Short-term photovoltaic power prediction method and device, medium and product

Abstract

The invention discloses a short-term photovoltaic power prediction method and device, a medium and a product, and relates to the field of photovoltaic prediction, and the method comprises the steps: obtaining a historical photovoltaic power time sequence of a photovoltaic power station; performing modal decomposition on the historical photovoltaic power time sequence to obtain a high-frequency component and a low-frequency component; performing feature extraction on the high-frequency component by adopting a high-frequency component prediction model to obtain a high-frequency feature vector; the high-frequency component prediction model is constructed based on a time sequence convolutional network and an Informer model; performing feature extraction on the low-frequency component by adopting a low-frequency component prediction model to obtain a low-frequency feature vector; the low-frequency component prediction model is constructed based on an xLstm model; and determining a photovoltaic power generation power prediction result in a future set time period according to the high-frequency feature vector and the low-frequency feature vector. According to the invention, the prediction precision and efficiency of the photovoltaic power generation power are improved.

Description

Technical Field

[0001] This application relates to the field of photovoltaic forecasting, and in particular to a method, equipment, medium, and product for short-term photovoltaic power forecasting. Background Technology

[0002] In its current development, photovoltaic (PV) power generation has gradually become a crucial step in the green transformation. However, the output power of PV power generation is highly random and volatile due to the influence of weather conditions and other factors. Mainstream PV forecasting methods, such as CNN (Convolutional Neural Networks), LSTM (Long Short-Term Memory), and Transformer neural network algorithms, have relatively low prediction accuracy. Furthermore, combining these prediction models with a single decomposition model makes it difficult to further improve the prediction accuracy and decomposition effectiveness, and can lead to information leakage and pseudo-modal problems. Summary of the Invention

[0003] The purpose of this application is to provide a method, device, medium, and product for short-term photovoltaic power prediction, which can improve the prediction accuracy and efficiency of photovoltaic power generation.

[0004] To achieve the above objectives, this application provides the following solution:

[0005] In a first aspect, this application provides a short-term photovoltaic power prediction method, including:

[0006] Obtain the historical photovoltaic power time series of photovoltaic power plants;

[0007] Modal decomposition was performed on the historical photovoltaic power time series to obtain high-frequency and low-frequency components;

[0008] The high-frequency component prediction model is used to extract features from the high-frequency components to obtain high-frequency feature vectors; the high-frequency component prediction model is constructed based on temporal convolutional networks and the Informer model.

[0009] The low-frequency component is used to extract features from the low-frequency component to obtain a low-frequency feature vector; the low-frequency component prediction model is constructed based on the xLstm model.

[0010] Based on the high-frequency feature vector and the low-frequency feature vector, the predicted photovoltaic power generation within a set future time period is determined.

[0011] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described short-term photovoltaic power prediction method.

[0012] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described short-term photovoltaic power prediction method.

[0013] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned short-term photovoltaic power prediction method.

[0014] According to the specific embodiments provided in this application, this application has the following technical effects:

[0015] This application provides a method, device, medium, and product for short-term photovoltaic power prediction. By performing modal decomposition on historical photovoltaic power time series and re-superimposing to obtain high-frequency and low-frequency components, and by modeling the high-frequency and low-frequency components separately, the method solves the problem that the prediction effect of traditional decomposition models will decrease in different seasons, thereby improving the prediction accuracy and efficiency of photovoltaic power generation. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is an application environment diagram of a short-term photovoltaic power prediction method according to an embodiment of this application;

[0018] Figure 2 This is an overall flowchart of a short-term photovoltaic power prediction method provided in an embodiment of this application;

[0019] Figure 3 A detailed flowchart of a short-term photovoltaic power prediction method provided in an embodiment of this application;

[0020] Figure 4 This is a schematic diagram of a high-frequency component prediction model in one embodiment of this application;

[0021] Figure 5 This is a schematic diagram of a low-frequency component prediction model in one embodiment of this application;

[0022] Figure 6This is a comparison diagram of the entropy values ​​of the original sequence after variational mode decomposition in one embodiment of this application;

[0023] Figure 7 This is a diagram of the high and low frequency component sequences after re-overlay in one embodiment of this application;

