A weather-based deduction method and system

Abstract

The present application relates to a meteorological-based deduction method, which includes the following steps: obtaining historical meteorological information and meteorological impact information corresponding to the target area, extracting meteorological features and impact features and packaging them into a first training feature set; training a meteorological deduction model using the first training feature set; setting initial conditions and boundary conditions for the deduction model, obtaining current meteorological information and meteorological impact information and inputting them into the meteorological deduction model for preliminary deduction; when deduction information is received, inputting it into a deviation correction model for deviation correction; when corrected deduction information is received, presenting it in a visual manner. The present application constructs a more accurate meteorological deduction model by comprehensively utilizing historical meteorological information and meteorological impact information, extracting key features for training, and at the same time, by introducing a deviation correction model to perform refined correction on the deduction results, it has the effect of significantly improving the accuracy and practicality of meteorological deduction.

Description

Technical Field

[0001] The present application relates to the technical field of meteorological deduction, and in particular to a meteorological-based deduction method and system. Background Art

[0002] Meteorological simulation is a technology used to predict and simulate future weather conditions. It is widely used in fields such as weather forecasting, climate research, agricultural production, aviation and navigation, and urban planning. Through meteorological simulation, people can better understand the patterns of meteorological changes, provide a scientific basis for decision-making, reduce losses caused by natural disasters, and improve social and economic benefits.

[0003] At present, there are still some problems that need to be solved in traditional meteorological deduction technology:

[0004] First, traditional meteorological deduction methods often only focus on past historical meteorological data itself, and tend to ignore factors that have indirect but important impacts on the weather, such as topography, vegetation cover, water distribution, human activities, etc., which can easily lead to inaccurate deduction results or even misleading conclusions; at the same time, in the meteorological deduction process, due to the limitations of the model itself, the incompleteness of the data and the complexity of meteorological phenomena, the deduction results are prone to certain deviations, and most existing meteorological deduction methods are based on simple statistical models and lack the ability to accurately correct the deviations of the deduction results. The existence of deduction deviations not only affects the accuracy of the deduction results, but also limits the reliability and practicality of meteorological deduction technology in practical applications. Summary of the Invention

[0005] In order to improve the accuracy and practicality of meteorological deduction, the present application provides a meteorological-based deduction method and system.

[0006] The above-mentioned invention objective of this application is achieved through the following technical solutions:

[0007] A weather-based deduction method comprises the following steps:

[0008] Obtain historical meteorological information and meteorological impact information corresponding to the target area;

[0009] Extracting meteorological features and impact features from historical meteorological information and meteorological impact information respectively, and packaging the extracted meteorological features and impact features into a first training feature set;

[0010] Training a pre-built meteorological deduction model using a first training feature set;

[0011] Set the initial conditions and boundary conditions of the deduction model, obtain current meteorological information and meteorological impact information and input them into the meteorological deduction model for preliminary deduction;

[0012] When receiving the deduction information output by the deduction model, the deduction information is input into the pre-trained bias correction model for bias correction;

[0013] When the correction deduction information output by the bias correction model is received, it is presented in a visual manner.

[0014] By adopting the above technical solution, historical meteorological data information and information on factors affecting meteorological changes in the target area are collected, and key meteorological features and impact features are extracted from the collected historical meteorological information and meteorological impact information and packaged into a first training feature set for training a pre-built meteorological deduction model; according to actual needs, initial conditions and boundary conditions are set for the meteorological deduction model, and the latest current meteorological information and meteorological impact information are obtained, and the current meteorological information and meteorological impact information are input into the meteorological deduction model for preliminary meteorological deduction. Since various uncertainties may exist in the meteorological deduction process, resulting in a certain deviation between the deduction results and the actual meteorological conditions, a pre-trained bias correction model is introduced to perform refined bias correction on the deduction information output by the deduction model to improve the accuracy of the prediction; finally, the bias-corrected deduction information is visualized in intuitive forms such as charts, maps, and animations; this application constructs a more accurate meteorological deduction model by comprehensively utilizing historical meteorological information and meteorological impact information, extracting key features for training, and at the same time, by introducing the bias correction model to perform refined correction on the deduction results, it has the effect of significantly improving the accuracy and practicality of meteorological deduction.

[0015] In a preferred example, the present application may be further configured as follows: the step of extracting meteorological features and impact features from historical meteorological information and meteorological impact information, respectively, and packaging the extracted meteorological features and impact features into a first training feature set, includes the following steps:

[0016] Extracting element features from historical meteorological information and performing trend analysis and interpolation analysis on the element features to obtain meteorological features, wherein the meteorological features include time trend features and spatial distribution features;

[0017] Identifying direct impact data, indirect impact data, and lagged effect data from meteorological impact information, and extracting impact features, wherein the impact features include direct impact features, indirect impact features, and lagged impact features;

[0018] Combining the extracted meteorological features and the corresponding associated influencing features into a plurality of training feature vectors, wherein the training feature vectors include the meteorological features corresponding to the time point and spatial position and the corresponding influencing features;

[0019] Several training feature vectors are packaged into a first training feature set.

