A Multi-Source Heterogeneous Meteorological Data Assimilation and Fusion Method

Abstract

The present invention discloses a multi-source heterogeneous meteorological data assimilation and fusion method, which includes the following steps: synchronously acquiring meteorological data from multiple data sources through a distributed data interface and performing preprocessing, where the data sources include satellite remote sensing data, ground meteorological station data, and meteorological radar data; after completing the preprocessing of the meteorological data from multiple data sources, based on the weighted least squares method, respectively fusing the data from multiple data sources into meteorological element values of each meteorological element; assimilating the obtained meteorological element values with a numerical weather prediction model. The present invention has the following advantages and effects: The present invention can accurately identify abnormal data and perform targeted repair, and dynamically adjust the weights of data sources through weighting, effectively avoiding the interference of low-quality data on the fusion result, and then improving the final prediction accuracy by improving the numerical weather prediction model assimilation step.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological data processing, and particularly relates to a method for assimilating and fusing multi-source heterogeneous meteorological data. Background Art

[0002] Meteorological data plays a crucial role in weather forecasting and climate research. Currently, meteorological data mainly comes from different data sources such as ground meteorological stations, satellite remote sensing, meteorological radars, and meteorological soundings. Ground meteorological stations provide local meteorological information with high spatio-temporal resolution, but their coverage is limited and they may be affected by equipment failures. Satellite remote sensing can provide global data, especially in aspects such as atmospheric temperature, humidity, and ocean state, but due to its low spatio-temporal resolution and cloud cover, it may bring inaccurate data. Meteorological radars are used to monitor meteorological dynamics such as precipitation and wind speed in real time, but their data may be interfered by terrain and climate, resulting in incomplete or noisy information. Reference can be made to the method for early warning of sparse spatial wind vector field anomalies based on multi-source heterogeneous spatio-temporal data disclosed in application number CN202411368950.X.

[0003] The differences in spatio-temporal resolution, measurement accuracy, and coverage of the above different data sources have caused inconsistencies and errors in meteorological data. In this context, meteorological data assimilation technology has emerged. The basic goal of data assimilation is to combine observational data from different data sources with numerical weather prediction models and, by adjusting the initial conditions of the models, make the output of the models closer to the actual meteorological conditions, thereby improving the prediction accuracy. Through data assimilation, not only can the numerical weather prediction models be optimized and the prediction errors be corrected in real time, but also the prediction results can be adjusted in a timely manner according to new observational data to improve the accuracy of early warnings.

[0004] However, the following defects still exist in the prior art:

[0005] Firstly, during the preprocessing of meteorological data from multiple data sources, due to equipment failures, environmental factors, etc., data missing or outliers often occur, and the prior art cannot accurately identify and process these abnormal data to ensure the data quality.

[0006] Secondly, in the subsequent fusion and assimilation process, traditional fusion methods such as Kalman filtering are used. There are limitations in the collaborative optimization of heterogeneous data, resulting in the fusion results being dominated by the errors of a single data source. Moreover, during the assimilation process with the numerical weather prediction model after fusion, accuracy also needs to be ensured to reduce prediction errors. Summary of the Invention

[0007] The purpose of the present invention is to provide a method for assimilating and fusing multi-source heterogeneous meteorological data to solve the problems raised in the background art.

[0008] The above technical objectives of the present invention are achieved through the following technical solutions:

[0009] A multi-source heterogeneous meteorological data assimilation and fusion method includes the following steps:

[0010] S100, synchronously acquiring meteorological data from multiple data sources through a distributed data interface and performing preprocessing, the data sources including satellite remote sensing data, ground meteorological station data, and meteorological radar data;

[0011] S200, after completing the preprocessing of the meteorological data from multiple data sources in S100, based on the weighted least squares method, respectively fuse the data from the multiple data sources into meteorological element values ​​of each meteorological element;

[0012] S300, assimilating the meteorological element values ​​obtained by fusion in S200 with the numerical weather forecast model;

[0013] S400, in conjunction with S100, S200 and S300, can dynamically optimize numerical weather forecast model parameters and generate executable weather warning instructions;

[0014] Among them, S200 is specifically:

[0015] Multiply the weights of data source type, spatiotemporal proximity, and data quality item by item to generate a comprehensive weight matrix;

[0016] The data of each meteorological element will correspond to a comprehensive weight matrix, and the fused meteorological element value will be obtained based on the weighted least squares method.

