Meteorological prediction precision improvement method and device, equipment, storage medium and computer program product

Abstract

The invention relates to the technical field of meteorological prediction, in particular to a meteorological prediction precision improvement method and device, equipment, a storage medium and a computer program product. The method comprises the following steps: constructing a plurality of initial weather prediction models based on a Transform structure in combination with a PIDL (Precision Independent Design Language) algorithm; taking the power curve of the wind turbine generator and the performance characteristic curve of the photovoltaic module as constraint conditions, and integrating the constraint conditions into each initial meteorological prediction model to obtain a plurality of basic meteorological prediction models; based on a preset automatic hyper-parameter tuning tool, performing hyper-parameter optimization on each basic meteorological prediction model in combination with a cross validation algorithm; fusing the plurality of basic meteorological prediction models after hyper-parameter optimization by adopting an integrated learning algorithm to obtain a target meteorological prediction model; and according to the meteorological characteristic data and the target meteorological prediction model, the target meteorological prediction result is determined, and the meteorological prediction precision is improved.

Description

Technical Field

[0001] This application relates to the field of meteorological prediction technologies, and in particular, to a method, device, equipment, storage medium, and computer program product for improving meteorological prediction accuracy. Background Art

[0002] With the rapid development of the new energy industry, the proportion of renewable energy such as wind energy and solar energy in the power system has been continuously increasing. However, new energy power generation is greatly affected by meteorological factors. Although traditional numerical weather prediction (NWP) methods can provide medium- and long-term meteorological predictions, they have limitations in short-term and ultra-short-term prediction accuracy and are difficult to meet the demand for accurate power prediction in new energy power stations. Therefore, how to improve meteorological prediction accuracy has become a technical problem to be solved urgently. Summary of the Invention

[0003] The main purpose of this application is to provide a method, device, equipment, storage medium, and computer program product for improving meteorological prediction accuracy, aiming to solve the technical problem of how to improve meteorological prediction accuracy.

[0004] To achieve the above object, this application provides a method for improving meteorological prediction accuracy, and the method includes the following steps:

[0005] Based on the Transformer structure and combined with the PIDL algorithm, construct multiple initial meteorological prediction models;

[0006] Use the power curve of the wind turbine and the performance characteristic curve of the photovoltaic module as constraint conditions and incorporate them into each initial meteorological prediction model to obtain multiple basic meteorological prediction models;

[0007] Based on a preset automatic hyperparameter tuning tool and combined with the cross-validation algorithm, optimize the hyperparameters of each basic meteorological prediction model;

[0008] Adopt an ensemble learning algorithm to fuse multiple basic meteorological prediction models with optimized hyperparameters to obtain a target meteorological prediction model;

[0009] According to the meteorological feature data and the target meteorological prediction model, determine the target meteorological prediction result.

[0010] In one embodiment, before the step of determining the target meteorological prediction result according to the meteorological feature data and the target meteorological prediction model, it further includes:

[0011] Obtain meteorological data, where the meteorological data includes satellite remote sensing data, ground meteorological station data, historical meteorological data, and new energy power station operation data;

[0012] The meteorological data is preprocessed to obtain the meteorological characteristic data, wherein the preprocessing includes one or more of data cleaning, missing value filling and standardization processing.

[0013] In one embodiment, after the step of using an ensemble learning algorithm to fuse multiple basic meteorological forecast models after hyperparameter optimization to obtain a target meteorological forecast model, the following step is further included:

[0014] The performance of the target meteorological forecast model is evaluated by the mean square error, the root mean square error and the determination coefficient respectively;

[0015] Based on the performance evaluation results, the target meteorological forecast model is optimized in combination with preset feature selection and automatic hyperparameter tuning strategies.

[0016] In one embodiment, after the step of using the wind turbine power curve and the photovoltaic module performance characteristic curve as constraints and integrating them into each initial meteorological forecast model to obtain multiple basic meteorological forecast models, the method further includes:

[0017] Based on the preset feature importance analysis algorithm, the contribution value of each feature to the prediction result is evaluated;

[0018] The features corresponding to the contribution values ​​in the evaluation results that are lower than the preset redundancy threshold are regarded as redundant features, and the redundant features are eliminated;

[0019] The features corresponding to the contribution values ​​in the evaluation results that are higher than the preset importance threshold are taken as important features, and the weights of the important features are strengthened.

[0020] In one embodiment, the step of constructing multiple initial meteorological forecast models based on the Transformer structure and the PIDL algorithm includes:

[0021] Build a model framework based on Transformer;

[0022] Based on the multi-layer self-attention mechanism in the model framework, the spatiotemporal dependencies in historical meteorological data are obtained;

[0023] Based on the PIDL algorithm, meteorological physical laws are used as constraints;

[0024] Based on the spatiotemporal dependency and the constraint conditions, the plurality of initial meteorological forecast models are determined.

[0025] In one embodiment, the step of using an ensemble learning algorithm to fuse multiple basic meteorological forecast models after hyperparameter optimization to obtain a target meteorological forecast model includes:

[0026] Using the integrated learning algorithm, perform weighted averaging on multiple basic meteorological prediction models after hyperparameter optimization;

[0027] According to the weighted average result, the historical meteorological data, and the meteorological feature data, determine a fusion strategy;

[0028] Based on the fusion strategy, fuse multiple basic meteorological prediction models after hyperparameter optimization to obtain the target meteorological prediction model.

[0029] In addition, to achieve the above object, the present application also proposes a device for improving meteorological prediction accuracy, and the device for improving meteorological prediction accuracy includes:

[0030] An initial model module, configured to construct multiple initial meteorological prediction models based on a Transformer structure and in combination with a PIDL algorithm;

[0031] A physical constraint module, configured to use the power curve of a wind turbine and the performance characteristic curve of a photovoltaic module as constraint conditions and incorporate them into each initial meteorological prediction model to obtain multiple basic meteorological prediction models;

[0032] A hyperparameter optimization module, configured to perform hyperparameter optimization on each basic meteorological prediction model based on a preset automatic hyperparameter tuning tool and in combination with a cross-validation algorithm;

[0033] A fusion module, configured to use an integrated learning algorithm to fuse multiple basic meteorological prediction models after hyperparameter optimization to obtain a target meteorological prediction model;

[0034] A target module, configured to determine a target meteorological prediction result according to meteorological feature data and the target meteorological prediction model.

[0035] In addition, to achieve the above object, the present application also proposes a device for improving meteorological prediction accuracy, and the device includes: a memory, a processor, and a meteorological prediction accuracy improvement program stored on the memory and executable on the processor, and the meteorological prediction accuracy improvement program is configured to implement the steps of the meteorological prediction accuracy improvement method as described above.

[0036] In addition, to achieve the above object, the present application also proposes a storage medium, on which a meteorological prediction accuracy improvement program is stored, and when the meteorological prediction accuracy improvement program is executed by a processor, it implements the steps of the meteorological prediction accuracy improvement method as described above.

[0037] In addition, to achieve the above object, the present application also proposes a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the meteorological prediction accuracy improvement method as described above.

