Extreme weather identification method and system for new energy power prediction

Abstract

The invention discloses an extreme weather recognition method and system for new energy power prediction, and belongs to the technical field of extreme weather recognition, and the method comprises the steps: constructing an initial extreme weather recognition model based on a random forest and a convolutional neural network, and inputting historical meteorological time series data into the initial extreme weather recognition model to obtain historical fusion features; constructing an initial feature evaluation model through a causal test method and SHAP analysis, training the initial feature evaluation model through historical fusion features and historical extreme weather to obtain a final feature evaluation model, and obtaining a final extreme weather recognition model based on the final feature evaluation model and the initial extreme weather recognition model; inputting the real-time meteorological time sequence data into the final extreme weather recognition model to obtain real-time extreme weather; and according to the characteristics of the real-time scene, the predicted power is obtained by combining the equipment state data under the real-time extreme weather. The technical problem that the accuracy of extreme weather recognition is difficult to improve in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of extreme weather identification technology, specifically to an extreme weather identification method and system for predicting new energy power. Background Technology

[0002] With the rapid development of the new energy industry, new energy power generation forecasting has become a core support for power system dispatch optimization. Accurate forecasting results can directly improve the grid's ability to absorb new energy power and ensure power supply stability. However, the strong weather dependence of new energy power generation means that its output is always significantly unstable, which has become a core bottleneck restricting the accuracy of power forecasting, especially under extreme weather scenarios such as strong winds and heavy rain. Existing technologies attempt to introduce random forest algorithms to identify extreme weather and thus improve the accuracy of predicted power. However, because the decision logic of random forests is coarse-grained classification, their ability to subdivide categories is insufficient. For extreme weather events with similar appearances, such as thunderstorms and short-term strong convection, random forests struggle to capture the essential differences in their temporal patterns. This fuzzy judgment of extreme weather types makes it difficult to obtain actual extreme weather data. Therefore, how to accurately identify extreme weather is a technical problem that existing technologies cannot solve.

[0003] The information disclosed in the background section is only intended to enhance the understanding of the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] To address the technical problem of existing technologies' inability to improve the accuracy of extreme weather identification, this invention provides an extreme weather identification method and system for new energy power prediction. It constructs an initial extreme weather identification model using random forests and convolutional neural networks, builds an initial feature evaluation model using causal testing and SHAP analysis, trains the initial feature evaluation model using historical fusion features and historical extreme weather data to obtain a final feature evaluation model, and finally obtains a final extreme weather identification model based on the final feature evaluation model and the initial extreme weather identification model. This solves the technical problem of existing technologies' inability to improve the accuracy of extreme weather identification.

[0005] To address the aforementioned technical problems, this invention provides an extreme weather identification method for new energy power prediction, comprising the following steps: An initial extreme weather identification model is constructed based on random forest and convolutional neural network. Historical meteorological time series data are input into the initial extreme weather identification model to obtain historical fusion features. An initial feature evaluation model is constructed using causal testing and SHAP analysis. The initial feature evaluation model is then trained using historical fusion features and historical extreme weather data to obtain the final feature evaluation model. Finally, the final extreme weather identification model is obtained based on the final feature evaluation model and the initial extreme weather identification model. Real-time meteorological time series data is input into the final extreme weather identification model to obtain real-time extreme weather; Based on the characteristics of the real-time scenario, the predicted power is obtained by combining real-time equipment status data under extreme weather conditions.

[0006] Preferably, the step of inputting historical meteorological time-series data into the initial extreme weather identification model to obtain historical fusion features includes: Historical meteorological time-series data are input into the initial extreme weather identification model to extract features and obtain the first time-series abrupt change features and the first basic time-series features. The historical fusion features are obtained by splicing the first time-series mutation features with the first basic time-series features.

[0007] Preferably, before inputting historical meteorological time-series data into the initial extreme weather identification model for feature extraction to obtain the first time-series abrupt change feature and the first basic time-series feature, the method further includes: Outliers in historical meteorological time-series data are removed using the 3σ criterion, denoising is performed on the historical meteorological time-series data using wavelet denoising, and missing values ​​are handled using adaptive interpolation; the historical meteorological time-series data includes at least wind speed, temperature, humidity, and air pressure.

