A short-term photovoltaic power generation power prediction method based on decomposition-prediction-set

Abstract

The application discloses a short-term photovoltaic power prediction method based on decomposition-prediction-set, which firstly performs preliminary decomposition on an original photovoltaic power sequence through variational mode decomposition to obtain a plurality of modal components; then, high-frequency modal components are identified from the modal components, and the high-frequency modal components are subjected to secondary decomposition by using adaptive noise complete ensemble empirical mode decomposition, so that random noise and fluctuation components are separated more finely; for the components with different characteristics after the decomposition, an asymmetric prediction strategy is adopted; low-frequency components with strong periodicity are input into a time convolution network for prediction, and high-frequency components after the secondary decomposition with high complexity and randomness are input into a long short-term memory network for prediction; finally, the prediction results of all the components are time-series superimposed to reconstruct the final photovoltaic power prediction value. Through the secondary decomposition strategy combined with multi-model prediction, the photovoltaic power prediction precision is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of new energy power generation technology, and in particular to a short-term photovoltaic power generation prediction method based on decomposition-prediction-ensemble. Background Technology

[0002] Amid the global wave of addressing climate change and transforming the energy structure, clean energy, represented by photovoltaic (PV) power generation, has developed rapidly and has become an important component of the power system. According to data from the National Energy Administration, my country's installed capacity and power generation of PV power generation have continued to grow rapidly, highlighting its increasingly prominent position in the energy system. However, the power output of PV power generation is strongly influenced by various meteorological factors such as solar irradiance, temperature, humidity, and cloud cover, exhibiting significant randomness, volatility, and intermittency. This inherent uncertainty poses significant challenges to grid dispatching plans, safe and stable operation, and power quality.

[0003] Therefore, achieving high-precision short-term forecasts of photovoltaic power generation is crucial for the power system. Accurate forecasts not only provide reliable decision-making support for grid dispatch centers, optimizing power generation plans and reserve capacity allocation, but also effectively reduce system operating costs, increase the absorption rate of photovoltaic power generation, and promote the healthy development of the electricity market.

[0004] Currently, photovoltaic (PV) power prediction methods are mainly divided into physical methods, statistical methods, and artificial intelligence methods. Physical methods rely on complex meteorological models and the physical parameters of PV panels, resulting in a complex modeling process and high requirements for weather forecast accuracy. Statistical methods, such as the ARIMA model, while simple, struggle to capture the strong nonlinear and non-stationary characteristics of PV power sequences. In recent years, artificial intelligence methods, represented by deep learning, such as recurrent neural networks (RNNs) and long short-term memory networks (LSTMs), have attracted considerable attention due to their powerful nonlinear fitting capabilities. However, directly inputting the raw, highly complex PV power sequence into a single deep learning model often fails to fully learn its inherent multi-scale fluctuation patterns, leading to difficulties in model training and limited prediction accuracy.

[0005] To address this issue, researchers introduced a "decomposition-prediction-assembly" framework. This involves first using signal decomposition techniques (such as wavelet transform and Empirical Mode Decomposition (EEMD)) to divide the original sequence into several relatively stationary subsequences, then predicting each subsequence separately, and finally summing the prediction results. Variational Mode Decomposition (VMD), as an advanced signal processing technique, effectively avoids mode aliasing compared to methods like EEMD. However, practical experience has shown that for signals like photovoltaic power, which contain a large amount of high-frequency noise and random fluctuations, the high-frequency components obtained after a single VMD decomposition still exhibit strong randomness. Directly predicting these components introduces significant errors, affecting the final prediction accuracy.

[0006] Therefore, how to more thoroughly decompose the photovoltaic power sequence and select the most suitable prediction model for subsequences with different characteristics is the key technical bottleneck for improving the accuracy of short-term photovoltaic power prediction. Summary of the Invention

[0007] The main objective of this invention is to provide a short-term photovoltaic power generation prediction method based on decomposition-prediction-ensemble, which aims to solve the technical problem in the prior art that the high frequency component has strong randomness and poor prediction accuracy due to incomplete single mode decomposition.