[0038] This application is based on the Transformer structure and combines the PIDL algorithm to construct multiple initial meteorological prediction models; uses the power curve of wind turbines and the performance characteristic curve of photovoltaic modules as constraint conditions and incorporates them into each initial meteorological prediction model to obtain multiple basic meteorological prediction models; based on a preset automatic hyperparameter tuning tool and combines the cross-validation algorithm to optimize the hyperparameters of each basic meteorological prediction model; uses the ensemble learning algorithm to fuse multiple basic meteorological prediction models with optimized hyperparameters to obtain the target meteorological prediction model; determines the target meteorological prediction result according to the meteorological feature data and the target meteorological prediction model. This application uses the Transformer structure to improve the feature extraction effect of meteorological data, combines physics-informed deep learning (PIDL) to ensure that the prediction results conform to meteorological physical laws, and at the same time introduces the power curve of wind turbines and the performance characteristic curve of photovoltaic modules as physical constraints to enhance the adaptability and physical consistency of the model for new energy power prediction; uses an automatic hyperparameter tuning tool combined with cross-validation to efficiently optimize model parameters, improve the generalization ability, and avoid overfitting; further uses the ensemble learning algorithm to fuse multiple optimized basic meteorological prediction models, combines the advantages of different models, reduces the instability of single-model prediction, and finally generates high-precision, highly stable and physically realistic meteorological prediction results through the target meteorological prediction model, improving the meteorological prediction accuracy in the new energy scenario. Description of the Drawings

[0039] Figure 1 It is a schematic flowchart of the first embodiment of the method for improving the meteorological prediction accuracy of this application;

[0040] Figure 2 It is a schematic sub-flowchart of the second embodiment of the method for improving the meteorological prediction accuracy of this application;

[0041] Figure 3 It is a schematic sub-flowchart of the third embodiment of the method for improving the meteorological prediction accuracy of this application;

[0042] Figure 4 It is a schematic module structure diagram of the device for improving the meteorological prediction accuracy in the embodiments of this application;

[0043] Figure 5 It is a schematic device structure diagram of the hardware operating environment involved in the method for improving the meteorological prediction accuracy in the embodiments of this application.

[0044] The realization, functional features and advantages of the purpose of this application will be further described with reference to the embodiments and the accompanying drawings. Detailed Embodiments

[0045] It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.

[0046] To better understand the technical solution of this application, the following will be described in detail in conjunction with the accompanying drawings of the specification and specific implementation manners.

[0047] It should be noted that with the rapid development of the new energy industry, the proportion of renewable energy such as wind energy and solar energy in the power system has been continuously increasing. However, new energy power generation is greatly affected by meteorological factors. Although traditional numerical weather prediction (NWP) methods can provide medium- and long-term meteorological forecasts, there are limitations in short-term and ultra-short-term prediction accuracy, making it difficult to meet the demand for accurate power prediction in new energy power stations. Therefore, how to improve the meteorological prediction accuracy has become a technical problem to be solved urgently.

[0048] The main solution of this application is as follows: Based on the Transformer structure and combined with the PIDL algorithm, multiple initial meteorological prediction models are constructed; the power curve of the wind turbine and the performance characteristic curve of the photovoltaic module are used as constraint conditions and incorporated into each initial meteorological prediction model to obtain multiple basic meteorological prediction models; based on a preset automatic hyperparameter tuning tool and combined with the cross-validation algorithm, the hyperparameters of each basic meteorological prediction model are optimized; an ensemble learning algorithm is used to fuse multiple basic meteorological prediction models with optimized hyperparameters to obtain a target meteorological prediction model; according to the meteorological feature data and the target meteorological prediction model, the target meteorological prediction result is determined.

[0049] This application uses the Transformer structure to improve the feature extraction effect of meteorological data, combines physics-informed deep learning (PIDL) to ensure that the prediction results conform to meteorological physical laws, and at the same time introduces the power curve of the wind turbine and the performance characteristic curve of the photovoltaic module as physical constraints to enhance the adaptability and physical consistency of the model for new energy power prediction; uses an automatic hyperparameter tuning tool combined with cross-validation to efficiently optimize model parameters, improve the generalization ability, and avoid overfitting; further uses an ensemble learning algorithm to fuse multiple optimized basic meteorological prediction models, combines the advantages of different models, reduces the instability of single-model prediction, and finally generates a meteorological prediction result with high accuracy, strong stability and conforming to actual physical laws through the target meteorological prediction model, improving the meteorological prediction accuracy in the new energy scenario.

[0050] It should be noted that the execution subject of the method in this embodiment can be a computing service device with data processing, network communication, and program running functions, or the above-mentioned meteorological prediction accuracy improvement device with the same or similar functions. This embodiment and the following embodiments will be described by taking the meteorological prediction accuracy improvement device as an example.

[0051] Based on this, the first embodiment of the method for improving the meteorological prediction accuracy of this application is proposed. Please refer to Figure 1 , Figure 1This is a schematic flowchart of the first embodiment of the method for improving the meteorological prediction accuracy of this application.

[0052] In this embodiment, the method for improving the meteorological prediction accuracy includes the following steps:

[0053] S1: Based on the Transformer structure and combined with the PIDL algorithm, construct multiple initial meteorological prediction models;

[0054] It should be noted that the Transformer structure is a deep learning architecture based on the self-attention mechanism, which was first used for natural language processing (NLP) tasks. Compared with traditional recurrent neural networks (RNNs), the Transformer can process data in parallel, capture long-range dependencies in time-series data, and achieve excellent results in time-series modeling tasks such as meteorological prediction. PIDL (Physics-Informed Deep Learning): PIDL is a method that combines physical laws and deep learning. It uses known physical laws (such as fluid mechanics, thermodynamics, etc.) as constraints to guide the model learning, making its prediction results conform to real physical laws, and improving the interpretability and generalization ability of the model. The initial meteorological prediction model refers to the most preliminary model constructed based on the original meteorological data in the meteorological prediction task. These models use the Transformer structure for modeling and introduce physical constraints by combining the PIDL method to ensure that the prediction results conform to meteorological theories.

[0055] Specifically, use the Transformer structure to establish multiple meteorological prediction models to make full use of its self-attention mechanism to process the time dependence and spatial correlation of meteorological data. Specifically, the input data includes historical meteorological observation data (such as temperature, humidity, wind speed, solar radiation, etc.), numerical weather prediction (NWP) data, and other relevant environmental variables. The data is processed by the encoder, and the multi-head attention mechanism is used to extract the global features of the meteorological data, enabling the model to focus on the key variables at different times and different spatial positions, and improving the understanding ability of time-series changes. The decoder part predicts the meteorological state at future times based on historical data, and introduces time information using positional encoding to ensure that the model can accurately model the time-series characteristics of meteorological variables.

[0056] Furthermore, on the basis of establishing the initial meteorological prediction model, physical-informed deep learning (PIDL) is further incorporated. During the model training process, meteorological physical equations (such as energy conservation and fluid mechanics equations) are used as additional loss terms to ensure that the model follows known physical laws during the optimization process. By combining the data-driven deep learning method with numerical weather prediction (NWP) based on physical equations and using the NWP results to guide the model learning, the prediction results can not only be based on historical data but also take into account physical mechanisms. During the training process, physical rationality screening is performed on the input data to eliminate data points that do not conform to meteorological laws, so as to improve the robustness and generalization ability of the model. Finally, by combining the time series modeling ability of the Transformer structure and the physical constraint ability of PIDL, multiple initial meteorological prediction models are constructed, each with different hyperparameter settings or training strategies, to enhance the diversity and adaptability of the model.

[0057] This step improves the time series modeling ability through the Transformer structure and combines the PIDL physical constraints to enhance physical consistency and generalization ability, enabling the multiple initial meteorological prediction models constructed to have both the strong prediction ability of deep learning and the interpretability of physical modeling, providing a solid foundation for subsequent hyperparameter optimization and ensemble learning, and ultimately achieving more accurate and stable meteorological prediction.

