A method and system for wind and solar power output prediction based on multi-algorithm complementary integration

Abstract

The present application relates to a kind of wind and light output prediction method and system of multi-algorithm complementary integration, the method includes: collecting the historical power data and meteorological data of wind farm and photovoltaic power station;Introduce successive variational mode decomposition, utilize ivy optimization algorithm to adaptively optimize mode number and penalty factor;The historical power data is input into successive variational mode decomposition using optimal parameter combination, obtains multiple different center frequency mode components;Each mode component is time-aligned with meteorological data and constructs multivariate feature matrix;iTransformer prediction model is constructed, the complete time series of each variable in feature matrix is used as independent variable token, is projected output prediction sequence by embedding mapping, multi-head self-attention mechanism extraction global dependence. Compared with prior art, the present application effectively avoids mode aliasing, reduces data complexity, fully captures the global correlation of multivariate, significantly improves the prediction accuracy of wind and light output.

Description

Technical Field

[0001] This invention relates to the field of new energy power output prediction technology, and in particular to a wind and solar power output prediction method and system that integrates multiple complementary algorithms. Background Technology

[0002] Against the backdrop of accelerated global energy structure transformation, the scale of new energy power generation, represented by wind and solar power, is rapidly expanding, and its penetration rate in the power system is constantly rising. However, the inherent volatility of wind and solar resources leads to significant nonlinear, non-stationary, and highly stochastic characteristics in their power output process, posing an unprecedented challenge to the safety, stability, and dispatch control of the power system. Therefore, improving the accuracy of wind and solar power prediction has become a core challenge that urgently needs to be addressed in the research field of ensuring the large-scale grid-connected operation of new energy sources.

[0003] Existing wind and solar power output prediction methods mainly include physical models, statistical models, and combined models. Physical models are highly dependent on meteorological conditions and equipment parameters, and are computationally complex; traditional statistical models have limited ability to represent nonlinearities in highly random data; combined models, by introducing neural networks and deep learning, can effectively capture nonlinear relationships in the data and improve prediction capabilities. Commonly used combined models include Long Short-Term Memory Neural Networks (LSTM), Recurrent Neural Networks (RNNs), and Gated Recurrent Units (GRUs). RNNs perform well in short-term memory but suffer from the vanishing gradient problem and have difficulty handling long-term dependencies; GUs improve training efficiency by simplifying the LSTM structure, while LSTMs are more advantageous in tracking long-term dependencies and mitigating gradient vanishing, but their parameter selection relies on subjective experience, limiting performance. Given the large prediction errors of single models, combined models have become the mainstream approach. Although single neural network models can capture some nonlinear features, they still have shortcomings in handling long-term dependencies, noise disturbances, and mode mixing problems.

[0004] Patent CN113343590A proposes a wind speed prediction method and system based on a combined model, relating to the field of wind speed prediction for wind power generation. The method includes: collecting historical wind speed data to construct an original wind speed dataset; using a variational mode decomposition (VMD) algorithm to decompose the wind speed time series of the original wind speed dataset into N modal components; the N modal components include K intrinsic mode components and 1 residual component; each modal component is individually input into a pre-trained improved Transformer model for prediction, obtaining the prediction results of the intrinsic mode components and the residual component; the prediction results of the intrinsic mode components and the residual component are superimposed to obtain the final wind speed prediction result. However, relying on manual observation of the center frequency to determine the number of VMD modes and lacking an adaptive optimization mechanism for the penalty factor easily leads to decomposition distortion and blind parameter tuning; simultaneously, the method of using multiple models to predict each component separately and then linearly superimposing them severs the global multivariate coupling dependency between each modal component and meteorological variables. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a wind and solar power output prediction method and system that uses the Ivy optimization algorithm to adaptively optimize the successive variational mode decomposition parameters to reduce data complexity and avoid mode aliasing, and combines iTransformer to model the global dependency relationship between each mode component and meteorological variables as independent tokens, which significantly improves the prediction accuracy and stability of wind and solar power output.

