A weather forecasting method and system based on data fusion in complex environment

Abstract

This invention provides a method and system for complex environmental weather forecasting based on data fusion. The method includes: collecting underlying surface data of a target area and determining an underlying surface feature set; dividing the target area into grids to determine multiple target sub-regions; determining multiple underlying surface driving datasets; then constructing a mapping relationship between the multiple target sub-regions and the multiple underlying surface driving datasets, performing data influence coefficient analysis, and constructing a data correction model using a BP neural network method; inputting real-time monitoring data into the data correction model and outputting corrected driving data; sending the corrected driving data to a weather forecasting system, and finally outputting the weather forecast result. This invention can solve the problem of inaccurate weather forecast results caused by the influence of underlying surfaces, thereby effectively improving the accuracy of weather forecasts.

Description

Technical Field

[0001] This application relates to the field of weather forecasting technology, specifically to a method and system for forecasting complex environmental weather based on data fusion. Background Technology

[0002] Weather forecasting is the process of making qualitative or quantitative predictions about the weather conditions of a specific region or location over a future period, based on meteorological observation data and applying principles and methods of synoptic dynamics and statistics. The results of weather forecasts directly affect the daily lives of people in the region; therefore, the accuracy of weather forecasts is of paramount importance.

[0003] China has a large land area and complex terrain. Especially in the Hengduan Mountains, there is a saying that "one mountain has four seasons and the weather can be different every ten miles." Weather forecasts in the region are mainly based on weather data obtained from meteorological stations. However, since meteorological stations may be far from the destination and the underlying surface conditions are complex, inaccurate weather forecasts often occur, which greatly affects people's daily travel.

[0004] In summary, existing technologies suffer from inaccurate weather forecasts due to the influence of the underlying surface. Summary of the Invention

[0005] Therefore, it is necessary to provide a method and system for forecasting complex environmental weather based on data fusion to address the aforementioned technical problems.

[0006] A complex environmental weather forecasting method based on data fusion includes: collecting underlying surface data of a target area and extracting features from the underlying surface data to determine an underlying surface feature set based on the feature extraction results; dividing the target area into grids based on the underlying surface feature set to determine multiple target sub-regions; using an underlying surface data acquisition module to retrieve driving data from the multiple target sub-regions to obtain multiple underlying surface driving datasets; constructing a mapping relationship between the multiple target sub-regions and the multiple underlying surface driving datasets, and performing data influence coefficient analysis based on the mapping relationship to obtain multiple complexities; constructing a data correction model based on the multiple underlying surface driving datasets and the multiple complexities; collecting real-time monitoring data of the multiple target sub-regions, inputting the real-time monitoring data into the data correction model, and outputting corrected driving data; and sending the corrected driving data to a weather forecasting system to output weather forecast results.

[0007] In one embodiment, the method further includes: extracting a river system feature set and a hydrological feature set from the underlying surface feature set, wherein the river system feature set includes flow path, flow direction, and catchment area, and the hydrological feature set includes flow rate, flood season, and water level; determining the river system distribution surface of the target area based on the flow path, flow direction, and catchment area, and identifying the river system regions; determining the hydrological influence coefficients of the multiple target sub-regions based on the flow rate, flood season, and water level; classifying the multiple target sub-regions into hydrological grades based on the hydrological influence coefficients and the river system distribution surface, and obtaining grade classification results; determining the prediction accuracy of the multiple target sub-regions based on the grade classification results, and providing weather forecasts for the target area based on the prediction accuracy.

[0008] In one embodiment, the method further includes: randomly selecting n experts from an expert database to score the hydrological impact of multiple target sub-regions based on the flow rate, flood season, and water level, respectively, to obtain multiple score result sets, wherein the multiple score result sets correspond one-to-one with the multiple target sub-regions; calculating the mean of the multiple score result sets to obtain multiple score mean values ​​corresponding to the flow rate, flood season, and water level; constructing multiple priority graph weight matrices based on the multiple score mean values, and calculating the weights to obtain multiple weight percentage value sets; and obtaining the hydrological impact coefficient based on the multiple weight percentage value sets and the multiple score mean values.