[0058] S2: Use the power curve of the wind turbine and the performance characteristic curve of the photovoltaic module as constraint conditions and incorporate them into each initial meteorological prediction model to obtain multiple basic meteorological prediction models;

[0059] It should be noted that the power curve of the wind turbine describes the relationship between the wind speed and the output power of the wind turbine. Generally, when the wind speed is below the rated wind speed, the output power increases with the increase of the wind speed; when the wind speed exceeds the rated wind speed, the output power tends to be stable; under extreme high wind speed conditions, to protect the unit, the wind turbine may stop, resulting in the power dropping to zero. The performance characteristic curve of the photovoltaic module describes the relationship between the solar radiation intensity, temperature and the power generation power of the photovoltaic module. The output power of the photovoltaic module usually increases with the increase of the solar radiation intensity, but the increase in temperature may reduce its conversion efficiency. Therefore, the combined influence of temperature and radiation needs to be considered in the prediction model. The constraint condition is a rule or function used to limit the model output. In this embodiment, the power curve of the wind turbine and the performance characteristic curve of the photovoltaic module are used as physical constraints to ensure that the prediction results of the model conform to the physical characteristics of new energy equipment and improve the physical consistency of the prediction. The basic meteorological prediction model refers to the model further optimized by combining physical constraints on the basis of the initial meteorological prediction model, enabling the model to not only have high prediction ability of deep learning but also conform to physical mechanisms, improving the interpretability and reliability of the prediction.

[0060] Specifically, extract the power curve of the wind turbine and the performance characteristic curve of the photovoltaic module, and introduce them as physical constraint conditions into the initial meteorological prediction model to ensure that the model prediction results conform to the physical laws of new energy equipment. Based on the actual operation data of the wind turbine, establish the wind speed-power mapping relationship and construct a mathematical function, such as a piecewise function or an interpolation method, to simulate the power output of the wind turbine under different wind speeds. Set the wind speed range to ensure that the power output of the model is close to zero at low wind speeds (e.g., less than 3 m / s), reaches the maximum power at the rated wind speed (e.g., around 12 m / s), and drops to zero due to safety shutdown at extreme wind speeds (e.g., above 25 m / s). Based on the relationship between the solar radiation intensity and the output power of the photovoltaic system, construct a mathematical model, such as the photovoltaic output curve based on standard test conditions (STC), and consider influencing factors such as ambient temperature, incident angle, and photovoltaic panel aging to optimize the model input variables. Set the power output range under different solar radiation intensities so that the model predicts a lower power generation at low light conditions and predicts close to the theoretical maximum power when the light is sufficient, while considering the impact of high temperature environment on the photovoltaic efficiency.

[0061] Furthermore, after using the power curve of the wind turbine and the performance characteristic curve of the photovoltaic module as constraint conditions, optimize the initial meteorological prediction model so that it can not only learn meteorological data but also follow physical constraints to obtain multiple basic meteorological prediction models. Add a physical constraint term to the loss function of the model so that the prediction results numerically approximate the power curve of the wind turbine and the performance curve of the photovoltaic module. For example, when predicting the wind speed, if the corresponding power prediction value is far from the physical curve, the loss function will increase a penalty term to guide the model to adjust the output. Add combined data of different wind speeds and solar radiation intensities to the training dataset, and screen the training samples through the physical model to ensure that the input data conforms to the physical laws of wind power and photovoltaics. Use the Transformer structure to perform time series modeling on meteorological data, and add a physical constraint layer in the decoding stage to make the output results conform to the operating characteristics of wind turbines and photovoltaic modules. Optimize the neural network weights through the backpropagation algorithm, and continuously adjust the prediction output during the training process to make it meet the physical consistency requirements based on data-driven. Finally, multiple meteorological prediction models optimized by physical constraints are trained to be able to accurately predict meteorological variables and conform to the actual operation laws of new energy equipment, forming multiple basic meteorological prediction models.

[0062] This step makes the final obtained basic meteorological prediction model have higher physical consistency, prediction accuracy, generalization ability, and robustness by using the power curve of the wind turbine and the performance characteristic curve of the photovoltaic module as physical constraint conditions and integrating them into the initial meteorological prediction model, effectively improving the reliability of new energy power prediction and laying a foundation for subsequent hyperparameter optimization and model fusion.

[0063] S3: Based on the preset automatic hyperparameter tuning tool and combined with the cross-validation algorithm, the hyperparameters of each basic meteorological forecast model are optimized;

[0064] It should be noted that the AutoML Hyperparameter Tuning Tools refer to tools used to automatically search for optimal hyperparameters, including Optuna, Hyperopt, Bayesian Optimization, Grid Search, etc. Compared with manual parameter adjustment, these tools can efficiently explore the hyperparameter space, find the optimal parameter combination, and improve the performance of the model. Cross-Validation (CV) is a model evaluation method that divides the data set into multiple subsets, which are used alternately as training sets and validation sets to reduce model overfitting and improve generalization ability. Common cross-validation methods include K-Fold Cross-Validation (K-FoldCross-Validation) and Leave-One-Out Cross-Validation (LOOCV).

[0065] Specifically, in the process of hyperparameter optimization, we first need to determine which hyperparameters need to be tuned and how to define the optimization goals. According to the characteristics of the Transformer structure and the PIDL algorithm, select key hyperparameters for optimization, such as: Learning Rate: affects the speed and convergence of the gradient descent; Batch Size: determines the number of samples used in each iterative training; Number of Layers: affects the model complexity and computational cost; Number of Attention Heads: affects the Transformer's ability to process multi-dimensional features; Regularization parameters: such as Dropout rate and L2 regularization term, to prevent model overfitting. Optimization goal: Set the optimization objective function, such as minimizing the mean square error (MSE), root mean square error (RMSE), or maximizing R 2 (coefficient of determination) to ensure that the model has better predictive performance after hyperparameter optimization.

[0066] Furthermore, Optuna, Hyperopt, or Bayesian optimization is used for hyperparameter search to reduce the attempts of invalid hyperparameter combinations and improve the tuning efficiency. Random Search or GridSearch can be selected, but these methods have a high computational cost and are suitable for cases with a small hyperparameter range. K-Fold Cross-Validation is adopted, where the training data is divided into K parts. Each time, K - 1 parts of the data are used to train the model, and the remaining 1 part is used for validation. This is repeated K times, and finally, the average error is calculated to evaluate the generalization ability of the model. During the cross-validation process, the hyperparameter search tool is used to dynamically adjust the parameters, and the hyperparameter combination that performs best in all K rounds of validation is selected. Combining the GPU parallel computing ability, different hyperparameter combinations are tested simultaneously on multiple computing nodes to improve the search efficiency and reduce the tuning time. Finally, through the combination of the automatic hyperparameter tuning tool and the cross-validation algorithm, the optimal hyperparameter combination is found and applied to each basic meteorological prediction model to improve the prediction accuracy and generalization ability of the model.

[0067] This step uses an automatic hyperparameter tuning tool (Optuna, Hyperopt) combined with a cross-validation algorithm (K-Fold Cross-Validation) to efficiently search for the optimal hyperparameter combination, improving the prediction accuracy and generalization ability of the model. It can not only prevent overfitting and improve the adaptability of the model under different meteorological conditions but also optimize the training efficiency, providing high-quality basic meteorological prediction models for subsequent model fusion, thus enhancing the final meteorological prediction effect.

