A method and system for intelligent complementary prediction of new energy power generation

Abstract

The application provides a new energy power intelligent complementary prediction method and system, the method comprises the following steps: acquiring wind speed, solar radiation, meteorological factors, numerical weather prediction and historical actual power; constructing wind-solar complementary correlation characteristics and extracting influence factors, OPMD decomposition and entropy constraint recombination are performed on the wind speed sequence and the solar radiation sequence; through the improved DOA and the introduction of the time-varying cooperative gate and the dynamic trend compensation mechanism, the TiDE model is improved to carry out double-branch prediction, and the initial wind speed prediction value and the initial solar radiation prediction value are obtained; the GPR residual correction model optimized by the improved DOA is used to correct the residual error of the initial prediction result; according to the dynamic updating of the mapping relationship from the wind speed and the solar radiation to the power, the wind power prediction value and the photovoltaic power prediction value are obtained, finally, fusion is carried out, and the new energy power prediction result is obtained, so as to improve the power prediction accuracy and the robustness under the wind-solar complementary scene.

Description

Technical Field

[0001] This invention relates to the field of new energy power generation prediction technology, and in particular to a method and system for intelligent complementary prediction of new energy power generation. Background Technology

[0002] New energy power generation, such as wind power and photovoltaic power, is developing rapidly. Wind and photovoltaic power generation have advantages such as being clean, renewable, and flexible in deployment; however, their power output is affected by multiple factors, including wind speed, solar irradiance, temperature, humidity, air pressure, and numerical weather forecasts, exhibiting characteristics of volatility, randomness, and intermittency. The accuracy of new energy power generation forecasting determines the safe and stable operation of the power grid dispatch, thus affecting wind and solar resource utilization, energy storage coordinated regulation, and electricity market trading decisions.

[0003] Most existing methods for predicting renewable energy power generation are designed to model wind power or solar power separately. Common methods include time series forecasting based on statistical analysis, physical modeling based on mechanistic analysis, and data-driven methods based on machine learning or deep learning. While these methods can achieve power prediction under certain conditions, for the joint prediction problem in wind-solar hybrid scenarios, they generally treat wind power and solar power predictions independently and then simply superimpose them for output. This approach fails to fully characterize the complementary relationship between wind speed series and solar irradiance series in a time-varying environment, resulting in overall prediction results that do not adequately reflect the synergistic output characteristics of the renewable energy system.

[0004] Furthermore, wind speed and solar irradiance sequences typically exhibit strong non-stationarity, multi-scale fluctuations, and noise disturbances. Directly modeling based on raw time-series data will lead to overreaction to high-frequency disturbances, thus reducing the accuracy and stability of wind and solar power predictions. While some studies have employed time-series decomposition methods or deep network models, their characterization of the differences in complexity levels, information entropy characteristics, and wind-solar complementarity constraints among the decomposed subsequences is incomplete, making it difficult to simultaneously depict the dynamic evolution patterns across various time scales.

[0005] Furthermore, existing methods often lack effective correction mechanisms for residual evolution after the initial prediction results are generated, or they only use a single error correction method, failing to comprehensively utilize key meteorological factors, complementary correlation characteristics, and the nonlinear mapping relationship between prediction residuals. Therefore, under complex meteorological conditions, sudden weather processes, and frequent changes in operating status, existing methods are prone to the accumulation of prediction biases, resulting in insufficient robustness of the final power output results.

[0006] On the other hand, in existing research on renewable energy power prediction, the conversion process from wind speed to wind power and from solar irradiance to photovoltaic power largely relies on relatively fixed mapping relationships. It has not fully incorporated the dynamic effects of current meteorological scenarios, historical operating conditions, and residual correction information on the power mapping process, thus resulting in discrepancies between the final prediction results and actual operating conditions. Especially in wind-solar hybrid scenarios, if static mapping or independent modeling of single branches is still used, it is difficult to reflect the synergistic compensation effect between wind power branches and photovoltaic branches.

[0007] Therefore, it is necessary to propose an intelligent complementary prediction method and system for new energy power generation, which incorporates the correlation characteristics of wind and solar complementarity into the entire process of key factor screening, time series decomposition and recombination, dual-branch prediction, residual correction and power synergy fusion. At the same time, it combines the improved DOA-optimized TiDE model, GPR residual correction model and dynamic power mapping mechanism to improve the accuracy, robustness and engineering applicability of new energy power generation prediction in wind and solar complementarity scenarios. Summary of the Invention

[0008] Purpose of the invention: The technical problem to be solved by the present invention is to address the shortcomings of the existing technology by providing a method and system for intelligent complementary prediction of new energy power generation. This method and system can achieve joint and coordinated prediction of wind speed, solar irradiance and power generation in wind-solar complementary scenarios, thereby improving the accuracy, robustness and engineering application value of new energy power generation prediction, and has strong practical significance.

[0009] The method includes the following steps:

[0010] Step 1: Obtain historical wind speed, solar irradiance, meteorological factors, numerical weather prediction and historical actual power data, and perform time alignment, missing value processing and normalization preprocessing.

[0011] Step 2: Construct wind-solar complementary correlation features based on wind speed, solar irradiance and meteorological factors, and extract key factors for wind speed prediction and solar irradiance prediction by combining random forest;

[0012] Step 3: Decompose the preprocessed wind speed sequence and solar irradiance sequence using the Ordinal Pattern-based Mode Decomposition (OPMD) method, calculate the information entropy and conditional entropy of each intrinsic mode component, and reconstruct the subsequences by combining the correlation features to obtain the wind speed reconstructed subsequence and the solar irradiance reconstructed subsequence.

[0013] Step 4: Input the wind speed reconstruction subsequence and the solar irradiance reconstruction subsequence into the improved time-series dense encoder (TiDE) model, which introduces time-varying collaborative gating and dynamic trend compensation mechanisms, respectively, to obtain the prediction results corresponding to each subsequence, and reconstruct the prediction results of each subsequence to obtain the initial wind speed prediction value and the initial solar irradiance prediction value.

[0014] Step 5: Based on historical real wind speed, historical real solar irradiance, and initial wind speed and solar irradiance predictions, construct historical residual samples. Use the improved Dragonfly Optimization Algorithm (DOA) to correct the parameters of the Gaussian Process Regression (GPR) model and establish a GPR residual correction model. Input the key factors at the current moment, wind-solar complementarity correlation features, and initial wind speed and solar irradiance predictions into the GPR residual correction model to obtain wind speed prediction residuals and solar irradiance prediction residuals, and obtain the corrected wind speed and solar irradiance predictions.