[0008] In a first aspect, the present invention provides a short-term photovoltaic power generation prediction method based on decomposition-prediction-ensembling, comprising the following steps:

[0009] Obtain raw time series data containing historical photovoltaic power generation data, and preprocess the raw time series data;

[0010] Variational mode decomposition (VMD) is performed on the preprocessed photovoltaic power generation data to obtain multiple first-order mode components. Based on the center frequency of each first-order mode component, it is divided into at least one high-frequency mode component and multiple low-frequency mode components.

[0011] For the at least one high-frequency modal component, perform adaptive noise complete set empirical mode decomposition (CEEMDAN) to obtain multiple secondary modal components;

[0012] The multiple low-frequency modal components are input into the temporal convolutional network (TCN) model for prediction to obtain the prediction results of each low-frequency modal component;

[0013] The multiple secondary modal components are input into a Long Short-Term Memory (LSTM) network model for prediction to obtain the prediction results of each secondary modal component;

[0014] The prediction results of each low-frequency mode component are time-series superimposed with the prediction results of each secondary mode component to obtain the final photovoltaic power generation prediction value.

[0015] As an optional implementation of the first aspect of this application, the step of obtaining raw time series data containing historical photovoltaic power generation data includes: obtaining environmental meteorological data and historical photovoltaic power generation data of the photovoltaic power station, wherein each data record includes temperature, wind speed, relative humidity, total horizontal solar irradiance, diffuse irradiance and effective power generation, and the data sampling interval is 5 minutes.

[0016] As an optional implementation of the first aspect of this application, the step of preprocessing the original time series data includes: imputing missing values ​​in the original time series data using the K-nearest neighbor algorithm, wherein one missing value is determined by the average of its K nearest neighbor data points, and the calculation formula is as follows: ,in, Indicates supplementary values ​​for missing data; This represents the first k missing data points; This represents the last k missing data points, where k is 12, indicating that the missing data was searched within one hour. After data imputation, a min-maximum normalization method is used to scale the data; the calculation formula is as follows: ,in, This represents the normalized value. This represents the value in the original data. Indicates the maximum value. This represents the minimum value.

[0017] As an optional implementation of the first aspect of this application, the step of performing variational mode decomposition on the preprocessed photovoltaic power generation data is achieved by constructing and solving a constrained variational problem. The objective of the constrained variational problem is to find K modal components such that the sum of the bandwidths of each modal component is minimized and the sum of all modal components is equal to the original signal. The solution process adopts the alternating direction multiplier method (ADMM), which iteratively updates each modal component, the center frequency of each mode, and the Lagrange multiplier operator until the convergence condition is met.

[0018] As an optional implementation of the first aspect of this application, the step of performing adaptive noise complete set empirical mode decomposition includes: at each stage of decomposition, adding a pair of Gaussian white noise of a specific amplitude to the at least one high-frequency mode component to be decomposed, the pair of Gaussian white noise having opposite signs, performing empirical mode decomposition (EMD) on the signal after adding noise to extract intrinsic mode functions (IMFs), and finally obtaining the plurality of secondary mode components by averaging the intrinsic mode functions obtained from multiple decompositions.

[0019] As an optional implementation of the first aspect of this application, the Temporal Convolutional Network (TCN) model adopts a one-dimensional dilated causal convolutional structure, wherein the dilation coefficient d increases exponentially with the number of network layers n, and the relationship is as follows: The receptive field is increased by interval sampling to capture the long-period dependence in the low-frequency modal components. The Long Short-Term Memory (LSTM) network model contains one memory unit and three control gates, namely the forget gate, the input gate, and the output gate. The three control gates work together to learn and predict time series patterns with high volatility and nonlinear characteristics in the secondary modal components.