[0009] In one embodiment, the method further includes: collecting multiple historical weather data sets of the multiple target sub-regions within the data retrieval window; performing data cleaning on the multiple historical weather data sets to obtain multiple standard historical weather data sets; and traversing the multiple historical weather data sets and the multiple underlying surface driving datasets to extract abnormal data and obtain an abnormal data set.

[0010] In one embodiment, the method further includes: matching the corresponding weather scene based on the abnormal data set to obtain a scene type; obtaining a deviation value set based on the difference between the abnormal data set and the standard data, wherein the deviation value set corresponds one-to-one with the scene type; assigning values ​​to the scene type based on the deviation value set to obtain a scene coefficient set; and obtaining the multiple complexities based on the scene coefficient set.

[0011] In one embodiment, the method further includes: constructing a network structure based on a BP neural network; constructing a training dataset based on the plurality of underlying surface driving datasets and the plurality of historical monitoring datasets; training and validating the network structure using the training dataset until the accuracy meets the requirements, thereby obtaining the data correction model; inputting the real-time monitoring data into the data correction model and outputting initial correction data; and optimizing the initial correction data according to the plurality of complexities to obtain correction driving data.

[0012] In one embodiment, the method further includes: setting a verification time window according to a preset verification rule; extracting the monitoring data within the verification time window and matching and verifying it with the correction driving data to obtain a verification result; determining whether the verification result is successful, and updating the data correction model if it is not successful.

[0013] A complex environmental weather forecasting system based on data fusion includes:

[0014] The underlying surface feature set determination module is used to collect underlying surface data of the target area, extract features from the underlying surface data, and determine the underlying surface feature set based on the feature extraction results.

[0015] A multiple target sub-region determination module is used to divide the target region into grids based on the underlying surface feature set to determine multiple target sub-regions.

[0016] A driving data retrieval module is used to retrieve driving data from the multiple target sub-regions using the underlying surface data acquisition module to obtain multiple underlying surface driving datasets.

[0017] The data impact coefficient analysis module is used to construct the mapping relationship between the multiple target sub-regions and the multiple underlying surface driving datasets, and to perform data impact coefficient analysis based on the mapping relationship to obtain multiple complexities.

[0018] A data correction model construction module is used to construct a data correction model based on the multiple underlying surface driving datasets and the multiple complexities.

[0019] A correction-driven data output module is used to collect real-time monitoring data of the multiple target sub-regions, input the real-time monitoring data into the data correction model, and output correction-driven data.

[0020] The weather forecast result output module is used to send the correction driving data to the weather forecast system and output the weather forecast result.

[0021] The aforementioned data fusion-based method and system for complex environmental weather forecasting can solve the problem of inaccurate weather forecasts caused by the influence of underlying surface. It divides the target area into multiple target sub-regions based on the characteristics of the underlying surface; obtains weather forecast data for each of these sub-regions; performs correlation analysis between the weather forecast data and historical weather data for each sub-region, further obtaining multiple complexities based on scenario type; corrects the weather forecast results using a neural network model, then optimizes the correction results based on the multiple complexities to obtain correction-driven data; finally, it obtains the weather forecast result based on the correction-driven data. This method can fuse monitored weather data with local underlying surface data, and obtain the weather forecast result based on the fusion result, thereby improving the accuracy of weather forecasts.

[0022] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0023] Figure 1 This application provides a flowchart illustrating a method for forecasting complex environmental weather based on data fusion.

[0024] Figure 2 This application provides a flowchart illustrating the process of obtaining anomaly data sets in a complex environmental weather forecasting method based on data fusion.

[0025] Figure 3 This application provides a flowchart illustrating the process of obtaining corrected driving data in a complex environmental weather forecasting method based on data fusion.

[0026] Figure 4 This application provides a schematic diagram of the structure of a complex environmental weather forecasting system based on data fusion.