[0006] The objective of this invention can be achieved through the following technical solutions: A wind and solar power output prediction method integrating multiple complementary algorithms includes the following steps: Collect historical power data and corresponding meteorological data for wind farms and photovoltaic power stations; The Ivy optimization algorithm is used to adaptively optimize the number of modes and the penalty factor in successive variational mode decomposition to determine the optimal parameter combination. The collected historical power data is input into the successive variational mode decomposition using the optimal parameter combination to decompose the signal, resulting in multiple mode components with different center frequencies. The obtained modal components are time-aligned with the collected meteorological data and combined to construct a multivariate feature matrix; The multivariate feature matrix is ​​input into the iTransformer prediction model. The complete time series of each variable in the multivariate feature matrix is ​​used as an independent variable token. The global dependencies are extracted through embedding layer mapping, multi-head self-attention mechanism and output projection to generate wind power and photovoltaic power output prediction sequences for future time moments.

[0007] Furthermore, the collection of historical power data and corresponding meteorological data from wind farms and photovoltaic power stations includes: Active power data, along with corresponding meteorological parameters such as wind speed, light intensity, and temperature, are collected at multiple time points throughout the day according to preset time intervals.

[0008] Furthermore, when performing optimization using the Ivy League optimization algorithm, the minimum envelope entropy is used as the fitness function, and the number of modes and penalty factor corresponding to the minimum envelope entropy are solved by iteratively updating the population position.

[0009] Furthermore, the population initialization formula for the Ivy League optimization algorithm is: in, For the first Each individual position vector represents a set of parameters of the successive variational mode decomposition model; and Let these represent the upper and lower bound vectors of the search space, respectively. Represents a uniformly distributed random vector in the range [0,1]. This represents the element-wise multiplication operator; This represents a population size parameter.

[0010] Furthermore, the Ivy optimization algorithm updates the candidate parameter position vector of the successive variational mode decomposition through phototactic motion, and the update formula is as follows: in, Indicates the reference location of the neighbor; Indicates a specific neighborhood location; Indicates the location of a special marker; This represents the global optimal parameter position vector that minimizes the envelope entropy of the successive variational mode decomposition. Represents an individual Updated position vector; Represents the absolute value of the normally distributed sampling vector; This represents the new parameter position vector used for the next round of decomposition evaluation after the update.

[0011] Furthermore, in the Ivy optimization algorithm, the spatial location of each individual corresponds to a set of mode numbers and penalty factor candidate parameters from the successive variational mode decomposition; the Ivy optimization algorithm dynamically adjusts the parameter optimization step size using the growth rate change formula: in, For growth rate variables; Indicates the basic growth coefficient; Represents the squared term of the random disturbance; The sampling vector is a standard normal distribution. For the first t In the nth iteration The change in growth rate of each parameter individual; the search position vector of the modality number and penalty factor is updated based on the change in growth rate.

[0012] Furthermore, in the successive variational mode decomposition process, the constructed augmented Lagrangian function is: in, For the first The center frequency of each modal component; It is a secondary penalty factor used to balance the trade-off between data fidelity and modal bandwidth; It is a pulse function; The original landscape output sequence; For each modal component; It is a Lagrange function.

[0013] Furthermore, the input variables included in the multivariate feature matrix are: each modal component of the wind power output, each modal component of the photovoltaic power output, and each meteorological variable in the collected meteorological data.

[0014] Furthermore, the iTransformer prediction model calculates the discrete Fourier transform of time-domain data using the following formula: The formula for recovering a time-domain signal using the inverse Fourier transform is: in, It is time series data. For time-domain landscape signals, For the frequency domain spectrum, The length of the time series; The calculation formula for the multi-head self-attention mechanism is as follows: in, For query matrix; The key matrix; It is a value matrix; is the number of column dimensions of the key matrix; This represents the transpose of the key matrix.