[0020] By adopting the above technical solution, element features are extracted from historical meteorological information. After extracting these element features, trend analysis is performed on them to reveal the temporal changes of meteorological elements. At the same time, interpolation analysis is performed to supplement and improve the spatial distribution information of meteorological elements. Direct influence data, indirect influence data, and lagged effect data are identified from meteorological influence information. Direct influence data include factors such as topography and vegetation cover that directly affect the weather. Indirect influence data include factors such as human activities and economic development that indirectly affect the weather through certain channels. Lagged effect data considers possible delayed effects in meteorological changes, such as the impact of early precipitation on later soil moisture. After identifying direct influence data, indirect influence data, and lagged effect data, their influence characteristics on the weather are extracted, including direct influence characteristics, indirect influence characteristics, and lagged influence characteristics. The extracted meteorological features and corresponding associated influence characteristics are combined into a number of training feature vectors. Each training feature vector contains the meteorological features and their corresponding influence characteristics at the corresponding time point and spatial location. This combination method ensures that the training feature vectors can fully reflect the complex relationship between meteorological changes and their influencing factors. The combined several training feature vectors are then packaged into a first training feature set.

[0021] In a preferred example, the present application can be further configured as follows: the steps of setting the initial conditions and boundary conditions of the deduction model, obtaining current meteorological information and meteorological impact information, and inputting them into the meteorological deduction model for preliminary deduction include the following steps:

[0022] Obtain deduction demand information and determine the initial conditions of the deduction model based on historical meteorological information and deduction demand information;

[0023] The boundary conditions of the deduction model are determined based on the deduction requirement information, and the boundary conditions include spatial boundary conditions, temporal boundary conditions and external boundary conditions.

[0024] By adopting the above technical solution, the deduction demand information is obtained, including the specific requirements of users or application scenarios for meteorological deduction, such as the geographical area of ​​concern, time range, meteorological elements, etc. Based on the combination of deduction demand information and historical meteorological information, the initial conditions of the deduction model are determined, and the boundary conditions of the deduction model are further refined according to the deduction demand information. Boundary conditions are crucial in meteorological deduction. They limit the spatial scope, time span and external factors that may be affected by the model deduction. Among them, the spatial boundary conditions define the geographical area of ​​the deduction to ensure that the model focuses on the area of ​​user concern. The time boundary conditions set the start and end time of the deduction so that the deduction results can cover the time period required by the user. The external boundary conditions take into account The application considers factors outside the deduction area that may affect the internal meteorological conditions, such as the meteorological conditions and terrain characteristics of the surrounding areas; this application has the effect of significantly improving the prediction accuracy and applicability of the meteorological deduction model by comprehensively considering the deduction needs, historical meteorological information and detailed boundary condition settings. First, by accurately setting the initial conditions, the model can start deduction from a meteorological state that is closer to the actual state, reducing the prediction error caused by inaccurate initial conditions. Secondly, the carefully divided boundary conditions ensure the rationality of the deduction process in space, time and external influences, making the deduction results closer to the actual situation, improving the accuracy of weather forecasts, and providing a more reliable scientific basis for disaster prevention and mitigation, agricultural production, urban planning and other fields.

[0025] In a preferred example, the present application may be further configured as follows: upon receiving the deduction information output by the deduction model, before inputting the deduction information into the pre-trained bias correction model for bias correction, the following steps are performed:

[0026] Obtain historical weather forecast information corresponding to the target area, and extract deviation characteristics based on historical weather information and historical weather forecast information;

[0027] Packing meteorological features and deviation features into a second training feature set in time sequence;

[0028] The bias correction models are trained using the second training feature set to determine the weights of the bias correction models, which include a linear regression model, a random forest model, and a neural network model.

[0029] By adopting the above technical solution, historical meteorological forecast information corresponding to the target area is obtained, and deviation features are extracted based on historical meteorological information (i.e., actually observed meteorological data) and historical meteorological forecast information (i.e., previously predicted meteorological data). The meteorological features and the extracted deviation features are packaged in time series as a second training feature set. The deviation correction model is trained using the second training feature set to determine the weight of each deviation correction model. The deviation correction model includes a linear regression model, a random forest model, and a neural network model, each of which has different advantages and characteristics and can capture different types of deviation patterns. Through training, the contribution (i.e., weight) of each model in the deviation correction can be determined, thereby constructing a more accurate and robust deviation correction system, which has the effect of improving the accuracy of meteorological deduction.

[0030] In a preferred example, the present application may be further configured as follows: the bias correction model is trained by the second training feature set to determine the weight of each bias correction model, wherein the bias correction model includes a linear regression model, a random forest model, and a neural network model, including the steps of:

[0031] Identifying the meteorological type of the historical meteorological forecast information corresponding to each deviation feature in the second training feature set;

[0032] Determine the dynamic weights of each bias correction model based on all identified weather types.

[0033] By adopting the above technical solution, the meteorological type of the historical meteorological forecast information corresponding to each deviation feature in the second training feature set is identified to understand the characteristics and laws of the prediction deviation under different meteorological conditions. For example, some meteorological types (such as heavy rain, typhoon, etc.) may have more complex and difficult to predict characteristics, so the corresponding deviation patterns and correction requirements may also be more special; based on the identified meteorological type, the dynamic weights of each deviation correction model (including linear regression model, random forest model and neural network model) are determined. Because different deviation correction models may have different advantages and effects when processing different types of meteorological data, by dynamically adjusting the weights of each model, the prediction deviation can be corrected more effectively and adaptively under different meteorological types, which has the effect of making the deviation correction more refined and enhancing the adaptability of the deviation correction model.

