Electric power meteorological prediction method and device based on adaptive enhancement

Abstract

The invention discloses an electric power meteorological prediction method and device based on adaptive enhancement, and relates to the technical field of electric power meteorological prediction, and the method comprises the following steps: collecting historical meteorological data and electric power data, and cleaning the collected data; according to the method, a model reasoning path tracing module is constructed based on Chain-of-Though logic, a prediction result is decomposed into subtasks such as meteorological factor contribution degree and power system response logic, interpretability of the prediction process is achieved, reliability of a generated result is evaluated by adopting sequence probability confidence, a visual interpretation report is dynamically generated, and the prediction efficiency is improved. According to the method, operation and maintenance personnel can visually acquire key information, sensitivity of the model to key meteorological events is optimized through adversarial training, prediction capability of the model to extreme meteorological conditions is improved, the trust problem of the black box model in power decision is finally effectively solved, a visual and reliable decision basis is provided for the operation and maintenance personnel, and the operation and maintenance efficiency is improved. And the operation safety and stability of the power system under the complex meteorological condition are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power meteorological forecasting, and in particular to an electric power meteorological forecasting method and device based on adaptive enhancement. Background Art

[0002] The power system is an electric energy production and consumption system composed of power generation, transmission, transformation, distribution and consumption. It converts natural primary energy into electrical energy through power generation equipment, transmits it to various regions through the transmission network, changes the voltage level through transformation equipment, and finally distributes it to various users through the distribution system to meet the electricity needs of social production and life. It is an important infrastructure for the development of modern social economy.

[0003] It is crucial to conduct power meteorological forecasts. Meteorological conditions have a direct and significant impact on the operation of the power system. For example, high temperatures will greatly increase the electricity demand for refrigeration equipment such as air conditioners, resulting in a sharp increase in power load; severe weather such as strong winds, heavy rains, and blizzards may damage power equipment and affect the stability of power supply. Accurate power meteorological forecasts can help power companies make power dispatching and equipment maintenance arrangements in advance, ensure the safe and stable operation of the power system, reduce operating costs, and improve power supply reliability. It is of great significance to ensure the normal operation of society and stable economic development.

[0004] However, there are many problems with existing power meteorological forecasting technologies. Most existing power meteorological forecasting models are black box models. Although they have certain performance in prediction accuracy, they cannot explain the generation process of the prediction results, making it difficult for operation and maintenance personnel to understand the logical relationship between meteorological factors and power system responses. When making power decisions, there is a lack of trust basis and it is difficult to formulate targeted response strategies, which may lead to unstable power supply and affect social production and life. Therefore, it is of great significance to develop power meteorological forecasting methods and devices based on adaptive enhancement. Summary of the Invention

[0005] The purpose of the present invention is to make up for the shortcomings of the existing technology and provide an electric power meteorological forecasting method and device based on adaptive enhancement. It can construct a model reasoning path tracing module based on Chain-of-Thought logic, decompose the forecast results into sub-tasks such as the contribution of meteorological factors and the response logic of the power system, realize the interpretability of the forecast process, use sequence probability confidence to evaluate the reliability of the generated results, and dynamically generate visual explanation reports so that operation and maintenance personnel can intuitively obtain key information. Through adversarial training, the model's sensitivity to key meteorological events is optimized, the model's predictive ability for extreme meteorological conditions is improved, and the trust problem of black box models in power decision-making is solved.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for power meteorological forecasting based on adaptive enhancement, the method comprising the following steps:

[0007] Collect historical meteorological data M={m1,m2,…,m n} and power data E={e1,e2,…,e n The meteorological monitoring station provides meteorological data, including temperature, humidity, and wind speed. The power system monitoring equipment provides power data, including power load and equipment operating status parameters. The collected data is cleaned, outliers are removed, and the data is normalized.