[0015] A wind and solar power output prediction system based on the complementary integration of multiple algorithms includes: The data acquisition module is used to collect historical power data and corresponding meteorological data from wind farms and photovoltaic power plants. The parameter optimization module is used to adaptively optimize the number of modes and penalty factor in successive variational mode decomposition using the Ivy optimization algorithm to determine the optimal parameter combination. The mode decomposition module is used to input historical power data into successive variational mode decomposition with optimal parameter combination to decompose the signal and obtain multiple mode components with different center frequencies. The feature construction module is used to time-align the obtained modal components with meteorological data and combine them to construct a multivariate feature matrix. The power output prediction module is used to construct the iTransformer prediction model. The multivariate feature matrix is ​​input into the iTransformer prediction model. The complete time series of each variable in the multivariate feature matrix is ​​used as an independent variable token. The global dependency is extracted through embedding layer mapping, multi-head self-attention mechanism and output projection to generate wind power and photovoltaic power output prediction sequences for future time moments.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. Strong adaptive decomposition capability, avoiding mode aliasing: The Ivy optimization algorithm is used to adaptively optimize the number of modes and penalty factor in successive variational mode decomposition, which gets rid of the dependence of traditional decomposition methods on manual parameter tuning experience, adaptively matches the optimal parameters, and avoids mode aliasing, under-decomposition and over-decomposition problems in the signal decomposition process.

[0017] 2. Effectively reduces data complexity and decreases the difficulty of prediction and fitting: The successive variational mode decomposition with optimal parameter combination is used to decompose the original historical power data into signal components with multiple different center frequencies. This decouples the highly fluctuating and strongly random non-stationary original power sequence into a stationary frequency band sequence with clear frequency localization, effectively reducing data complexity and nonlinearity, and alleviating the difficulty of prediction and fitting of subsequent models.

[0018] 3. Rich feature representation dimensions, realizing deep fusion of multi-source data: The modal components obtained by decomposition are time-aligned with the collected meteorological data and a multivariate feature matrix is ​​constructed, so that the intrinsic frequency domain characteristics of power generation are deeply integrated with external meteorological driving factors such as wind speed, light, and temperature, which comprehensively enriches the representation dimensions of input features.

[0019] 4. Complete preservation of long-term and short-term time series evolution, improving the ability to resist noise disturbances: In the iTransformer prediction model, the complete time series of each variable in the multivariate feature matrix is ​​used as an independent variable token, which preserves the global evolution trend of each variable within the complete time window. This avoids the problems of traditional time series models being sensitive to local noise due to single time step segmentation and easy decay due to long-term dependence.

[0020] 5. Superior global multivariate correlation mining significantly improves power output prediction accuracy: By utilizing embedding layer mapping and multi-head self-attention mechanism to extract global dependencies between variables and output prediction sequences, the deep coupling relationship between each modal component and each meteorological variable is fully captured, significantly reducing prediction errors and improving the accuracy and stability of wind and solar power output prediction. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a diagram of the wind and solar power output prediction model architecture of the present invention; Figure 3 This is a flowchart of the wind and solar power output prediction model algorithm of the present invention; Figure 4 A comparison chart of wind power forecast accuracy; Figure 5 A comparison chart of photovoltaic forecast accuracy; Figure 6 This is a block diagram of the system modules of the present invention. Detailed Implementation

[0022] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0023] Example 1 This embodiment provides a wind and solar power output prediction method that integrates multiple complementary algorithms, such as... Figure 1 As shown, it includes the following steps: Collect historical power data and corresponding meteorological data for wind farms and photovoltaic power stations; A joint solution framework of Ivy League Optimization Algorithm (IVYA) and Successive Variational Mode Decomposition (SVMD) is constructed, with minimum envelope entropy as the fitness function. Ivy League Optimization Algorithm iteratively updates the population position update parameters, calculates the envelope entropy corresponding to the signal after SVMD, and seeks the minimum value based on the envelope entropy calculation result, thereby achieving adaptive optimization of the number of modes and penalty factor in SVMD and determining the optimal parameter combination. The collected historical power data is input into the SVMD using the optimal parameter combination for signal decomposition, resulting in multiple modal components with different center frequencies. like Figure 2 This is an architecture diagram of the Ivy League optimization algorithm combined with successive variational mode decomposition (SMD) combined with time-series transformation prediction model constructed in this embodiment. The top part is the data processing framework. After inputting the historical wind and solar power data sequence, it sequentially goes through the initialization of SMD parameters, the dynamic adjustment of SMD parameters by the Ivy League optimization algorithm, and the execution of SMD, finally obtaining the intrinsic mode components and residuals. On the lower left side of the architecture, the extracted feature time-series waveforms are used as independent variable tokens for feature value input. The embedding block maps each numerical sequence into a high-dimensional feature representation and generates feature vectors. Then, the features are input into the processing block on the right, and sequentially go through a multivariate attention mechanism, a normalization layer, a feedforward neural network, and a normalization layer for global feature extraction and deep correlation modeling. Finally, the output layer generates the wind and solar power output prediction result sequence.