[0015] Step 6: Based on the current meteorological scenario and historical operating status, dynamically update the mapping relationship between wind speed and wind power and the mapping relationship between solar irradiance and photovoltaic power, and convert the corrected wind speed prediction value and the corrected solar irradiance prediction value into wind power prediction value and photovoltaic power prediction value, respectively.

[0016] Step 2 includes: constructing a wind-solar complementary driving index C based on the changing trends, fluctuation amplitudes, and complementarity of wind speed and solar irradiance sequences at the same time scale. t And based on the wind-solar complementary driving index C t Candidate meteorological influencing factors are screened, ranked, and weighted to form a set of key factors for subsequent wind speed and solar irradiance prediction; the wind-solar hybrid driving index C t The expression is:

[0017] ,

[0018] Among them, C t ρ represents the wind-solar hybrid driving index at time t; λ1, λ2, and λ3 represent weighting coefficients; satisfying λ1+λ2+λ3=1 and λ1, λ2, λ3∈[0,1]; t It is the Pearson correlation coefficient between the wind speed series and the solar irradiance series within a sliding time window; v t G represents the wind speed at time t; t Δv represents the solar irradiance at time t. t =v t -vt-1 ΔG represents the wind speed increment at time t. t =G t -G t-1 This represents the increase in solar irradiance at time t; This represents the local standard deviation of the wind speed sequence within the current sliding time window; This represents the local standard deviation of the solar irradiance sequence within the current sliding time window; exp represents the smallest positive constant that prevents the denominator from being zero; exp represents the natural exponential function.

[0019] The key factor set is based on the wind-solar hybrid driving index C. t The key factors are determined together with the feature importance of the random forest output. A predetermined number of key factors are selected from the comprehensive evaluation value of the key factors in descending order. The key factors include one or more of the following: wind speed, solar irradiance, temperature, humidity, air pressure, wind direction, and numerical weather prediction data. These factors are further used as constraint information for subsequent subsequence recombination, residual correction, and power fusion to achieve collaborative prediction between wind power branches and photovoltaic branches.

[0020] Step 3 includes:

[0021] For wind speed and solar irradiance sequences, OPMD decomposition is performed to obtain two or more intrinsic mode components (IMFi). The information entropy, conditional entropy, and correlation between each IMFi and the original sequence are calculated. A recombination criterion Ω is then constructed based on the complexity and complementary correlation of each IMFi. i The expression is:

[0022] ,

[0023] Among them, Ω i The recombination criterion for the i-th intrinsic mode component; H i H represents the information entropy of the i-th intrinsic mode component; i∣C The i-th intrinsic mode component is represented by the wind-solar hybrid drive index C. t Conditional entropy; Let represent the correlation coefficient between the i-th intrinsic mode component and the original sequence; α, β, and γ represent weight coefficients, satisfying α+β+γ=1, and α, β, γ∈[0,1];

[0024] Will satisfy The intrinsic mode components are merged into the q-th reconstructed subsequence, and the reconstructed subsequence is obtained according to the following formula:

[0025] ,

[0026] Where, τ q-1 and τq S represents the lower and upper thresholds of the q-th reconstruction interval, respectively; q Z represents the set of eigenmode component indices that satisfy the interval conditions; q Represents the q-th reconstructed subsequence;

[0027] The wind speed sequence and solar irradiance sequence are decomposed and recombined according to step 3 to obtain the wind speed reconstructed subsequence set and the solar irradiance reconstructed subsequence set, thereby reducing the non-stationarity and high-frequency disturbance of the original time series signal and improving the ability of the subsequent prediction model to represent the fluctuation characteristics of different time scales.

[0028] In step 4, the improved DOA algorithm is used to optimize the key parameters of the TiDE model and the GPR residual correction model. The improved DOA algorithm simultaneously constrains the prediction error, fitting bias and statistical consistency through a composite fitness function to reduce the risk of overfitting caused by a single error index.

[0029] The expression for the composite fitness function J is:

[0030] ,

[0031] Where ω1, ω2, ω3, and ω4 represent weighting coefficients, satisfying ω1+ω2+ω3+ω4=1, and ω1, ω2, ω3, ω4∈[0,1]; RMSE represents root mean square error; MAE represents mean absolute error; KGE represents the Kling-Gupta Efficiency index; and B represents prediction bias. Represents the true mean; RMSE0 is the root mean square error of the baseline model; MAE0 is the mean absolute error of the baseline model;

[0032] The expression for the Kling-Gupta Efficiency index KGE is:

[0033] ,

[0034] Where r represents the linear correlation coefficient between the predicted value sequence and the actual value sequence; It is the standard deviation of the predicted value series; σ y The standard deviation of the true value sequence; μ represents the mean of the predicted value sequence. y This represents the mean of the sequence of true values;

[0035] In the improved DOA algorithm, the update expression for the p-th candidate solution in the q-th iteration is:

[0036] ,

[0037] in, This represents the p-th candidate solution selected in the q-th iteration; This represents the globally optimal solution obtained in the q-th iteration; This represents the randomly selected reference solution in the q-th iteration; r1 and r2 are random numbers in the range [0,1]. and b q Let represent the adaptive compression coefficient and expansion coefficient of the q-th iteration, respectively;

[0038] The adaptive compression coefficient and expansion coefficient satisfy the following conditions:

[0039] ,

[0040] Among them, a max This represents the maximum value of the adaptive compression factor; a min This represents the minimum value of the adaptive compression factor; b max This represents the maximum value of the coefficient of thermal expansion; b min The minimum value of the expansion coefficient is represented by ; Q represents the maximum number of iterations; q represents the current number of iterations. By adjusting these parameters, candidate solutions can achieve good global search performance in the early iteration stages while maintaining good local convergence in later iterations. Based on this, the stability and accuracy of the TiDE model parameters and the GPR residual correction model parameters are further improved.