[0008] Based on the Chain-of-Thought logic, the model reasoning path tracing module is constructed, and the processed data is used to calculate the path tracing module through the formula Calculate meteorological factors m i Contribution C i , where m ij is the value of the i-th meteorological factor in the j-th sample, and Meteorological factors are i and the mean value of the power data e, simulate the power system topology and operation rules, and establish a power system response logic model;

[0009] Taking into account the uncertainty of meteorological data, the dynamic changes of the power system, and the model error, the reliability of the forecast results is calculated using the formula R = α × U + β × D + γ × Err, where U represents the quantitative value of the uncertainty of meteorological data, D represents the quantitative value of the dynamic changes of the power system, Err represents the quantitative value of the model error, and α, β, and γ are weight coefficients.

[0010] Generate adversarial samples, train the model together with the original data, and adjust the model parameters.

[0011] Furthermore, in determining the contribution C i During the calculation, the process of optimizing the calculation process and determining the relevant parameters is as follows: first, extract the features of historical meteorological data and power data, and use the principal component analysis method to convert multiple meteorological and power features into a few unrelated comprehensive features. Then, use the stepwise regression algorithm to screen out the features that have a significant impact on the calculation of the contribution of meteorological factors from these comprehensive features, and then determine the coefficients in the formula. In the feature extraction process, the weight of each comprehensive feature is determined based on the variance contribution rate of the feature. The larger the variance contribution rate, the more important the comprehensive feature is in subsequent calculations.

[0012] Furthermore, when determining the weight coefficients α, β, and γ, the particle swarm optimization algorithm is used. First, the initial position and speed of the particle swarm are set. Each particle represents a set of possible values ​​of the weight coefficients α, β, and γ. In each iteration, the particle updates its speed and position based on its own historical optimal position and the global optimal position. The sum of the squares of the errors between the historical prediction results and the actual results is used as the fitness function. The iterative calculation is continuously performed until the preset number of iterations and the convergence conditions of the fitness function are met. The weight coefficients corresponding to the particle positions obtained at this time are the determined α, β, and γ.

[0013] Furthermore, when constructing the model reasoning path tracing module based on Chain-of-Thought logic, the calculated contribution of meteorological factors C i Sort in descending order and select the top k meteorological factors with the highest contribution as key meteorological factors. k is determined based on statistical analysis of historical data and actual business needs, and 1≤k≤n, where n is the total number of meteorological factors. At the same time, obtain the geographic location information of each device in the power system and establish a geography-meteorology-power impact model based on the geographic information data. This model considers the impact of meteorological differences in different geographic locations on power equipment, and refines the power system response logic model through this model.

[0014] Furthermore, when generating adversarial samples, different processing methods are adopted for different types of key meteorological events. For strong wind events, the wind speed and direction are adjusted according to the wind speed and wind direction variation range of historical strong wind data to generate adversarial samples. For heavy rain events, the rainfall and rainfall duration are adjusted according to historical rainfall and rainfall duration data to generate adversarial samples. The adversarial samples generated in the above way cover various possible situations of key meteorological events.

[0015] Furthermore, in the process of optimizing the model's sensitivity to key meteorological events through adversarial training, a multi-model fusion strategy is adopted to select multiple models, including convolutional neural network CNN, recurrent neural network RNN ​​and support vector machine SVM models. In the early stage of training, the same weight is assigned to each model. After each iterative training, the fusion weight is adjusted according to the prediction accuracy of each model on the validation set. The specific adjustment method is: calculate the ratio of the prediction accuracy of each model to the sum of the prediction accuracy of all models, and use this ratio as the new fusion weight of the model. Through continuous iterative training, the advantages of multiple models are complementary to improve the overall model performance.

[0016] An adaptive enhancement-based power meteorological forecasting device is applicable to the above-mentioned adaptive enhancement-based power meteorological forecasting method, and includes a data preprocessing module, a model reasoning path tracing module, a prediction result reliability assessment module, and a model sensitivity optimization module;

[0017] The data preprocessing module is used to collect historical meteorological data and power data, clean the data to remove outliers, and complete data normalization;

[0018] The model reasoning path tracing module is constructed based on Chain-of-Thought logic, uses pre-processed data to calculate the contribution of meteorological factors, simulates the power system topology and operation rules, and constructs a power system response logic model;

[0019] The forecast result reliability assessment module comprehensively considers the uncertainty of meteorological data, dynamic changes of the power system and model errors to complete the calculation of the reliability of the forecast result;

[0020] The model sensitivity optimization module is responsible for generating adversarial samples and using them together with original data to train the model. During the training process, the model parameters are adjusted to enhance the model's ability to identify and predict key meteorological events.