[0024] like Figure 3 As shown, the algorithm flow of the ivy optimization algorithm combined with successive variational mode decomposition and time-series transformation prediction model constructed in this embodiment is as follows: First, the parameter initialization stage is entered to initialize the algorithm parameters and input initial data to load historical wind and solar power and meteorological data; then, the ivy optimization algorithm enters the iterative optimization stage, sequentially executing growth rate modeling, phototactic motion, propagation evolution trend, and survival strategy selection, and then outputs model parameters to obtain the candidate mode number and penalty factor combination; next, successive variational mode decomposition signal decomposition is executed to decompose the initial data, and the minimum envelope entropy fitness function is calculated to evaluate the decomposition effect; then, a convergence determination is made. If convergence is not achieved, the growth rate modeling step is returned to continue iterative optimization. If convergence is achieved, the feature data obtained by the optimal parameter decomposition is input into the iTransformer prediction model for global time-series modeling, and finally, the wind and solar power output prediction sequence is output and the process ends.

[0025] The steps of the Ivy League optimization algorithm are as follows: Initialization phase: In the formula, Let be the position vector of the kth individual, representing a set of parameters for SVMD; and These represent the search space boundary vectors, respectively. Represents a uniformly distributed random vector in the range [0,1]. This represents the element-wise multiplication operator; This represents a population size parameter.

[0026] Growth rate modeling: In the formula, For growth rate variables; Indicates the basic growth coefficient; Represents the squared term of the random disturbance; The sampling vector is a standard normal distribution. This represents the change in growth rate; This represents the current iteration number.

[0027] Phototaxis: In the formula, Indicates the reference location of the neighbor; Indicates a specific neighborhood location; Indicates the location of a special marker; Indicates the globally optimal position; Represents an individual Updated position vector; The absolute value of the sample vector from the normal distribution.

[0028] Evolutionary trends in propagation: Survival strategy selection: By dynamic threshold The system enables the switching of growth stages. When an individual's performance exceeds a threshold, an expansion mode is triggered; otherwise, the climbing mode is maintained, simulating the stage characteristics of plant growth. In the formula, Let be the objective function, if If the branches and leaves expand, the plant will begin to widen; otherwise, the plant will begin to grow upwards.

[0029] In the coupled optimization framework of the Ivy League optimization algorithm and successive variational mode decomposition, the Ivy League optimization algorithm iterates continuously, repeatedly executing steps such as population initialization, growth rate adjustment, phototactic movement, propagation evolution, and survival selection, until the change in the set fitness function is less than a preset convergence threshold. At this point, the algorithm determines that the optimization process has converged and outputs the parameter combination corresponding to the current best individual, which serves as the optimal number of modes and penalty factor for the successive variational mode decomposition.

[0030] The successive variational mode decomposition algorithm is as follows: 1) Let the original wind and solar power output sequence be... The goal of successive variational mode decomposition is to classify it into successive levels. Modal components This allows each modal component to represent a different frequency component in the signal that characterizes a specific feature. In each decomposition process, a modal component is obtained by solving an optimization problem, and then it is subtracted from the original signal. This process is repeated for the remaining signal until the stopping condition is met.