[0027] Figure labeling: 1. Underlying surface feature set determination module; 2. Multiple target sub-region determination module; 3. Driving data retrieval module; 4. Data influence coefficient analysis module; 5. Data correction model construction module; 6. Correction driving data output module; 7. Weather forecast result output module. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0029] like Figure 1 As shown, this application provides a method for forecasting complex environmental weather based on data fusion, the method comprising:

[0030] Step S100: Collect underlying surface data of the target area, extract features from the underlying surface data, and determine the underlying surface feature set based on the feature extraction results;

[0031] Step S200: Divide the target region into grids based on the underlying surface feature set to determine multiple target sub-regions;

[0032] Step S300: Use the underlying surface data acquisition module to retrieve driving data from the multiple target sub-regions respectively to obtain multiple underlying surface driving datasets;

[0033] Specifically, underlying surface data is collected for the target area, which refers to the region where weather forecast results need to be corrected. The underlying surface data includes data such as elevation, relative height, slope, water system, and hydrology. Then, key feature data, including elevation, relative height, and slope, are extracted from the underlying surface data. Based on the feature extraction results, an underlying surface feature set is obtained. This feature set refers to the underlying surface and its characteristic range within the region. For example, for plains, the elevation is 0-200 meters, the relative height is less than or equal to 50 meters, and the slope is less than 5 degrees; for hills, the elevation is 0-500 meters, the relative height is less than or equal to 300 meters, and the slope is greater than 10 degrees and less than 40 degrees. The characteristic range can be customized by those skilled in the art according to the required level of regional division.

[0034] The target area is divided into multiple target sub-regions based on the underlying surface features and their ranges. These sub-regions share the same underlying surface. A underlying surface data acquisition module (referring to weather stations within the target sub-regions) retrieves driving data from these sub-regions. This module retrieves historical weather forecast data from the weather stations, resulting in multiple historical weather forecast datasets, i.e., the multiple underlying surface driving datasets. These datasets include data on temperature, air pressure, humidity, wind speed, precipitation, and radiation. Dividing the target area according to the underlying surface provides support for further refined analysis of weather forecast results for the target area, indirectly improving the accuracy of weather forecasts.

[0035] Step S400: Construct a mapping relationship between the multiple target sub-regions and the multiple underlying surface driving datasets, and perform data influence coefficient analysis based on the mapping relationship to obtain multiple complexities;

[0036] like Figure 2 As shown, in one embodiment, step S400 of this application further includes:

[0037] Step S410: Collect data to retrieve multiple historical weather data sets for the multiple target sub-regions within the data acquisition window;

[0038] Step S420: Perform data cleaning on the multiple historical weather datasets to obtain multiple standard historical weather datasets;

[0039] Step S430: Traverse the multiple historical weather data sets and the multiple underlying surface driving datasets to extract abnormal data and obtain an abnormal data set.

[0040] Specifically, a preset data retrieval window is used, which refers to an interval of time from the present, such as the last 15 days or the last 30 days. Those skilled in the art can customize this setting. Multiple historical weather data sets for multiple target sub-regions within the data retrieval window are collected to obtain multiple historical weather data sets. Missing, omitted, and invalid data in these multiple historical weather data sets are cleaned to obtain multiple standard historical weather data sets. Then, these multiple historical weather data sets and the multiple underlying surface driving datasets are compared and traversed according to data collection time and data type. Data with different values ​​are extracted as outliers to obtain an outlier data set. Obtaining this outlier data set provides data support for the next step of analyzing outlier data.

[0041] In one embodiment, step S400 of this application further includes:

[0042] Step S440: Based on the abnormal data set, match the corresponding weather scene to obtain the scene type;

[0043] Step S450: Obtain a set of deviation values ​​based on the difference between the abnormal data set and the standard data, wherein the set of deviation values ​​corresponds one-to-one with the scene type;

[0044] Step S460: Assign values ​​to the scene type according to the deviation value set to obtain the scene coefficient set;

[0045] Step S470: Obtain the multiple complexities based on the set of scene coefficients.