[0020] As an optional implementation of the first aspect of this application, before the step of time-series superimposing the prediction results of each low-frequency mode component with the prediction results of each secondary mode component to obtain the final photovoltaic power generation prediction value, the method further includes a step of automatically optimizing the hyperparameters of the temporal convolutional network model and the long short-term memory network model using the Hyperband algorithm; the Hyperband algorithm treats hyperparameter optimization as a resource allocation problem, and through a successive halving strategy, terminates the poor-performing hyperparameter combinations in advance during the training process, and concentrates the computing resources on the high-performing hyperparameter combinations to determine the optimal hyperparameter configuration.

[0021] Secondly, embodiments of this application provide a short-term photovoltaic power generation prediction system based on decomposition-prediction-ensembling, comprising:

[0022] The data acquisition and preprocessing module is used to acquire raw time series data containing historical photovoltaic power generation data and to preprocess the raw time series data.

[0023] The signal decomposition module is used to perform variational mode decomposition (VMD) on the preprocessed photovoltaic power generation data to obtain multiple first-order mode components, and divide them into at least one high-frequency mode component and multiple low-frequency mode components according to the center frequency of each first-order mode component; for the at least one high-frequency mode component, adaptive noise complete set empirical mode decomposition (CEEMDAN) is performed to obtain multiple second-order mode components.

[0024] The model prediction module is used to input the multiple low-frequency modal components into a temporal convolutional network (TCN) model for prediction, and obtain the prediction results of each low-frequency modal component; and to input the multiple secondary modal components into a long short-term memory network (LSTM) model for prediction, and obtain the prediction results of each secondary modal component.

[0025] The ensemble prediction module is used to time-series superimpose the prediction results of each low-frequency mode component with the prediction results of each secondary mode component to obtain the final photovoltaic power generation prediction value.

[0026] Thirdly, embodiments of this application provide an electronic device, which includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the steps of the method described in the first aspect.

[0027] Fourthly, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0029] 1. Improve the thoroughness of decomposition: By using the two-stage cascade decomposition strategy of VMD and CEEMDAN, especially by performing secondary fine decomposition on the high-frequency random parts that are difficult for VMD to handle, the complexity and randomness of the subsequences are effectively reduced, laying a solid foundation for subsequent accurate prediction.

[0030] 2. Enhanced model applicability: By adopting the asymmetric prediction architecture of TCN and LSTM, the modal components with different characteristics are assigned to the model that is best at handling that characteristic for prediction, realizing "teaching according to aptitude", giving full play to the advantages of deep learning models and improving the overall prediction performance.

[0031] 3. Improved prediction accuracy and stability: Through the above-mentioned combination of innovations, this invention can more accurately capture the deterministic trend and random fluctuations in photovoltaic power sequences, effectively solving the problem of low prediction accuracy of traditional methods. The final output prediction results have a higher degree of agreement with the actual values, and the model is more robust. Attached Figure Description

[0032] Figure 1 This is a flowchart of a short-term photovoltaic power generation prediction method based on decomposition-prediction-set according to an embodiment of the present invention;

[0033] Figure 2 This is a design block diagram corresponding to a short-term photovoltaic power generation prediction method based on decomposition-prediction-set according to an embodiment of the present invention;

[0034] Figure 3 This is a flowchart of the overall prediction model corresponding to a short-term photovoltaic power generation prediction method based on decomposition-prediction-set according to an embodiment of the present invention.

[0035] Figure 4 This is a schematic diagram of the results obtained by VMD decomposition of the original photovoltaic power sequence in an embodiment of the present invention;

[0036] Figure 5 This is a schematic diagram showing the result of performing CEEMDAN secondary decomposition on the high-frequency modes decomposed by VMD in an embodiment of the present invention.

[0037] Figure 6 This is a comparison chart of the prediction curve and the actual value curve of the overall prediction model designed in the embodiments of the present invention;

[0038] Figure 7 This is a schematic diagram of a short-term photovoltaic power generation prediction system based on decomposition-prediction-set provided by an embodiment of the present invention.

[0039] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0040] 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, 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.