[0041] In step 4, the wind speed reconstruction subsequence and the solar irradiance reconstruction subsequence are respectively input into the improved TiDE model, which incorporates time-varying collaborative gating and dynamic trend compensation mechanisms, to obtain the prediction results corresponding to each subsequence. Furthermore, the prediction results of each subsequence are reconstructed to obtain the initial wind speed prediction value and the initial solar irradiance prediction value; specifically including:

[0042] The wind speed reconstruction subsequence and corresponding key factors are input into the improved TiDE prediction model of the wind speed branch, and the solar irradiance reconstruction subsequence and corresponding key factors are input into the improved TiDE prediction model of the solar irradiance branch. In each branch, a time-varying collaborative gating vector is constructed based on the historical window of the current branch, the historical window of another complementary branch, the historical window of the key factors, and the historical window of the wind-solar complementary driving index. The input sequence of the current branch is then collaboratively enhanced, and the process satisfies the following formula:

[0043] ,

[0044] ,

[0045] Where m∈{v,G} represents the current predicted branch, v represents the wind speed branch, G represents the solar irradiance branch, and n represents another complementary branch corresponding to the current predicted branch; This represents the historical window of the reconstructed subsequence of the m-th branch with length L before time t; This indicates the historical window corresponding to another complementary branch; This represents a key factor history window of length L before time t; This represents a historical window of length L for the wind-solar hybrid drive index before time t. Represents a time-varying cooperative gating vector; This represents the Sigmoid activation function; This represents element-wise multiplication; , and Indicates the parameters to be optimized; This represents the input sequence after wind-solar hybrid enhancement.

[0046] The input sequence, after co-enhancement, is fed into the TiDE prediction model of the corresponding branch, and a dynamic trend compensation term is introduced to enhance the response to short-term fluctuations and abrupt changes. The process satisfies the following formula:

[0047] ,

[0048] ,

[0049] in, TiDE represents the predicted subsequence value of the m-th branch at time t. m Denotes the TiDE prediction model for the m-th branch; θ m This represents the parameter vector of the corresponding TiDE model; This represents the local change of the m-th branch between the last two moments; This represents the reconstructed subsequence value of the m-th branch at time t-1; This represents the reconstructed subsequence value of the m-th branch at time t-2; The dynamic trend compensation coefficient is represented by tanh(.); the hyperbolic tangent activation function is represented by tanh(.). and Indicates the parameters to be optimized; The key factor vector representing time t;

[0050] The wind speed reconstruction subsequence and its corresponding key factors are input into the TiDE prediction model of the wind speed branch to obtain the prediction results of each wind speed subsequence; the solar irradiance reconstruction subsequence and its corresponding key factors are input into the TiDE prediction model of the solar irradiance branch to obtain the prediction results of each solar irradiance subsequence. The process is as follows:

[0051] ,

[0052] ,

[0053] ,

[0054] ,

[0055] in, This represents the predicted value of the k-th wind speed reconstruction subsequence at time t; This represents the predicted value of the l-th solar irradiance reconstruction subsequence at time t; The TiDE prediction model represents the wind speed branch; The TiDE prediction model representing the solar irradiance branch; θ v θ represents the parameter vector of the TiDE prediction model for wind speed branches. G This represents the parameter vector of the TiDE prediction model for the solar irradiation branch; This represents the historical window of the k-th wind speed reconstruction subsequence of length L before time t; K represents the l-th solar irradiance reconstruction subsequence history window of length L before time t; v K represents the number of wind speed reconstructed subsequences; G Indicates the number of solar irradiation reconstructed subsequences; This represents the initial wind speed forecast value; This represents the initial solar irradiance forecast. The wind speed branch and the solar irradiance branch use the same forecasting framework and parameter settings, allowing the two different types of time series to share the overall forecasting system while retaining the same dynamic response characteristics. The initial wind speed forecast and the initial solar irradiance forecast are the basic inputs in the residual correction process.

[0056] In step 5, a residual correction model is established using an improved DOA-optimized GPR residual correction model. This GPR residual correction model takes the current key factors, the wind-solar hybrid driving index, and the initial prediction results as inputs, and the wind speed prediction residuals and solar irradiance prediction residuals as outputs, thereby performing online corrections to the initial wind speed and solar irradiance prediction values. The input vector of the residual correction model is:

[0057] ,

[0058] in, This represents the input vector of the GPR residual correction model at time t; This represents the initial wind speed forecast value; This represents the initial solar irradiance prediction.

[0059] Residual prediction using the GPR residual correction model:

[0060] ,

[0061] in, The parameter is GPR residual correction model;

[0062] The corrected wind speed prediction and solar irradiance prediction respectively satisfy:

[0063] ,

[0064] ,

[0065] in, This represents the residual for wind speed prediction; This represents the residual from the solar irradiance prediction; This represents the corrected wind speed forecast; This represents the revised solar irradiance forecast.

[0066] The kernel function of the GPR residual correction model is:

[0067] ,

[0068] Where, k(z) i ,z j ) represents the i-th input sample z i With the j-th input sample z j The kernel function value; Z represents the signal variance; D represents the input vector dimension; z i,d and z j,d Let represent the value of the i-th input sample in the d-th dimension and the value of the j-th input sample in the d-th dimension, respectively; This represents the length scale parameter of the d-th dimension; Indicates the noise variance; denoted as the Kronecker function, which takes the value 1 when i=j and 0 otherwise. This residual correction model constructs training samples using historical true values ​​and initial predicted values, and predicts the residuals based on the current input vector during network prediction, thus avoiding the problem of directly using future true values ​​as online input.

[0069] In step 6, based on the current meteorological scenario and historical operating status, the mapping relationship between wind speed and wind power, as well as the mapping relationship between solar irradiance and photovoltaic power, are dynamically updated, and the prediction results of new energy power generation are output. Specifically, to characterize the coupling relationship between residual correction information and physical mechanism mapping, a dynamic mechanism-residual coupling correction coefficient is constructed. The expression is:

[0070] ,

[0071] Among them, Λ tκ represents the dynamic mechanism-residual coupling correction coefficient at time t; κ represents the corrected weighting coefficient. This represents the residual for wind speed prediction; σ represents the residual of solar irradiance prediction; e This represents the recent residual fluctuation standard deviation;

[0072] The predicted wind power output satisfies:

[0073] ,

[0074] in, This represents the predicted wind power output at time t; v ci Represents the cut-in wind speed; v r Represents the rated wind speed; v co P represents the cut-out wind speed; r The value represents the rated power of the fan; μ represents the power exponent of the power curve.

[0075] The photovoltaic power forecast meets the following requirements:

[0076] ,

[0077] in, P represents the predicted photovoltaic power at time t; stc It is the rated power of photovoltaic power under standard testing conditions; G stc It is the irradiance under standard test conditions; β T It is the temperature correction factor; T t T is the component temperature at time t; stc It is the standard test temperature; η t It is the comprehensive efficiency correction coefficient, which is used to represent the effects of shading attenuation, pollution attenuation and amplitude limiting.

