the search direction, α is the convergence speed factor, τ4 and τ5 are random numbers in the interval [0,1]. It is the inelastic collision force generated by an individual in the population attacking a predator.

4. The photovoltaic power generation prediction method based on spatiotemporal feature fusion according to claim 1 is characterized in that: The step of using the mode number K and the bandwidth constraint α as input parameters of the VMD algorithm to decompose the original signal into multiple modal components includes: Signal decomposition: The original signal is decomposed into K modal components by VMD algorithm, and the IMF of each mode is obtained, including IMF1, IMF2, ...IMF k ; Each modal component contains information about the set frequency band of the signal; Modal Image Generation: Generate a two-dimensional image for each modal component, and use visualization to further understand and analyze the frequency characteristics of the signal.

5. The photovoltaic power generation prediction method based on spatiotemporal feature fusion according to claim 4 is characterized in that: The VMD algorithm is used to decompose the original signal into K modal components, and the IMF of each mode is obtained, including IMF1, IMF2, ...IMF k In the step, VMD decomposes the input signal into K eigenmode components with limited bandwidth through alternating iterative updates through the variational constraint model; the variational constraint model of VMD is: Among them, {u i } are k eigenmodal components, {ω i } is the center frequency of k eigenmode components, δ(t) represents the Dirac function, u i (t) represents the value of the component at time t, Indicates u i (t) Signal after Hilbert transform.

6. The photovoltaic power generation prediction method based on spatiotemporal feature fusion according to claim 1 is characterized in that: The step of extracting spatial features and temporal features by combining ST-GCN with LSTM for multiple modal components, and fusing the spatial features and temporal features to obtain spatiotemporal features includes: Spatial feature extraction: The graph convolution layer in ST-GCN is used to extract spatial dependency information from the data matrix. Temporal feature extraction: Use LSTM networks to model time series data and capture long-term and short-term temporal dependencies; Feature fusion: The time feature and the space feature are weighted and summed to obtain the fused space-time feature. The feature fusion method adopts weighted fusion, and the fusion formula is: y=αI s +(1-a)I t Among them, y represents the fusion result, I s Represents spatial features, I t represents the time feature, and α is the weight.

7. The photovoltaic power generation prediction method based on spatiotemporal feature fusion according to claim 6 is characterized in that: In the step of extracting spatial dependency information from the data matrix through the graph convolution layer in ST-GCN, ST-GCN is composed of multiple spatiotemporal convolution modules stacked together, and the spatial features in the input are extracted through the graph convolution layer. The graph convolution is the product of the input signal x and the graph kernel Θ: Θ*x=Θ(UΛU T )x=UΘ(Λ)U T x Where "*" represents the convolution operator, U∈R n×n is the normalized graph Laplace matrix, Λ is the diagonal matrix composed of the eigenvalues ​​of the Laplace matrix, where the graph Laplace matrix L is: Among them, I is the identity matrix, A is the node adjacency matrix, and D is the weighted degree matrix of the graph, which represents the weight between nodes; The convolution kernel Θ is approximately fitted by Chebyshev polynomials to reduce the complexity of calculation. Then: Among them, K is the size of the graph convolution kernel, k represents the order of Chebyshev polynomial, θ k are the Chebyshev polynomial coefficients, is the eigenvector matrix after L scaling, and λ max is the maximum eigenvalue of L; The recursive formula of Chebyshev polynomial is: T k (i)=2iT k-1 (i)-T k-2 (i) Among them, T0(i)=1, T1(i)=i, then the graph convolution formula can be expressed as:

8. The photovoltaic power generation prediction method based on spatiotemporal feature fusion according to claim 1 is characterized in that: The step of mapping the fused spatiotemporal features to a photovoltaic power prediction space through a fully connected layer and a regression layer and outputting a predicted value includes: The fused spatiotemporal features are processed through the fully connected layer to extract deeper features and further enhance the expressive power of the model; The extracted features are mapped to the photovoltaic power prediction result space through the regression layer to generate the final photovoltaic power prediction value. At the same time, the prediction results of each modal component are weighted and superimposed to obtain a more accurate prediction result.

9. A photovoltaic power generation prediction device based on spatiotemporal feature fusion, characterized in that: include: The optimization module is used to obtain the original signal and optimize the key parameters of VMD using the CPO algorithm; wherein the key parameters include the mode number K and the bandwidth constraint α A decomposition module, configured to use the mode number K and the bandwidth constraint α as input parameters of a VMD algorithm to decompose the original signal into multiple modal components; The extraction module is used to extract spatial features and temporal features from multiple modal components by combining ST-GCN and LSTM, and fuse the spatial features and temporal features to obtain spatiotemporal features; The prediction module is used to map the fused spatiotemporal features to the photovoltaic power prediction space through a fully connected layer and a regression layer, and output a predicted value.

10. 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 8 are implemented.

Images