[0034] In a preferred example, the present application may be further configured as follows: the step of determining the dynamic weight of each deviation correction model based on all identified weather types includes the steps of:

[0035] Based on the preset preliminary weight allocation strategy, the physical mechanism of the identified meteorological type is analyzed and a preliminary weight range is assigned to each bias correction model;

[0036] Each bias correction model is trained and cross-validated using the second training feature set, and the preliminary weight range of each bias correction model is adjusted based on the cross-validation result to determine the dynamic weight of each bias correction model based on each meteorological type.

[0037] By adopting the above technical solution, the physical mechanism of the identified meteorological type is analyzed based on the preset preliminary weight allocation strategy, so as to assign a preliminary weight range that matches the meteorological type to each model to ensure that the weight allocation is consistent with the physical mechanism of meteorology, thereby improving the physical rationality and accuracy of the model. The second training feature set is used to train and cross-validate each bias correction model. The second training feature set contains meteorological features and bias features packaged in time series, which can fully reflect the changing laws and bias characteristics of meteorological data. Through training and cross-validation, the performance of each model under different weights is evaluated, and the preliminary weight range of each model is adjusted based on the verification results, so as to optimize the weight allocation through actual data to ensure the accuracy and stability of the model in actual application. After multiple iterations and adjustments, the dynamic weight of each bias correction model based on each meteorological type is determined, which has the effect of improving the adaptability and physical rationality of the bias correction model.

[0038] The second object of the present invention is achieved through the following technical solutions:

[0039] A weather-based deduction system, comprising:

[0040] Information acquisition module, used to obtain historical meteorological information and meteorological impact information corresponding to the target area;

[0041] A feature extraction module is used to extract meteorological features and impact features from historical meteorological information and meteorological impact information respectively, and package the extracted meteorological features and impact features into a first training feature set;

[0042] A first training module, configured to train a pre-built meteorological deduction model using a first training feature set;

[0043] The deduction model input module is used to set the initial conditions and boundary conditions of the deduction model, obtain current meteorological information and meteorological impact information, and input them into the meteorological deduction model for preliminary deduction;

[0044] The bias model input module is used to input the deduction information output by the deduction model into the pre-trained bias correction model for bias correction when receiving the deduction information;

[0045] The visualization module is used to present the correction deduction information output by the deviation correction model in a visual manner when it is received.

[0046] By adopting the above technical solution, the information acquisition module is used to obtain historical meteorological information and meteorological impact information corresponding to the target area; the feature extraction module is used to extract meteorological features and impact features from the historical meteorological information and meteorological impact information respectively, and package the extracted meteorological features and impact features into a first training feature set; the first training module is used to train a pre-built meteorological deduction model through the first training feature set; the deduction model input module is used to set the initial conditions and boundary conditions of the deduction model, obtain current meteorological information and meteorological impact information and input them into the meteorological deduction model for preliminary deduction; the deviation model input module is used to input the deduction information output by the deduction model into a pre-trained deviation correction model for deviation correction when receiving the deduction information output by the deduction model; the visualization module is used to present the corrected deduction information output by the deviation correction model in a visual manner when receiving it.

[0047] In a preferred example, the present application can be further configured as follows: the feature extraction module includes:

[0048] The meteorological feature acquisition submodule is used to extract element features from historical meteorological information and perform trend analysis and interpolation analysis on the element features to obtain meteorological features, which include time trend features and spatial distribution features;

[0049] An image feature extraction submodule is used to identify direct impact data, indirect impact data, and hysteresis effect data from meteorological impact information, and extract impact features, which include direct impact features, indirect impact features, and hysteresis impact features;

[0050] A feature vector combination submodule is used to combine the extracted meteorological features and the corresponding associated impact features into a plurality of training feature vectors, wherein the training feature vectors include the meteorological features corresponding to the time point and spatial position and the corresponding impact features;

[0051] The feature packaging submodule is used to package a plurality of training feature vectors into a first training feature set.

[0052] By adopting the above technical solution, the meteorological feature acquisition submodule is used to extract element features from historical meteorological information, and perform trend analysis and interpolation analysis on the element features to obtain meteorological features, wherein the meteorological features include time trend features and spatial distribution features; the image feature extraction submodule is used to identify direct impact data, indirect impact data and lag effect data from meteorological impact information, and extract impact features, wherein the impact features include direct impact features, indirect impact features and lag impact features; the feature vector combination submodule is used to combine the extracted meteorological features and the corresponding associated impact features into a number of training feature vectors, wherein the training feature vectors include meteorological features corresponding to time points and spatial positions and their corresponding impact features; the feature packaging submodule is used to package a number of training feature vectors into a first training feature set.

[0053] In summary, this application includes at least one of the following beneficial technical effects:

[0054] 1. This application uses historical meteorological information and meteorological impact information to extract key features for training, thereby constructing a more accurate meteorological deduction model. At the same time, by introducing a bias correction model to fine-tune the deduction results, it has the effect of significantly improving the accuracy and practicality of meteorological deduction.