[0078] The present invention also provides a new energy power generation intelligent complementary prediction system based on the method, comprising:

[0079] The multi-source time series conditioning unit is used to acquire historical wind speed data, historical solar irradiance data, meteorological influence factor data, numerical weather prediction data, and historical actual power generation data, and to perform time alignment, missing value processing, and normalization preprocessing.

[0080] The wind-solar feedback sensing unit is used to determine the wind-solar complementary drive index C. t Constructing association features to characterize the complementary relationship between wind and solar power;

[0081] The dominant factor analysis unit is used to extract key factors affecting wind speed prediction and solar irradiance prediction based on correlation features and random forest output results.

[0082] Heterogeneous wave decomposition and recompilation units are used to perform OPMD decomposition on wind speed and solar irradiance sequences respectively, and to recombinate them according to the recombination criterion Ω. i Perform intrinsic mode component merging and reconstruction;

[0083] The dual-branch time series extrapolation unit is used to predict the wind speed reconstruction subsequence and the solar irradiance reconstruction subsequence based on the improved DOA-optimized TiDE model, and obtain the initial wind speed prediction value and the initial solar irradiance prediction value.

[0084] The bias feedback correction unit is used to perform residual correction on the initial wind speed prediction and the initial solar irradiance prediction based on the GPR residual correction model.

[0085] The present invention also provides an electronic device, including a processor and a memory, the memory storing program code that, when executed by the processor, causes the processor to perform the steps of the method.

[0086] The present invention also provides a storage medium storing a computer program or instructions that, when the computer program or instructions are run on a computer, execute the steps of the method described.

[0087] Beneficial Effects: This invention constructs a wind-solar complementary correlation-driven collaborative prediction mechanism. Unlike the conventional approach of simply paralleling and superimposing wind and solar power predictions, this invention first constructs wind-solar complementary correlation features based on wind speed sequences, solar irradiance sequences, and their meteorological influencing factors. These features are then used in key factor screening, subsequent decomposition and recombination, and the final power fusion process. This shifts the "complementary relationship" from the result layer to the modeling layer, thereby achieving intelligent complementary prediction of new energy power generation.

[0088] This invention constructs a prediction framework that integrates multi-scale decomposition and recombination with dual-branch time-series extrapolation. Specifically, for wind speed and solar irradiance sequences, OPMD is used for multi-scale decomposition, and information entropy, conditional entropy, and wind-solar complementarity features are combined to complete subsequence recombination. The reconstructed wind speed and solar irradiance subsequences are then input into a TiDE model optimized with improved DOA to achieve high-precision prediction of these two key meteorological variables. Compared to directly using a single time-series model, this invention integrates "decomposition-recombination-prediction" into a unified process and improves DOA using a composite fitness function, further enhancing the model's predictive adaptability and stability under complex fluctuation scenarios.

[0089] This invention establishes a power generation mechanism that couples residual feedback correction with dynamic updating of physical mapping. After obtaining initial wind speed and solar irradiance predictions, this invention constructs residual samples using historical real values ​​and initial predictions. It then employs a GPR residual correction model optimized with improved DOA to perform residual correction on the initial prediction results. Simultaneously, it incorporates the dynamic mapping relationship between wind speed and wind power, and between solar irradiance and photovoltaic power, to adaptively update the power conversion formula based on specific scenarios. In other words, this invention does not simply rely on the output power of the data model, but rather coordinates "residual prediction correction" and "physical mechanism mapping update" to improve the accuracy, robustness, and engineering applicability of the final new energy power generation prediction results. Attached Figure Description

[0090] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.

[0091] Figure 1 This is a system framework diagram of the present invention. Detailed Implementation

[0092] This embodiment provides a method for intelligent complementary prediction of new energy power generation. Taking a wind-solar hybrid power plant as an example, it first collects data on wind speed, solar irradiance, temperature, humidity, air pressure, wind direction, numerical weather forecasts, and actual power generation. The method includes the following steps:

[0093] Step 1: Data Preprocessing; Since data preprocessing involves performing various operations on the collected multi-source data, such as time alignment, missing value imputation, outlier removal, and normalization, the final result is an input dataset with a unified time scale. Normalization can be naturally described as follows:

[0094] ,

[0095] Where x′ is the normalized data, x min and x max These represent the minimum and maximum values ​​of the feature, respectively. Because the temporal correspondence of the data is more consistent after preprocessing, and noise interference is well suppressed, it naturally provides stable input for subsequent modeling work.

[0096] Step 2: Construction of Wind-Solar Complementary Correlation Features; Since the wind-solar complementary correlation features are constructed from wind speed series, solar radiation series, and meteorological factors, a random forest model is first used to extract the main influencing factors related to wind-solar complementarity, thereby naturally and reasonably deriving the wind-solar complementarity driving index C. t :

[0097] ,

[0098] Among them, C t λt represents the wind-solar hybrid driving index at time t; λ1, λ2, and λ3 represent the weighting coefficients; satisfying λ1 + λ2 + λ3 = 1 and λ1, λ2, λ3 ∈ [0, 1]; ρt represents the Pearson correlation coefficient between the wind speed sequence and the solar irradiance sequence within the sliding time window; Δv t =v t -v t-1 ΔGt represents the wind speed increment at time t; t -G t-1 This represents the increase in solar irradiance at time t; This represents the local standard deviation of the wind speed sequence within the current sliding time window; This represents the local standard deviation of the solar irradiance sequence within the current sliding time window; This represents a very small positive number to prevent the denominator from being zero; through this step, a set of features that can characterize the synergistic relationship between wind and solar power can be obtained.

[0099] Step 3: Sequence Decomposition and Recombination; Since the sequence decomposition and recombination performed OPMD decomposition on the wind speed sequence and solar irradiance sequence respectively, obtaining several intrinsic mode components, they can be naturally and reasonably grouped and recombined according to information entropy, conditional entropy, and correlation index. The corresponding recombination criterion Ωi is thus defined: , Among them, Ω i H represents the recombination criterion for the i-th intrinsic mode component; i H represents the information entropy of the i-th intrinsic mode component; i∣C Let represent the conditional entropy of the i-th intrinsic mode component relative to the wind-solar hybrid drive index Ct; denoted as the correlation coefficient between the i-th intrinsic mode component and the original sequence; α, β, and γ are the weight coefficients that satisfy α+β+γ=1 and α, β, γ∈[0,1]; therefore, this step naturally and appropriately obtains the wind speed reconstruction subsequence and the solar irradiance reconstruction subsequence, and thus effectively eliminates the high-frequency disturbances in the original sequence.