[0031] In each decomposition When separating the modal components, the objective function is to minimize the correlation term of the modal component's center frequency, while satisfying a certain constraint relationship between the modal components and the remaining signal. The goal of successive variational mode decomposition is to separate individual modal components from the original signal one by one. In separating the _th modal component... When dealing with a single component, it requires solving a constrained optimization problem, which is quite difficult. To facilitate the solution, Lagrange multipliers and a quadratic penalty factor are introduced to add constraints to the objective function. Therefore, this paper constructs an augmented Lagrange function, thus transforming it into an unconstrained problem. Let the current residual signal be... For the first The augmented Lagrangian function constructed by the second decomposition is as follows. In the formula, the first term is the original signal decomposition objective function, where, For the first The center frequency of each modal component; It is a secondary penalty factor used to balance the trade-off between data fidelity and modal bandwidth; It is a pulse function; It is the Lagrangian function, used to incorporate constraints into the objective function.

[0032] The successive variational mode decomposition model uses the alternating direction multiplier method to alternately update in the frequency domain. until the constructed objective function converges, thus finding each and And generate corresponding components of different modes.

[0033] Suppose the initial wind and solar data input to the model and the predicted wind and solar data output are as shown in the following formula: The obtained modal components are time-aligned with the collected meteorological data and combined to construct a multivariate feature matrix; The multivariate feature matrix is ​​input into the iTransformer prediction model. The complete time series of each variable in the multivariate feature matrix is ​​used as an independent variable token. The global dependencies are extracted through embedding layer mapping, multi-head self-attention mechanism and output projection to generate wind power and photovoltaic power output prediction sequences for future time moments.

[0034] The iTransform prediction model's prediction process is as follows: Frequency domain conversion: In the iTransformer prediction model, frequency domain transformation is a crucial initial step for efficient prediction, converting wind and solar power output signals into spectral and phase information via Fourier transform. For a given length of... time domain signal Its discrete Fourier transform is defined as: In the formula, The spectrum is represented. The inverse Fourier transform is used to recover the time-domain signal, and it is defined as: Frequency domain processing: The frequency domain representation is input to a Transformer encoder consisting of M stacked layers, each containing two sub-layers. Multi-head self-attention captures global dependencies, while a feedforward network models feature patterns. The iTransformer encoding block consists of L stacked modules containing layers for normalization, feedforward networks, and self-attention modules. The normalization layers ensure a Gaussian distribution for univariate sequences, reducing measurement variability. The feedforward network provides a complex representation of the time series. The self-attention module models the correlations between different variables, i.e., different features of the wind and light data. It obtains the values ​​of the query matrix Q, key matrix K, and value matrix V through linear projection and calculates the softmax and concat scores to reveal the correlations between variables.

[0035] Time Domain Recovery: After the encoder outputs the frequency domain representation, the decoder uses an inverse Fourier transform to convert it back to the time domain landscape prediction map. The mathematical process of recovering the time domain signal using the inverse Fourier transform is as follows: In the formula, For predicting the spectrum.

[0036] To verify the superiority of the proposed method, comparative experiments were conducted on the following four models: traditional LSTM, the basic iTransformer and SVMD-iTransformer combined model, and the IVYA-SVMD-iTransformer combined model.

[0037] The comparison and prediction results are shown in Table 1: Table 1 Prediction results of various models In the table, MSE, MAE, MAPE, and R2 represent mean squared error, mean absolute error, mean absolute percentage error, and coefficient of determination, respectively.

[0038] As can be seen from the comparison data in Table 1, the traditional LSTM model is significantly inferior to other models in all aspects because it is difficult to capture complex nonlinearities and long-term dependencies. Although the traditional iTransformer model is better than LSTM, it still has shortcomings in predicting wind and solar power output under the influence of strong and variable factors such as wind speed and illumination.

