A high-precision prediction and analysis method for integrated source-grid-load-storage renewable energy generation

By separating normal and extreme weather samples and combining multiple algorithms and models, high-precision prediction of new energy power generation has been achieved, solving the problem of insufficient prediction accuracy of new energy power generation and promoting the stable operation of the power system and the efficient consumption of new energy.

CN123121476APending Publication Date: 2026-11-17XUCHANG POWER SUPPLY COMPANY OF STATE GRID HENAN ELECTRIC POWER
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611010124.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-08
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Insufficient accuracy in predicting renewable energy generation leads to poor power system stability, difficulty in integrating renewable energy, and high activation costs for backup power sources, resulting in energy waste.

Method used

We employ multi-scale difference and sliding window anomaly detection methods to separate normal and extreme weather samples. We combine variational mode decomposition and temporal difference to extract weather features, use support vector machine to identify extreme weather types, and construct a power generation prediction model based on XGBoost and Grey Wolf optimization algorithms to perform high-precision predictions in stages.

Benefits of technology

It has improved the accuracy of new energy power generation forecasting, reduced the complexity of model calculations, and enabled the safe and stable operation of the power system and the efficient consumption of new energy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123121476A_ABST
    Figure CN123121476A_ABST
Patent Text Reader

Abstract

This invention discloses a high-precision prediction and analysis method for renewable energy generation integrating source, grid, load, and storage, comprising the following analysis steps: S1, constructing a power generation dataset and a meteorological dataset respectively; S2, dividing the data into normal weather samples and extreme weather samples; S3, extracting extreme weather characteristics and identifying extreme weather types; S4, when no extreme weather is identified, constructing a basic power generation prediction model; S5, training the basic power generation prediction model using normal weather samples and predicting power generation; S6, when extreme weather is identified, constructing a basic power generation prediction model and selecting hyperparameters using an optimization algorithm; S7, training the optimized prediction model using extreme weather samples and predicting power generation; S8, performing power dispatch based on the power generation prediction results in step S5 or S7. This invention can improve the accuracy of renewable energy generation prediction, promote renewable energy consumption, and ensure the safe and stable operation of the power system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of new energy power generation prediction technology, specifically involving a high-precision prediction and analysis method for new energy power generation that integrates source, grid, load, and storage. Background Technology

[0002] With the advancement of the "dual carbon" goals, the installed capacity of wind power and photovoltaic power nationwide has reached 40% of the total installed power capacity, marking the accelerated formation of an energy system dominated by new energy. Against this backdrop, how to resolve the contradiction between the intermittency and volatility of new energy power generation and the stable operation of the power system has become the main challenge of energy transformation. The integrated generation, grid, load and storage system, as a systematic solution that integrates new energy power generation (source), grid transmission (grid), electricity load (load), and energy storage regulation (storage), effectively improves the flexibility and risk resistance of the power system through multi-link synergistic optimization, and has become a key technical path for energy transformation.

[0003] In the process of promoting integrated power generation, grid, load and energy storage projects, the insufficient accuracy of new energy power generation forecasting has become a key bottleneck restricting the efficient operation of the system. The core of the integrated power generation, grid, load and energy storage system is to achieve coordinated optimization of power sources, grid, load and energy storage. Accurate new energy power generation forecasting is the foundation for achieving this goal. However, the randomness and intermittency of new energy power generation such as wind power and photovoltaic power generation lead to a large deviation between the forecast and the actual output of new energy power generation. This has a series of adverse effects on the stable operation of the power system, the consumption of new energy and the economic benefits of the project. When the forecast output is higher than the actual output, the power system will experience power shortage. In order to maintain the stability of the system frequency, the dispatch center has to call backup power sources, such as gas turbines and diesel generators. The start-up and operation costs of these backup power sources are high and they will increase carbon emissions. When the forecast output is lower than the actual output, it will lead to power surplus. With the limited capacity of the energy storage system, wind and solar power curtailment has to be carried out, resulting in energy waste.