[0068] S4: An ensemble learning algorithm is adopted to fuse multiple basic meteorological prediction models with optimized hyperparameters to obtain the target meteorological prediction model.

[0069] It should be noted that Ensemble Learning is a machine learning method that combines the prediction results of multiple models to improve the stability and accuracy of the overall prediction and reduce the error of a single model. Common ensemble learning methods include Weighted Averaging, Voting, Stacking, and Boosting. The basic meteorological prediction models with optimized hyperparameters refer to multiple high-precision and low-error meteorological prediction models selected through hyperparameter optimization and cross-validation in the previous step (S3). These models perform excellently under different hyperparameter settings and can be used for ensemble learning. The target meteorological prediction model is a high-precision and stable prediction model obtained by fusing multiple optimized basic meteorological prediction models through ensemble learning and is used for actual meteorological prediction tasks.

[0070] Specifically, calculate the prediction results of multiple models, and assign weights according to their prediction performance (such as mean squared error MSE, root mean squared error RMSE), and finally obtain the weighted average prediction result. Gradually train multiple weak models (such as decision trees), so that subsequent models focus on optimizing the error part in the prediction of the previous models, and improve the overall prediction accuracy. Calculate the evaluation metrics (such as RMSE, MSE, R 2 ) of each basic meteorological prediction model, and select the model with the best performance for weighted fusion. Determine the weight assignment method, usually based on the historical performance of the model or use an optimization algorithm (such as Bayesian optimization) to automatically adjust the weights. If the Stacking method is adopted, select a meta-learner with strong generalization ability, such as a neural network, support vector machine (SVM) or random forest, to further optimize the fusion effect.

[0071] Furthermore, use multiple optimized basic meteorological prediction models to predict the same meteorological data, and obtain multiple prediction results as input features. Use cross-validation (K-Fold Cross-Validation) to evaluate the individual performance of each model to determine the best fusion method. Perform weighted calculation on the prediction results of multiple models to obtain the final predicted value; use a meta-model to learn the combination methods of different basic models and optimize the final prediction result. Through historical error analysis, use an error correction model (such as LSTM, Bayesian regression) to fine-tune the fused prediction result to further improve the prediction accuracy. For extreme weather conditions, adopt an outlier detection algorithm to eliminate abnormal predicted values and ensure the stability of the model.

[0072] This step uses an ensemble learning algorithm to fuse multiple basic meteorological prediction models optimized with hyperparameters, and uses methods such as weighted average, voting mechanism, Stacking or Boosting to improve the stability and accuracy of the prediction. Through intelligent fusion strategies, error correction and cross-validation optimization, finally obtain the target meteorological prediction model, ensuring that it has higher prediction accuracy, stronger generalization ability and better stability under different meteorological conditions, and is particularly suitable for new energy power prediction and extreme weather warning tasks.

[0073] S5: Determine the target meteorological prediction result according to the meteorological feature data and the target meteorological prediction model.

[0074] It should be noted that meteorological feature data refers to the input data used for meteorological prediction, including variables such as temperature, humidity, wind speed, wind direction, air pressure, and solar radiation. These data can be sourced from surface meteorological stations, satellite observations, numerical weather prediction (NWP), and historical meteorological data. The target meteorological prediction result refers to the predicted value of future meteorological variables calculated by the target meteorological prediction model, such as temperature, wind speed, precipitation, etc. for a future period of time.

[0075] Specifically, real-time meteorological observation data and numerical weather prediction (NWP) data are collected, and the data is cleaned, denoised, and completed to ensure data quality. Data standardization (such as Z-Score standardization or Min-Max normalization) is performed to make the numerical ranges of meteorological feature data consistent and improve the stability of model calculations. The meteorological feature data after standardization is used as input and transmitted to the target meteorological prediction model. The model uses the self-attention mechanism of the Transformer structure and PIDL physical constraints, combined with historical time series information, to calculate the predicted values of meteorological variables at a future moment.

[0076] Furthermore, the target meteorological prediction model performs forward calculations on the input data and outputs the predicted values of meteorological variables such as temperature, wind speed, and precipitation at a future moment. If a probability prediction method is used, the predicted distribution is output and a confidence interval is provided to improve the ability to quantify the uncertainty of the prediction. Combining historical error analysis, error correction algorithms (such as Bayesian regression, time series residual model LSTM) are used to fine-tune the prediction results to reduce systematic errors; for extreme weather predictions (such as typhoons, heavy rains), outlier detection algorithms are used to eliminate possible abnormal predicted values and improve the robustness of the model. Finally, through the calculation and post-processing of the target meteorological prediction model, the target meteorological prediction results for a future specific time period are obtained, and these results can be used in practical applications such as new energy power prediction and meteorological warnings.

[0077] This step ensures that the prediction results have higher accuracy, stability, and interpretability by inputting meteorological feature data into the target meteorological prediction model and calculating and outputting the target meteorological prediction results. These prediction results can not only optimize new energy power prediction but also be used for meteorological warnings and power grid dispatching, improving the operation efficiency and safety of new energy power stations.

[0078] This embodiment is based on the Transformer structure and combines the PIDL algorithm to construct multiple initial meteorological prediction models. The power curve of the wind turbine and the performance characteristic curve of the photovoltaic module are used as constraint conditions and incorporated into each initial meteorological prediction model to obtain multiple basic meteorological prediction models. Based on a preset automatic hyperparameter tuning tool and combined with the cross-validation algorithm, the hyperparameters of each basic meteorological prediction model are optimized. The ensemble learning algorithm is used to fuse multiple basic meteorological prediction models with optimized hyperparameters to obtain the target meteorological prediction model. According to the meteorological feature data and the target meteorological prediction model, the target meteorological prediction result is determined. This embodiment uses the Transformer structure to improve the feature extraction effect of meteorological data, combines physics-informed deep learning (PIDL) to ensure that the prediction results conform to meteorological physical laws, and at the same time introduces the power curve of the wind turbine and the performance characteristic curve of the photovoltaic module as physical constraints to enhance the adaptability and physical consistency of the model for new energy power prediction. The automatic hyperparameter tuning tool combined with cross-validation is used to efficiently optimize the model parameters, improve the generalization ability, and avoid overfitting. Furthermore, the ensemble learning algorithm is used to fuse multiple optimized basic meteorological prediction models, integrate the advantages of different models, reduce the instability of single-model prediction, and finally generate high-precision, stable and physically realistic meteorological prediction results through the target meteorological prediction model, improving the meteorological prediction accuracy in the new energy scenario.

[0079] Based on the above first embodiment, a second embodiment of the method for improving the meteorological prediction accuracy of this application is proposed. Please refer to Figure 2 , Figure 2 which is a schematic diagram of a sub-process in the second embodiment of the method for improving the meteorological prediction accuracy of this application.

[0080] As Figure 2 shown, in this embodiment, before step S5, it further includes:

[0081] S5a: Obtain meteorological data, where the meteorological data includes satellite remote sensing data, ground meteorological station data, historical meteorological data, and new energy power station operation data;

[0082] S5b: Preprocess the meteorological data to obtain the meteorological feature data, and the preprocessing includes one or more of data cleaning, missing value filling, and normalization processing.