[0024] Figure 8 This is a comparison chart of the prediction results of different models in spring according to one embodiment of this application;

[0025] Figure 9 This is a comparison chart of the prediction results of different models in summer in one embodiment of this application;

[0026] Figure 10 This is a comparison chart of the prediction results of different models in autumn in one embodiment of this application;

[0027] Figure 11 This is a comparison chart of the prediction results of different models in winter in one embodiment of this application;

[0028] Figure 12 This is a comparison chart of evaluation metrics for different models in one embodiment of this application;

[0029] Figure 13 This is a comparison chart of the mean absolute percentage error of different models in one embodiment of this application;

[0030] Figure 14 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0031] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0032] This application solves the problem that traditional decomposition models experience a decline in prediction performance in different seasons by performing a secondary decomposition on the historical photovoltaic power time series and re-superimposing it to obtain high-frequency and low-frequency components as inputs to the model, and by modeling the high-frequency and low-frequency components separately.

[0033] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0034] The short-term photovoltaic power prediction method provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on other servers. Terminal 102 can send the historical photovoltaic power time series of the photovoltaic power plant to server 104. After receiving the historical photovoltaic power time series, server 104 performs mode decomposition on the historical photovoltaic power time series, and uses high-frequency component prediction models and low-frequency component prediction models to extract features to determine the photovoltaic power generation prediction result for a future set time period. Server 104 can feed back the photovoltaic power generation prediction result for the future set time period to terminal 102. Furthermore, in some embodiments, the short-term photovoltaic power prediction method can also be implemented independently by server 104 or terminal 102.

[0035] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0036] In one exemplary embodiment, such as Figure 2 and Figure 3 As shown, a short-term photovoltaic power prediction method is provided. This method is executed by computer equipment, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 205.

[0037] Step 201: Obtain the historical photovoltaic power time series of the photovoltaic power plant.

[0038] Specifically, Pearson correlation analysis was performed on the factors influencing photovoltaic power to identify weakly correlated influencing factors. The values ​​of these weakly correlated influencing factors for the photovoltaic power plant within a set historical time period were obtained, resulting in a historical photovoltaic power time series.

[0039] The historical photovoltaic power time series includes the total radiation level, direct radiation level, global radiation level, air temperature, humidity, and actual photovoltaic power generation within a set historical period.

[0040] Because photovoltaic power is affected by multiple external factors such as solar irradiance, weather temperature, air humidity, global radiation levels, direct sunlight, and diffuse reflection, considering all characteristics would drastically increase the computational load of the model. Therefore, Pearson correlation analysis was used to screen out strongly correlated influencing factors, while the influence of weakly correlated features was not considered. This reduces the computational load of the model, improves the prediction accuracy, and prevents overfitting. The final analysis yielded a 15-minute time series of photovoltaic power from a specific photovoltaic power plant over two years.

[0041] The Pearson correlation coefficient ranges from -1 to 1. A higher correlation coefficient indicates a stronger association between the feature variables, while a lower correlation coefficient indicates a weaker association between the feature variables. The formula for calculating the correlation coefficient is:

[0042]

[0043] Where r(I) φ J φ ) for I φ With J φ Pearson correlation coefficient, I φ J represents the observed values ​​of the characteristic variable. φ For the observed values ​​of the dependent variable, The average value of the characteristic variable. This represents the average value of the dependent variable.

[0044] Step 202: Perform mode decomposition on the historical photovoltaic power time series to obtain high-frequency components and low-frequency components.

[0045] In this application, step 202 includes steps 21 to 24.

[0046] Step 21: Perform VMD (Variational Mode Decomposition) on the historical photovoltaic power time series to obtain multiple IMFs (Intrinsic Mode Functions). Each IMF has different frequency characteristics. By minimizing the bandwidth constraints of the mode solutions, VMD can better extract the intrinsic features of the input signal, thereby improving the expressive power of the historical photovoltaic power time series and thus improving the prediction accuracy of the model.

[0047] The steps of variational mode decomposition are as follows:

[0048] (1) The core of VMD is to decompose the historical photovoltaic power time series into several intrinsic mode functions by establishing a constrained optimization problem. First, the objective function of each subsequence is established:

[0049]

[0050] Where, m k (t) is the k-th eigenmode function, ω k Let be the center frequency of the k-th eigenmode function, δ(t) be the Dirac function, B be the number of eigenmode functions, t be time, and j be an imaginary number.