[0100] Step 4: Dual-branch prediction; Input the wind speed reconstruction subsequence and the solar irradiance reconstruction subsequence into the improved TiDE model, which is modified by the improved DOA algorithm and incorporates time-varying collaborative gating and dynamic trend compensation mechanisms, respectively, to obtain the initial wind speed prediction value and the initial solar irradiance prediction value. The process can be represented as follows:

[0101] The wind speed reconstruction subsequence and its corresponding key factors are input into the improved TiDE prediction model of the wind speed branch, and the solar irradiance reconstruction subsequence and its corresponding key factors are input into the improved TiDE prediction model of the solar irradiance branch. In each branch, a time-varying collaborative gating vector is constructed based on the historical window of the current branch, the historical window of another complementary branch, the historical window of the key factors, and the historical window of the wind-solar complementary driving index. The input sequence of the current branch is then collaboratively enhanced. The process satisfies the following:

[0102] ,

[0103] ,

[0104] Where m∈{v,G} represents the current predicted branch, v represents the wind speed branch, G represents the solar irradiance branch, and n represents another complementary branch corresponding to the current predicted branch; This represents the historical window of the reconstructed subsequence of the m-th branch with length L before time t; This indicates the historical window corresponding to another complementary branch; This represents a key factor history window of length L before time t; This represents a historical window of length L for the wind-solar hybrid drive index before time t. Represents a time-varying cooperative gating vector; This represents the Sigmoid activation function; This represents element-wise multiplication; , as well as Indicates the parameters to be optimized; This represents the input sequence after being enhanced by wind-solar hybrid synergy.

[0105] Furthermore, the input sequence after collaborative enhancement is fed into the TiDE prediction model of the corresponding branch, and a dynamic trend compensation term is introduced to enhance the response capability to short-term fluctuations and abrupt changes. The process satisfies:

[0106] ,

[0107] ,

[0108] in, TiDE represents the predicted subsequence value of the m-th branch at time t. m Denotes the TiDE prediction model for the m-th branch; θ m This represents the parameter vector of the corresponding TiDE model; This represents the local change of the m-th branch between the last two moments; The dynamic trend compensation coefficient is represented by tanh(.); the hyperbolic tangent activation function is represented by tanh(.). and Indicates the parameter to be optimized; F t Represents the key factor vector at time t; C t This represents the wind-solar hybrid drive index at time t.

[0109] The wind speed reconstruction subsequence and its corresponding key factors are input into the TiDE prediction model of the wind speed branch to obtain the prediction results of each wind speed subsequence; the solar irradiance reconstruction subsequence and its corresponding key factors are input into the TiDE prediction model of the solar irradiance branch to obtain the prediction results of each solar irradiance subsequence; the process can be expressed as:

[0110] ,

[0111] ,

[0112] ,

[0113] ,

[0114] in, This represents the predicted value of the k-th wind speed reconstruction subsequence at time t; This represents the predicted value of the l-th solar irradiance reconstruction subsequence at time t; The TiDE prediction model represents the wind speed branch; The TiDE prediction model representing the solar irradiance branch; θ v and θ G These represent the parameter vectors of the corresponding TiDE models; This represents the historical window of the k-th wind speed reconstruction subsequence of length L before time t; K represents the l-th solar irradiance reconstruction subsequence history window of length L before time t; v K represents the number of wind speed reconstructed subsequences; G Indicates the number of solar irradiation reconstructed subsequences; This represents the initial wind speed forecast value; This represents the initial solar irradiance forecast. The wind speed branch and the solar irradiance branch use the same forecasting framework and parameter settings, allowing the two different types of time series to share the overall forecasting system while retaining the same dynamic response characteristics. The initial wind speed forecast and the initial solar irradiance forecast are the basic inputs in the residual correction process.

[0115] Step 5: Residual Correction; Since residual correction constructs residual samples using historical true values ​​and initial predicted values, and then uses the improved DOA-optimized GPR residual correction model to correct the initial prediction results, the mathematical expression is:

[0116] ,

[0117] ,

[0118] ,

[0119] ,

[0120] in, This represents the input vector of the residual correction model at time t; This represents the initial wind speed forecast value; This represents the initial solar irradiance prediction. The parameter is GPR residual correction model; This represents the residual for wind speed prediction; This represents the residual from the solar irradiance prediction; This represents the corrected wind speed forecast value; This represents the corrected solar irradiance forecast. After correction, the model's adaptability to complex weather changes is improved.

[0121] Step 6: Power mapping and fusion output;

[0122] Since the power mapping and fusion output requires inputting the corrected wind speed prediction value into the wind power mapping model and the corrected solar irradiance prediction value into the photovoltaic power mapping model, the wind power prediction value and photovoltaic power prediction value can be obtained as follows:

[0123] ,

[0124] Among them, Λ t κ represents the dynamic mechanism-residual coupling correction coefficient at time t; κ represents the corrected weighting coefficient. This represents the residual for wind speed prediction; σ represents the residual of solar irradiance prediction; e This represents the recent residual fluctuation standard deviation;

[0125] The predicted wind power output satisfies:

[0126] ,

[0127] in, This represents the predicted wind power output at time t; v ci Represents the cut-in wind speed; v r Represents the rated wind speed; v co P represents the cut-out wind speed; r The value represents the rated power of the wind turbine; μ represents the power exponent of the power curve.

[0128] The photovoltaic power forecast meets the following requirements:

[0129] ,

[0130] in, This represents the predicted photovoltaic power at time t; This is a corrected predicted value for solar irradiance; P stc It is the rated power of photovoltaic power under standard testing conditions; G stc It is the irradiance under standard test conditions; β T It is the temperature correction factor; T t T is the component temperature at time t; stc It is the standard test temperature; η t It is the comprehensive efficiency correction coefficient, which is used to represent the effects of shading attenuation, pollution attenuation and amplitude limiting.