[0055] 2. This application significantly improves the prediction accuracy and applicability of meteorological deduction models by comprehensively considering deduction needs, historical meteorological information, and detailed boundary condition settings. First, by accurately setting the initial conditions, the model can start deduction from a meteorological state that is closer to the actual state, reducing the prediction error caused by inaccurate initial conditions. Secondly, the carefully divided boundary conditions ensure the rationality of the deduction process in space, time, and external influences, making the deduction results closer to the actual situation, improving the accuracy of weather forecasts, and providing a more reliable scientific basis for disaster prevention and mitigation, agricultural production, urban planning and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a flow chart of an embodiment of a weather-based deduction method of the present application;

[0057] Figure 2 This is a flowchart for implementing step S20 in an embodiment of a weather-based deduction method of the present application;

[0058] Figure 3 This is an implementation flow chart before step S50 in an embodiment of a weather-based deduction method of the present application. DETAILED DESCRIPTION

[0059] The following is combined with Figure 1-3 This application is described in further detail.

[0060] In one embodiment, if Figure 1 As shown, the present application discloses a weather-based deduction method, which specifically includes the following steps:

[0061] S10: Obtain historical meteorological information and meteorological impact information corresponding to the target area;

[0062] In this embodiment, the target area is the area where the weather simulation is to be conducted; the historical weather information is the weather data recorded in the past, including but not limited to temperature, humidity, precipitation, wind speed, wind direction, air pressure, etc.; the weather impact information is the relevant data that may affect the weather, including but not limited to geographical and environmental data, ocean and atmospheric data, solar radiation and astronomical data, human activity data, and other weather observation data;

[0063] Furthermore, geographic and environmental data include:

[0064] Topographic data: including altitude, mountain distribution, and the location of rivers and lakes, which can affect local climate and weather patterns;

[0065] Land use and vegetation cover data: Different types of land cover (such as forests, grasslands, and urban areas) have different effects on climate. For example, forests can affect rainfall and temperature.

[0066] Soil type and moisture data: Soil moisture and type can affect surface albedo and water evaporation, thus affecting local climate;

[0067] Ocean and atmospheric data include:

[0068] Sea Surface Temperature (SST): The ocean is an important heat source for the atmosphere, and changes in SST can affect atmospheric circulation and weather patterns;

[0069] Ocean currents and current data: Ocean currents carry heat and moisture and have a significant impact on the climate of coastal areas;

[0070] Atmospheric circulation indices: such as the El Niño-Southern Oscillation (ENSO) and the Arctic Oscillation (AO). These indices can reflect changes in atmospheric circulation and have a significant impact on climate;

[0071] Solar radiation and astronomical data include:

[0072] Solar radiation intensity: Solar radiation is the main source of energy for the Earth, and its changes can affect climate and weather;

[0073] The distance between the Earth and the Sun and the speed of Earth's rotation: Although this factor has little impact on long-term climate change, it may have some influence on weather patterns under certain conditions (such as changes in the speed of Earth's rotation);

[0074] Human activity data include:

[0075] Greenhouse gas emissions data: The emission of greenhouse gases such as carbon dioxide and methane is one of the main causes of global warming;

[0076] Land use change data: such as urbanization and agricultural expansion, which can alter surface albedo and the water cycle, thus affecting climate;

[0077] Energy use and industrial emissions data: Pollutants and particulate matter emitted by industry can affect local climate and weather;

[0078] Other meteorological observations include:

[0079] Meteorological observation data of neighboring areas: Meteorological conditions in neighboring areas can affect the meteorological changes in the target area, so they can be used as a reference for meteorological deduction and correction;

[0080] Satellite remote sensing data: Satellites can provide large-scale, high-resolution meteorological observation data, such as cloud maps, temperature maps, water vapor maps, etc., which are helpful for meteorological deduction and correction;

[0081] Specifically, collect historical meteorological data information of the target area and information on factors affecting meteorological changes.

[0082] S20: extracting meteorological features and impact features from the historical meteorological information and the meteorological impact information respectively, and packaging the extracted meteorological features and impact features into a first training feature set;

[0083] In this embodiment, meteorological characteristics are representative or key meteorological characteristics extracted through in-depth analysis of historical meteorological information, including daily temperature variations, precipitation intensity and distribution, wind direction and speed, etc. They can reflect the essential attributes and changing patterns of meteorological phenomena and are important inputs to meteorological deduction models. Influencing characteristics are key features that have a significant impact on meteorological deduction, extracted from meteorological influencing information, including topographic and geomorphic characteristics, land use and vegetation cover characteristics, soil type and humidity characteristics, etc. They can reflect the interaction between meteorological conditions and the geographical environment, providing necessary background information for meteorological deduction.

[0084] Specifically, key meteorological features and impact features are extracted from the collected historical meteorological information and meteorological impact information respectively and packaged into a first training feature set for training a pre-built meteorological deduction model.

[0085] S30: training the pre-built meteorological deduction model using the first training feature set;

[0086] In this embodiment, the meteorological deduction model is a model constructed based on historical meteorological information and meteorological principles, and is used to predict future meteorological conditions and simulate the development process of meteorological events;

[0087] Specifically, a meteorological deduction model pre-built based on historical meteorological information and meteorological principles is trained using the first training feature set.

[0088] S40: setting initial conditions and boundary conditions of the deduction model, obtaining current meteorological information and meteorological impact information and inputting them into the meteorological deduction model for preliminary deduction;

[0089] In this embodiment, the initial conditions are the starting states set when the meteorological deduction model starts running, including the spatial distribution and temporal evolution of meteorological elements; the boundary conditions are the constraints followed by the meteorological deduction model during operation, including spatial boundary conditions, temporal boundary conditions, and external boundary conditions; the current meteorological information is the current meteorological data, including but not limited to temperature, humidity, precipitation, wind speed, wind direction, air pressure, etc.; the preliminary deduction is the deduction performed by the meteorological deduction model based on the current meteorological information and meteorological impact information;

[0090] Specifically, according to actual needs, initial conditions and boundary conditions are set for the meteorological deduction model, and the latest current meteorological information and meteorological impact information are obtained, and the current meteorological information and meteorological impact information are input into the meteorological deduction model for preliminary meteorological deduction.