[0039] like Figure 4 As shown in the radar chart comparing wind power output predictions, the evaluation indices of each model exhibit a clear hierarchical distribution. While the MAE index of the SVMD-iTransformer model is slightly higher than that of the single iTransformer model, the coefficient of determination... A slight optimization was achieved; however, the IVYA-SVMD-iTransformer prediction model proposed in this invention performed best, with MSE, MAE, and MAPE decreasing by approximately 0.77%, 1.18%, and 6.84% respectively compared to the SVMD-iTransformer model. Figure 4 The intuitive representation is that the vertices corresponding to the mean squared error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) all converge significantly towards the central axis, and the coefficient of determination... The corresponding vertex is closest to the outermost boundary.

[0040] like Figure 5 As shown in the radar chart comparing photovoltaic power output predictions, the stepwise improvement in model performance is more significant. The SVMD-iTransformer model's predicted MSE for photovoltaic power output decreased to 3.165, a 21.54% reduction compared to the iTransformer. The accuracy reached 0.986, an improvement of 0.41%; and the IVYA-SVMD-iTransformer prediction model was further optimized on this basis, with MSE, MAE, and MAPE decreasing significantly by 37.25%, 24.15%, and 24.36% respectively compared to the SVMD-iTransformer model. An increase of 0.51%, in Figure 5 The most intuitive manifestation of this is that the coverage area of ​​each error dimension shrinks significantly inward, demonstrating excellent error suppression capabilities.

[0041] The aforementioned indicators and the changing distribution of the spectrum strongly confirm that the IVYA algorithm, by improving SVMD decomposition to make the modal components more closely match the actual characteristics of the data, combined with the powerful learning capabilities of iTransformer, significantly improves the accuracy of wind and solar power prediction. The combined model has outstanding advantages in predicting new energy output.

[0042] Example 2 This embodiment provides a wind and solar power output prediction system that integrates multiple complementary algorithms, such as... Figure 6 As shown, it includes the following modules: Data acquisition module 1 is used to collect historical power data and corresponding meteorological data of wind farms and photovoltaic power stations; Parameter optimization module 2 is used to adaptively optimize the number of modes and penalty factor of successive variational mode decomposition using the Ivy optimization algorithm to determine the optimal parameter combination; The mode decomposition module 3 is used to input historical power data into successive variational mode decomposition with optimal parameter combination to decompose the signal and obtain multiple mode components with different center frequencies. Feature construction module 4 is used to time-align the obtained modal components with meteorological data and combine them to construct a multivariate feature matrix; The power output prediction module 5 is used to construct the iTransformer prediction model. The multivariate feature matrix is ​​input into the iTransformer prediction model. The complete time series of each variable in the multivariate feature matrix is ​​used as an independent variable token. The global dependency is extracted through embedding layer mapping, multi-head self-attention mechanism and output projection to generate the wind power and photovoltaic power output prediction sequence for future time.

[0043] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A wind and solar power output prediction method integrating multiple complementary algorithms, characterized in that, Includes the following steps: Collect historical power data and corresponding meteorological data for wind farms and photovoltaic power stations; The Ivy optimization algorithm is used to adaptively optimize the number of modes and the penalty factor in successive variational mode decomposition to determine the optimal parameter combination. The collected historical power data is input into the successive variational mode decomposition using the optimal parameter combination to decompose the signal, resulting in multiple mode components with different center frequencies. The obtained modal components are time-aligned with the collected meteorological data and combined to construct a multivariate feature matrix; The multivariate feature matrix is ​​input into the iTransformer prediction model. The complete time series of each variable in the multivariate feature matrix is ​​used as an independent variable token. The global dependencies are extracted through embedding layer mapping, multi-head self-attention mechanism and output projection to generate wind power and photovoltaic power output prediction sequences for future time moments.

2. The wind and solar power output prediction method based on multi-algorithm complementary integration according to claim 1, characterized in that, The collection of historical power data and corresponding meteorological data from wind farms and photovoltaic power stations includes: Active power data, along with corresponding meteorological parameters such as wind speed, light intensity, and temperature, are collected at multiple time points throughout the day according to preset time intervals.

3. The wind and solar power output prediction method based on multi-algorithm complementary integration according to claim 1, characterized in that, The Ivy League optimization algorithm, when performing optimization, uses the minimum envelope entropy as the fitness function and solves for the number of modes and penalty factor corresponding to the minimum envelope entropy by iteratively updating the population position.