[0017] By adopting the above technical scheme, the present invention can effectively integrate data from different sources through the assimilation and fusion of multi-source heterogeneous meteorological data, and make up for the shortcomings of a single data source; by fusing multiple data sources through the weighted least squares method, the differences between different data sources can be eliminated, and the fusion results of the same meteorological element can be guaranteed to have higher accuracy and consistency. At the same time, through the weighted fusion of multiple data sources, even if a data source is missing, the data of other data sources can still supplement the relevant information, thereby ensuring the integrity of the final fused data, and finally improving the final prediction accuracy by improving the assimilation step of the numerical weather forecast model.

[0018] It is further configured that the preprocessing process in S100 is:

[0019] S101, constructing a meteorological element threshold rule base, wherein the meteorological element threshold rule base includes the minimum and maximum values ​​of each meteorological element to preliminarily filter out abnormal data;

[0020] S102. Perform quality inspection on the data points within each meteorological element, and further screen out abnormal data. Specifically, adopt a sliding time window mechanism to dynamically detect the data points within each meteorological element, and based on the window length and step size corresponding to each meteorological element, calculate the corresponding local outlier factor according to all the data points within each window. If the local outlier factor is greater than the set threshold, all the data points within that window are determined to be abnormal data.

[0021] Among them, the sliding time window mechanism can automatically adjust the window length and step size according to the volatility of each meteorological element. Specifically: calculate the standard deviation and mean of each data point within the current window of each meteorological element, and then divide the calculated standard deviation by the mean to obtain the coefficient of variation. When the coefficient of variation is higher than the set maximum threshold, indicating that the data changes violently, shorten the window length; when the coefficient of variation is lower than the set minimum threshold, indicating that the data changes gently, extend the window length.

[0022] Extract the values of the data points at the current moment and the previous moment in each meteorological element, and divide them by the time interval between the current moment and the previous moment to obtain the change rate at the current moment. When the change rate at the current moment is greater than the set threshold, shorten the step size to more accurately capture the rapidly changing trend; when the change rate at the current moment is less than the set threshold, extend the step size to reduce the calculation amount.

[0023] S103. Repair the screened abnormal data.

[0024] S104. Perform normalization processing on each meteorological element one by one to map different meteorological elements to the same numerical range.

[0025] By adopting the above technical solutions, by constructing a meteorological element threshold rule base and performing quality inspection and repair, these abnormal data can be initially screened out, thereby improving the reliability of the data. Then, through quality inspection and repair, it is ensured that all data are within the effective numerical range, avoiding the impact of abnormal data on the assimilation process, thereby improving the quality of the final fused data. Finally, through the normalization processing of each meteorological element, data from different sources can be compared and fused within a unified numerical range. Through the above steps, the process of data integration is greatly simplified, and the compatibility between data is ensured.

[0026] Through the sliding time window mechanism, abnormal data caused by external interference or equipment failure can be detected in real time, and the abnormality of the data can be identified by calculating the local outlier factor. It has strong dynamic adaptability and can effectively handle situations where the data changes greatly.

[0027] It can automatically adjust the size of the sliding window according to the volatility of the data, which helps to more accurately capture rapidly changing meteorological data, such as the changes within a short period of extreme weather events; this adjustment can flexibly respond to various meteorological changes, improving the accuracy and real-time performance of data processing. Specifically, when the meteorological data changes violently, by calculating the coefficient of variation and the change rate to adjust the window length and step size, it is possible to more accurately capture rapidly changing meteorological elements, which is of great significance for short-term severe weather events in weather forecasting; when the data changes little or is stable, extending the window length and step size can reduce the computational amount and improve the processing efficiency; this dynamic adjustment mechanism can operate efficiently in different scenarios, avoiding unnecessary computational burdens.

[0028] A further setting is that S103 is specifically:

[0029] If the abnormal data is ground meteorological station data; use the Delaunay triangulation algorithm to perform spatial triangulation on the ground stations. Each station is connected to other stations to form triangle sides, and then determine the adjacency relationship between the stations of the abnormal ground meteorological station and the stations of other ground meteorological stations. Then select the 5 stations of ground meteorological stations that are closest to the station of the abnormal ground meteorological station; for the selected 5 stations of ground meteorological stations, use the weighted average method to replace the abnormal data, and determine the weight according to the distance between the stations of these ground meteorological stations and the station of the abnormal ground meteorological station. The weight is inversely proportional to the distance; calculate the weighted average value according to the weights of the selected 5 stations of ground meteorological stations to replace the abnormal data;

[0030] If the abnormal data is satellite remote sensing data or meteorological radar data, then use the data points within 3 time instants before and after in the time dimension for time interpolation repair.

[0031] By adopting the above technical solution, repairing the ground meteorological station data through the Delaunay triangulation algorithm can find the most suitable replacement value within the neighborhood of the abnormal data, thus ensuring that the repaired data is as close as possible to the true observed value and avoiding error accumulation during the data repair process; by time interpolation to repair the missing parts of satellite remote sensing data and meteorological radar data, it can effectively fill the data gap, reduce the impact of data missing on the final fusion result, and ensure the integrity and high quality of the data.