[0021] Furthermore, the device also includes a data storage module for storing collected historical meteorological data, power data, as well as intermediate data and result data generated during data processing, model training and prediction.

[0022] Compared with the existing technology, the power meteorological forecasting method and device based on adaptive enhancement have the following beneficial effects:

[0023] The present invention constructs a model reasoning path tracing module based on Chain-of-Thought logic, decomposes the prediction results into sub-tasks such as the contribution of meteorological factors and the response logic of the power system, thereby achieving the interpretability of the prediction process, using sequence probability confidence to evaluate the reliability of the generated results, and dynamically generating visual explanation reports, so that operation and maintenance personnel can intuitively obtain key information. Through adversarial training, the model's sensitivity to key meteorological events is optimized, and the model's prediction ability for extreme meteorological conditions is improved. Ultimately, it effectively solves the trust problem of black box models in power decision-making, provides intuitive and reliable decision-making basis for operation and maintenance personnel, and improves the safety and stability of power system operation under complex meteorological conditions.

[0024] Other advantages, objects and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be learned from the practice of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort.

[0026] Figure 1 It is a flowchart of the power meteorological forecasting method based on adaptive enhancement;

[0027] Figure 2 This is a structural diagram of an electric power meteorological forecasting device based on adaptive enhancement. DETAILED DESCRIPTION

[0028] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0029] Example 1

[0030] In a large city with a dense population and diversified industries, its power supply network has a wide coverage and complex structure. The city includes multiple commercial areas, residential areas, industrial areas and a large number of public facilities. The electricity consumption characteristics of different areas vary significantly. At the same time, the city's geographical location has complex meteorological conditions with significant changes in the four seasons. The city is often threatened by extreme weather disasters such as strong winds, heavy rains, and blizzards. These meteorological factors will directly affect the normal operation of power equipment. Accurate power meteorological forecasts are crucial to ensuring the stability of the city's power supply, meeting the electricity demand of different areas, rationally arranging power resources, and reducing the damage of meteorological disasters to the power system.

[0031] The power company has deployed a complete data collection system to obtain historical meteorological data M = {m1,m2,…,m n These meteorological data cover various meteorological factors such as temperature, humidity, wind speed, wind direction, rainfall, air pressure, etc. At the same time, the corresponding power data E = {e1, e2, …, e n}, including parameters such as power load, active power, reactive power, voltage, current, etc. at different times.

[0032] After data collection is completed, the collected data is cleaned. First, by setting a reasonable data threshold range, data that is obviously beyond the normal range is eliminated, such as abnormally high or low temperature values, unreasonable sudden increases or decreases in power load, etc. Then, data interpolation is used to supplement the missing data to ensure data integrity. Next, the data is normalized so that different types of data have the same scale, which is convenient for subsequent analysis and model training.

[0033] Based on the Chain-of-Thought logic, a model reasoning path tracing module is constructed, and the contribution of meteorological factors is calculated using the pre-processed data. Calculate each meteorological factor m i Contribution to power data C i , where m ij is the value of the i-th meteorological factor in the j-th sample, and Meteorological factors are i and the mean of the power data e. For example, when calculating the contribution of temperature to power load, substitute the temperature data into the m in the formula i , the power load data is substituted into e, and the relative importance of temperature as a meteorological factor in affecting the change of power load is calculated through the formula.

[0034] After the calculation is completed, the contribution of meteorological factors C i Sort in descending order and select the top k (k is determined based on historical data statistical analysis and actual business needs, and 1≤k≤n) meteorological factors with the highest contribution as key meteorological factors. In this embodiment, analysis shows that temperature, wind speed, and rainfall have a greater impact on power load and equipment operation, and are therefore identified as key meteorological factors.