[0091] S50: When receiving the deduction information output by the deduction model, input the deduction information into the pre-trained bias correction model for bias correction;

[0092] In this embodiment, the deduction information is the result information of the predicted future weather conditions and simulated weather events output by the meteorological deduction model after preliminary deduction; the deviation correction model is a model used to correct the output results of the meteorological deduction model;

[0093] Specifically, since there may be various uncertain factors in the meteorological deduction process, which may lead to certain deviations between the deduction results and the actual meteorological conditions, a pre-trained bias correction model is introduced to perform refined bias correction on the deduction information output by the deduction model to improve the accuracy of the prediction.

[0094] S60: When receiving the correction deduction information output by the deviation correction model, presenting it in a visual manner;

[0095] In this embodiment, the corrected deduction information is the result information output after the deviation correction model corrects the deduction information; the visual presentation is to display the data or results output by the meteorological deduction model in the form of graphics, images, or animations, so as to facilitate understanding and analysis;

[0096] Specifically, the bias-corrected deduction information is visualized in intuitive forms such as charts, maps, and animations.

[0097] In one embodiment, if Figure 2 As shown, step S20 includes the steps of:

[0098] S21: extracting element features from historical meteorological information, and performing trend analysis and interpolation analysis on the element features to obtain meteorological features, wherein the meteorological features include time trend features and spatial distribution features;

[0099] S22: identifying direct impact data, indirect impact data, and hysteresis effect data from the meteorological impact information, and extracting impact features, wherein the impact features include direct impact features, indirect impact features, and hysteresis impact features;

[0100] S23: combining the extracted meteorological features and the corresponding associated influencing features into a plurality of training feature vectors, wherein the training feature vectors include meteorological features corresponding to time points and spatial locations and their corresponding influencing features;

[0101] S24: Packing a plurality of training feature vectors into a first training feature set;

[0102] In this embodiment, element features are features of basic meteorological elements extracted from historical meteorological information, including but not limited to temperature, humidity, wind speed, wind direction, precipitation, etc., which are basic components of meteorological data and are used to reflect the basic features of meteorological conditions; trend analysis is a statistical method for identifying long-term trends or patterns in data, specifically for identifying the evolution of meteorological element features over time, such as whether the temperature rises year by year, seasonal changes in precipitation, etc.; interpolation analysis is a mathematical method for estimating values ​​at unknown points, which is usually based on data at known points, specifically used to fill gaps in meteorological observation data, or to generate higher-resolution meteorological data; time trend features are features that describe how meteorological elements change over time, such as seasonal changes, interannual changes, etc.; spatial distribution features are features that describe the spatial distribution of meteorological elements, such as the geographical distribution of precipitation, the vertical gradient of temperature, etc.; directly affecting data are data that have a direct and immediate impact on meteorological conditions, such as topography, land use type, etc.; indirect affecting data are data that have a direct and immediate impact on meteorological conditions, such as topography, land use type, etc. Data refers to data that indirectly affects meteorological conditions by affecting other factors, such as sea surface temperature, atmospheric circulation index, etc.; hysteresis effect data refers to data that has a lagged effect on meteorological conditions, that is, its impact only appears after a period of time, such as the long-term impact of greenhouse gas emissions on the global climate; direct impact features are features extracted from direct impact data, which directly reflect the immediate impact on meteorological conditions; indirect impact features are features extracted from indirect impact data, which reflect the characteristics of indirect impact on meteorological conditions through other factors; lagged impact features are features extracted from lagged effect data, which reflect its long-term or delayed impact on meteorological conditions; a training feature vector is a data structure containing multiple features, which is used to train machine learning models. Specifically, each training feature vector contains the meteorological features and their corresponding impact features at the corresponding time point and spatial position, where the corresponding time point refers to the specific time corresponding to the meteorological features and the impact features, and the corresponding spatial position refers to the specific geographical location corresponding to the meteorological features and the impact features;

[0103] Specifically, element features are extracted from historical meteorological information. After extracting these element features, trend analysis is performed on them to reveal the changing patterns of meteorological elements over time. At the same time, interpolation analysis is performed to supplement and improve the spatial distribution information of meteorological elements. Direct impact data, indirect impact data, and lag effect data are identified from meteorological impact information. Direct impact data include factors such as topography and vegetation cover that directly affect the weather. Indirect impact data involve factors such as human activities and economic development that indirectly affect the weather through certain channels. Lag effect data considers possible delayed effects in meteorological changes, such as the impact of early precipitation on later soil moisture. After identifying direct impact data, indirect impact data, and lag effect data, their impact characteristics on the weather are extracted, including direct impact characteristics, indirect impact characteristics, and lag impact characteristics. The extracted meteorological features and corresponding associated impact characteristics are combined into several training feature vectors. Each training feature vector contains the meteorological features and their corresponding impact characteristics at the corresponding time point and spatial location. This combination method ensures that the training feature vectors can fully reflect the complex relationship between meteorological changes and their influencing factors. The combined several training feature vectors are packaged into a first training feature set.