[0051] (2) VMD decomposes the signal by performing frequency domain transformation. To solve the decomposition problem, VMD represents each eigenmode function through frequency modulation:

[0052]

[0053] Among them, a k (t) is the complex envelope function, representing the spectrum of the local signal.

[0054] (3) Using the Lagrange multiplier λ, the quadratic penalty term β, and the constraints VMD transforms the constrained variational problem into an unconstrained variational problem, where s(t) is the historical photovoltaic power time series. By minimizing the bandwidth constraint and the regularization term, the optimization problem becomes:

[0055]

[0056] (4) Optimizing process variables is mainly to solve the above-mentioned constrained optimization problem. In order to reduce computational complexity, the time-domain equality constraints are transformed into frequency-domain equality constraints. And the alternating direction multiplier method was used to respectively... and These process variables are updated using the following formula:

[0057]

[0058] in, This indicates that a Fourier transform is performed on the eigenmode functions. This represents performing a Fourier transform on the Lagrange multipliers, where η is the learning rate or step size to control the update speed of the Lagrange multipliers, ω is the center frequency of each mode, and n is the number of iterations.

[0059] This application ultimately decomposes the historical photovoltaic power time series into four intrinsic mode functions, namely...

[0060] Step 22: Calculate the entropy value of each intrinsic mode function (EMF). EEMFs with entropy values ​​greater than a set threshold are designated as high-entropy mode sequences, while those with entropy values ​​less than or equal to the set threshold are designated as low-entropy mode sequences. The entropy results of the original sequence after variational mode decomposition are as follows: Figure 6 As shown.

[0061] To evaluate the modal complexity of each variational mode decomposition, entropy calculation can identify which modes contain more ordered information or noise. Specifically, the k-th intrinsic mode function m is calculated using the following formula. k The entropy value of (t):

[0062]

[0063] Wherein, H(m) k ) represents the k-th intrinsic mode function m k The entropy value of p(t), l Let L be the probability of each discrete event in the probability distribution of the k-th intrinsic mode function, and L be the number of discretization intervals of the signal value, which is set to 100 here.

[0064] Step 23: Perform adaptive noise set empirical mode decomposition on the high entropy mode sequence and re-superimpose it to obtain high frequency components.

[0065] Low-complexity modes may be useful signals, while high-complexity modes may be noise. Therefore, high-entropy mode sequences are further decomposed using ICEEMDAN (Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise) to reduce noise in the signal and enhance the robustness of the model. The specific steps of ICEEMDAN are as follows:

[0066] (1) Add Gaussian white noise that follows a normal distribution to the high-entropy mode sequence to obtain the updated signal m. new (t):

[0067] m new (t)=m high (t)+ε0ξ e (t), e = 1, 2, 3, ..., N;

[0068] Where, m high (t) is a high-entropy modal sequence, ε0 is the standard deviation of Gaussian white noise, and ξ e (t) represents Gaussian white noise, and N represents the number of experiments.

[0069] (2) For the updated signal m new (t) Perform EMD (Empirical Mode Decomposition). EMD decomposes the updated signal into several empirical eigenmode functions and residual signals. The update formula is as follows:

[0070]

[0071] Among them, IMF f r'(t) is the f-th empirical eigenmode function, F is the total number of empirical eigenmode functions decomposed, and r'(t) is the residual signal.

[0072] (3) To enhance the stability and accuracy of the decomposition, repeat steps (1) and (2) multiple times, adding different Gaussian white noise each time, and performing EMD decomposition. Through multiple noise injections and decompositions, a set of different second eigenmode functions is obtained:

[0073]

[0074] Where M represents the number of noise injections and decompositions.

[0075] (4) For each empirical eigenmode function (IMF) f (t), the results of multiple decompositions are averaged to obtain the final empirical intrinsic mode (IMF). f '(t): Then all empirical intrinsic modes IMF f '(t) are superimposed as high-frequency components.

[0076] Step 24: Superimpose the low-entropy mode sequences to obtain low-frequency components.