[0035] In order to deeply understand the response logic of the power system under different meteorological conditions and simulate the topological structure and operation rules of the power system, with the help of geographic information system (GIS) data, the geographical distribution of the power system is accurately mapped, including the location information of power plants, substations, transmission lines, distribution stations and major power consumption areas. Combined with the electrical parameters and operation principles of the power system, a power system response logic model is established. Taking into account the different impacts of meteorological differences in different geographical locations on power equipment, for example, substations located on the coast are more susceptible to erosion by sea breezes, a geography-meteorology-power impact model is established. Through this model, the changes in the operating status of power equipment in different regions under specific meteorological conditions are analyzed, such as the impact of temperature on transformer heat dissipation, the impact of wind speed on transmission line losses, etc., to further refine the power system response logic model.

[0036] Taking into account the uncertainty U of meteorological data, the dynamic changes D of the power system and the model error Err, the reliability R of the prediction result is calculated using the formula R = α × U + β × D + γ × Err. Among them, the uncertainty U of meteorological data is quantified by analyzing the fluctuation range and change trend of historical meteorological data, such as statistical indicators such as the standard deviation and coefficient of variation of meteorological factors such as temperature and wind speed within a certain time range; the dynamic changes D of the power system are determined by monitoring real-time load changes, equipment operating status switching and other data in the power system; the model error Err is calculated by comparing and analyzing the model prediction results with historical actual data to calculate statistical indicators of the prediction error, such as the mean square error (MSE) and the mean absolute error (MAE).

[0037] α, β, and γ are weight coefficients, which are determined by the particle swarm optimization algorithm. First, the initial position and speed of the particle swarm are set. Each particle represents a possible value of a set of weight coefficients (α, β, γ). In each iteration, the particle updates its speed and position based on its own historical optimal position and the global optimal position. The sum of squares of the errors between the historical prediction results and the actual results is used as the fitness function. The iterative calculation is continuously performed until the preset number of iterations or the convergence condition of the fitness function is met. The weight coefficients corresponding to the particle positions obtained at this time are α, β, and γ in the formula for determining the reliability of the prediction results.

[0038] Adversarial samples are generated for different types of key meteorological events. For strong wind events, the wind speed adjustment ratio range is determined based on the wind speed and wind direction variation range of historical strong wind data. When generating adversarial samples, the wind speed adjustment ratio and wind direction rotation angle are randomly selected within this range to perturb the original strong wind data and simulate strong winds of different intensities and directions.

[0039] For rainstorm events, the adjustment range of rainfall is determined based on historical rainfall and rainfall duration data. When generating adversarial samples, adjustment values ​​are randomly selected within these ranges to modify the original rainstorm data and simulate rainstorms of different intensities and durations.

[0040] The generated adversarial samples are used together with the original data to train the model, and the backpropagation algorithm in deep learning is used to adjust the model parameters. During the training process, the network structure and weights of the model are continuously optimized so that the model can better identify and predict the impact of key meteorological events on the power system. For example, in a model based on the combination of convolutional neural network CNN and recurrent neural network RNN, the weight parameters of the convolutional layer, pooling layer and recurrent layer in the network are adjusted through the backpropagation algorithm to improve the model's ability to extract and analyze features of complex meteorological data and dynamic changes in the power system.

[0041] The trained model is used to predict future power and meteorological conditions. The model can not only predict the trend of power load changes in the future, but also predict the failure risks of power equipment under different meteorological conditions. For example, when it is predicted that high temperatures will continue in the next few days, the model will predict that the power load in various regions will increase significantly, especially the air-conditioning power load in residential and commercial areas. At the same time, based on the power system response logic model and the geography-meteorology-power impact model, it is predicted that high temperatures may cause some power equipment to overheat, posing a risk of failure. For example, the increase in transformer oil temperature may affect its insulation performance and normal operation.

[0042] The generated forecast results will be presented to the power company's operation and maintenance personnel in the form of a visual explanatory report. The report uses charts, graphs, and other intuitive methods to show the path and extent of the impact of meteorological factors on the power system. Based on these forecast results and explanatory reports, the power company's operation and maintenance personnel will formulate targeted power dispatch plans and equipment maintenance plans in advance.