[0008] Preferably, the construction of the initial feature evaluation model through causal testing and SHAP analysis includes: An initial association identification model is constructed using the Pearson correlation coefficient in the causality test. An initial association verification model is constructed using the Granger causality test in the causality test method; An initial feature evaluation model is constructed based on the initial association identification model, the initial association verification model, and SHAP analysis.

[0009] Preferably, the step of training the initial feature evaluation model using historical fusion features and historical extreme weather data to obtain the final feature evaluation model includes: The historical fusion features are input into the initial feature evaluation model to obtain the correlation coefficient between each feature in the historical fusion features and historical extreme weather. Features with correlation coefficients less than the preset correlation coefficient are removed to obtain process features. The process features are compared with historical extreme weather data to obtain the final features. The output of the initial feature evaluation model and the final features are used to obtain the predicted extreme weather. If the predicted extreme weather does not match the historical extreme weather, the parameters of the initial feature evaluation model are adjusted according to the deviation between the predicted extreme weather and the historical extreme weather to obtain the final feature evaluation model. Otherwise, the initial feature evaluation model is used as the final feature evaluation model.

[0010] Preferably, the step of inputting real-time meteorological time-series data into the final extreme weather identification model to obtain real-time extreme weather includes: Real-time meteorological time-series data are input into the final extreme weather identification model to extract features, thereby obtaining the second time-series abrupt change features and the second basic time-series features. The second time-series abrupt change feature is concatenated with the second basic time-series feature to obtain the real-time fusion feature. The real-time fusion feature is then input into the final extreme weather recognition model to obtain the weight of each feature in the real-time fusion feature. Real-time extreme weather is then obtained based on the weight of each feature in the real-time fusion feature.

[0011] Preferably, the step of obtaining the predicted power based on the characteristics of the real-time scene and combined with the device status data under real-time extreme weather conditions includes: The characteristic interference factor is determined based on the real-time scene characteristics, the correction direction of the initial power is determined based on the characteristic interference factor, and the initial power is corrected according to the correction direction to obtain the process power. The abnormal state is identified by comparing the equipment status data under extreme weather conditions with the normal threshold of the equipment under extreme weather conditions. The process power is then corrected according to the type of abnormal state identified to obtain the predicted power.

[0012] Preferably, the real-time scene includes at least mountainous areas, islands, and oceans.

[0013] Preferably, the characteristic interference factors of the mountainous area include at least the terrain shading coefficient, the turbulence intensity index, and the local precipitation enhancement rate; The characteristic interference factors of the island include at least the concentration of salt spray in the air, the duration of relative humidity exceeding the standard, and the frequency of sea and land breeze switching. The characteristic disturbance factors of the ocean include at least the wind-wave coupling coefficient and the ocean current velocity.

[0014] By adopting the above technical solution, the present invention has the following advantages: An initial extreme weather identification model is constructed using random forest and convolutional neural network. An initial feature evaluation model is constructed using causal testing and SHAP analysis. The initial feature evaluation model is then trained with historical fusion features and historical extreme weather data to obtain a final feature evaluation model. Finally, a final extreme weather identification model is derived based on the final feature evaluation model and the initial extreme weather identification model. This approach solves the technical problem of existing technologies failing to improve the accuracy of extreme weather identification. Based on the characteristics of real-time scenarios, the predicted power is obtained by combining real-time equipment status data under extreme weather conditions, thereby solving the technical problem that existing technologies are unable to improve the accuracy of power prediction. By integrating Pearson correlation coefficient, Granger causality test, and SHAP analysis to construct an initial feature evaluation model, we can accurately screen out pseudo-meteorological features that have no substantial correlation with extreme weather, ensuring that the weights of meteorological features strictly match the physical mechanism of meteorological conditions-extreme weather, thus fundamentally solving the distortion problem caused by traditional methods that rely solely on statistical correlation to calculate weights. On the other hand, we can objectively quantify the true impact of different meteorological features on extreme weather, effectively avoiding weight mismatch caused by pseudo-feature interference, and ultimately achieving a significant improvement in the accuracy of extreme weather prediction.