[0077] Step 203: High-frequency component prediction model is used to extract features from the high-frequency components to obtain high-frequency feature vectors. For example... Figure 4 As shown, the high-frequency component prediction model is built based on TCN (Temporal Convolutional Network) and the Informer model.

[0078] High-frequency signals still exhibit strong volatility. To fully extract the features of high-frequency components, the high-frequency component prediction model not only combines the time dependence capability of TCN in capturing long sequences, but also leverages the structural characteristics of TCN to improve the prediction accuracy and computational efficiency of the model when combined with Informer, thus enhancing its ability to handle complex sequences. The structures of TCN and Informer models are described below.

[0079] Compared to traditional convolutional networks, TCN captures the temporal dependencies of long sequences through causal convolution, dilated convolution, and residual connections instead of recurrent connections. Firstly, causal convolution is a special form of one-dimensional convolution. Its main characteristic is that when the convolution kernel slides, it only uses data from the current and previous time steps, while the right side of the kernel is padded with zeros. Therefore, the output value o(τ) is calculated using the following formula:

[0080]

[0081] Where τ is the time step, g is the index number, G is the kernel size, w(g) is the weight value of the kernel at the g-th index position, and s1(τ-g) is the data point of the first g time steps in the high-frequency component.

[0082] To improve the model's ability to handle long-term sequence tasks, an inflation factor is introduced in causal convolution to increase the receptive field, enabling the network to cover data over a longer time span. The inflation factor increases exponentially with the number of convolutional layers.

[0083] As a variant of the Transformer model, the Informer model's most significant characteristic lies in addressing the challenges of traditional attention mechanisms. When dealing with long input sequences, the computational complexity of the weight matrix increases with the sequence length E, requiring the calculation of the attention matrix (E×E) between the current time step and other time steps. Furthermore, some query vectors and key vectors in the attention scoring mechanism exhibit weak correlation. To address these issues, the Informer model reduces computational complexity through a sparse self-attention mechanism and a distillation mechanism in the encoder. The formula for the sparse attention scoring mechanism is as follows:

[0084]

[0085] Where Q is the query matrix, K is the key matrix, V is the value matrix, and d is the dimension of the input sequence. Each K is only associated with a strongly correlated Q. Q is transformed into a mapping of the first h strongly correlated query vectors after sparse metric. Therefore, the size of the sparse matrix is ​​equal to the size of the query vectors. The size of h is determined by constants c1 and L. K control.

[0086] To further improve computational efficiency, it is necessary to set limits for the sparsity metric. The empirical formula for the sparsity metric is as follows:

[0087]

[0088] Where, q u For the u-th row of Q, k v For the v-th row of K, L K The length E is equal to that of the input sequence. The score for each query vector can be obtained using the above formula, and then h query vectors are selected for QKV matrix multiplication.

[0089] The distillation operation includes a one-dimensional convolutional layer (Conv1D) and a max-pooling layer (Maxpool), with ELU as the activation function. The max-pooling layer performs downsampling, so the sequence length is halved after each distillation layer. The distillation process is as follows:

[0090]

[0091] in, Let τ be the input to the κ-th layer of the network at time step τ, [·] AB This indicates the Probsparse attention module.

[0092] To maintain the same output dimension across all encoder stacks, an encoder stack approach is used for layering. This operation significantly reduces the temporal dimension of the input while still preserving key features. Similar to the TCN structure, it expands the receptive field to capture greater contextual information.

[0093] In a specific application example, a temporal convolutional network is used to transform and extract the feature dimensions of the high-frequency components to obtain an enhanced feature sequence. This enhanced feature sequence is used as the input to the encoder of the Informer model, the enhanced sequence of the encoder at the last E / 2 time steps is used as the input to the decoder of the Informer model, and the output of the decoder is used as the high-frequency feature vector; where E is the sequence length.

[0094] Step 204: The low-frequency component prediction model is used to extract features from the low-frequency component to obtain a low-frequency feature vector. The low-frequency component prediction model is constructed based on the xLstm model.

[0095] Specifically, such as Figure 5 As shown, the low-frequency component prediction model includes a first linear layer, a first normalized layer, an xLstm model, a second linear layer, and a second normalized layer connected in sequence.