[0041] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0042] Example 1

[0043] Please see Figure 1 , Figure 2 and Figure 3 This document provides a flowchart of a short-term photovoltaic power generation prediction method based on decomposition-prediction-ensemble model, a design block diagram of the method, and a flowchart of the overall prediction model for the method, all provided in this embodiment of the invention. The method may include the following steps:

[0044] S1: Obtain raw time series data containing historical photovoltaic power generation data, and preprocess the raw time series data.

[0045] 1. Data Acquisition and Preparation

[0046] This embodiment uses historical operating data from a photovoltaic power station, with a data time span of one year. The data collection frequency is once every 5 minutes, and each record includes: air temperature (°C), wind speed (m / s), relative humidity (%), total horizontal solar irradiance (W / m²), diffuse irradiance (W / m²), and the effective power generation of the photovoltaic power station (kW).

[0047] 2. Data Preprocessing

[0048] In this embodiment, a photovoltaic power prediction dataset is constructed after data cleaning and preprocessing using the K-nearest neighbor method. Data is then supplemented using the K-nearest neighbor algorithm to avoid data loss affecting subsequent decomposition and prediction. The calculation expression is as follows:

[0049] (1)

[0050] in, Indicates the missing data's supplementary value; This represents the first k missing data points; This represents the k last few missing data points. In the model, k is set to 12, meaning it searches for missing data within one hour.

[0051] After completing the data, MIN-MAX is used for data normalization. The calculation formula is as follows:

[0052] (2)

[0053] in, This represents the normalized value. This represents the value in the original data. Indicates the maximum value. This represents the minimum value.

[0054] S2: Perform variational mode decomposition on the preprocessed photovoltaic power generation data to obtain multiple first-order mode components, and divide them into at least one high-frequency mode component and multiple low-frequency mode components according to the center frequency of each first-order mode component; perform adaptive noise complete set empirical mode decomposition on the at least one high-frequency mode component to obtain multiple second-order mode components.

[0055] Specifically, the decomposition process includes initial mode decomposition using VMD, followed by secondary mode decomposition using CEEMD. The mode decomposition process comprises both VMD and CEEMD decomposition. The preprocessed data is first decomposed using VMD, the expression of which is:

[0056] (3)

[0057] (4)

[0058] Where k represents the number of modes in the sequence decomposition; This represents the k-th modal component; This represents the center frequency of the k-th mode; Represents the original sequence;

[0059] The expression for the unconstrained variational model is shown below:

[0060] (5)

[0061] in: Indicates the quadratic multiplication factor; This represents the Lagrange multiplication operator.

[0062] During the iterative search process, the Alternating Direction Multiplier Method (ADMM) is used to solve Equation (5) to obtain the optimal solution of the constrained variational model, which is then iteratively updated. for:

[0063] (6)

[0064] Iterative update for:

[0065] (7)

[0066] After classifying the multimodal sub-signals following the initial decomposition by VMD, the high-frequency modes are input into CEEMDAN for secondary decomposition based on their center frequencies. The decomposition expression is as follows:

[0067] (8)

[0068] in, Represents the IMF obtained from the k-th decomposition; This represents the noise amplitude coefficient.

[0069] In this step, the decomposition part includes preliminary mode decomposition using VMD, resulting in 5 subsequences, as shown below. Figure 4 As shown, after the multimodal sub-signals are initially decomposed by VMD, they are classified. Based on the center frequency, the highest mode subsequence is selected. The high-frequency modes are then input into CEEMDAN for secondary decomposition. The decomposition results are shown below. Figure 5 As shown.

[0070] S3: Input the multiple low-frequency modal components into a temporal convolutional network model for prediction to obtain the prediction results of each low-frequency modal component; input the multiple secondary modal components into a long short-term memory network model for prediction to obtain the prediction results of each secondary modal component.

[0071] Specifically, the prediction part includes inputting the low-frequency part decomposed by VMD into the TCN neural network for training, inputting the high-frequency part decomposed by CEEMDAN into the LSTM neural network for training, using the Hyperband algorithm to optimize the model hyperparameters, and the ensemble part is ensembled into the final prediction result by time-series addition.