[0104] In one embodiment, step S40 includes the steps of:

[0105] S41: Obtaining deduction requirement information, and determining initial conditions of the deduction model based on historical meteorological information and the deduction requirement information;

[0106] S42: Determine boundary conditions of the deduction model based on the deduction requirement information, where the boundary conditions include spatial boundary conditions, temporal boundary conditions, and external boundary conditions;

[0107] In this embodiment, the deduction requirement information is the specific requirements for weather deduction proposed by the user or the system, including the deduction time range, spatial range, meteorological elements, etc., and further includes specific meteorological events or phenomena, such as typhoon path prediction and rainstorm warning. The spatial boundary conditions define the geographical area of ​​the deduction to ensure that the model focuses on the area of ​​user interest. The temporal boundary conditions set the start and end time of the deduction so that the deduction results can cover the time period required by the user. The external boundary conditions take into account factors outside the deduction area that may affect the internal meteorological conditions, such as the meteorological conditions and terrain characteristics of the surrounding areas.

[0108] Specifically, the simulation requirement information is obtained, including the specific requirements of the user or application scenario for meteorological simulation, such as the geographical area of ​​interest, time range, and meteorological elements. The initial conditions of the simulation model are determined by combining the simulation requirement information with historical meteorological information. Statistical analysis is performed on this data to understand the average value, change trend, and extreme values ​​of the meteorological element characteristics. The initial conditions of the simulation model are determined based on the time range, spatial range, and meteorological elements in the simulation requirement information. If the simulation requirement information is to predict temperature changes in the next few days, the initial conditions may include the temperature distribution at the start of the simulation and the recent temperature change trend. The boundary conditions of the simulation model are further refined based on the simulation requirement information. Boundary conditions are crucial in meteorological simulation. They limit the spatial scope and time span of the model simulation, as well as the external factors that may be affected. The spatial boundary conditions define the geographical area of ​​the simulation, ensuring that the model focuses on the area of ​​interest to the user. The temporal boundary conditions set the start and end times of the simulation, so that the simulation results cover the time period required by the user. The external boundary conditions consider factors outside the simulation area that may affect the internal meteorological conditions, such as the meteorological conditions and terrain characteristics of the surrounding area.

[0109] In one embodiment, if Figure 3 As shown, before step S50, the following steps are performed:

[0110] S47: Obtain historical meteorological forecast information corresponding to the target area, and extract deviation characteristics based on the historical meteorological information and the historical meteorological forecast information;

[0111] S48: Packing the meteorological features and the deviation features into a second training feature set in time sequence;

[0112] S49: Training the bias correction model using the second training feature set to determine the weight of each bias correction model, wherein the bias correction model includes a linear regression model, a random forest model, and a neural network model;

[0113] In this embodiment, historical weather forecast information is weather forecast data released in the past for the target area; deviation features are features extracted from the difference or deviation between historical weather forecast information and actual weather information; time-series packaging is to organize weather features and deviation features in chronological order to form a series of feature sets with a time-series relationship; the second training feature set is a data set composed of weather features and deviation features combined in chronological order, which is used to train the deviation correction model; the linear regression model is a machine learning model used to establish a linear relationship between input features (such as weather features and deviation features) and output targets (such as deviation correction values); the random forest model is an ensemble learning method that improves the accuracy of prediction by constructing multiple decision trees and integrating their prediction results; the neural network model is a machine learning model that simulates the working mode of the human brain neural network. The neural network model can learn the complex relationship between input features and output prediction results through nonlinear transformation of multiple layers of neurons;

[0114] Specifically, historical meteorological forecast information corresponding to the target area is obtained, and deviation features are extracted based on the historical meteorological information (i.e., actually observed meteorological data) and the historical meteorological forecast information (i.e., previously predicted meteorological data). The meteorological features and the extracted deviation features are packaged in time series as a second training feature set, and the deviation correction model is trained using the second training feature set to determine the weight of each deviation correction model. The deviation correction model includes a linear regression model, a random forest model, and a neural network model, each of which has different advantages and characteristics and can capture different types of deviation patterns. Through training, the contribution (i.e., weight) of each model in the deviation correction can be determined, thereby constructing a more accurate and robust deviation correction system, which has the effect of improving the accuracy of meteorological deduction.

[0115] In one embodiment, step S43 includes the steps of:

[0116] S431: Identifying the meteorological type of the historical meteorological forecast information corresponding to each deviation feature in the second training feature set;

[0117] S432: Determine the dynamic weight of each deviation correction model based on all identified weather types;

[0118] In this embodiment, the weather type is a classification of weather based on meteorological elements and weather phenomena, including sunny, rainy, snowy, windy, etc.; the dynamic weight is a weight dynamically adjusted according to the weather type and other conditions to more accurately correct the deviation;

[0119] Specifically, the meteorological type of the historical meteorological forecast information corresponding to each deviation feature in the second training feature set is identified to understand the characteristics and laws of the prediction deviation under different meteorological conditions. For example, some meteorological types (such as heavy rain, typhoon, etc.) may have more complex and difficult to predict characteristics, so the corresponding deviation patterns and correction requirements may also be more special; based on the identified meteorological type, the dynamic weights of each deviation correction model (including linear regression model, random forest model and neural network model) are determined. Because different deviation correction models may have different advantages and effects when processing different types of meteorological data, by dynamically adjusting the weights of each model, the prediction deviation can be corrected more effectively and adaptively under different meteorological types, which has the effect of making the deviation correction more refined and enhancing the adaptability of the deviation correction model.