[0004] Therefore, in order to solve the above problems, it is necessary to develop a high-precision prediction and analysis method for new energy power generation that integrates source, grid, load and storage. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a high-precision prediction and analysis method for new energy power generation that integrates source, grid, load and storage, thereby improving the prediction accuracy of new energy power generation, promoting the consumption of new energy, and ensuring the safe and stable operation of the power system.

[0006] The objective of this invention is achieved as follows: a high-precision predictive analysis method for new energy power generation integrating source, grid, load, and storage, comprising the following analytical steps:

[0007] S1. Acquire historical wind power data and historical photovoltaic data respectively, and construct a power generation dataset; acquire historical weather data, future forecast weather data, and real-time observed meteorological data respectively, and construct a meteorological dataset;

[0008] S2. Based on the power generation dataset and meteorological dataset, the data is divided into normal weather samples and extreme weather samples;

[0009] S3. Extract extreme weather features based on extreme weather samples and identify extreme weather types;

[0010] S4. When extreme weather is not identified, construct a basic prediction model for power generation.

[0011] S5. Train the basic power generation prediction model using normal weather samples, and use the current model to predict power generation under normal weather conditions based on future weather prediction data.

[0012] S6. When extreme weather is identified, a basic prediction model for power generation is constructed, and an optimization algorithm is used to select the hyperparameters of the basic prediction model. A physical model is introduced into the error function to perform multi-physics coupling modeling, and an optimized prediction model for power generation is obtained.

[0013] S7. Use extreme weather samples to train the power generation optimization prediction model, and use the power generation optimization prediction model to predict power generation under extreme weather conditions based on future weather forecast data.

[0014] S8. Power dispatch is carried out based on the power generation prediction results in step S5 or S7.

[0015] Furthermore, in step S1, historical weather data and future forecast weather data are centrally obtained through the sixth Coupled Model Comparison Project (CMIP6) dataset.

[0016] Furthermore, in step S2, anomaly power extraction based on multi-scale difference and anomaly weather discrimination method based on sliding window are specifically adopted to extract meteorological and power data corresponding to normal weather and extreme weather, respectively, and to obtain normal weather samples and extreme weather samples.

[0017] Furthermore, the extraction of normal weather samples and extreme weather samples in step S2 specifically includes the following steps: ① Constructing anomaly power sample criteria: In the formula, express Power at any moment This indicates the width of the sliding window. Indicates the scaling factor. ① Represent the installed capacity of new energy; ② Set the window width and scaling factor parameters, and extract abnormal power samples at different time scales based on the set parameters; ③ Merge the extracted abnormal power samples; ④ Merge adjacent discrete samples after merging to obtain continuous abnormal power samples; ⑤ Construct extreme weather criteria, and based on the extreme weather criteria, select extreme weather samples from the merged continuous abnormal power samples or screen them out.

[0018] Furthermore, in step S3, variational mode decomposition (VMD) and temporal difference are specifically used to extract features from meteorological data in extreme weather samples to obtain small-scale fluctuation features and temporal fluctuation features of extreme weather.

[0019] Furthermore, the step S3, which uses variational mode decomposition to extract small-scale fluctuation features, is specifically expressed as follows: Let the original meteorological data be... The duration is Then the constrained variational model is expressed as: In the formula: Indicates the first One modal component, Indicates the first The center frequency of each modal component Represents the gradient. Represents the Dirac function, The modal components and center frequencies after n iterations are expressed as follows: The augmented Lagrange equation is used to transform the constrained optimization problem into an unconstrained optimization problem. Then, the alternating direction multiplier method is used to search for the saddle points of the unconstrained optimization model. , In the formula, This represents the secondary penalty factor. The Lagrange operator is represented; the extraction of temporal fluctuation features using temporal difference in step S3 is specifically expressed as follows: , , Raw meteorological data, Duration.

[0020] Furthermore, in step S3, a convolutional neural network with a support vector machine as the classifier is specifically used to identify extreme weather types; specifically, it includes an input layer, convolutional layer 1, pooling layer 1, convolutional layer 2, pooling layer 2, a fully connected layer, and a support vector machine classification layer, directly connecting the convolutional neural network and the support vector machine to enhance the classification effect while extracting deep features.