[0015] This invention also provides an extreme weather identification system for new energy power forecasting, applicable to the aforementioned extreme weather identification method for new energy power forecasting, comprising: The historical fusion feature acquisition module is used to build an initial extreme weather identification model based on random forest and convolutional neural network, and to obtain historical fusion features by inputting historical meteorological time series data into the initial extreme weather identification model; The final extreme weather identification model acquisition module is used to construct an initial feature evaluation model through causal testing and SHAP analysis, train the initial feature evaluation model with historical fusion features and historical extreme weather to obtain the final feature evaluation model, and obtain the final extreme weather identification model based on the final feature evaluation model and the initial extreme weather identification model. The real-time extreme weather acquisition module is used to input real-time meteorological time series data into the final extreme weather recognition model to obtain real-time extreme weather. The predicted power acquisition module is used to obtain predicted power based on the characteristics of the real-time scene and combined with the equipment status data under real-time extreme weather conditions.

[0016] By adopting the above technical solution, the present invention has the following advantages: An initial extreme weather identification model is constructed using random forest and convolutional neural network. An initial feature evaluation model is constructed using causal testing and SHAP analysis. The initial feature evaluation model is then trained with historical fusion features and historical extreme weather data to obtain a final feature evaluation model. Finally, a final extreme weather identification model is derived based on the final feature evaluation model and the initial extreme weather identification model. This approach solves the technical problem of existing technologies failing to improve the accuracy of extreme weather identification. Based on the characteristics of real-time scenarios, and combined with equipment status data under real-time extreme weather conditions, predicted power is obtained, thereby solving the technical problem that existing technologies struggle to improve the accuracy of power prediction. Attached Figure Description

[0017] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings.

[0018] Figure 1 This is a flowchart illustrating the extreme weather identification method for new energy power prediction according to the present invention. Figure 2 This is an architecture diagram of the extreme weather identification system for predicting new energy power, based on the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only one preferred embodiment of this invention and are only used to explain this invention. They do not limit the scope of protection of this invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0020] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (or steps) as sequential processes, many of the operations (or steps) can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but it may also have additional steps not included in the figures; the process may correspond to a method, function, procedure, subroutine, subroutine, etc.

[0021] Example 1: like Figure 1 As shown, the extreme weather identification method for new energy power forecasting includes the following steps: S1: Construct an initial extreme weather identification model based on random forest and convolutional neural network, and input historical meteorological time series data into the initial extreme weather identification model to obtain historical fusion features.

[0022] In this embodiment, the number of trees in the random forest is set to 200, the maximum depth is limited to 10 layers, the Gini coefficient is used for feature selection, a 32-dimensional meteorological feature vector is input, and the probability distribution of extreme weather types is output, enabling rapid identification of 7 types of extreme weather, including typhoons and thunderstorms. The convolutional neural network is specifically a one-dimensional convolutional neural network with a network structure containing 3 convolutional layers. The kernel size is uniformly set to 3. Each convolutional layer is followed by a ReLU activation function and a max-pooling layer, focusing on extracting local features from meteorological time series to enhance the ability to identify abrupt weather change patterns.

[0023] As an optional embodiment, the step of inputting historical meteorological time-series data into the initial extreme weather identification model to obtain historical fusion features includes: Historical meteorological time-series data are input into the initial extreme weather identification model to extract features and obtain the first time-series abrupt change features and the first basic time-series features. The historical fusion features are obtained by splicing the first time-series mutation features with the first basic time-series features.

[0024] Understandably, the first time-series abrupt change feature is a detailed feature of abrupt change, fluctuation, and trend extracted from the time-series evolution of meteorological data. It reflects the dynamic evolution of meteorological factors over time, such as "wind speed suddenly jumps from 5 m / s to 10 m / s" or "air pressure first drops slowly for 2 hours and then drops sharply for 1 hour." The first basic time-series feature is a static quantitative indicator obtained after global statistics of meteorological time-series data. It does not focus on the dynamic process of meteorological factors changing over time, but only on the statistical results within the entire time window, such as "average wind speed 9 m / s" or "maximum wind speed 15 m / s."

[0025] In some embodiments, before inputting historical meteorological time-series data into the initial extreme weather identification model for feature extraction to obtain the first time-series abrupt change feature and the first basic time-series feature, the method further includes: Outliers in historical meteorological time-series data are removed using the 3σ criterion, denoising is performed on the historical meteorological time-series data using wavelet denoising, and missing values ​​are handled using adaptive interpolation; the historical meteorological time-series data includes at least wind speed, temperature, humidity, and air pressure.