[0096] Considering the dimensionality issue, the low-frequency component is first input to the first linear layer for dimensionality transformation, and then input to the first normalization layer. The method of stabilizing feature distribution is used to solve the noise problem in the low-frequency component and prevent gradient explosion. The historical photovoltaic power time series obtained in this application has a large time span, so mLstm is finally selected for prediction. Then, each feature before output is independently normalized to eliminate the spatiotemporal differences on each feature sensor.

[0097] The xLSTM model is an improved model based on LSTM (Long Short-Term Memory). When dealing with longer time series, traditional LSTM still has difficulty capturing dependencies over long time spans. Therefore, for low-frequency components, LSTM cannot capture long-term trends between sequences.

[0098] xLstm, also known as extended LSTM, mainly utilizes the serial structure of sLstm and mLstm modules for arbitrary combinations.

[0099] The sLstm module is an improvement upon LSTM by adding exponential activation functions to the input and forget gates, introducing normalized states and establishing stabilization operations in the hidden states. The exponential gate structure formula is as follows:

[0100] i τ =exp(w ix x τ +w ih h τ-1 +b i );

[0101] f τ =exp(w fx x τ +w fh h τ-1 +b f );

[0102] Among them, i τ f is the input gate of the sLstm network. τ For the forget gate of the sLstm network, x τ For the input features of the low-frequency components at time step τ, w ix The weight matrix w between the input gate and the input features ih Let w be the weight matrix between the input gate and the hidden state. fx w is the weight matrix between the forget gate and the input features. fh Let b be the weight matrix between the forget gate and the hidden state. i For the bias term of the input gate, b f h is the bias term for the forget gate. τ-1 Let τ be the hidden state at time step τ-1.

[0103] Compared to the sigmoid activation function in traditional LSTM, the exponential activation function exp in the sLSTM module can capture more fine-grained feature changes. The normalization operation is mainly reflected in the hidden state h. τ and normalized state n τ The calculation is as follows:

[0104] n τ =f τ n τ-1 +i τ ;

[0105]

[0106] Among them, o τ For the parameters of the output gate, c τ-1 For the memory cell state of features before update, n τ-1z represents the state quantity of the memory cell before normalization. τ This is the input for memory cells.

[0107] The mLstm module adds an attention mechanism similar to the Transformer model to the traditional LSTM. By introducing the calculation of k, q, and v, it updates the memory cell c to capture the complex relationships between features. The model structure is similar to the sLStm module mentioned above, but the state variables of each cell are changed from vectors to matrices, which greatly enhances the storage capacity of the memory cells. The specific formula is as follows:

[0108] q τ =w q x τ +b q ;

[0109]

[0110] v τ =w v x τ +b v ;

[0111] c τ =f τ c τ-1 +i τ v τ k τ T ;

[0112]

[0113] Where, q τ Let k be the query vector at time step τ. τ Let v be the key vector at time step τ. τ Let w be the value vector at time step τ. q w is the weight matrix of the query vector. k Let w be the weight matrix of the key vectors. v Let w be the weight matrix of the value vector. f Let b be the weight matrix of the forget gate. q For the bias term of the query vector, b k b is the bias term of the key vector. v b is the bias term of the value vector. f For the bias term of the forget gate, n τ The normalized state variable is a weighted sum of key vectors, where each key vector is weighted by the input gate and all future forget gates. It can be clearly observed that the memory cell capacity is controlled by the key vector k. τ Sum vector v τ The introduction of this technology increases the number of memory cells from the traditional c∈R to c∈Rd×d Furthermore, it improves the separation of information extraction and reduces signal noise.

[0114] Step 205: Determine the photovoltaic power generation prediction result within the future set time period based on the high-frequency feature vector and the low-frequency feature vector.

[0115] Specifically, after concatenating the high-frequency feature vector and the low-frequency feature vector, a fully connected layer is used to determine the photovoltaic power generation prediction result within a set future time period. The re-superimposed high- and low-frequency component sequences are as follows: Figure 7 As shown.

[0116] like Figures 8 to 13 As shown in the simulation analysis, relying solely on a single model results in poor fitting performance, especially at peak values. If the dataset is not preprocessed, the model may overfit, reducing its generalization performance. After dual decomposition, the predicted curve is smoother and closer to the true value compared to the single model. The ICEEMDAN decomposition and prediction model structure result in the best overall robustness. Furthermore, due to its structural characteristics, the TCN-Informer model has shorter computation time and higher prediction accuracy and computational efficiency compared to the TCN-Transformer model. Parameter optimization of the model in this application using the RIME algorithm further improves the prediction accuracy of photovoltaic power generation.