[0043] In summary, the interpretability of power meteorological forecasts is achieved through this embodiment, and operation and maintenance personnel can clearly understand the logical relationship between meteorological factors and power system responses, thereby enhancing their trust in the forecast results. When faced with complex meteorological conditions, power companies can carry out power dispatching and equipment maintenance in advance based on accurate forecast results, effectively reducing the operating risks of the power system under complex meteorological conditions. At the same time, by evaluating the reliability of forecast results and optimizing model sensitivity, the accuracy of the forecast is improved, so that power resources can be allocated more reasonably, operating costs are reduced, and the economic benefits of power companies are improved. In addition, since the stability of power supply is guaranteed, it also provides strong support for the city's economic development and residents' lives, and promotes the sustainable development of the city.

[0044] Example 2

[0045] In a large energy base with a vast territory and diversified energy structure, there are not only thermal power generation and hydropower generation, but also gradually developed new energy projects such as wind power generation and photovoltaic power generation. There are many power plants and substations of different sizes and types distributed in the base, and the power transmission network is crisscrossed. The meteorological conditions in the area where the energy base is located are complex and changeable, with large temperature differences between day and night, and it is often subject to special meteorological disasters such as sandstorms and cold waves.

[0046] For thermal power plants, high temperatures can affect the efficiency of the cooling system and increase energy consumption, while cold snaps can cause equipment pipes to freeze and crack, impacting normal power generation. Hydropower generation is significantly affected by precipitation and temperature. Insufficient precipitation can cause reservoir water levels to drop, reducing power generation. Floods caused by heavy rains can threaten the safety of hydropower station facilities. Wind and photovoltaic power generation are more sensitive to meteorological conditions. Dust storms can reduce the power generation efficiency of photovoltaic modules, while unstable wind speeds can affect the normal operation and power generation stability of wind turbines. Therefore, accurate power meteorological forecasts are crucial to energy bases' power production planning, equipment maintenance management, and efficient energy output.

[0047] The data acquisition system of the energy base comprehensively collects various data and obtains historical meteorological data from densely distributed meteorological monitoring stations around the base. n}, covering meteorological factors such as temperature, humidity, wind speed, wind direction, precipitation, dust concentration, etc. At the same time, the corresponding power data E = {e1, e2, …, e n}, including power generation power, equipment operating parameters, transmission line losses, etc.

[0048] After collecting the data, a cleaning operation is performed. By setting reasonable data filtering rules, outliers caused by sensor failure, communication interference, etc. are removed, such as temperature data that deviates significantly from the normal range or sudden changes in power generation data. Data smoothing algorithms are used to process data with large fluctuations to improve data quality. Afterwards, the data is normalized and data of different magnitudes are mapped to the [0, 1] interval to make the data comparable and facilitate subsequent analysis.

[0049] Based on the Chain-of-Thought logic, the model reasoning path tracing module is constructed, and the pre-processed data is used to calculate the path tracing module. Calculate meteorological factors m i Contribution C i For example, when calculating the contribution of dust concentration to photovoltaic power generation, substitute the dust concentration data into m i , the photovoltaic power generation data is substituted into e to obtain the quantitative index of the impact of dust concentration on photovoltaic power generation.

[0050] The contribution of meteorological factors was ranked and the top-ranked key meteorological factors were selected. After analysis, it was found that wind speed, temperature and dust concentration had significant impacts on wind power generation, thermal power generation and photovoltaic power generation respectively, and were identified as key meteorological factors.

[0051] Simulate the topological structure and operating rules of the power system within the energy base, combine the geographical location, equipment parameters and energy conversion principles of each power plant and substation, establish a power system response logic model, take into account the impact of meteorological differences in different regions on power equipment, and construct a geographical-meteorological-power impact model. For example, analyze the differences in the impact of wind speed shear on wind turbines at different altitudes, and the differences in power generation caused by different sunlight duration and intensity in photovoltaic power stations at different latitudes, and further refine the power system response logic model.