[0021] Furthermore, in step S4 or S6, the extreme gradient boosting XGBoost algorithm is used to construct the basic prediction model for power generation, specifically including the following model construction steps: ① Set the predicted values ​​of all samples in the training data to 0 as the initial values ​​of the model, which are used to train the first decision tree and serve as the initial weak classifier. ② Based on the current prediction results of the model Calculate the residual for each sample; the residual is specifically the true value of each sample. Compared with the current prediction results of the model Error between Calculated; ③ Calculate the first gradient of the residual. and second derivative The gradient of each sample is obtained. and Hessian matrix ④ Calculate the loss function of the model; ⑤ Based on the gradient of the samples and Hessian matrix And the loss function of the current model, to generate a new decision tree. and add it to the existing decision tree set. In the middle; ⑥ A weighted update method is used, assigning a weight to each sample and adaptively adjusting the weights according to the magnitude of its error; ⑦ The generated decision tree and existing decision tree sets Integrate the models to obtain new model outputs; ⑧ Repeat steps ②-⑦ until the model error reaches the preset threshold or the maximum number of iterations is reached.

[0022] Furthermore, in step S6, the Grey Wolf optimization algorithm is specifically used to optimize and select the parameters in the basic power generation prediction model.

[0023] The beneficial effects of this invention, achieved by adopting the above technical solution, are as follows: By collecting power generation data and meteorological data, obtaining data samples, and classifying and extracting the data, extreme weather predictions are made, thus dividing power generation predictions into predictions under normal weather conditions and predictions under extreme weather conditions. By constructing a basic power generation prediction model, predictions under normal weather conditions are directly performed using the basic prediction model without complex calculations, achieving rapid prediction. Under extreme weather conditions, an optimized power generation prediction model, optimized by an optimization algorithm, is used, effectively ensuring prediction accuracy. This reduces the computational complexity of the model while improving prediction accuracy, achieving a balance between prediction efficiency and accuracy, promoting the consumption of new energy sources, and ensuring the safe and stable operation of the power system. Attached Figure Description

[0024] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0025] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0026] like Figure 1 As shown, a high-precision predictive analysis method for new energy power generation integrating source, grid, load, and storage includes the following analysis steps:

[0027] S1. Obtain historical wind power data and historical photovoltaic data respectively, and construct a power generation dataset; obtain historical weather data, future forecast weather data, and real-time observed meteorological data respectively, and construct a meteorological dataset.

[0028] Preferably, in step S1, historical weather data and future forecast weather data are obtained centrally through the sixth Coupled Model Comparison Project (CMIP6) dataset.

[0029] S2. Based on the power generation dataset and meteorological dataset, the data is divided into normal weather samples and extreme weather samples.

[0030] Preferably, in step S2, anomaly power extraction based on multi-scale difference and anomaly weather discrimination method based on sliding window are specifically adopted to extract meteorological and power data corresponding to normal weather and extreme weather, respectively, and to obtain normal weather samples and extreme weather samples.

[0031] Preferably, the extraction of normal weather samples and extreme weather samples in step S2 specifically includes the following steps: ① Constructing anomaly power sample criteria: In the formula, express Power at any moment This indicates the width of the sliding window. Indicates the scaling factor. The process involves: ① representing the installed capacity of new energy sources; ② setting window width and scaling factor parameters, and extracting abnormal power samples at different time scales based on these parameters; ③ merging the extracted abnormal power samples; ④ merging adjacent discrete samples to obtain continuous abnormal power samples; ⑤ constructing extreme weather criteria, and selecting extreme weather samples from the merged continuous abnormal power samples based on these criteria; taking cold waves and strong winds as examples of extreme weather affecting wind power generation, the criteria for cold wave extreme weather are: using a 48-hour sliding window... , express Temperature at any moment express Temperature 48 hours prior to the time of the event express If both the time and the lowest temperature within the preceding 48 hours are met, then the highest temperature within that time window is recorded as the "7 o'clock cold wave," and the end point is when the temperature recovers from the lowest temperature to above 4°C. The criterion for extreme wind weather is: using a 24-hour sliding window, the average wind speed over 24 hours... At that time, it is considered a strong wind extreme weather event; taking high temperature extreme weather, which affects photovoltaic power generation, as an example, the criterion for high temperature extreme weather is: [The text abruptly ends here, so the translation stops as well.] Indicates the first Heavenly The temperature of the hour, , For the first Temperature difference of the day Represented as: .