[0026] S2: Construct an initial feature evaluation model using causal testing and SHAP analysis. Train the initial feature evaluation model with historical fusion features and historical extreme weather data to obtain the final feature evaluation model. Based on the final feature evaluation model and the initial extreme weather identification model, obtain the final extreme weather identification model.

[0027] In some embodiments, the construction of the initial feature evaluation model through causal testing and SHAP analysis includes: An initial association identification model is constructed using the Pearson correlation coefficient in the causality test. An initial association verification model is constructed using the Granger causality test in the causality test method; An initial feature evaluation model is constructed based on the initial association identification model, the initial association verification model, and SHAP analysis.

[0028] Understandably, the Pearson correlation coefficient is used to filter out features with a clear statistical association with extreme weather (such as wind speed and pressure drop) and eliminate completely irrelevant redundant features (such as cloud cover fluctuations), thus significantly reducing the amount of subsequent computation. The Granger causality test is used to perform a secondary verification of the selected associated features, thereby addressing the Pearson correlation coefficient's inability to distinguish between causal and spurious associations. This ensures that the features ultimately included are the core features that truly influence extreme weather identification, rather than interference features with accidental associations. Furthermore, the introduction of SHAP analysis transforms the contribution of each feature to extreme weather identification from a black-box weighted allocation into a quantifiable and traceable numerical value, thereby improving the credibility of the weight allocation.

[0029] As an optional embodiment, the step of training the initial feature evaluation model with historical fusion features and historical extreme weather data to obtain the final feature evaluation model includes: The historical fusion features are input into the initial feature evaluation model to obtain the correlation coefficient between each feature in the historical fusion features and historical extreme weather. Features with correlation coefficients less than the preset correlation coefficient are removed to obtain process features. The process features are compared with historical extreme weather data to obtain the final features. The output of the initial feature evaluation model and the final features are used to obtain the predicted extreme weather. If the predicted extreme weather does not match the historical extreme weather, the parameters of the initial feature evaluation model are adjusted according to the deviation between the predicted extreme weather and the historical extreme weather to obtain the final feature evaluation model. Otherwise, the initial feature evaluation model is used as the final feature evaluation model.

[0030] In this embodiment, the preset correlation coefficient is specifically 0.6, but it can also be flexibly set according to user needs. During the causal test, the preset significance threshold is set to 0.05. When the significance level between the feature and historical extreme weather is less than 0.05, it indicates that there is a real causal relationship between the feature and the extreme weather, and the feature is then included in the final feature.

[0031] S3: Input real-time meteorological time series data into the final extreme weather identification model to obtain real-time extreme weather.

[0032] In some embodiments, inputting real-time meteorological time-series data into the final extreme weather identification model to obtain real-time extreme weather includes: Real-time meteorological time-series data are input into the final extreme weather identification model to extract features, thereby obtaining the second time-series abrupt change features and the second basic time-series features. The second time-series abrupt change feature is concatenated with the second basic time-series feature to obtain the real-time fusion feature. The real-time fusion feature is then input into the final extreme weather recognition model to obtain the weight of each feature in the real-time fusion feature. Real-time extreme weather is then obtained based on the weight of each feature in the real-time fusion feature.

[0033] S4: Based on the characteristics of the real-time scene, obtain the predicted power by combining the equipment status data under real-time extreme weather conditions.

[0034] In some preferred embodiments, obtaining the predicted power based on the characteristics of the real-time scene and combined with equipment status data under real-time extreme weather conditions includes: The characteristic interference factor is determined based on the real-time scene characteristics, the correction direction of the initial power is determined based on the characteristic interference factor, and the initial power is corrected according to the correction direction to obtain the process power. The abnormal state is identified by comparing the equipment status data under extreme weather conditions with the normal threshold of the equipment under extreme weather conditions. The process power is then corrected according to the type of abnormal state identified to obtain the predicted power.

[0035] Specifically, the real-time scenarios include at least mountainous areas, islands, and oceans.

[0036] Specifically, the characteristic interference factors of the mountainous area include at least the terrain shading coefficient, turbulence intensity index, and local precipitation enhancement rate; The characteristic interference factors of the island include at least the concentration of salt spray in the air, the duration of relative humidity exceeding the standard, and the frequency of sea and land breeze switching. The characteristic disturbance factors of the ocean include at least the wind-wave coupling coefficient and the ocean current velocity.