[0131] In one specific embodiment, a wind-solar hybrid power plant is selected as the prediction target, and data on wind speed, solar irradiance, temperature, humidity, air pressure, wind direction, numerical weather prediction, and historical actual power are collected over a continuous operating cycle. In this embodiment, the sampling interval is 15 minutes, and a rolling prediction method is used to predict the total power generation of new energy sources at the next sampling time.

[0132] In step 1, the collected wind speed, solar irradiance, temperature, humidity, air pressure, wind direction, numerical weather prediction, and historical actual power data undergo time alignment, missing value processing, outlier removal, and normalization preprocessing to construct an input sample sequence with a unified time scale. This step yields standardized input data for subsequent wind speed prediction, solar irradiance prediction, and power mapping.

[0133] In step 2, wind-solar complementarity correlation features are constructed based on wind speed, solar irradiance, and meteorological factors. A random forest is then used to rank the candidate influencing factors by importance, extracting a set of key factors. Optionally, the number of decision trees in the random forest can be set to 200, and the maximum tree depth can be set to 10. Through this step, the wind-solar complementarity driving index Ct and the set of key factors used for subsequent sequence reconstruction, dual-branch prediction, and residual correction can be obtained.

[0134] In step 3, the preprocessed wind speed and solar irradiance sequences are decomposed using the Ordinal Mode Decomposition (OPMD) method, and the intrinsic mode components are reconstructed using information entropy, conditional entropy, and correlation indices. Optionally, the number of decomposed modes for both the wind speed and solar irradiance sequences can be set to 5. This step yields sets of reconstructed wind speed and solar irradiance subsequences, respectively, thereby reducing the impact of non-stationary disturbances and random noise in the original sequences.

[0135] In step 4, the wind speed reconstruction subsequence and the solar irradiance reconstruction subsequence are respectively input into the improved time series dense encoder TiDE model, which incorporates time-varying collaborative gating and dynamic trend compensation mechanisms, for prediction, and the prediction results of each subsequence are reconstructed. Optionally, the input time window length can be set to 16 sampling points, and the prediction step size can be set to 1 sampling point. Through this step, the initial wind speed prediction value and the initial solar irradiance prediction value can be obtained.

[0136] In step 5, historical residual samples are constructed based on historical real wind speed, historical real solar irradiance, and initial predicted wind speed and solar irradiance. A GPR residual correction model optimized using the improved Dragonfly Algorithm (DOA) is then employed. Optionally, the DOA population size can be set to 30, and the maximum number of iterations can be set to 100. This step yields corrected predicted wind speed and corrected predicted solar irradiance, further reducing the initial prediction error.

[0137] In step 6, based on the current meteorological scenario and historical operating status, the mapping relationship between wind speed and wind power, as well as the mapping relationship between solar irradiance and photovoltaic power, are dynamically updated. The corrected wind speed prediction value and the corrected solar irradiance prediction value are then converted into wind power prediction value and photovoltaic power prediction value, respectively, and finally fused to obtain the predicted total power output of new energy generation. In this embodiment, the prediction results for some times are shown in Table 1: at time t1, the actual total power is 68.4MW, and the predicted total power is 69.1MW; at time t2, the actual total power is 74.2MW, and the predicted total power is 73.6MW; at time t3, the actual total power is 61.7MW, and the predicted total power is 62.5MW. Through this step, joint prediction of wind power and photovoltaic output can be achieved, improving the accuracy and stability of the prediction of the total power output of new energy generation in wind-solar complementary scenarios.

[0138] In the above embodiments, step 1 obtains standardized input data under a unified time scale; step 2 obtains the wind-solar complementary driving index Ct and the key factor set; step 3 obtains the wind speed reconstruction subsequence set and the solar irradiance reconstruction subsequence set; step 4 obtains the initial wind speed prediction value and the initial solar irradiance prediction value; step 5 obtains the corrected wind speed prediction value and the corrected solar irradiance prediction value; and step 6 obtains the wind power prediction value, the photovoltaic power prediction value, and the prediction result of the total power generation of new energy after fusion, thereby forming a new energy power generation intelligent complementary prediction process that combines wind speed prediction, solar irradiance prediction, residual correction, and power fusion.

[0139] The prediction results are shown in Table 1 below.

[0140] Table 1

[0141]

[0142] Overall performance: MAE: 0.023; RMSE: 0.034; R²: 0.966. Because the predicted total power obtained by the method of this invention is very close to the actual total power and reflects the power variation trend under wind-solar hybrid operation well, its prediction effect is excellent and its stability is extremely good. Detailed implementation method

[0143] This invention provides a method and system for intelligent complementary prediction of new energy power generation. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. A method for intelligent complementary prediction of new energy power generation, characterized in that, Includes the following steps: Step 1: Obtain historical wind speed, solar irradiance, meteorological factors, numerical weather prediction and historical actual power data, and perform time alignment, missing value processing and normalization preprocessing. Step 2: Construct wind-solar complementary correlation features based on wind speed, solar irradiance and meteorological factors, and extract key factors for wind speed prediction and solar irradiance prediction by combining random forest; Step 3: Decompose the preprocessed wind speed sequence and solar irradiance sequence using the Ordinal Mode Decomposition (OPMD) method, calculate the information entropy and conditional entropy of each intrinsic mode component, and reconstruct the subsequences by combining the correlation features to obtain the wind speed reconstructed subsequence and the solar irradiance reconstructed subsequence. Step 4: Input the wind speed reconstruction subsequence and the solar irradiance reconstruction subsequence into the improved time series dense encoder TiDE model, which introduces time-varying collaborative gating and dynamic trend compensation mechanism, respectively, to obtain the prediction results corresponding to each subsequence, and reconstruct the prediction results of each subsequence to obtain the initial wind speed prediction value and the initial solar irradiance prediction value. Step 5: Construct historical residual samples based on historical real wind speed, historical real solar irradiance, and initial wind speed and solar irradiance predictions. Use the improved Dragonfly Optimization Algorithm (DOA) to correct the parameters of the Gaussian Process Regression (GPR) model and establish a GPR residual correction model. Input the key factors at the current moment, wind-solar complementary correlation features, and initial wind speed and solar irradiance predictions into the GPR residual correction model to obtain wind speed prediction residuals and solar irradiance prediction residuals, and obtain the corrected wind speed and solar irradiance predictions. Step 6: Based on the current meteorological scenario and historical operating status, dynamically update the mapping relationship between wind speed and wind power and the mapping relationship between solar irradiance and photovoltaic power, and convert the corrected wind speed prediction value and the corrected solar irradiance prediction value into wind power prediction value and photovoltaic power prediction value, respectively.