[0046] Specifically, the weather scenarios corresponding to the time nodes of the abnormal data collection in the abnormal data set are obtained. These weather scenarios include sunny days, light rain, heavy rain, light snow, heavy snow, and hail, thus obtaining the weather scenario types. The difference between the abnormal data and the underlying surface driving data is used as the deviation value, and these deviation values ​​are categorized according to the scenario type to obtain a deviation value set, where the deviation value set and the scenario type have a one-to-one correspondence. The average value of the deviation values ​​in the deviation value set is obtained, and the scenario type is assigned a value based on this average value. The average value of the deviation values ​​is used as the coefficient of each data type in the scenario to obtain a scenario coefficient set. For example, assuming the scenario type is light rain, the average temperature deviation value is -0.1 degrees Celsius, and the average precipitation deviation value is 1 mm / hour, then the temperature coefficient for light rain is -0.1, and the precipitation coefficient is 1. Then, the scenario coefficient set is used as multiple complexities, where the complexity includes the scenario type and the scenario coefficient. By obtaining multiple complexities, support is provided for the next step of weather data correction, thereby improving the accuracy of weather forecast results.

[0047] Step S500: Construct a data correction model based on the multiple underlying surface driving datasets and the multiple complexities;

[0048] like Figure 3 As shown, in one embodiment, step S500 of this application further includes:

[0049] Step S510: Construct the network structure based on the BP neural network;

[0050] Step S520: Construct a training dataset based on the multiple underlying surface driving datasets and multiple historical monitoring datasets;

[0051] Step S530: Train and validate the network structure using the training dataset until the accuracy meets the requirements, and obtain the data correction model;

[0052] Step S540: After inputting the real-time monitoring data into the data correction model, output the initial correction data;

[0053] Step S550: Optimize the initial correction data according to the multiple complexities to obtain correction driving data.

[0054] Specifically, a data correction model is constructed based on a backpropagation (BP) neural network. This model is a neural network model in machine learning that can be iteratively optimized and is obtained through supervised training using a training dataset. Multiple underlying surface driving datasets and multiple historical monitoring datasets are used as sample datasets. The historical monitoring datasets consist of historical real weather data for the target sub-region. A preset data partitioning rule is used to divide the sample datasets into training and validation datasets. The data correction model is then trained under supervised conditions using the training dataset. When the model's output results tend to converge, the output results are validated using the validation dataset. When the accuracy of the model's output results reaches a preset accuracy index, the data correction model is obtained.

[0055] Weather forecast data is input into the data correction model to obtain the model output, which is the initial correction data. Then, the initial correction data is optimized based on several complexity factors. This optimization involves adding scenario coefficients from the complexity factors to the initial correction data to obtain the correction-driving data. By constructing a data correction model to correct the weather forecast results, and then further optimizing the initial correction results based on the complexity factors, the accuracy of the weather forecast results can be further improved.

[0056] Step S600: Collect real-time monitoring data of the multiple target sub-regions, input the real-time monitoring data into the data correction model, and output correction driving data;

[0057] In one embodiment, step S600 of this application further includes:

[0058] Step S610: Set the verification time window according to the preset verification rules;

[0059] Step S620: Extract the monitoring data within the verification time window and match and verify it with the correction driving data to obtain the verification result;

[0060] Step S630: Determine whether the verification result is successful. If the verification is unsuccessful, update the data correction model.

[0061] Specifically, weather forecast data for the multiple target sub-regions is obtained, and this weather forecast data is input into the data correction model to obtain the model output, i.e., the correction driving data. A preset verification time window is established, which can be customized by those skilled in the art. The weather forecast data and correction driving data within the verification time window are extracted and matched to obtain the verification result, i.e., the data deviation degree. A preset data deviation danger value is also established. When the verification result is less than or equal to the data deviation danger value, the verification passes, indicating that the model output is normal. When the verification result is greater than the data deviation danger value, the verification fails, indicating that the model output is abnormal and the data correction model needs to be updated. The update refers to training the model with a larger amount of training data to improve the accuracy of the model output.

[0062] Step S700: Send the correction driving data to the weather forecast system and output the weather forecast result.