[0052] The reliability R of the prediction result is calculated by integrating the uncertainty of meteorological data U, the dynamic change of the power system D, and the model error Err using the formula R = α × U + β × D + γ × Err. The uncertainty of meteorological data U is quantified by analyzing the fluctuation characteristics of meteorological data and the prediction error, such as the standard deviation of wind speed and the mean absolute error of temperature prediction. The dynamic change of the power system D is determined based on data such as power adjustment and equipment switching of different power generation methods within the energy base. The model error Err is calculated by comparing the model prediction value with the actual power data.

[0053] The weight coefficients α, β, and γ are determined with the help of the particle swarm optimization algorithm. The initial state of the particle swarm is set. Each particle represents a set of weight coefficients. The sum of squares of the errors between historical prediction results and actual results is used as the optimization target. During the iteration process, the particles update their positions according to themselves and the global optimal solution until the convergence conditions are met and the optimal weight coefficients are obtained.

[0054] Adversarial samples are generated for special meteorological events such as sandstorms and cold waves. For sandstorm events, adversarial samples are generated by increasing or decreasing the dust concentration by a certain proportion and extending or shortening the duration accordingly based on the concentration and duration range of historical sandstorm data. For cold wave events, adversarial samples are generated by adjusting the cooling amplitude and duration based on the cooling amplitude and duration of historical cold wave data.

[0055] The generated adversarial samples are used together with the original data to train the model, and optimization algorithms such as gradient descent are used to adjust the model parameters. For example, in a model that integrates a convolutional neural network (CNN) and a long short-term memory (LSTM) network, the weights in the network are optimized through the backpropagation algorithm, so that the model can better capture the impact characteristics of special meteorological events on the power system and improve the ability to identify and predict key meteorological events.

[0056] The trained model is used to predict future power and meteorological conditions. The prediction results include the power generation trend of different power generation methods, equipment failure risks, etc. For example, if sandstorms are predicted in the future, the model will give the degree of reduction in power generation efficiency of each photovoltaic power station and the affected power generation value, and at the same time predict the potential failure risks that sandstorms may cause to photovoltaic modules.

[0057] The forecast results are presented in the form of a visual report, which includes a meteorological factor impact analysis chart, a power system operation status change chart, and risk warning information. The energy base's operation and maintenance personnel formulate corresponding strategies based on the forecast results and report.

[0058] In summary, through the method and process of this embodiment, the interpretability of the energy base power meteorological forecast is achieved, the operation and maintenance personnel can clearly understand the impact mechanism of meteorological factors on power production, and enhance their trust in the forecast results. Accurate forecasts help the energy base plan power production in advance and optimize energy distribution. At the same time, based on the reliability assessment of the forecast results and the optimization of the model sensitivity, the accuracy of the forecast is improved, the operating cost of the energy base is reduced, and the overall energy production efficiency and economic benefits are improved.

[0059] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as above in terms of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make some changes or modifications to equivalent embodiments using the technical contents disclosed above. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. The power meteorological forecasting method based on adaptive enhancement is characterized by: The method comprises the following steps: Collect historical meteorological data M={m1,m2,…,m n } and power data E={e1,e2,…,e n }, clean the collected data, remove outliers, and normalize the data; Based on the Chain-of-Thought logic, the model reasoning path tracing module is constructed, and the processed data is used to calculate the path tracing module through the formula Calculate meteorological factors m i Contribution C i , where m ij is the value of the i-th meteorological factor in the j-th sample, and Meteorological factors are i and the mean value of the power data e, simulate the power system topology and operation rules, and establish a power system response logic model; Taking into account the uncertainty of meteorological data, the dynamic changes of the power system, and the model error, the reliability of the forecast results is calculated using the formula R = α × U + β × D + γ × Err, where U represents the quantitative value of the uncertainty of meteorological data, D represents the quantitative value of the dynamic changes of the power system, Err represents the quantitative value of the model error, and α, β, and γ are weight coefficients. Generate adversarial samples, train the model together with the original data, and adjust the model parameters.