[0117] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 14 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores historical photovoltaic power time series data from photovoltaic power plants. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a short-term photovoltaic power prediction method.

[0118] Those skilled in the art will understand that Figure 14The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0119] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0120] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0121] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0122] In this application, all actions to acquire signals, information, or data are carried out in compliance with the relevant data protection laws and policies of the country where the location is situated, and with the authorization granted by the owner of the relevant device.

[0123] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0124] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0125] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0126] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A short-term photovoltaic power prediction method, characterized in that, The method includes: Obtain the historical photovoltaic power time series of photovoltaic power plants; Modal decomposition was performed on the historical photovoltaic power time series to obtain high-frequency and low-frequency components; The high-frequency component prediction model is used to extract features from the high-frequency components to obtain high-frequency feature vectors; the high-frequency component prediction model is constructed based on temporal convolutional networks and the Informer model. The low-frequency component is used to extract features from the low-frequency component to obtain a low-frequency feature vector; the low-frequency component prediction model is constructed based on the xLstm model. Based on the high-frequency feature vector and the low-frequency feature vector, the predicted photovoltaic power generation within a set future time period is determined.

2. The short-term photovoltaic power prediction method according to claim 1, characterized in that, The historical photovoltaic power time series includes the total radiation level, direct radiation level, global radiation level, air temperature, humidity, and actual photovoltaic power generation within a set historical period.

3. The short-term photovoltaic power prediction method according to claim 1, characterized in that, Obtain the historical photovoltaic power time series of photovoltaic power plants, specifically including: Pearson correlation analysis was performed on the factors affecting photovoltaic power to identify weakly correlated influencing factors. By obtaining the values ​​of weakly correlated influencing factors of photovoltaic power plants within a set historical time period, a historical photovoltaic power time series can be obtained.

4. The short-term photovoltaic power prediction method according to claim 1, characterized in that, Modal decomposition is performed on the historical photovoltaic power time series to determine the high-frequency and low-frequency components, specifically including: Variational mode decomposition was performed on the historical photovoltaic power time series to obtain multiple intrinsic mode functions; Calculate the entropy value of each intrinsic mode function, and take the intrinsic mode functions with entropy values ​​greater than a set threshold as high entropy mode sequences, and take the intrinsic mode functions with entropy values ​​less than or equal to the set threshold as low entropy mode sequences; The high-entropy mode sequence is subjected to adaptive noise set empirical mode decomposition and re-superimposed to obtain high-frequency components; The low-entropy mode sequences are superimposed to obtain low-frequency components.

5. The short-term photovoltaic power prediction method according to claim 1, characterized in that, The high-frequency components are used to extract features from the high-frequency components using a high-frequency component prediction model to obtain high-frequency feature vectors, specifically including: A temporal convolutional network is used to transform and extract the feature dimensions of the high-frequency components to obtain an enhanced feature sequence. The enhanced feature sequence is used as the input to the encoder of the Informer model, the enhanced sequence of the encoder at the last E / 2 time steps is used as the input to the decoder of the Informer model, and the output of the decoder of the Informer model is used as the high-frequency feature vector; where E is the sequence length.

6. The short-term photovoltaic power prediction method according to claim 1, characterized in that, The low-frequency component prediction model includes a first linear layer, a first normalized layer, an xLstm model, a second linear layer, and a second normalized layer connected in sequence.

7. The short-term photovoltaic power prediction method according to claim 1, characterized in that, Based on the high-frequency feature vector and the low-frequency feature vector, the photovoltaic power generation prediction result for a future set time period is determined, specifically including: After concatenating the high-frequency feature vector with the low-frequency feature vector, a fully connected layer is used to determine the photovoltaic power generation prediction result within a set future time period.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the short-term photovoltaic power prediction method according to any one of claims 1-7.

9. 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 short-term photovoltaic power prediction method according to any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the short-term photovoltaic power prediction method according to any one of claims 1-7.

Images