[0037] It is understandable that the initial power is a theoretically calculated power that does not consider real-time scene feature interference or the actual state deviation of the equipment under extreme weather conditions. In this embodiment, taking a mountainous scene as an example, it is understandable that the higher the terrain shading coefficient, the stronger the reduction of wind speed by the terrain. Therefore, when the terrain shading coefficient is high, the initial power is reduced accordingly. When the local precipitation enhancement rate is high, precipitation will affect air humidity, which in turn will affect the operation of the wind turbine. Therefore, when the local precipitation enhancement rate is high, the initial power is reduced accordingly.

[0038] Understandably, equipment status data refers to real-time collected values ​​of various monitoring indicators that reflect the operating status of the equipment. These data directly reflect the current health and performance of the equipment and serve as the basis for determining whether the equipment is abnormal. In this embodiment, if the equipment status data shows a blade wear rate of 15% and the normal threshold is 5%, then the blade wear is considered abnormal. Since blade wear reduces the wind turbine's sensitivity to wind speed changes, the process power is reduced. If the equipment status data shows a salt spray concentration of 8 mg / cm² and the normal threshold is 5 mg / cm², the high salt spray concentration on the island leads to generator and photovoltaic module corrosion, reducing the wind turbine's sensitivity to wind speed changes and significantly decreasing the photovoltaic module's sensitivity to dynamic changes in irradiance. Therefore, the process power is also reduced in this case. In this embodiment, the predicted power is obtained by combining equipment status data under extreme weather conditions with the characteristics of the real-time scenario, thereby significantly improving the accuracy of the predicted power.

[0039] In this embodiment, the robustness of each model is ensured through 5-fold cross-validation, achieving an extreme weather recognition accuracy of ≥90% on the test set. In actual deployment, the system has achieved accurate predictions in multiple scenarios including mountainous areas, islands, and oceans, providing reliable technical support for the safe operation of new energy power plants under extreme weather conditions.

[0040] Example 2: This embodiment also provides an extreme weather identification system for new energy power prediction, applicable to the aforementioned extreme weather identification method for new energy power prediction, characterized by including: The historical fusion feature acquisition module is used to build an initial extreme weather identification model based on random forest and convolutional neural network, and to obtain historical fusion features by inputting historical meteorological time series data into the initial extreme weather identification model; The final extreme weather identification model acquisition module is used to construct an initial feature evaluation model through causal testing and SHAP analysis, train the initial feature evaluation model with historical fusion features and historical extreme weather to obtain the final feature evaluation model, and obtain the final extreme weather identification model based on the final feature evaluation model and the initial extreme weather identification model. The real-time extreme weather acquisition module is used to input real-time meteorological time series data into the final extreme weather recognition model to obtain real-time extreme weather. The predicted power acquisition module is used to obtain predicted power based on the characteristics of the real-time scene and combined with the equipment status data under real-time extreme weather conditions.

[0041] In this embodiment, the system also includes a data acquisition module for collecting meteorological time-series data. The system adopts an advanced cloud-edge-device collaborative deployment architecture, such as... Figure 2 As shown, the architecture realizes the entire chain of functions from data acquisition to intelligent application, ensuring the real-time performance and reliability of power prediction under extreme weather conditions. The architecture is divided into five core parts from top to bottom. The bottom layer is the data acquisition module, i.e., the field acquisition terminal, which consists of meteorological sensors, power acquisition devices, and equipment status monitors deployed at renewable energy power plants. It acts as the system's sensory nerves, responsible for capturing real-time meteorological time-series data such as wind speed, irradiance, and actual power generation. The acquired data is uploaded to the edge computing node via communication networks (including 4G / 5G wireless and Ethernet). The edge computing node consists of a historical fusion feature acquisition module, a final extreme weather identification model acquisition module, a real-time extreme weather acquisition module, and a predicted power acquisition module. This node is the system's local brain, with a built-in ARM computing unit and a dedicated neural network accelerator, capable of independently running a real-time prediction engine to quickly identify extreme weather and complete second-level power prediction, greatly reducing the risks caused by cloud communication latency. The edge node not only reports the processed key data and models to the cloud platform center but also directly serves external systems such as power grid dispatching and power plant monitoring through standard application interfaces. The cloud platform, acting as the intelligent hub, bears the heavy responsibility of storing massive amounts of data, training complex models, and managing global monitoring. It can also continuously distribute optimized prediction models to edge nodes, enabling the system to self-evolve. This layered and distributed deployment ensures independent and stable operation at the edge under harsh communication conditions while fully utilizing the powerful computing capabilities of the cloud.