[0063] In one embodiment, step S700 of this application further includes:

[0064] Step S710: Extract the water system feature set and the hydrological feature set from the underlying surface feature set, wherein the water system feature set includes flow path, flow direction and drainage area, and the hydrological feature set includes flow rate, flood season and water level;

[0065] Step S720: Determine the water system distribution area of ​​the target region based on the process, flow direction and watershed area, and identify the water system area connections;

[0066] Step S730: Determine the hydrological influence coefficients of the multiple target sub-regions based on the flow rate, flood season, and water level;

[0067] In one embodiment, step S730 of this application further includes:

[0068] Step S731: Randomly select n experts from the expert database to score the degree of hydrological impact on multiple target sub-regions based on the flow rate, flood season, and water level, and obtain multiple score result sets, wherein each of the multiple score result sets corresponds one-to-one with the multiple target sub-regions;

[0069] Step S732: Calculate the mean value based on the multiple score result sets to obtain the mean values ​​of multiple scores corresponding to the flow rate, flood season, and water level;

[0070] Step S733: Construct multiple priority graph weight matrices based on the multiple average scores, and perform weight calculations to obtain multiple sets of weight percentage values;

[0071] Step S734: Obtain the hydrological impact coefficient based on the multiple sets of weight percentage values ​​and the multiple average scores.

[0072] Specifically, data on flow path, flow direction, and catchment area from the underlying surface feature set are extracted to construct a river system feature set. Data on flow rate, flood season, and water level from the underlying surface feature set are also extracted to construct a river system feature set. The river system distribution area of ​​the target region is determined based on the flow path, flow direction, and catchment area. Then, the river system regions are identified based on the differences in the distribution area; for example, regions with short flow paths and small catchment areas are defined as mountain river systems, while regions with long flow paths and large catchment areas are defined as plain river systems.

[0073] An expert database is constructed based on an artificial neural network. This database contains multiple experts in the field of hydrology. N experts score the hydrological impact of multiple target sub-regions from three perspectives: flow rate, flood season, and water level. Multiple score result sets are obtained, each with a one-to-one correspondence with a target sub-region. The average scores for flow rate, flood season, and water level in these result sets are calculated to obtain multiple average flow rate scores, flood season scores, and water level scores. Multiple pecking order graph weight matrices are constructed based on these average scores. The weight matrix compares the average values ​​of flow rate, flood season, and water level pairwise: a larger average score is scored as 1 point, equal average scores as 0.5 points, and a smaller average score as 0 points. The weight value of each indicator is obtained by summing the scores of each column and summing the scores of all parameters. This results in multiple sets of weight percentage values. The sum of the average scores for each target sub-region multiplied by the corresponding weight percentage is then used as the hydrological impact coefficient for that sub-region.

[0074] Step S740: Based on the hydrological influence coefficient and the river system distribution, classify the multiple target sub-regions into hydrological levels to obtain the level classification results;

[0075] Step S750: Determine the prediction accuracy of the multiple target sub-regions based on the classification results, and make a weather forecast for the target region based on the prediction accuracy.

[0076] Specifically, a pre-defined hydrological classification rule is used. This rule can be customized by those skilled in the art based on the actual regional hydrological conditions. For example, when the regional water system distribution surface is a mountainous water system and the hydrological influence coefficient is greater than 5, it is set to hydrological level one. Hydrological levels are classified for the multiple target sub-regions based on the hydrological influence coefficient and the water system distribution surface, obtaining the classification results. Then, the prediction accuracy for the multiple target sub-regions is set according to the classification results. The higher the classification result, the larger the difference in weather data within the sub-region, and the higher the prediction accuracy setting. Finally, a weather forecast is made for the target area based on the prediction accuracy. By setting different prediction accuracy for multiple target sub-regions, the accuracy of weather forecasts can be ensured while reducing operating costs and avoiding resource waste. Finally, the correction driving data is sent to the weather forecasting system, and the weather forecast results for the target sub-regions are obtained through the weather forecasting system. The above method solves the problem of inaccurate weather forecast results caused by the influence of the underlying surface. By fusing the monitored weather data with the local underlying surface data, the weather forecast results are obtained based on the weather data fusion results, thereby improving the accuracy of weather forecasts.