2. The method according to claim 1, characterized in that, Step 2 includes: constructing a wind-solar complementary driving index C based on the changing trends, fluctuation amplitudes, and complementarity of wind speed and solar irradiance sequences at the same time scale. t And based on the wind-solar complementary driving index C t Candidate meteorological influencing factors are screened, ranked, and weighted to form a set of key factors for subsequent wind speed and solar irradiance prediction; the wind-solar hybrid driving index C t The expression is: ， Among them, C t ρ represents the wind-solar hybrid driving index at time t; λ1, λ2, and λ3 represent weighting coefficients; satisfying λ1+λ2+λ3=1 and λ1, λ2, λ3∈[0,1]; t It is the Pearson correlation coefficient between the wind speed series and the solar irradiance series within a sliding time window; v t G represents the wind speed at time t; t Δv represents the solar irradiance at time t. t =v t -v t-1 ΔG represents the wind speed increment at time t. t =G t -G t-1 This represents the increase in solar irradiance at time t; This represents the local standard deviation of the wind speed sequence within the current sliding time window; This represents the local standard deviation of the solar irradiance sequence within the current sliding time window; exp represents the smallest positive constant that prevents the denominator from being zero; exp represents the natural exponential function. The key factor set is based on the wind-solar hybrid driving index C. t The key factors are determined together with the feature importance of the random forest output. A predetermined number of key factors are selected from the comprehensive evaluation value of the key factors in descending order. The key factors include one or more of the following: wind speed, solar irradiance, temperature, humidity, air pressure, wind direction, and numerical weather prediction data. These factors are further used as constraint information for subsequent subsequence recombination, residual correction, and power fusion to achieve collaborative prediction between wind power branches and photovoltaic branches.

3. The method according to claim 2, characterized in that, Step 3 includes: For wind speed and solar irradiance sequences, OPMD decomposition is performed to obtain two or more intrinsic mode components (IMFi). The information entropy, conditional entropy, and correlation between each IMFi and the original sequence are calculated. Based on the complexity and complementary correlation of each IMFi, a recombination criterion Ω is constructed. i The expression is: ， Among them, Ω i The recombination criterion for the i-th intrinsic mode component; H i H represents the information entropy of the i-th intrinsic mode component; i∣C The i-th intrinsic mode component is represented by the wind-solar hybrid drive index C. t Conditional entropy; Let represent the correlation coefficient between the i-th intrinsic mode component and the original sequence; α, β, and γ represent weight coefficients, satisfying α+β+γ=1, and α, β, γ∈[0,1]; Will satisfy The intrinsic mode components are merged into the q-th reconstructed subsequence, and the reconstructed subsequence is obtained according to the following formula: ， Where, τ q-1 and τ q S represents the lower and upper thresholds of the q-th reconstruction interval, respectively; q Z represents the set of eigenmode component indices that satisfy the interval conditions; q Represents the q-th reconstructed subsequence; The wind speed sequence and solar irradiance sequence are decomposed and recombined according to step 3 to obtain the wind speed reconstructed subsequence set and the solar irradiance reconstructed subsequence set, respectively.

4. The method according to claim 3, characterized in that, In step 4, the improved DOA algorithm is used to optimize the key parameters of the TiDE model and the GPR residual correction model. The improved DOA algorithm simultaneously constrains the prediction error, fitting bias and statistical consistency through a composite fitness function to reduce the risk of overfitting caused by a single error index. The expression for the composite fitness function J is: ， Where ω1, ω2, ω3, and ω4 represent weighting coefficients, satisfying ω1+ω2+ω3+ω4=1, and ω1, ω2, ω3, ω4∈[0,1]; RMSE represents root mean square error; MAE represents mean absolute error; KGE represents the Kling-Gupta Efficiency index; and B represents prediction bias. Represents the true mean; RMSE0 is the root mean square error of the baseline model; MAE0 is the mean absolute error of the baseline model; The expression for the Kling-Gupta Efficiency index KGE is: ， Where r represents the linear correlation coefficient between the predicted value sequence and the actual value sequence; It is the standard deviation of the predicted value series; σ y The standard deviation of the true value sequence; μ represents the mean of the predicted value sequence. y This represents the mean of the sequence of true values; In the improved DOA algorithm, the update expression for the p-th candidate solution in the q-th iteration is: ， in, This represents the p-th candidate solution selected in the q-th iteration; This represents the globally optimal solution obtained in the q-th iteration; This represents the randomly selected reference solution in the q-th iteration; r1 and r2 are random numbers in the range [0,1]. and b q Let represent the adaptive compression coefficient and expansion coefficient of the q-th iteration, respectively; The adaptive compression coefficient and expansion coefficient satisfy the following conditions: ， Among them, a max This represents the maximum value of the adaptive compression factor; a min This represents the minimum value of the adaptive compression factor; b max This represents the maximum value of the coefficient of thermal expansion; b min The minimum value of the expansion coefficient is represented by Q; the maximum number of iterations is represented by Q; and the current number of iterations is represented by q.