[0032] S3. Extract extreme weather features based on extreme weather samples and identify extreme weather types.

[0033] Preferably, in step S3, variational mode decomposition (VMD) and temporal difference are used to extract features from meteorological data in extreme weather samples to obtain small-scale fluctuation features and temporal fluctuation features of extreme weather.

[0034] Preferably, the step S3, which uses variational mode decomposition to extract small-scale fluctuation features, is specifically expressed as follows: Let the original meteorological data be... The duration is Then the constrained variational model is expressed as: In the formula: Indicates the first One modal component, Indicates the first The center frequency of each modal component Represents the gradient. Represents the Dirac function, The modal components and center frequencies after n iterations are expressed as follows: The augmented Lagrange equation is used to transform the constrained optimization problem into an unconstrained optimization problem. Then, the alternating direction multiplier method is used to search for the saddle points of the unconstrained optimization model. , In the formula, This represents the secondary penalty factor. This represents the Lagrange operator.

[0035] Preferably, the extraction of temporal fluctuation features using temporal difference in step S3 is specifically expressed as follows: , , Raw meteorological data, Duration.

[0036] Preferably, in step S3, a convolutional neural network with a support vector machine as the classifier is used to identify extreme weather types; specifically, it includes an input layer, convolutional layer 1, pooling layer 1, convolutional layer 2, pooling layer 2, a fully connected layer, and a support vector machine classification layer, directly connecting the convolutional neural network and the support vector machine to enhance the classification effect while extracting deep features.

[0037] S4. When extreme weather is not identified, construct a basic prediction model for power generation.

[0038] S5. Train the basic power generation prediction model using normal weather samples, and use the current model to predict power generation under normal weather conditions based on future weather prediction data.

[0039] S6. When extreme weather is identified, a basic prediction model for power generation is constructed. An optimization algorithm is used to select the hyperparameters of the basic prediction model, and a physical model is introduced into the error function to perform multi-physics coupling modeling, thereby obtaining an optimized prediction model for power generation.

[0040] Preferably, steps S4 and S6 employ the Extreme Gradient Boosting (XGBoost) algorithm to construct the basic prediction model for power generation, specifically including the following model construction steps: ① Set the predicted values ​​of all samples in the training data to 0 as the initial values ​​for training the first decision tree, and use it as the initial weak classifier. ② Based on the current prediction results of the model Calculate the residual for each sample; the residual is specifically the true value of each sample. Compared with the current prediction results of the model Error between Calculated; ③ Calculate the first gradient of the residual. and second derivative The gradient of each sample is obtained. and Hessian matrix ④ Calculate the loss function of the model; ⑤ Based on the gradient of the samples and Hessian matrix And the loss function of the current model, to generate a new decision tree. and add it to the existing decision tree set. In the middle; ⑥ A weighted update method is used, assigning a weight to each sample and adaptively adjusting the weights according to the magnitude of its error; ⑦ The generated decision tree and existing decision tree sets Integrate the models to obtain new model outputs; ⑧ Repeat steps ②-⑦ until the model error reaches the preset threshold or the maximum number of iterations is reached.

[0041] Preferably, in steps S4 and S6, the Extreme Gradient Boosting (XGBoost) algorithm is used to construct the basic prediction model for power generation. Its loss function, regularization penalty function, and objective function are expressed as follows: In the formula, Represents the loss function. , They respectively represent joining the first , Predicted values ​​after [number] trees This represents the regularization penalty function. Indicates the number of trees. express The penalty coefficient, This represents the coefficient of the L2 regularization penalty term. Indicates the first The weight values ​​of each leaf node. This represents the objective function.