[0077] In one embodiment, such as Figure 4 The system provides a complex environmental weather forecasting system based on data fusion, comprising: 1. a surface feature set determination module; 2. a multiple target sub-region determination module; 3. a driving data retrieval module; 4. a data influence coefficient analysis module; 5. a data correction model construction module; 6. a correction driving data output module; and 7. a weather forecast result output module.

[0078] The underlying surface feature set determination module 1 is used to collect underlying surface data of the target area, extract features from the underlying surface data, and determine the underlying surface feature set based on the feature extraction results.

[0079] Multiple target sub-region determination module 2, which is used to divide the target area into grids based on the underlying surface feature set to determine multiple target sub-regions;

[0080] The driving data retrieval module 3 is used to retrieve driving data from the multiple target sub-regions using the underlying surface data acquisition module to obtain multiple underlying surface driving datasets.

[0081] The data impact coefficient analysis module 4 is used to construct the mapping relationship between the multiple target sub-regions and the multiple underlying surface driving datasets, and to perform data impact coefficient analysis based on the mapping relationship to obtain multiple complexities.

[0082] Data correction model construction module 5, which is used to construct a data correction model based on the multiple underlying surface driving datasets and the multiple complexities;

[0083] The correction driving data output module 6 is used to collect real-time monitoring data of the multiple target sub-regions, input the real-time monitoring data into the data correction model, and output correction driving data.

[0084] Weather forecast result output module 7 is used to send the correction driving data to the weather forecast system and output the weather forecast result.

[0085] In one embodiment, the system further includes:

[0086] A water system hydrological feature set extraction module is used to extract a water system feature set and a hydrological feature set from the underlying surface feature set, wherein the water system feature set includes flow path, flow direction and drainage area, and the hydrological feature set includes flow rate, flood season and water level;

[0087] A water system area connection identification module is used to determine the water system distribution area of ​​the target region based on the process, flow direction and watershed area, and to perform water system area connection identification.

[0088] A hydrological impact coefficient determination module is used to determine the hydrological impact coefficients of the multiple target sub-regions based on the flow rate, flood season, and water level.

[0089] A hydrological classification module is used to classify the hydrological levels of the multiple target sub-regions based on the hydrological influence coefficient and the river system distribution surface, and obtain the classification results.

[0090] A prediction accuracy determination module is used to determine the prediction accuracy of the multiple target sub-regions based on the classification results, and to make weather forecasts for the target regions based on the prediction accuracy.

[0091] In one embodiment, the system further includes:

[0092] The scoring result set acquisition module is used to randomly select n experts from the expert database to score the degree of hydrological impact of the flow rate, flood season and water level on multiple target sub-regions, and obtain multiple scoring result sets, wherein the multiple scoring result sets correspond one-to-one with the multiple target sub-regions;

[0093] A module for obtaining multiple average scores is provided, which is used to calculate the average score based on the multiple score result sets to obtain multiple average scores corresponding to the flow rate, flood season, and water level.

[0094] The weight calculation module is used to construct multiple aptitude graph weight matrices based on the multiple average scores, and to perform weight calculations to obtain multiple sets of weight percentage values.

[0095] The hydrological impact coefficient acquisition module is used to obtain the hydrological impact coefficient based on the multiple sets of weight percentage values ​​and the multiple average scores.

[0096] In one embodiment, the system further includes:

[0097] A historical weather data set acquisition module is used to acquire multiple historical weather data sets of the multiple target sub-regions within the data retrieval window.

[0098] A data cleaning module is used to clean the multiple historical weather data sets to obtain standard multiple historical weather data sets.

[0099] An abnormal data set acquisition module is used to traverse the multiple historical weather data sets and the multiple underlying surface driving datasets to extract abnormal data and obtain an abnormal data set.

[0100] In one embodiment, the system further includes:

[0101] A scene type acquisition module is used to match the corresponding weather scene based on the abnormal data set to obtain the scene type.