[0042] The substantial effects of this invention are: In terms of extreme weather identification and feature extraction capabilities, this invention integrates a hybrid identification model that combines random forest and one-dimensional convolutional neural network with SHAP interpretability analysis. This not only accurately identifies extreme weather types but also deeply analyzes the contribution of various meteorological elements, thereby improving the extraction accuracy of key features such as typhoon path, intensity change, and duration. In terms of prediction accuracy and stability, this invention deeply analyzes the physical impact mechanism of extreme weather on power generation equipment, including the stress effect of strong winds on wind turbine blades, the corrosive effect of salt spray on photovoltaic modules, and the distortion effect of turbulence on power curves, thereby improving the power prediction accuracy under extreme weather conditions. In terms of multi-scenario adaptability, this invention fully considers the unique factors of the region, such as terrain undulation, sea and land breeze effect, and wave height, which greatly reduces the workload of customized development. This strong adaptability makes this invention not only suitable for large-scale new energy bases, but also well meet the forecasting needs of distributed power generation projects, greatly expanding the application scope of the technology. In terms of real-time performance and reliability, the edge computing architecture significantly improves the system's responsiveness under extreme weather conditions. The system pushes the core identification and prediction algorithms down to the edge nodes, reducing the reliance on cloud communication. Even in the event of a network outage, it can still maintain basic prediction functions and meet the needs of the power grid for rapid response. In addition, the distributed architecture also improves the system's robustness. The failure of a single node will not cause the entire system to fail, ensuring the continuity of prediction services under extreme weather conditions. In terms of economic benefits and operation and maintenance costs, this invention brings significant value enhancement through accurate prediction. The system can help operating units formulate scientific power generation plans before extreme weather arrives, effectively reducing deviation assessment costs. Through real-time monitoring and early warning of equipment status, the system can guide operation and maintenance personnel to take protective measures in a timely manner, thereby reducing the risk of equipment damage and extending the service life of critical equipment.

[0043] The specific embodiments described above are preferred embodiments of the extreme weather identification method and system for new energy power prediction of the present invention, and are not intended to limit the specific scope of the present invention. The scope of the present invention includes but is not limited to the specific embodiments described above. All equivalent changes made in accordance with the shape and structure of the present invention are within the protection scope of the present invention.

Claims

1. An extreme weather identification method for new energy power forecasting, characterized in that, Includes the following steps: An initial extreme weather identification model is constructed based on random forest and convolutional neural network. Historical meteorological time series data are input into the initial extreme weather identification model to obtain historical fusion features. An initial feature evaluation model is constructed using causal testing and SHAP analysis. The initial feature evaluation model is then trained using historical fusion features and historical extreme weather data to obtain the final feature evaluation model. Finally, the final extreme weather identification model is obtained based on the final feature evaluation model and the initial extreme weather identification model. Real-time meteorological time series data is input into the final extreme weather identification model to obtain real-time extreme weather; Based on the characteristics of the real-time scenario, the predicted power is obtained by combining real-time equipment status data under extreme weather conditions.

2. The extreme weather identification method for new energy power prediction according to claim 1, characterized in that, The step of inputting historical meteorological time-series data into the initial extreme weather identification model to obtain historical fusion features includes: Historical meteorological time-series data are input into the initial extreme weather identification model to extract features and obtain the first time-series abrupt change features and the first basic time-series features. The historical fusion features are obtained by splicing the first time-series mutation features with the first basic time-series features.

3. The extreme weather identification method for new energy power prediction according to claim 2, characterized in that, Before inputting historical meteorological time-series data into the initial extreme weather identification model for feature extraction to obtain the first time-series abrupt change feature and the first basic time-series feature, the process also includes: Outliers in historical meteorological time-series data are removed using the 3σ criterion, denoising is performed on the historical meteorological time-series data using wavelet denoising, and missing values ​​are handled using adaptive interpolation; the historical meteorological time-series data includes at least wind speed, temperature, humidity, and air pressure.