[0072] The low-frequency components after VMD decomposition are input into the TCN neural network for prediction. Its algorithm structure is naturally suited to fixed-length periodic sequences. The TCN addresses gradient vanishing by using dilated convolution for interval sampling. In this process, the dilation coefficient d is used to adjust the sampling rate, and the expression for d is:

[0073] (9)

[0074] Where n represents the number of dilated convolutional layers.

[0075] The expression for dilated convolution treaty sampling to expand the receptive field is:

[0076] (10)

[0077] in, Indicates the coefficient of thermal expansion. This indicates the kernel size.

[0078] The subsequences obtained from CEEMDAN decomposition are input into an LSTM neural network for prediction, as shown in the following expression:

[0079] The forgetting gate determines how much historical information is retained; its expression is:

[0080] (11)

[0081] The input gate determines the cell update state as follows:

[0082] (12)

[0083] (13)

[0084] Cell status updated to:

[0085] (14)

[0086] The expression for the output gate to generate the output at the current moment is as follows:

[0087] (15)

[0088] (16)

[0089] in, This represents the sigmoid function.

[0090] Furthermore, model optimization is performed. To avoid manually configuring LSTM and TCN hyperparameters and affecting model prediction accuracy, the Hyperband algorithm is used to optimize hyperparameters in the neural network prediction part. This algorithm can efficiently find the optimal or near-optimal hyperparameter combination with limited computational resources. Hyperparameter optimization is treated as a resource allocation problem, reducing ineffective training through early stopping and low-potential configuration. The calculation expression is:

[0091] (17)

[0092] Where R represents the maximum number of resources, which is the epoch in the model; η represents the reduction ratio.

[0093] S4: The prediction results of each low-frequency mode component and the prediction results of each secondary mode component are time-series superimposed to obtain the final photovoltaic power generation prediction value.

[0094] Predicting future photovoltaic power generation, such as Figure 6 The image shown is the prediction curve in this example. (The text repeats itself here.) Figure 6 Comparative experiments demonstrate that the model designed in this invention has high accuracy and applicability in short-term photovoltaic power generation prediction, and can meet the requirements of short-term photovoltaic power generation prediction.

[0095] In summary, this invention provides a short-term photovoltaic power generation prediction method based on decomposition-prediction-ensemble. It preprocesses the collected data using the K-nearest neighbor algorithm, decomposes the photovoltaic power generation signal using VMD, and then applies CEEMDAN secondary decomposition to the high-frequency modes obtained after decomposition to extract the frequency and time domain information of the corresponding time series signals. Based on the different characteristics of the modes, the high-frequency modes are input into an LSTM model, and the low-frequency modes are input into a TCN model for prediction. The hyperparameters are optimized using the Hyperband algorithm. Finally, the prediction results from different neural networks are superimposed in a time-series manner to obtain the final prediction result. By combining decomposition-prediction-ensemble approach with mode decomposition and deep learning technology, the operational safety of the photovoltaic power generation system and the accuracy and reliability of short-term photovoltaic power generation prediction are improved.

[0096] Example 2

[0097] Please see Figure 7 The diagram shown is a schematic representation of a short-term photovoltaic power generation prediction system based on decomposition-prediction-set according to the second embodiment of this application. The system includes the following key modules:

[0098] The data acquisition and preprocessing module 100 is used to acquire raw time series data containing historical photovoltaic power generation data and to preprocess the raw time series data.

[0099] The signal decomposition module 200 is used to perform variational mode decomposition (VMD) on the preprocessed photovoltaic power generation data to obtain multiple first-order mode components, and divide them into at least one high-frequency mode component and multiple low-frequency mode components according to the center frequency of each first-order mode component; for the at least one high-frequency mode component, adaptive noise complete set empirical mode decomposition (CEEMDAN) is performed to obtain multiple second-order mode components.