4. The wind and solar power output prediction method based on multi-algorithm complementary integration according to claim 3, characterized in that, The population initialization formula for the Ivy League optimization algorithm is: in, For the first Each individual position vector represents a set of parameters of the successive variational mode decomposition model; and Let these represent the upper and lower bound vectors of the search space, respectively. Represents a uniformly distributed random vector in the range [0,1]. This represents the element-wise multiplication operator; This represents a population size parameter.

5. The wind and solar power output prediction method based on multi-algorithm complementary integration according to claim 3, characterized in that, The ivy optimization algorithm updates the candidate parameter position vector of the successive variational mode decomposition through phototactic motion, and the update formula is as follows: in, Indicates the reference location of the neighbor; Indicates a specific neighborhood location; Indicates the location of a special marker; This represents the global optimal parameter position vector that minimizes the envelope entropy of the successive variational mode decomposition. Represents an individual Updated position vector; Represents the absolute value of the normally distributed sampling vector; This represents the new parameter position vector used for the next round of decomposition evaluation after the update.

6. The wind and solar power output prediction method based on multi-algorithm complementary integration according to claim 3, characterized in that, In the Ivy optimization algorithm, the spatial location of each individual corresponds to a set of mode numbers and penalty factor candidate parameters in the successive variational mode decomposition; the Ivy optimization algorithm dynamically adjusts the parameter optimization step size through the growth rate change formula: in, For growth rate variables; Indicates the basic growth coefficient; Represents the squared term of the random disturbance; The sampling vector is a standard normal distribution. For the first t In the nth iteration The change in growth rate of each parameter individual; the search position vector of the modality number and penalty factor is updated based on the change in growth rate.

7. The wind and solar power output prediction method based on multi-algorithm complementary integration according to claim 1, characterized in that, The augmented Lagrangian function constructed during the successive variational mode decomposition process is: in, For the first The center frequency of each modal component; It is a secondary penalty factor used to balance the trade-off between data fidelity and modal bandwidth; It is a pulse function; The original landscape output sequence; For each modal component; It is a Lagrange function.

8. The wind and solar power output prediction method based on multi-algorithm complementary integration according to claim 1, characterized in that, The input variables included in the multivariate feature matrix are: each modal component of the wind power output, each modal component of the photovoltaic power output, and each meteorological variable in the collected meteorological data.

9. The wind and solar power output prediction method based on multi-algorithm complementary integration according to claim 1, characterized in that, The iTransformer prediction model calculates the discrete Fourier transform of time-domain data using the following formula: The formula for recovering a time-domain signal using the inverse Fourier transform is: in, It is time series data. For time-domain landscape signals, For the frequency domain spectrum, The length of the time series; The calculation formula for the multi-head self-attention mechanism is as follows: in, For query matrix; The key matrix; It is a value matrix; is the number of column dimensions of the key matrix; This represents the transpose of the key matrix.

10. A wind and solar power output prediction system based on the complementary integration of multiple algorithms, characterized in that, include: The data acquisition module is used to collect historical power data and corresponding meteorological data from wind farms and photovoltaic power plants. The parameter optimization module is used to adaptively optimize the number of modes and penalty factor in successive variational mode decomposition using the Ivy optimization algorithm to determine the optimal parameter combination. The mode decomposition module is used to input historical power data into successive variational mode decomposition with optimal parameter combination to decompose the signal and obtain multiple mode components with different center frequencies. The feature construction module is used to time-align the obtained modal components with meteorological data and combine them to construct a multivariate feature matrix. The power output prediction module is used to construct the iTransformer prediction model. The multivariate feature matrix is ​​input into the iTransformer prediction model. The complete time series of each variable in the multivariate feature matrix is ​​used as an independent variable token. The global dependency is extracted through embedding layer mapping, multi-head self-attention mechanism and output projection to generate wind power and photovoltaic power output prediction sequences for future time moments.

Images