[0083] It should be noted that satellite remote sensing data refers to the atmospheric and surface information obtained through meteorological satellites, including cloud cover, temperature, humidity, precipitation, solar radiation, etc. These data have a global coverage and are applicable to the monitoring and prediction of large-scale weather systems. Ground meteorological station data refers to the measured data collected from ground meteorological observation stations, such as air temperature, wind speed, wind direction, humidity, air pressure, etc. These data have a high spatio-temporal resolution and can reflect local meteorological changes. Historical meteorological data refers to past meteorological observation records, including long-term weather conditions, extreme weather events, etc., which can be used to train deep learning models to improve prediction accuracy. New energy power station operation data refers to the operation data of new energy power generation sites such as wind power and photovoltaic power, including power generation power, wind speed, solar radiation intensity, etc., which is used to construct a new energy power prediction model. Data cleaning (DataCleaning): refers to removing or correcting outliers, duplicate values, and incomplete data in meteorological data to ensure the quality of input data. Missing value imputation refers to using methods such as interpolation, mean filling, and KNN filling to supplement the missing parts in the dataset to reduce information loss. Standardization refers to converting the data into a standard normal distribution with a mean of 0 and a variance of 1 or scaling the data to a specific range (such as [0,1]) to improve the training stability and convergence speed of the model.

[0084] Specifically, global meteorological data such as atmospheric temperature, humidity, cloud distribution, and solar radiation are obtained through a satellite remote sensing system; local meteorological data such as wind speed, air temperature, humidity, and air pressure are collected through ground meteorological stations; historical meteorological data is obtained through meteorological databases (such as NOAA, ECMWF, and domestic meteorological centers) for deep learning model training; the operation data of wind power and photovoltaic equipment are obtained through a new energy power station monitoring system to analyze the relationship between meteorological conditions and new energy power generation. The data formats from different sources may not be consistent. For example, satellite remote sensing data is usually raster data (GeoTIFF), ground meteorological station data is time series data (CSV, JSON), and historical meteorological data may be in tabular format (NetCDF, HDF5). A data conversion tool (such as Pandas, Xarray) is used to unify the data format to ensure that the data can be used for model training.

[0085] Furthermore, detect and remove outliers, such as when the wind speed exceeds the reasonable range (>100 m / s) or the temperature data shows extreme anomalies (>60°C or < -80°C). Process duplicate data to ensure the consistency of time series data and avoid model bias caused by duplicate data. For time series data, use interpolation methods (linear interpolation, spline interpolation) to fill short-term missing values and ensure temporal continuity; for randomly missing data points, use mean filling, KNN filling, or deep learning interpolation methods (such as Autoencoder) to fill them. Use Z-score normalization (i.e., mean of 0 and standard deviation of 1) to ensure that data with different dimensions can be calculated on the same scale, which is applicable to physical variables such as wind speed, air pressure, and temperature. Use Min-Max normalization (scaled to [0, 1]), which is applicable to neural network models to accelerate convergence speed and improve training stability. Finally, the meteorological feature data after data cleaning, missing value filling, and normalization processing will be used as model input to improve the training quality and prediction accuracy of the model.

[0086] This step ensures the high quality and consistency of the input data through multi-source data collection and data preprocessing (data cleaning, missing value filling, normalization), provides more reliable training data for the meteorological prediction model, improves prediction accuracy and stability, and at the same time optimizes the new energy power prediction and improves the operation efficiency of the power system.

[0087] Based on the above first embodiment, in this embodiment, after step S4, it further includes:

[0088] S4a: Evaluate the performance of the target meteorological prediction model through mean squared error, root mean squared error, and coefficient of determination respectively;

[0089] S4b: Based on the performance evaluation results, optimize the target meteorological prediction model in combination with the preset feature selection and automatic hyperparameter tuning strategy.

[0090] It should be noted that the mean squared error (MSE) is used to measure the average error between the predicted value and the true value. The smaller the error, the better the model prediction effect. The root mean squared error (RMSE) is the square root of MSE, which can provide an error measure consistent with the unit of meteorological data and make the prediction error more intuitive. The coefficient of determination (R 2 ) is used to measure the fitting ability of the model, and its value range is between 0 and 1. The closer it is to 1, the better the model prediction effect. Feature selection refers to selecting variables that have a greater impact on the prediction results from the dataset to reduce unnecessary inputs and improve the calculation efficiency and accuracy of the model. Automatic hyperparameter tuning is to use automatic search algorithms (such as Optuna, Hyperopt) to adjust the key parameters of the model to improve model performance and reduce the workload of manual parameter tuning.

[0091] Specifically, the accuracy of model prediction is evaluated by calculating the error between the predicted value and the true value of the model. As the square root of MSE, RMSE makes the error unit consistent with the original meteorological data, facilitating interpretation and comparison. The coefficient of determination (R 2 ) is used to measure the fitting degree of the model to meteorological variables. The closer R 2 is to 1, the stronger the model's prediction ability. If R 2 is too low, it indicates that the model is underfitting and needs further optimization. If MSE and RMSE are too large, it means that the model error is high, and it may be necessary to optimize the input features or adjust the model parameters. If R 2 is lower than expected, it shows that the model performs poorly under complex weather conditions and needs to optimize the model structure.

[0092] Furthermore, feature importance analysis (such as SHAP, LIME) is adopted to find out the meteorological variables that have the greatest impact on the prediction, and redundant variables are removed to improve the calculation efficiency. If some variables contribute little to the prediction result, they can be considered for removal to reduce the model complexity. An automatic search tool (such as Optuna, Hyperopt) is used to optimize the model parameters, such as the learning rate, batch size, number of neural network layers, etc. Combining with the Bayesian optimization method, the parameter combination is intelligently adjusted to improve the model training efficiency and prediction accuracy. If the error is still large, the Transformer structure can be adjusted, such as increasing or decreasing the number of layers, adjusting the attention mechanism, etc., to improve the model's expressive ability. Regularization methods (such as Dropout) are used to prevent overfitting and improve the generalization ability of the model. The optimized model is used to retrain, and MSE, RMSE, and R 2 are calculated again to evaluate whether the optimization is effective. If the model metrics meet the expected standards, the optimization plan is determined to be effective; otherwise, the features and parameters are continuously adjusted.

[0093] This step evaluates the model performance through MSE, RMSE, and R 2 , combines feature selection and automatic hyperparameter tuning to optimize the target meteorological prediction model, improve the prediction accuracy, reduce the consumption of computing resources, and at the same time enhance the reliability and application value of new energy power prediction.

[0094] This embodiment is based on the Transformer structure and combines the PIDL algorithm to construct multiple initial meteorological prediction models. The power curve of the wind turbine and the performance characteristic curve of the photovoltaic module are used as constraint conditions and incorporated into each initial meteorological prediction model to obtain multiple basic meteorological prediction models. Based on a preset automatic hyperparameter tuning tool and combined with the cross-validation algorithm, the hyperparameters of each basic meteorological prediction model are optimized. The integrated learning algorithm is used to fuse the multiple basic meteorological prediction models with optimized hyperparameters to obtain the target meteorological prediction model. According to the meteorological feature data and the target meteorological prediction model, the target meteorological prediction result is determined. This embodiment uses the Transformer structure to improve the feature extraction effect of meteorological data, combines physics-informed deep learning (PIDL) to ensure that the prediction results conform to meteorological physical laws, and at the same time introduces the power curve of the wind turbine and the performance characteristic curve of the photovoltaic module as physical constraints to enhance the adaptability and physical consistency of the model for new energy power prediction. The automatic hyperparameter tuning tool combined with cross-validation is used to efficiently optimize the model parameters, improve the generalization ability, and avoid overfitting. Further, the integrated learning algorithm is used to fuse multiple optimized basic meteorological prediction models, synthesize the advantages of different models, reduce the instability of single-model prediction, and finally generate high-precision, stable and physically realistic meteorological prediction results through the target meteorological prediction model, improving the meteorological prediction accuracy in the new energy scenario.