[0100] The model prediction module 300 is used to input the multiple low-frequency modal components into a temporal convolutional network (TCN) model for prediction, and obtain the prediction results of each low-frequency modal component; and to input the multiple secondary modal components into a long short-term memory network (LSTM) model for prediction, and obtain the prediction results of each secondary modal component.

[0101] The ensemble prediction module 400 is used to time-series superimpose the prediction results of each low-frequency mode component with the prediction results of each secondary mode component to obtain the final photovoltaic power generation prediction value.

[0102] The short-term photovoltaic power generation prediction system based on decomposition-prediction-ensembling in this application embodiment can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, mobile electronic devices can be mobile phones, tablets, laptops, PDAs, in-vehicle electronic devices, wearable devices, ultra-mobile personal computers (UMPCs), netbooks, or personal digital assistants (PDAs), etc., while non-mobile electronic devices can be servers, network-attached storage (NAS), personal computers (PCs), etc. This application embodiment does not impose specific limitations.

[0103] The short-term photovoltaic power generation prediction system based on decomposition-prediction-ensembling in this application embodiment can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit it.

[0104] The short-term photovoltaic power generation prediction system based on decomposition-prediction-ensembling provided in this application embodiment can achieve... Figure 1 The various processes implemented in the method embodiment of the short-term photovoltaic power generation prediction method based on decomposition-prediction-set are not described in detail here to avoid repetition.

[0105] Optionally, embodiments of this application also provide an electronic device, including a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the various processes of the above-described embodiment of a short-term photovoltaic power generation prediction method based on decomposition-prediction-set, and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0106] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described embodiment of a short-term photovoltaic power generation prediction method based on decomposition-prediction-set, and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0107] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0108] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0109] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0110] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A short-term photovoltaic power generation prediction method based on decomposition-prediction-ensemble, characterized in that, Includes the following steps: Obtain raw time series data containing historical photovoltaic power generation data, and preprocess the raw time series data; Variational mode decomposition is performed on the preprocessed photovoltaic power generation data to obtain multiple first-order mode components. Based on the center frequency of each first-order mode component, it is divided into at least one high-frequency mode component and multiple low-frequency mode components. For the at least one high-frequency modal component, perform adaptive noise complete set empirical mode decomposition to obtain multiple secondary modal components; The multiple low-frequency modal components are input into a temporal convolutional network model for prediction, and the prediction results of each low-frequency modal component are obtained. The multiple secondary modal components are input into the long short-term memory network model for prediction, and the prediction results of each secondary modal component are obtained. The prediction results of each low-frequency mode component are time-series superimposed with the prediction results of each second-order mode component to obtain the final photovoltaic power generation prediction value. Before the step of time-series superimposing the prediction results of each low-frequency mode component with the prediction results of each secondary mode component to obtain the final photovoltaic power generation prediction value, the algorithm further includes a step of automatically optimizing the hyperparameters of the temporal convolutional network model and the long short-term memory network model using the Hyperband algorithm. The Hyperband algorithm treats hyperparameter optimization as a resource allocation problem. Through a successive halving strategy, it terminates the poor-performing hyperparameter combinations in advance during training and concentrates the computing resources on the high-performing hyperparameter combinations to determine the optimal hyperparameter configuration.

2. The short-term photovoltaic power generation prediction method based on decomposition-prediction-ensemble as described in claim 1, characterized in that, The steps to obtain raw time-series data containing historical photovoltaic power generation data include: The system acquires environmental meteorological data and historical photovoltaic power generation data for photovoltaic power plants. Each data record includes temperature, wind speed, relative humidity, total horizontal solar irradiance, diffuse irradiance, and effective power generation, with a data sampling interval of 5 minutes.