[0102] A deviation value set acquisition module is used to obtain a deviation value set based on the difference between the abnormal data set and the standard data, wherein the deviation value set corresponds one-to-one with the scene type;

[0103] A scene coefficient set acquisition module, wherein the scene data set acquisition module is used to assign values ​​to the scene type according to the deviation value set to obtain the scene coefficient set;

[0104] Multiple complexity acquisition modules are used to obtain the multiple complexities based on the set of scene coefficients.

[0105] In one embodiment, the system further includes:

[0106] A network structure construction module, which is used to construct a network structure based on a BP neural network;

[0107] A training dataset construction module is used to construct a training dataset based on the multiple underlying surface driving datasets and multiple historical monitoring datasets;

[0108] A data correction model acquisition module is used to train and validate the network structure using the training dataset until the accuracy meets the requirements, thereby obtaining the data correction model.

[0109] An initial correction data acquisition module is used to input the real-time monitoring data into the data correction model and then output the initial correction data.

[0110] A correction driving data obtaining module is used to optimize the initial correction data according to the multiple complexities to obtain correction driving data.

[0111] In one embodiment, the system further includes:

[0112] A verification time window setting module is used to set a verification time window according to a preset verification rule.

[0113] The verification result acquisition module is used to extract the monitoring data within the verification time window and match and verify it with the correction driving data to obtain the verification result.

[0114] The data correction model update module is used to determine whether the verification result has passed. If the verification has not passed, the data correction model is updated.

[0115] In summary, this application provides a method and system for forecasting complex environmental weather based on data fusion, which has the following technical effects:

[0116] 1. It solves the problem of inaccurate weather forecasts caused by the influence of the underlying surface. By fusing the monitored weather data with the local underlying surface data, the weather forecast results are obtained based on the fusion results, thereby improving the accuracy of weather forecasts.

[0117] 2. By constructing a data correction model to correct weather forecast results, and then optimizing the initial correction results again based on the complexity, the accuracy of weather forecast results can be further improved.

[0118] 3. By setting different prediction accuracy for multiple target sub-regions, it is possible to reduce operating costs and avoid resource waste while ensuring the accuracy of weather forecasts.

[0119] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0120] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for forecasting complex environmental weather based on data fusion, characterized in that, include: Collect underlying surface data of the target area, extract features from the underlying surface data, and determine the underlying surface feature set based on the feature extraction results; Based on the underlying surface feature set, the target region is divided into grids to determine multiple target sub-regions; The underlying surface data acquisition module is used to retrieve driving data from the multiple target sub-regions to obtain multiple underlying surface driving datasets. A mapping relationship is constructed between the multiple target sub-regions and the multiple underlying surface driving datasets. Based on the mapping relationship, data influence coefficient analysis is performed to obtain multiple complexities. A data correction model is constructed based on the multiple underlying surface driving datasets and multiple historical monitoring datasets, and the output results of the data correction model are optimized using multiple complexity levels. Real-time monitoring data of the multiple target sub-regions is collected, the real-time monitoring data is input into the data correction model, and correction driving data is output. The correction driving data is sent to the weather forecasting system, and the weather forecast result is output. The process involves constructing a mapping relationship between the multiple target sub-regions and the multiple underlying surface driving datasets, performing data influence coefficient analysis based on the mapping relationship, and obtaining multiple complexities, including: The data collection window retrieves multiple sets of historical weather data for the multiple target sub-regions. Data cleaning is performed on the multiple historical weather datasets to obtain multiple standard historical weather datasets; By traversing the multiple historical weather data sets and the multiple underlying surface driving datasets, abnormal data extraction is performed to obtain an abnormal data set; Based on the abnormal data set, the corresponding weather scene is matched to obtain the scene type; Based on the difference between the abnormal data set and the standard data, a deviation value set is obtained, wherein the deviation value set corresponds one-to-one with the scenario type; The scene type is assigned a value based on the set of deviation values ​​to obtain a set of scene coefficients; The multiple complexities are obtained based on the set of scenario coefficients.