[0120] In one embodiment, step S432 includes the following steps:

[0121] S4321: Based on a pre-set preliminary weight allocation strategy, analyze the physical mechanism of the identified meteorological type and assign preliminary weight ranges to each bias correction model;

[0122] S4322: Training and cross-validating each bias correction model using the second training feature set, and adjusting a preliminary weight range of each bias correction model based on the cross-validation results to determine a dynamic weight of each bias correction model based on each weather type;

[0123] In this embodiment, the preliminary weight allocation strategy is to analyze the physical mechanism of meteorological types based on experience and preset rules, and to assign preliminary weight ranges to each bias correction model based on the analysis results of the physical mechanism. The physical mechanism is the physical principle or law behind the meteorological phenomenon, which is used to guide the preliminary allocation of weights. Cross-validation is a model evaluation method that divides a dataset into a training set and a validation set, trains the model with the training set, and evaluates the performance of the bias correction model with the validation set to adjust the parameters or weights of the bias correction model. Adjusting the preliminary weight range is to optimize the preliminary weight range based on the cross-validation results to determine more accurate dynamic weights.

[0124] Specifically, the physical mechanism of the identified meteorological type is analyzed based on the preset preliminary weight allocation strategy, so as to assign a preliminary weight range that matches the meteorological type to each model to ensure that the weight distribution is consistent with the physical mechanism of meteorology, thereby improving the physical rationality and accuracy of the model. The second training feature set is used to train and cross-validate each bias correction model. The second training feature set contains meteorological features and bias features packaged in time series, which can fully reflect the changing laws and bias characteristics of meteorological data. Through training and cross-validation, the performance of each model under different weights is evaluated, and the preliminary weight range of each model is adjusted based on the verification results, so as to optimize the weight distribution through actual data to ensure the accuracy and stability of the model in actual application. After multiple iterations and adjustments, the dynamic weight of each bias correction model based on each meteorological type is determined, which has the effect of improving the adaptability and physical rationality of the bias correction model.

[0125] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0126] In one embodiment, a weather-based deduction system is provided, and the weather-based deduction system corresponds one-to-one to the weather-based deduction method in the above embodiment.

[0127] A weather-based deduction system, comprising:

[0128] Information acquisition module, used to obtain historical meteorological information and meteorological impact information corresponding to the target area;

[0129] A feature extraction module is used to extract meteorological features and impact features from historical meteorological information and meteorological impact information respectively, and package the extracted meteorological features and impact features into a first training feature set;

[0130] A first training module, configured to train a pre-built meteorological deduction model using a first training feature set;

[0131] The deduction model input module is used to set the initial conditions and boundary conditions of the deduction model, obtain current meteorological information and meteorological impact information, and input them into the meteorological deduction model for preliminary deduction;

[0132] The bias model input module is used to input the deduction information output by the deduction model into the pre-trained bias correction model for bias correction when receiving the deduction information;

[0133] A visualization module is used to present the correction deduction information output by the bias correction model in a visual manner when receiving the correction deduction information;

[0134] Optionally, the feature extraction module includes:

[0135] The meteorological feature acquisition submodule is used to extract element features from historical meteorological information and perform trend analysis and interpolation analysis on the element features to obtain meteorological features, which include time trend features and spatial distribution features;

[0136] An image feature extraction submodule is used to identify direct impact data, indirect impact data, and hysteresis effect data from meteorological impact information, and extract impact features, which include direct impact features, indirect impact features, and hysteresis impact features;

[0137] A feature vector combination submodule is used to combine the extracted meteorological features and the corresponding associated impact features into a plurality of training feature vectors, wherein the training feature vectors include the meteorological features corresponding to the time point and spatial position and the corresponding impact features;

[0138] A feature packaging submodule, configured to package a plurality of training feature vectors into a first training feature set;

[0139] Optionally, the deduction model input module includes:

[0140] The initial condition determination submodule is used to obtain deduction requirement information and determine the initial conditions of the deduction model based on historical meteorological information and deduction requirement information;

[0141] A boundary condition determination submodule is used to determine the boundary conditions of the deduction model based on the deduction requirement information, wherein the boundary conditions include spatial boundary conditions, temporal boundary conditions and external boundary conditions;

[0142] Optionally, also include:

[0143] Deviation feature extraction module, used to obtain historical meteorological forecast information corresponding to the target area, and extract deviation features based on historical meteorological information and historical meteorological forecast information;

[0144] A feature packaging module is used to package meteorological features and deviation features into a second training feature set in time sequence;

[0145] a second training module, configured to train the bias correction models using a second training feature set to determine weights of the bias correction models, wherein the bias correction models include a linear regression model, a random forest model, and a neural network model;

[0146] Optionally, the second training module includes:

[0147] A meteorological type identification submodule, configured to identify the meteorological type of the historical meteorological forecast information corresponding to each deviation feature in the second training feature set;

[0148] A weight determination submodule is used to determine the dynamic weights of each bias correction model based on all identified meteorological types;

[0149] Optionally, the weight determination submodule includes:

[0150] A weight range allocation submodule is used to analyze the physical mechanism of the identified meteorological type and allocate preliminary weight ranges to each bias correction model based on a preset preliminary weight allocation strategy;

[0151] The weight adjustment submodule is used to train and cross-validate each deviation correction model through the second training feature set, and adjust the preliminary weight range of each deviation correction model based on the cross-validation result to determine the dynamic weight of each deviation correction model based on each meteorological type.