[0095] Based on the above second embodiment, a third embodiment of the method for improving the meteorological prediction accuracy of this application is proposed. Please refer to Figure 3 , Figure 3 which is a schematic diagram of a sub-process in the third embodiment of the method for improving the meteorological prediction accuracy of this application.

[0096] In this embodiment, after step S2, it further includes:

[0097] S2a: Based on a preset feature importance analysis algorithm, evaluate the contribution value of each feature to the prediction result;

[0098] S2b: Take the features corresponding to the contribution values lower than the preset redundancy threshold in the evaluation result as redundant features and remove the redundant features;

[0099] S2c: Take the features corresponding to the contribution values higher than the preset importance threshold in the evaluation result as important features and strengthen the weights of the important features.

[0100] It should be noted that the feature importance analysis algorithm is an algorithm used to evaluate the contribution degree of model input features to the prediction result. Common methods include SHAP (Shapley Additive Explanations), LIME (Local Interpretable Model-Agnostic Explanations), and feature importance analysis based on decision trees. The contribution value measures the influence degree of a single feature on the model prediction result. The higher the contribution value, the more important the feature is in the prediction process. Redundant features are features that have little impact on the prediction result, that is, features with contribution values lower than the preset redundancy threshold. These features may contain noise or be highly correlated with other features. Retaining them may increase the model calculation complexity without improving the prediction accuracy. Important features are features with contribution values higher than the preset importance threshold, indicating that they play a key role in the model prediction. Optimizing the weights of these features can enhance the prediction ability of the model and improve the final prediction accuracy. Weight strengthening refers to giving higher attention to important features during the model training process, such as enhancing their influence in the loss function or attention mechanism to ensure that the model learns key features more fully.

[0101] Specifically, use the SHAP method to calculate the contribution value of each feature to the prediction output and draw the feature influence ranking. Or use LIME to determine the importance of features by locally perturbing the input features and observing the changes in model predictions; it is also possible to use feature importance analysis based on decision trees, such as XGBoost and random forests, to calculate the frequency of each feature being used in the decision-making process. Input meteorological feature data, run the target meteorological prediction model, and calculate the average contribution value of each feature in multiple prediction tasks. Set the preset redundancy threshold and preset importance threshold, sort the contribution values, and compare them with the thresholds.

[0102] Furthermore, identify features with contribution values lower than the preset redundancy threshold, and judge whether they are highly correlated with other features or have little impact on the prediction. Delete these features from the input dataset to reduce the data dimension, improve the model calculation efficiency, and at the same time reduce the risk of overfitting. Identify features with contribution values higher than the preset importance threshold and give them higher attention during the training process. For key features, increase their influence so that the model pays more attention to learning these features. Optimize the attention mechanism (such as Transformer) to increase the weights of key features to ensure that important features obtain more computing resources during the deep learning process. For important features, perform data augmentation, such as adding feature interaction terms, to improve the model's learning ability for this feature. Finally, the optimized feature set retains the features most important for prediction, improving the model's training efficiency and prediction accuracy.

[0103] This step evaluates the contribution values of each feature to the prediction result through feature importance analysis (such as SHAP, LIME, etc.), eliminates redundant features with low contribution, and enhances the weights of important features, enabling the model to improve the prediction accuracy while ensuring computational efficiency, and providing a more accurate decision-making basis for new energy power prediction and meteorological early warning.

[0104] Based on the above second embodiment, in this embodiment, step S1 includes:

[0105] S11: Based on Transformer, construct a model framework;

[0106] S12: Based on the multi-head self-attention mechanism in the model framework, obtain the spatio-temporal dependencies in historical meteorological data;

[0107] S13: Based on the PIDL algorithm, use meteorological physical laws as constraint conditions;

[0108] S14: Based on the spatio-temporal dependencies and the constraint conditions, determine the multiple initial meteorological prediction models.

[0109] It should be noted that the model framework refers to the overall architecture of the meteorological prediction model designed based on the Transformer structure, including the input layer, the encoding layer (Encoder), the decoding layer (Decoder), and the output layer. The multi-head self-attention mechanism is the core mechanism of the Transformer structure, which can focus on key features in meteorological data at different scales and extract the spatio-temporal dependencies of historical meteorological data. Spatio-temporal dependencies refer to the correlations of meteorological variables in time (past, present, future) and space (different meteorological stations, geographical regions), which are important features of the evolution of weather systems.

[0110] Specifically, adopt the Transformer structure, design the input layer, the encoding layer (Encoder), the decoding layer (Decoder), and the output layer to construct a deep learning model for meteorological prediction. Use the multi-head self-attention mechanism (Multi-Head Self-Attention) in the encoding layer to enhance the model's ability to model the spatio-temporal relationships of meteorological data. Use positional encoding (Positional Encoding) in the decoding layer to ensure that the time information of time series data will not be lost and improve the model's ability to capture the future weather change trend. Collect and preprocess multi-source meteorological data (such as historical weather data, satellite remote sensing data, ground observation data), and convert it into a time series data format suitable for the Transformer structure. Adopt the sliding window method (Sliding Window) to process time series data to ensure that the model can learn long-term and short-term meteorological patterns.

[0111] Furthermore, in the Transformer Encoder part, multiple Multi-Head Attention mechanisms are used to process historical meteorological data, analyzing the correlations between different time steps. Through the Attention Weights mechanism, it is identified which historical time points of meteorological data are most important for the current prediction, improving the model's understanding ability of weather evolution trends. Through the attention mechanism, meteorological station data in different geographical regions are analyzed to learn the regional climate change laws. Combining geographical information such as altitude, latitude, and longitude ensures that the model can accurately model at different spatial scales.

[0112] Furthermore, combined with Physics-Informed Deep Learning (PIDL), meteorological dynamic equations such as energy conservation, mass conservation, and hydrodynamics are introduced as physical constraints for the model. A Physics-based Loss Function is adopted to ensure that the prediction results conform to meteorological principles and reduce physically unreasonable prediction values during the model optimization process. In the Transformer structure, physical constraint information is fused so that the model can not only learn meteorological data patterns but also follow meteorological laws. Through Physics-Regularized Training, the risk of overfitting of the neural network model is reduced and the generalization ability is improved. Multiple initial models are trained by combining different input feature combinations (such as different time window lengths, different meteorological variables). The hyperparameters of the Transformer structure (such as the number of attention heads, the number of layers, and the number of hidden units) are adjusted to obtain multiple initial meteorological prediction models with different architectures. Metrics such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Coefficient of Determination (R 2 ) are used to evaluate the prediction performance of each initial model. The multiple initial meteorological prediction models with the best performance are selected to lay the foundation for subsequent hyperparameter optimization and model fusion.

[0113] This step constructs a meteorological prediction model framework based on the Transformer structure, combines multi-layer self-attention mechanisms to extract spatio-temporal dependencies, and introduces PIDL physical constraints, finally generating multiple initial meteorological prediction models. This not only improves the accuracy of meteorological prediction but also ensures the physical rationality of the prediction results, providing high-quality basic models for subsequent optimization and ensemble learning.