5. The method according to claim 4, characterized in that, In step 4, the wind speed reconstruction subsequence and the solar irradiance reconstruction subsequence are respectively input into the improved TiDE model, which incorporates time-varying collaborative gating and dynamic trend compensation mechanisms, to obtain the prediction results corresponding to each subsequence. Furthermore, the prediction results of each subsequence are reconstructed to obtain the initial wind speed prediction value and the initial solar irradiance prediction value; specifically including: The wind speed reconstruction subsequence and corresponding key factors are input into the improved TiDE prediction model of the wind speed branch, and the solar irradiance reconstruction subsequence and corresponding key factors are input into the improved TiDE prediction model of the solar irradiance branch. In each branch, a time-varying collaborative gating vector is constructed based on the historical window of the current branch, the historical window of another complementary branch, the historical window of the key factors, and the historical window of the wind-solar complementary driving index. The input sequence of the current branch is then collaboratively enhanced, and the process satisfies the following formula: ， ， Where m∈{v,G} represents the current predicted branch, v represents the wind speed branch, G represents the solar irradiance branch, and n represents another complementary branch corresponding to the current predicted branch; This represents the historical window of the reconstructed subsequence of the m-th branch with length L before time t; This indicates the historical window corresponding to another complementary branch; This represents a key factor history window of length L before time t; This represents a historical window of length L for the wind-solar hybrid drive index before time t. Represents a time-varying cooperative gating vector; This represents the Sigmoid activation function; This represents element-wise multiplication; , and Indicates the parameters to be optimized; This represents the input sequence after wind-solar hybrid enhancement. The input sequence, after co-enhancement, is fed into the TiDE prediction model of the corresponding branch, and a dynamic trend compensation term is introduced to enhance the response to short-term fluctuations and abrupt changes. The process satisfies the following formula: ， ， in, TiDE represents the predicted subsequence value of the m-th branch at time t. m Denotes the TiDE prediction model for the m-th branch; θ m This represents the parameter vector of the corresponding TiDE model; This represents the local change of the m-th branch between the last two moments; This represents the reconstructed subsequence value of the m-th branch at time t-1; This represents the reconstructed subsequence value of the m-th branch at time t-2; The dynamic trend compensation coefficient is represented by tanh(.); the hyperbolic tangent activation function is represented by tanh(.). and Indicates the parameters to be optimized; The key factor vector representing time t; The wind speed reconstruction subsequence and its corresponding key factors are input into the TiDE prediction model of the wind speed branch to obtain the prediction results of each wind speed subsequence; the solar irradiance reconstruction subsequence and its corresponding key factors are input into the TiDE prediction model of the solar irradiance branch to obtain the prediction results of each solar irradiance subsequence. The process is as follows: ， ， ， ， in, This represents the predicted value of the k-th wind speed reconstruction subsequence at time t; This represents the predicted value of the l-th solar irradiance reconstruction subsequence at time t; The TiDE prediction model represents the wind speed branch; The TiDE prediction model representing the solar irradiance branch; θ v θ represents the parameter vector of the TiDE prediction model for wind speed branches. G This represents the parameter vector of the TiDE prediction model for the solar irradiation branch; This represents the historical window of the k-th wind speed reconstruction subsequence of length L before time t; K represents the l-th solar irradiance reconstruction subsequence history window of length L before time t; v K represents the number of wind speed reconstructed subsequences; G Indicates the number of solar irradiation reconstructed subsequences; This represents the initial wind speed forecast value; This represents the initial solar irradiance prediction.

6. The method according to claim 5, characterized in that, In step 5, a residual correction model is established using an improved DOA-optimized GPR residual correction model. This GPR residual correction model takes the current key factors, the wind-solar hybrid driving index, and the initial prediction results as inputs, and the wind speed prediction residuals and solar irradiance prediction residuals as outputs, thereby performing online corrections to the initial wind speed and solar irradiance prediction values. The input vector of the residual correction model is: ， in, This represents the input vector of the GPR residual correction model at time t; This represents the initial wind speed forecast value; This represents the initial solar irradiance prediction. Residual prediction using the GPR residual correction model: ， in, The parameter is GPR residual correction model; The corrected wind speed prediction and solar irradiance prediction respectively satisfy: ， ， in, This represents the residual for wind speed prediction; This represents the residual from the solar irradiance prediction; This represents the corrected wind speed forecast; This represents the revised solar irradiance forecast; The kernel function of the GPR residual correction model is: ， Where, k(z) i ,z j ) represents the i-th input sample z i With the j-th input sample z j The kernel function value; Z represents the signal variance; D represents the input vector dimension; z i,d and z j,d Let represent the value of the i-th input sample in the d-th dimension and the value of the j-th input sample in the d-th dimension, respectively; This represents the length scale parameter of the d-th dimension; Indicates the noise variance; This represents the Kronecker function, which takes the value 1 when i=j and 0 otherwise.

7. The method according to claim 6, characterized in that, In step 6, based on the current meteorological scenario and historical operating status, the mapping relationship between wind speed and wind power, as well as the mapping relationship between solar irradiance and photovoltaic power, are dynamically updated, and the prediction results of new energy power generation are output. Specifically, to characterize the coupling relationship between residual correction information and physical mechanism mapping, a dynamic mechanism-residual coupling correction coefficient is constructed. The expression is: ， Among them, Λ t κ represents the dynamic mechanism-residual coupling correction coefficient at time t; κ represents the corrected weighting coefficient. This represents the residual for wind speed prediction; σ represents the residual of solar irradiance prediction; e This represents the recent residual fluctuation standard deviation; The predicted wind power output satisfies: ， in, This represents the predicted wind power output at time t; v ci Represents the cut-in wind speed; v r Represents the rated wind speed; v co P represents the cut-out wind speed; r The value represents the rated power of the fan; μ represents the power exponent of the power curve. The photovoltaic power forecast meets the following requirements: ， in, P represents the predicted photovoltaic power at time t; stc It is the rated power of photovoltaic power under standard testing conditions; G stc It is the irradiance under standard test conditions; β T It is the temperature correction factor; T t T is the component temperature at time t; stc It is the standard test temperature; η t It is the comprehensive efficiency correction coefficient, which is used to represent the effects of shading attenuation, pollution attenuation and amplitude limiting.

8. A new energy power generation intelligent complementary prediction system based on the method described in any one of claims 1 to 7, characterized in that, include: The multi-source time series conditioning unit is used to acquire historical wind speed data, historical solar irradiance data, meteorological influence factor data, numerical weather prediction data, and historical actual power generation data, and to perform time alignment, missing value processing, and normalization preprocessing. The wind-solar feedback sensing unit is used to determine the wind-solar complementary drive index C. t Constructing association features to characterize the complementary relationship between wind and solar power; The dominant factor analysis unit is used to extract key factors affecting wind speed prediction and solar irradiance prediction based on correlation features and random forest output results. Heterogeneous wave decomposition and recompilation units are used to perform OPMD decomposition on wind speed and solar irradiance sequences respectively, and to recombinate them according to the recombination criterion Ω. i Perform intrinsic mode component merging and reconstruction; The dual-branch time series extrapolation unit is used to predict the wind speed reconstruction subsequence and the solar irradiance reconstruction subsequence based on the improved DOA-optimized TiDE model, and obtain the initial wind speed prediction value and the initial solar irradiance prediction value. The bias feedback correction unit is used to perform residual correction on the initial wind speed prediction and the initial solar irradiance prediction based on the GPR residual correction model.

9. An electronic device, characterized in that, It includes a processor and a memory, the memory storing program code that, when executed by the processor, causes the processor to perform the steps of the method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, It stores a computer program or instructions that, when run on a computer, perform the steps of the method as described in any one of claims 1 to 7.

Images