[0152] The specific definition of a weather-based deduction system can be found in the definition of a weather-based deduction method above and will not be repeated here. Each module in the weather-based deduction system described above can be implemented in whole or in part through software, hardware, or a combination thereof. Each of the modules can be embedded in or independent of a processor in a computer device in hardware form, or stored in a computer device memory in software form, so that the processor can call and execute the corresponding operations of each module.

[0153] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. A weather-based deduction method, characterized by: Including steps: Obtain historical meteorological information and meteorological impact information corresponding to the target area; Extracting meteorological features and impact features from historical meteorological information and meteorological impact information respectively, and packaging the extracted meteorological features and impact features into a first training feature set; Training a pre-built meteorological deduction model using a first training feature set; Set the initial conditions and boundary conditions of the deduction model, obtain current meteorological information and meteorological impact information and input them into the meteorological deduction model for preliminary deduction; When receiving the deduction information output by the deduction model, the deduction information is input into the pre-trained bias correction model for bias correction; When receiving the correction deduction information output by the bias correction model, present it in a visual manner; When receiving the deduction information output by the deduction model, before inputting the deduction information into the pre-trained bias correction model for bias correction, the following steps are performed: Obtain historical weather forecast information corresponding to the target area, and extract deviation features based on the historical weather information and historical weather forecast information; Packing meteorological features and deviation features into a second training feature set in time sequence; Training the bias correction models using the second training feature set to determine weights of the bias correction models, the bias correction models including a linear regression model, a random forest model, and a neural network model; The bias correction model is trained by the second training feature set to determine the weight of each bias correction model, wherein the bias correction model includes a linear regression model, a random forest model, and a neural network model, including the steps of: Identifying the meteorological type of the historical meteorological forecast information corresponding to each deviation feature in the second training feature set; Determine the dynamic weights of each bias correction model based on all identified meteorological types; The step of determining the dynamic weight of each bias correction model based on all identified weather types comprises the steps of: Based on the preset preliminary weight allocation strategy, the physical mechanism of the identified meteorological type is analyzed and a preliminary weight range is assigned to each bias correction model; Each bias correction model is trained and cross-validated using the second training feature set, and the preliminary weight range of each bias correction model is adjusted based on the cross-validation result to determine the dynamic weight of each bias correction model based on each meteorological type.

2. The weather-based deduction method according to claim 1, characterized in that: The step of extracting meteorological features and impact features from historical meteorological information and meteorological impact information, respectively, and packaging the extracted meteorological features and impact features into a first training feature set, comprises the steps of: Extracting element features from historical meteorological information and performing trend analysis and interpolation analysis on the element features to obtain meteorological features, wherein the meteorological features include time trend features and spatial distribution features; Identifying direct impact data, indirect impact data, and lagged effect data from meteorological impact information, and extracting impact features, wherein the impact features include direct impact features, indirect impact features, and lagged impact features; Combining the extracted meteorological features and the corresponding associated influencing features into a plurality of training feature vectors, wherein the training feature vectors include the meteorological features corresponding to the time point and spatial position and the corresponding influencing features; Several training feature vectors are packaged into a first training feature set.

3. The weather-based deduction method according to claim 1, characterized in that: The steps of setting the initial conditions and boundary conditions of the deduction model, obtaining current meteorological information and meteorological impact information and inputting them into the meteorological deduction model for preliminary deduction include the following steps: Obtain deduction demand information and determine the initial conditions of the deduction model based on historical meteorological information and deduction demand information; The boundary conditions of the deduction model are determined based on the deduction requirement information, and the boundary conditions include spatial boundary conditions, temporal boundary conditions, and external boundary conditions.

4. A weather-based deduction system, configured to execute the steps of a weather-based deduction method according to any one of claims 1 to 3, characterized in that: include: Information acquisition module, used to obtain historical meteorological information and meteorological impact information corresponding to the target area; A feature extraction module is used to extract meteorological features and impact features from historical meteorological information and meteorological impact information respectively, and package the extracted meteorological features and impact features into a first training feature set; A first training module, configured to train a pre-built meteorological deduction model using a first training feature set; The deduction model input module is used to set the initial conditions and boundary conditions of the deduction model, obtain current meteorological information and meteorological impact information, and input them into the meteorological deduction model for preliminary deduction; The bias model input module is used to input the deduction information output by the deduction model into the pre-trained bias correction model for bias correction when receiving the deduction information; The visualization module is used to present the correction deduction information output by the deviation correction model in a visual manner when it is received.

5. The weather-based deduction system according to claim 4, characterized in that: The feature extraction module includes: The meteorological feature acquisition submodule is used to extract element features from historical meteorological information and perform trend analysis and interpolation analysis on the element features to obtain meteorological features, which include time trend features and spatial distribution features; An image feature extraction submodule is used to identify direct impact data, indirect impact data, and hysteresis effect data from meteorological impact information, and extract impact features, which include direct impact features, indirect impact features, and hysteresis impact features; A feature vector combination submodule is used to combine the extracted meteorological features and the corresponding associated impact features into a plurality of training feature vectors, wherein the training feature vectors include the meteorological features corresponding to the time point and spatial position and the corresponding impact features; The feature packaging submodule is used to package a plurality of training feature vectors into a first training feature set.

Images