[0114] Based on the above second embodiment, in this embodiment, step S4 includes:

[0115] S41: Using the ensemble learning algorithm, perform weighted averaging on the multiple basic meteorological prediction models after hyperparameter optimization;

[0116] S42: Determine a fusion strategy based on the weighted average result, the historical meteorological data, and the meteorological feature data;

[0117] S43: Based on the fusion strategy, fuse multiple basic meteorological prediction models with optimized hyperparameters to obtain the target meteorological prediction model.

[0118] It should be noted that weighted averaging is an ensemble method that assigns different weights according to the prediction performance of each model to calculate the final prediction result. The weights are usually determined based on the error magnitude of the model (such as RMSE, MSE). The fusion strategy refers to the strategy of formulating how to combine the prediction results of multiple basic models during the ensemble learning process to improve the final prediction accuracy. The fusion strategy can include methods such as dynamic weight adjustment, error correction, and probability-weighted fusion.

[0119] Specifically, evaluate the prediction performance of multiple basic meteorological prediction models with optimized hyperparameters through mean squared error (MSE), root mean squared error (RMSE), and coefficient of determination (R 2 )). Models with smaller errors are given higher weights, and models with larger errors have lower weights. Use the weighted average method to calculate the prediction outputs of multiple basic models. Calculate the predicted values of each model and perform weighted summation according to the weights to obtain a preliminary ensemble prediction result. Evaluate the fusion effect of the weighted average through the cross-validation method (such as K-fold cross-validation) and adjust the weights to optimize the final prediction accuracy.

[0120] Furthermore, combine the weighted average result, historical meteorological data, and meteorological feature data to construct a fusion strategy to ensure that the prediction model can maintain high accuracy under different meteorological conditions. Use the dynamic weight adjustment method to dynamically adjust the fusion method of the model according to different weather conditions or seasonal changes. For example, during a period when wind speed prediction is more important, increase the weight of the wind speed prediction model. Through historical error analysis, establish an error correction model (such as LSTM, Bayesian regression) to further optimize the fusion strategy. If there is a large deviation, use error-weighted correction (such as Kalman filtering or residual correction) to improve the prediction stability.

[0121] Furthermore, use the fusion strategy to combine multiple weighted-averaged basic meteorological prediction models to generate the final target meteorological prediction model. Ensure that the prediction results after the final model fusion meet the optimization objectives in all evaluation metrics (MSE, RMSE, R 2 ). Run the target meteorological prediction model on the test dataset and compare it with the real meteorological data to ensure that the prediction error is within an acceptable range. If the prediction accuracy does not meet the expectations, adjust the fusion strategy or optimize the weights of the basic models and retrain and fuse them.

[0122] This step performs weighted average fusion through an ensemble learning algorithm, combines historical meteorological data and meteorological feature data, formulates a fusion strategy, and finally generates a target meteorological prediction model, improving the stability and accuracy of meteorological prediction and optimizing new energy power prediction.

[0123] In this embodiment, based on the Transformer structure and combined with the PIDL algorithm, multiple initial meteorological prediction models are constructed; the power curve of the wind turbine and the performance characteristic curve of the photovoltaic module are used as constraint conditions and incorporated into each initial meteorological prediction model to obtain multiple basic meteorological prediction models; based on a preset automatic hyperparameter tuning tool and combined with the cross-validation algorithm, the hyperparameters of each basic meteorological prediction model are optimized; an ensemble learning algorithm is used to fuse the multiple basic meteorological prediction models with optimized hyperparameters to obtain a target meteorological prediction model; according to the meteorological feature data and the target meteorological prediction model, the target meteorological prediction result is determined. In this embodiment, the Transformer structure is used to improve the feature extraction effect of meteorological data, and the physical-guided deep learning (PIDL) is combined to ensure that the prediction result conforms to the meteorological physical laws. At the same time, the power curve of the wind turbine and the performance characteristic curve of the photovoltaic module are introduced as physical constraints to enhance the adaptability and physical consistency of the model for new energy power prediction; an automatic hyperparameter tuning tool is combined with cross-validation to efficiently optimize the model parameters, improve the generalization ability, and avoid overfitting; further, an ensemble learning algorithm is used to fuse multiple optimized basic meteorological prediction models, integrate the advantages of different models, reduce the instability of single-model prediction, and finally generate a meteorological prediction result with high accuracy, strong stability and conforming to the actual physical laws through the target meteorological prediction model, improving the meteorological prediction accuracy in the new energy scenario.

[0124] The embodiment of the present application also provides a device for improving meteorological prediction accuracy. Please refer to Figure 4 , Figure 4 which is a schematic diagram of the module structure of the device for improving meteorological prediction accuracy in the embodiment of the present application. The device for improving meteorological prediction accuracy includes:

[0125] An initial model module 401, configured to construct multiple initial meteorological prediction models based on the Transformer structure and combined with the PIDL algorithm;

[0126] A physical constraint module 402, configured to use the power curve of the wind turbine and the performance characteristic curve of the photovoltaic module as constraint conditions and incorporate them into each initial meteorological prediction model to obtain multiple basic meteorological prediction models;

[0127] A hyperparameter optimization module 403, configured to optimize the hyperparameters of each basic meteorological prediction model based on a preset automatic hyperparameter tuning tool and combined with the cross-validation algorithm;

[0128] A fusion module 404, configured to fuse multiple basic meteorological prediction models after hyperparameter optimization by using an ensemble learning algorithm to obtain a target meteorological prediction model;

[0129] A target module 405, configured to determine a target meteorological prediction result according to meteorological feature data and the target meteorological prediction model.

[0130] The meteorological prediction accuracy improvement device provided by the embodiments of the present application adopts the meteorological prediction accuracy improvement method in the above embodiments, and can solve the technical problem of how to improve the meteorological prediction accuracy. Compared with the prior art, the beneficial effects of the meteorological prediction accuracy improvement device provided by the embodiments of the present application are the same as those of the meteorological prediction accuracy improvement method provided by the above embodiments, and other technical features in the meteorological prediction accuracy improvement device are the same as the features disclosed in the above embodiment method, which will not be elaborated here.

[0131] The present application provides a meteorological prediction accuracy improvement device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the meteorological prediction accuracy improvement method in the above embodiments.

[0132] Next, refer to Figure 5 , Figure 5 , which is a schematic structural diagram of a device for the hardware operating environment involved in the meteorological prediction accuracy improvement method in the embodiments of the present application, and shows a schematic structural diagram of a device suitable for implementing the meteorological prediction accuracy improvement device in the embodiments of the present application. Figure 5 The shown meteorological prediction accuracy improvement device is only an example, and should not bring any limitation to the functions and usage scopes of the embodiments of the present application.

[0133] Such as Figure 5As shown, the weather prediction accuracy improvement device may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM: Read Only Memory) 1002 or the program loaded from the storage device 1003 into the random access memory (RAM: Random Access Memory) 1004. In the RAM 1004, various programs and data required for the operation of the weather prediction accuracy improvement device are also stored. The processing device 1001, the ROM 1002, and the RAM 1004 are connected to each other through a bus 1005. The input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the weather prediction accuracy improvement device to communicate with other devices wirelessly or wiredly to exchange data. Although the weather prediction accuracy improvement device with various systems is shown in the figure, it should be understood that it is not required to implement or have all the shown systems. More or fewer systems can be alternatively implemented or had.

[0134] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program codes for executing the methods shown in the flowcharts. When the computer program is executed by the processing device 1001, the above functions defined in the methods of the embodiments disclosed in the present application are executed.