2. The method as described in claim 1, characterized in that, include: The river system feature set and the hydrological feature set are extracted from the underlying surface feature set, wherein the river system feature set includes flow path, flow direction and drainage area, and the hydrological feature set includes flow rate, flood season and water level; The water system distribution area of ​​the target region is determined based on the process, flow direction, and watershed area, and water system area connection identification is performed; The hydrological impact coefficients of the multiple target sub-regions are determined based on the flow rate, flood season, and water level. Based on the hydrological influence coefficient and the river system distribution, the multiple target sub-regions are classified into hydrological levels to obtain the level classification results. The forecast accuracy of the multiple target sub-regions is determined based on the classification results, and the weather forecast is performed on the target region based on the forecast accuracy.

3. The method as described in claim 2, characterized in that, The determination of the hydrological impact coefficients of the multiple target sub-regions based on the flow rate, flood season, and water level includes: n experts are randomly selected from the expert database to score the degree of hydrological impact on multiple target sub-regions based on the flow rate, flood season, and water level, respectively, to obtain multiple score result sets, wherein each of the multiple score result sets corresponds one-to-one with the multiple target sub-regions; The mean of the multiple score result sets is calculated to obtain the mean of multiple scores corresponding to the flow rate, flood season, and water level; Multiple priority graph weight matrices are constructed based on the multiple average scores, and multiple sets of weight percentage values ​​are obtained by weight calculation. The hydrological impact coefficient is obtained based on the set of multiple weight percentage values ​​and the average of the multiple scores.

4. The method as described in claim 1, characterized in that, include: The network structure is built based on the BP neural network. A training dataset is constructed based on the multiple underlying surface driving datasets and multiple historical monitoring datasets; The network structure is trained and validated using the training dataset until the accuracy meets the requirements, thereby obtaining the data correction model; After the real-time monitoring data is input into the data correction model, the initial correction data is output. The initial correction data is optimized based on the multiple complexities to obtain correction driving data.

5. The method as described in claim 1, characterized in that, include: Set the verification time window according to the preset verification rules; The monitoring data within the verification time window is extracted and matched with the correction driving data to obtain the verification result. Determine whether the verification result is successful. If the verification is unsuccessful, update the data correction model.

6. A complex environmental weather forecasting system based on data fusion, characterized in that, include: The underlying surface feature set determination module is used to collect underlying surface data of the target area, extract features from the underlying surface data, and determine the underlying surface feature set based on the feature extraction results. A multiple target sub-region determination module is used to divide the target region into grids based on the underlying surface feature set to determine multiple target sub-regions. A driving data retrieval module is used to retrieve driving data from the multiple target sub-regions using the underlying surface data acquisition module to obtain multiple underlying surface driving datasets. The data impact coefficient analysis module is used to construct the mapping relationship between the multiple target sub-regions and the multiple underlying surface driving datasets, and to perform data impact coefficient analysis based on the mapping relationship to obtain multiple complexities. A data correction model building module is used to build a data correction model based on the multiple underlying surface driving datasets and multiple historical monitoring datasets, and to optimize the output of the data correction model using multiple complexity levels. A correction-driven data output module is used to collect real-time monitoring data of the multiple target sub-regions, input the real-time monitoring data into the data correction model, and output correction-driven data. A weather forecast result output module is used to send the correction driving data to the weather forecast system and output the weather forecast result; The system also includes: A historical weather data set acquisition module is used to acquire multiple historical weather data sets of the multiple target sub-regions within the data retrieval window. A data cleaning module is used to clean the multiple historical weather data sets to obtain standard multiple historical weather data sets. An abnormal data set acquisition module is used to traverse the multiple historical weather data sets and the multiple underlying surface driving datasets to extract abnormal data and obtain an abnormal data set. A scene type acquisition module is used to match the corresponding weather scene based on the abnormal data set to obtain the scene type. A deviation value set acquisition module is used to obtain a deviation value set based on the difference between the abnormal data set and the standard data, wherein the deviation value set corresponds one-to-one with the scenario type; A scene coefficient set acquisition module, wherein the scene data set acquisition module is used to assign values ​​to the scene type according to the deviation value set to obtain the scene coefficient set; Multiple complexity acquisition modules are used to obtain the multiple complexities based on the set of scene coefficients.

Images