[0042] Preferably, in step S6, the Grey Wolf Optimization Algorithm is specifically used to optimize the selection of parameters such as the maximum spanning tree count, learning rate, maximum tree depth, and minimum sum of leaf node sample weights in the basic prediction model XGBoost model.

[0043] Preferably, step S6, which uses the gray wolf optimization algorithm to optimize the parameters in the XGBoost model, specifically includes the following steps: ① Encircling the prey, the gray wolf pack gradually approaches and locks onto the target: In the formula, Indicates the current iteration number. Represents the position vector of the prey. This represents the position vector of the gray wolf. This represents the coordinate distance between the gray wolf and its prey. , For the coordination coefficient vector, , Indicates the convergence factor. , for ① Calculate random numbers between individual gray wolves and the population. , , The distance of the wolf: In the formula, , , They are respectively During iteration , , The wolf's position; then, with , , The wolf's location is constantly being updated: In the formula, for This causes other gray wolves to continuously approach their prey: .

[0044] Preferably, the physical model, specifically addressing the dust accumulation effect under extreme high-temperature weather conditions, can be represented by a sol-radiation-dust coupling equation, specifically as follows: In the formula, This represents the intensity of solar radiation under the influence of aerosol particles. It is the surface albedo. Represents solar spectral radiance. Indicates the wavelength of the emitted radiation. Represents the blackbody temperature. , Represents the speed of light. Represents Boltzmann's constant. Optical thickness, representing the attenuation of direct solar radiation. , They represent heights respectively. The extinction cross section of a single particle and the number of particles at that location, here It is the wavelength of the incident wave; This indicates the light transmittance of the dust-covered solar panel. Indicates the light transmittance of sand and dust particles. This represents the mass density of deposited particles per unit area. Indicates the mass density of the sediment particles. This indicates the average particle size of the particles deposited on the surface of the photovoltaic module. The distribution of the function representing the particle size of the sedimentary particles, , These represent the maximum and minimum particle sizes of the deposited particles, respectively.

[0045] S7. Use extreme weather samples to train the power generation optimization prediction model, and use the optimized prediction model to predict power generation under extreme weather conditions based on future weather forecast data.

[0046] S8. Power dispatch is carried out based on the power generation prediction results in step S5 or S7.

[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of the claims of the present invention.

Claims

1. A source, network, load and storage integrated new energy high-precision power generation prediction analysis method, characterized by: The analysis includes the following steps: S1. Acquire historical wind power data and historical photovoltaic data respectively, and construct a power generation dataset; acquire historical weather data, future forecast weather data, and real-time observed meteorological data respectively, and construct a meteorological dataset; S2. Based on the power generation dataset and meteorological dataset, the data is divided into normal weather samples and extreme weather samples; S3. Extract extreme weather features based on extreme weather samples and identify extreme weather types; S4. When extreme weather is not identified, construct a basic prediction model for power generation. S5. Train the basic power generation prediction model using normal weather samples, and use the current model to predict power generation under normal weather conditions based on future weather prediction data. S6. When extreme weather is identified, a basic prediction model for power generation is constructed, and an optimization algorithm is used to select the hyperparameters of the basic prediction model. A physical model is introduced into the error function to perform multi-physics coupling modeling, and an optimized prediction model for power generation is obtained. S7. The power generation optimization prediction model is trained using extreme weather samples, and the power generation optimization prediction model is used to predict power generation under extreme weather conditions based on future weather forecast data. S8. Power dispatch is carried out based on the power generation prediction results in step S5 or S7.

2. The source-network-load-storage integrated new energy high-precision power generation prediction analysis method according to claim 1, characterized in that: In step S1, historical weather data and future forecast weather data are centrally obtained through the sixth Coupled Model Comparison Project (CMIP6) dataset.

3. The source-network-load-storage integrated new energy high-precision power generation prediction analysis method of claim 1, characterized in that: In step S2, anomaly power extraction based on multi-scale difference and anomaly weather discrimination method based on sliding window are specifically adopted to extract meteorological and power data corresponding to normal weather and extreme weather, respectively, and obtain normal weather samples and extreme weather samples.