4. The extreme weather identification method for new energy power prediction according to claim 1, characterized in that, The initial feature evaluation model is constructed using causal testing and SHAP analysis, including: An initial association identification model is constructed using the Pearson correlation coefficient in the causality test. An initial association verification model is constructed using the Granger causality test in the causality test method; An initial feature evaluation model is constructed based on the initial association identification model, the initial association verification model, and SHAP analysis.

5. The extreme weather identification method for new energy power prediction according to claim 1, characterized in that, The process of training the initial feature evaluation model using historical fusion features and historical extreme weather data to obtain the final feature evaluation model includes: The historical fusion features are input into the initial feature evaluation model to obtain the correlation coefficient between each feature in the historical fusion features and historical extreme weather. Features with correlation coefficients less than the preset correlation coefficient are removed to obtain process features. The process features are compared with historical extreme weather data to obtain the final features. The output of the initial feature evaluation model and the final features are used to obtain the predicted extreme weather. If the predicted extreme weather does not match the historical extreme weather, the parameters of the initial feature evaluation model are adjusted according to the deviation between the predicted extreme weather and the historical extreme weather to obtain the final feature evaluation model. Otherwise, the initial feature evaluation model is used as the final feature evaluation model.

6. The extreme weather identification method for new energy power prediction according to claim 1, characterized in that, The step of inputting real-time meteorological time-series data into the final extreme weather identification model to obtain real-time extreme weather includes: Real-time meteorological time-series data are input into the final extreme weather identification model to extract features, thereby obtaining the second time-series abrupt change features and the second basic time-series features. The second time-series abrupt change feature is concatenated with the second basic time-series feature to obtain the real-time fusion feature. The real-time fusion feature is then input into the final extreme weather recognition model to obtain the weight of each feature in the real-time fusion feature. Real-time extreme weather is then obtained based on the weight of each feature in the real-time fusion feature.

7. The extreme weather identification method for new energy power prediction according to claim 1, characterized in that, The method of obtaining predicted power based on the characteristics of the real-time scene and combined with equipment status data under real-time extreme weather conditions includes: The characteristic interference factor is determined based on the real-time scene characteristics, the correction direction of the initial power is determined based on the characteristic interference factor, and the initial power is corrected according to the correction direction to obtain the process power. The abnormal state is identified by comparing the equipment status data under extreme weather conditions with the normal threshold of the equipment under extreme weather conditions. The process power is then corrected according to the type of abnormal state identified to obtain the predicted power.

8. The extreme weather identification method for new energy power prediction according to claim 7, characterized in that, The real-time scenarios include at least mountainous areas, islands, and oceans.

9. The extreme weather identification method for new energy power prediction according to claim 8, characterized in that, The characteristic disturbance factors in the mountainous area include at least the terrain shading coefficient, turbulence intensity index, and local precipitation enhancement rate. The characteristic interference factors of the island include at least the concentration of salt spray in the air, the duration of relative humidity exceeding the standard, and the frequency of sea and land breeze switching. The characteristic disturbance factors of the ocean include at least the wind-wave coupling coefficient and the ocean current velocity.

10. An extreme weather identification system for new energy power forecasting, applicable to the extreme weather identification method for new energy power forecasting as described in any one of claims 1-9, characterized in that, include: The historical fusion feature acquisition module is used to build an initial extreme weather identification model based on random forest and convolutional neural network, and to obtain historical fusion features by inputting historical meteorological time series data into the initial extreme weather identification model; The final extreme weather identification model acquisition module is used to construct an initial feature evaluation model through causal testing and SHAP analysis, train the initial feature evaluation model with historical fusion features and historical extreme weather to obtain the final feature evaluation model, and obtain the final extreme weather identification model based on the final feature evaluation model and the initial extreme weather identification model. The real-time extreme weather acquisition module is used to input real-time meteorological time series data into the final extreme weather recognition model to obtain real-time extreme weather. The predicted power acquisition module is used to obtain predicted power based on the characteristics of the real-time scene and combined with the equipment status data under real-time extreme weather conditions.

Images