2. The power meteorological forecasting method based on adaptive enhancement according to claim 1, characterized in that: In determining the contribution C i During the calculation, the process of optimizing the calculation process and determining the relevant parameters is as follows: first, extract the features of historical meteorological data and power data, and use the principal component analysis method to convert multiple meteorological and power features into a few unrelated comprehensive features. Then, use the stepwise regression algorithm to screen out the features that have a significant impact on the calculation of the contribution of meteorological factors from these comprehensive features, and then determine the coefficients in the formula. In the feature extraction process, the weight of each comprehensive feature is determined based on the variance contribution rate of the feature. The larger the variance contribution rate, the more important the comprehensive feature is in subsequent calculations.

3. The power meteorological forecasting method based on adaptive enhancement according to claim 1, characterized in that: When determining the weight coefficients α, β, and γ, the particle swarm optimization algorithm is used. First, the initial position and speed of the particle swarm are set. Each particle represents a set of possible values ​​of the weight coefficients α, β, and γ. In each iteration, the particle updates its speed and position based on its own historical optimal position and the global optimal position. The sum of the squares of the errors between the historical prediction results and the actual results is used as the fitness function. The iterative calculation is continuously performed until the preset number of iterations and the convergence conditions of the fitness function are met. The weight coefficients corresponding to the particle positions obtained at this time are the determined α, β, and γ.

4. The power meteorological forecasting method based on adaptive enhancement according to claim 1, characterized in that: When constructing the model reasoning path tracing module based on Chain-of-Thought logic, the contribution of the calculated meteorological factors C i Sort in descending order and select the top k meteorological factors with the highest contribution as key meteorological factors. k is determined based on statistical analysis of historical data and actual business needs, and 1≤k≤n, where n is the total number of meteorological factors. At the same time, obtain the geographic location information of each device in the power system and establish a geography-meteorology-power impact model based on the geographic information data. This model considers the impact of meteorological differences in different geographic locations on power equipment, and refines the power system response logic model through this model.

5. The power meteorological forecasting method based on adaptive enhancement according to claim 1, characterized in that: When generating adversarial samples, different processing methods are used for different types of key meteorological events. For strong wind events, the wind speed and direction are adjusted according to the wind speed and wind direction variation range of historical strong wind data to generate adversarial samples. For heavy rain events, the rainfall and rainfall duration are adjusted according to historical rainfall and rainfall duration data to generate adversarial samples. The adversarial samples generated in the above way cover various possible situations of key meteorological events.

6. The power meteorological forecasting method based on adaptive enhancement according to claim 1, characterized in that: In the process of optimizing the model's sensitivity to key meteorological events through adversarial training, a multi-model fusion strategy is adopted to select multiple models, including convolutional neural network (CNN), recurrent neural network (RNN) and support vector machine (SVM) models. In the initial training phase, the same weight is assigned to each model. After each iterative training, the fusion weight is adjusted according to the prediction accuracy of each model on the validation set.

7. An electric power meteorological forecasting device based on adaptive enhancement, applicable to the electric power meteorological forecasting method based on adaptive enhancement according to claims 1-6, characterized in that: The device includes a data preprocessing module, a model reasoning path tracing module, a prediction result reliability assessment module, and a model sensitivity optimization module; The data preprocessing module is used to collect historical meteorological data and power data, clean the data to remove outliers, and complete data normalization; The model reasoning path tracing module is constructed based on Chain-of-Thought logic, uses pre-processed data to calculate the contribution of meteorological factors, simulates the power system topology and operation rules, and constructs a power system response logic model; The forecast result reliability assessment module comprehensively considers the uncertainty of meteorological data, dynamic changes of the power system and model errors to complete the calculation of the reliability of the forecast result; The model sensitivity optimization module is responsible for generating adversarial samples and using them together with original data to train the model. During the training process, the model parameters are adjusted to enhance the model's ability to identify and predict key meteorological events.

8. The power meteorological forecasting device based on adaptive enhancement according to claim 7, characterized in that: The device also includes a data storage module for storing collected historical meteorological data, power data, as well as intermediate data and result data generated during data processing, model training and prediction.

Images