3. A short-term photovoltaic power generation prediction method based on decomposition-prediction-ensemble as described in claim 1 or 2, characterized in that, The preprocessing steps for the raw time series data include: The K-nearest neighbor algorithm is used to impute missing values ​​in the original time series data. Each missing value is determined by the average of its K nearest neighbors, calculated using the following formula: in, Indicates supplementary values ​​for missing data; This represents the first k missing data points; This represents the last k missing data points, where k is 12, meaning it searches for missing data within one hour. After data imputation is completed, the data is scaled using a minimum-maximum normalization method, the calculation formula of which is: in, This represents the normalized value. This represents the value in the original data. Indicates the maximum value. This represents the minimum value.

4. The short-term photovoltaic power generation prediction method based on decomposition-prediction-ensemble as described in claim 1, characterized in that, The step of performing variational mode decomposition on the preprocessed photovoltaic power generation data is achieved by constructing and solving a constrained variational problem. The objective of the constrained variational problem is to find K modal components such that the sum of the bandwidths of each modal component is minimized and the sum of all modal components is equal to the original signal. The solution process adopts the alternating direction multiplier method, and iteratively updates each modal component, the center frequency of each mode, and the Lagrange multiplier operator until the convergence condition is met.

5. The short-term photovoltaic power generation prediction method based on decomposition-prediction-ensemble as described in claim 1, characterized in that, The steps for performing adaptive noisy complete set empirical mode decomposition include: In each stage of the decomposition, a pair of Gaussian white noise with a specific amplitude is added to the at least one high-frequency modal component to be decomposed. The pair of Gaussian white noise has opposite signs. Empirical mode decomposition is performed on the signal after adding noise to extract the intrinsic mode functions. Finally, the multiple secondary mode components are obtained by averaging the intrinsic mode functions obtained from multiple decompositions.

6. The short-term photovoltaic power generation prediction method based on decomposition-prediction-ensemble as described in claim 1, characterized in that, The temporal convolutional network model employs a one-dimensional dilated causal convolutional structure, where the dilation coefficient d increases exponentially with the number of network layers n, and the relationship is as follows: By using interval sampling, the receptive field is increased to capture the long-period dependence in the low-frequency modal components; The Long Short-Term Memory (LSTM) network model includes a memory unit and three control gates: a forget gate, an input gate, and an output gate. These three control gates work together to learn and predict time series patterns with high volatility and nonlinear characteristics in the quadratic modal components.

7. A short-term photovoltaic power generation prediction system based on decomposition-prediction-ensemble, characterized in that, include: The data acquisition and preprocessing module is used to acquire raw time series data containing historical photovoltaic power generation data and to preprocess the raw time series data. The signal decomposition module is used to perform variational mode decomposition on the preprocessed photovoltaic power generation data to obtain multiple first-order mode components, and divide them into at least one high-frequency mode component and multiple low-frequency mode components according to the center frequency of each first-order mode component. For the at least one high-frequency modal component, perform adaptive noise complete set empirical mode decomposition to obtain multiple secondary modal components; The model prediction module is used to input the multiple low-frequency modal components into a temporal convolutional network model for prediction, and obtain the prediction results of each low-frequency modal component; and to input the multiple secondary modal components into a long short-term memory network model for prediction, and obtain the prediction results of each secondary modal component. The ensemble prediction module is used to time-series superimpose the prediction results of each low-frequency mode component with the prediction results of each secondary mode component to obtain the final photovoltaic power generation prediction value. Before the ensemble prediction module, a step is also included to automatically optimize the hyperparameters of the temporal convolutional network model and the long short-term memory network model using the Hyperband algorithm. The Hyperband algorithm treats hyperparameter optimization as a resource allocation problem. Through a successive halving strategy, it terminates the poor-performing hyperparameter combinations in advance during the training process and concentrates the computing resources on the high-performing hyperparameter combinations to determine the optimal hyperparameter configuration.

8. An electronic device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the steps of a short-term photovoltaic power generation prediction method based on decomposition-prediction-set as described in any one of claims 1-6.

9. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions, which, when executed by a processor, implement the steps of a short-term photovoltaic power generation prediction method based on decomposition-prediction-set as described in any one of claims 1-6.

Images