[0135] The weather prediction accuracy improvement device provided by the present application adopts the weather prediction accuracy improvement method in the above embodiments and can solve the technical problem of how to improve the weather prediction accuracy. Compared with the prior art, the beneficial effects of the weather prediction accuracy improvement device provided by the present application are the same as those of the weather prediction accuracy improvement method provided by the above embodiments, and other technical features in the weather prediction accuracy improvement device are the same as the features disclosed in the method of the previous embodiment, and will not be elaborated here.

[0136] It should be understood that the various parts disclosed in the present application can be implemented by hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in a suitable manner in any one or more embodiments or examples.

[0137] As described above, this is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application. Therefore, the protection scope of this application shall be subject to the protection scope of the claimed rights.

[0138] This application provides a computer-readable storage medium with computer-readable program instructions (i.e., computer programs) stored thereon, and the computer-readable program instructions are used to execute the method for improving meteorological prediction accuracy in the above embodiments.

[0139] The above computer-readable storage medium carries one or more programs. When the above one or more programs are executed by a device for improving meteorological prediction accuracy, the device for improving meteorological prediction accuracy is caused to: based on the Transformer structure, combine with the PIDL algorithm to construct multiple initial meteorological prediction models; use the power curve of the wind turbine and the performance characteristic curve of the photovoltaic module as constraint conditions and incorporate them into each initial meteorological prediction model to obtain multiple basic meteorological prediction models; based on a preset automatic hyperparameter tuning tool, combine with the cross-validation algorithm to optimize the hyperparameters of each basic meteorological prediction model; use the ensemble learning algorithm to fuse the multiple basic meteorological prediction models with optimized hyperparameters to obtain a target meteorological prediction model; and determine the target meteorological prediction result according to the meteorological feature data and the target meteorological prediction model. Computer program code for performing the operations of this application can be written in one or more programming languages or combinations thereof. The above programming languages include object-oriented programming languages such as Java, Smalltalk, and C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, by using an Internet service provider to connect through the Internet).

[0140] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a part of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, as well as combinations of blocks in the block diagram and / or flowchart, may be implemented by a dedicated hardware-based system that performs the specified functions or operations, or may be implemented by a combination of dedicated hardware and computer instructions.

[0141] The modules described in the embodiments of the present application can be implemented in software or in hardware. In some cases, the name of the module does not constitute a limitation on the unit itself.

[0142] The readable storage medium provided by the present application is a computer-readable storage medium that stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned method for improving meteorological prediction accuracy, and can solve the technical problem of how to improve meteorological prediction accuracy. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as those of the method for improving meteorological prediction accuracy provided by the above embodiments, and will not be elaborated here.

[0143] An embodiment of the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the steps of the method for improving meteorological prediction accuracy as described above.

[0144] The computer program product provided by the present application can solve the technical problem of how to improve meteorological prediction accuracy. Compared with the prior art, the beneficial effects of the computer program product provided by the embodiments of the present application are the same as those of the method for improving meteorological prediction accuracy provided by the above embodiments, and will not be elaborated here.

[0145] The above are only the preferred embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be included in the patent scope of the present application by the same token.

Claims

1. A method for improving weather forecast accuracy, characterized in that: The method comprises: Based on the Transformer structure and combined with the PIDL algorithm, multiple initial meteorological forecast models are constructed; The wind turbine power curve and photovoltaic module performance characteristic curve are used as constraints and integrated into each initial meteorological forecast model to obtain multiple basic meteorological forecast models; Based on the preset automatic hyperparameter tuning tool and combined with the cross-validation algorithm, the hyperparameters of each basic meteorological forecast model are optimized; An integrated learning algorithm is used to fuse multiple basic meteorological forecast models after hyperparameter optimization to obtain the target meteorological forecast model; A target meteorological forecast result is determined based on the meteorological characteristic data and the target meteorological forecast model.

2. The method according to claim 1, characterized in that Before the step of determining the target meteorological forecast result according to the meteorological characteristic data and the target meteorological forecast model, the method further includes: Acquiring meteorological data, including satellite remote sensing data, ground meteorological station data, historical meteorological data, and new energy station operation data; The meteorological data is preprocessed to obtain the meteorological characteristic data, wherein the preprocessing includes one or more of data cleaning, missing value filling and standardization processing.

3. The method according to claim 1, characterized in that After the step of adopting an ensemble learning algorithm to fuse multiple basic meteorological forecast models after hyperparameter optimization to obtain a target meteorological forecast model, the method further includes: The performance of the target meteorological forecast model is evaluated by the mean square error, the root mean square error and the determination coefficient respectively; Based on the performance evaluation results, the target meteorological forecast model is optimized in combination with preset feature selection and automatic hyperparameter tuning strategies.

4. The method according to claim 1, characterized in that After the step of taking the wind turbine power curve and the photovoltaic module performance characteristic curve as constraint conditions and integrating them into each initial meteorological forecast model to obtain multiple basic meteorological forecast models, the method further includes: Based on the preset feature importance analysis algorithm, the contribution value of each feature to the prediction result is evaluated; The features corresponding to the contribution values ​​in the evaluation results that are lower than the preset redundancy threshold are regarded as redundant features, and the redundant features are eliminated; The features corresponding to the contribution values ​​in the evaluation results that are higher than the preset importance threshold are taken as important features, and the weights of the important features are strengthened.

5. The method according to claim 1, characterized in that The steps of constructing multiple initial meteorological forecast models based on the Transformer structure and the PIDL algorithm include: Build a model framework based on Transformer; Based on the multi-layer self-attention mechanism in the model framework, the spatiotemporal dependencies in historical meteorological data are obtained; Based on the PIDL algorithm, meteorological physical laws are used as constraints; Based on the spatiotemporal dependency and the constraint conditions, the plurality of initial meteorological forecast models are determined.

6. The method according to claim 5, characterized in that The step of using an integrated learning algorithm to fuse multiple basic meteorological prediction models after hyperparameter optimization to obtain a target meteorological prediction model includes: Using the ensemble learning algorithm, weighted averaging the multiple basic meteorological forecast models after the hyperparameters are optimized; Determining a fusion strategy according to the weighted average result, the historical meteorological data, and the meteorological characteristic data; Based on the fusion strategy, the multiple basic meteorological prediction models after the hyperparameter optimization are fused to obtain the target meteorological prediction model.

7. A device for improving weather forecast accuracy, characterized in that: The device comprises: The initial model module is used to build multiple initial meteorological forecast models based on the Transformer structure and the PIDL algorithm; The physical constraint module is used to use the wind turbine power curve and the photovoltaic module performance characteristic curve as constraint conditions and integrate them into various initial meteorological forecast models to obtain multiple basic meteorological forecast models; Hyperparameter optimization module, which is used to optimize the hyperparameters of various basic meteorological forecast models based on the preset automatic hyperparameter tuning tool and the cross-validation algorithm; A fusion module is used to fuse multiple basic meteorological prediction models after hyperparameter optimization using an integrated learning algorithm to obtain a target meteorological prediction model; The target module is used to determine the target meteorological forecast result according to the meteorological characteristic data and the target meteorological forecast model.

8. A computer device, characterized in that: The device comprises: a memory, a processor, and a weather forecast accuracy improvement program stored in the memory and executable on the processor, wherein the weather forecast accuracy improvement program is configured to implement the steps of the weather forecast accuracy improvement method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium stores a weather forecast accuracy improvement program, and when the weather forecast accuracy improvement program is executed by the processor, the steps of the weather forecast accuracy improvement method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the method for improving the accuracy of meteorological forecasting according to any one of claims 1 to 6 are implemented.

Images