4. The integrated source-grid-load-storage new energy high-precision power generation prediction and analysis method according to claim 3, characterized in that: The extraction of the normal weather sample and the extreme weather sample in the step S2 specifically comprises the following steps: ① constructing an abnormal power sample criterion: , wherein, represents the power at the time t, represents the window width of the sliding window, represents a proportion factor, represents the installed capacity of the new energy, represents the installed capacity of the new energy; ② setting the window width and the proportion factor parameters, and extracting the abnormal power samples of different time scales based on the set parameters; ③ merging the extracted abnormal power samples; ④ merging the adjacent discrete samples after the merging to obtain continuous abnormal power samples; and ⑤ constructing an extreme weather criterion, and screening out the extreme weather samples based on the extreme weather criterion.

5. The integrated source-grid-load-storage new energy high-precision power generation prediction and analysis method according to claim 1, characterized in that: In step S3, variational mode decomposition (VMD) and temporal difference are used to extract features from meteorological data in extreme weather samples, thereby obtaining small-scale fluctuation features and temporal fluctuation features of extreme weather.

6. The integrated source-grid-load-storage new energy high-precision power generation prediction and analysis method according to claim 5, characterized in that: The step S3, which uses variational mode decomposition to extract small-scale fluctuation features, is specifically expressed as follows: Let the original meteorological data be... The duration is Then the constrained variational model is expressed as: In the formula: Indicates the first One modal component, Indicates the first The center frequency of each modal component Represents the gradient. Represents the Dirac function, The modal components and center frequencies after n iterations are expressed as follows: The augmented Lagrange equation is used to transform the constrained optimization problem into an unconstrained optimization problem. Then, the alternating direction multiplier method is used to search for the saddle points of the unconstrained optimization model. , In the formula, This represents the secondary penalty factor. The Lagrange operator is represented; the extraction of temporal fluctuation features using temporal difference in step S3 is specifically expressed as follows: , , Raw meteorological data, Duration.

7. The integrated source-grid-load-storage new energy high-precision power generation prediction and analysis method according to claim 1, characterized in that: In step S3, a convolutional neural network with a support vector machine as the classifier is used to identify extreme weather types. Specifically, it includes an input layer, convolutional layer 1, pooling layer 1, convolutional layer 2, pooling layer 2, a fully connected layer, and a support vector machine classification layer. The convolutional neural network is directly connected to the support vector machine, which enhances the classification effect while extracting deep features.

8. The integrated source-grid-load-storage new energy high-precision power generation prediction and analysis method according to claim 1, characterized in that: In step S4 or S6, the Extreme Gradient Boosting (XGBoost) algorithm is used to construct the basic prediction model for power generation. Specifically, the model construction steps are as follows: ① Set the predicted values ​​of all samples in the training data to 0 as the initial values ​​of the model, which are used to train the first decision tree and serve as the initial weak classifier. ② Based on the current prediction results of the model Calculate the residual for each sample; The residuals are specifically taken from the true values ​​of each sample. Compared with the current prediction results of the model Error between Calculated; ③ Calculate the first gradient of the residual. and second derivative The gradient of each sample is obtained. and Hessian matrix ④ Calculate the loss function of the model; ⑤ Based on the gradient of the samples and Hessian matrix And the loss function of the current model, to generate a new decision tree. and add it to the existing decision tree set. middle; ⑥ A weighted update method is used, assigning a weight to each sample and adaptively adjusting the weights based on the magnitude of the error; ⑦ The generated decision tree... and existing decision tree sets Integrate the models to obtain new model outputs; ⑧ Repeat steps ②-⑦ until the model error reaches the preset threshold or the maximum number of iterations is reached.

9. The integrated source-grid-load-storage new energy high-precision power generation prediction and analysis method according to claim 1, characterized in that: In step S6, the Grey Wolf optimization algorithm is specifically used to optimize and select the parameters in the basic power generation prediction model.