[0032] A further setting is that S300 is specifically:

[0033] S301. Use bicubic interpolation to upgrade the low-resolution data of the numerical weather prediction model to the same resolution as the meteorological data, so that the meteorological data and the data of the numerical weather prediction model can be compared on the same spatial grid;

[0034] S302. Calculate the difference between the predicted value of the numerical weather prediction model and the meteorological element value.

[0035] By adopting the above technical solution, by assimilating the fused meteorological element values with the numerical weather prediction model, the initial conditions of the model can be corrected, significantly reducing the prediction error. Especially in extreme weather prediction, data assimilation can effectively improve the accuracy of the forecast; the assimilation process can adjust the model prediction according to the actual observed data, reduce the error introduced by the model assumptions, and enhance the stability and reliability of the forecast results.

[0036] A further setting is that S302 is specifically as follows:

[0037] S3021. For the same meteorological element at each grid point, calculate the direct difference between the predicted value and the meteorological element value.

[0038] S3022. Calculate the average value of the differences of all grid points to evaluate the systematic bias of the numerical weather prediction model; then calculate the root mean square error to measure the dispersion degree between the predicted value and the meteorological element value, reflecting the uncertainty of the numerical weather prediction model.

[0039] S3023. Combine the generated comprehensive weight matrix to adjust the difference of each grid point according to the weight of the data source, avoiding the interference of low-quality data on the difference calculation.

[0040] By adopting the above technical solution, by calculating the difference between the predicted value and the meteorological element value, the error of the numerical weather prediction model can be quantified, and the bias of the model prediction can be evaluated; the difference analysis and error evaluation can accurately identify which parts of the forecast have biases, so as to improve the model and forecast algorithm targeted, and further improve the accuracy of the forecast.

[0041] A further setting is that after S3023, the following steps are also included:

[0042] S3024. Visualize the difference result in the form of a spatial heat map, where red represents the positive deviation where the predicted value is higher than the meteorological element value, and blue represents the negative deviation where the predicted value is lower than the meteorological element value, facilitating the intuitive positioning of the deviation concentration area and marking the grid points exceeding the preset threshold.

[0043] By adopting the above technical solution, by visualizing the difference result in the form of a spatial heat map, the deviation between the predicted value and the actual observed value can be intuitively presented, helping meteorologists and forecasters quickly locate the area where the deviation is concentrated and optimize the subsequent forecasting work.

[0044] A further setting is that S301 is specifically as follows:

[0045] S3011. Obtain the original grid resolution of the numerical weather prediction model and the actual grid resolution of the meteorological data, and calculate the proportionality coefficient between the two:

[0046] Based on the original row resolution and original column resolution of the numerical weather prediction model, and the actual row resolution and actual column resolution of the meteorological data, calculate the resolution proportionality coefficients in the row direction and column direction respectively; the row direction proportionality coefficient is equal to the meteorological data row resolution divided by the original model row resolution, and the column direction proportionality coefficient is equal to the meteorological data column resolution divided by the original model column resolution;

[0047] S3012. Dynamically adjust the kernel function parameters of the bicubic interpolation algorithm based on the proportionality coefficient:

[0048] According to the larger value of the row direction proportionality coefficient and the column direction proportionality coefficient, set the sampling radius of the kernel function to 1.5 times of the larger value, and adaptively adjust the smoothness coefficient in combination with the spatial distribution density of the meteorological data; when the meteorological data density is higher than the preset threshold, the smoothness coefficient is reduced to retain the detailed features; when the meteorological data density is lower than the preset threshold, the smoothness coefficient is increased to enhance the data smoothness;

[0049] S3013. Perform layer-by-layer interpolation processing on the low-resolution data of the numerical weather prediction model:

[0050] Adopt a hierarchical interpolation strategy to divide the original low-resolution data into a coarse-grained layer and a fine-grained layer:

[0051] For the interpolation operation of the coarse-grained layer, perform preliminary interpolation on the original grid based on the bicubic interpolation algorithm to generate intermediate-resolution data;

[0052] For the interpolation operation of the fine-grained layer, on the basis of the intermediate-resolution data, dynamically adjust the interpolation step according to the local gradient change of the meteorological data, and finally generate high-resolution grid data consistent with the meteorological data resolution;

[0053] S3014. Verify the spatial coordinate consistency between the interpolated high-resolution grid data and the meteorological data:

[0054] Randomly select a preset number of key grid points from the meteorological data as reference points;

[0055] Calculate the coordinate offset between the interpolated data and the reference points. If the offset exceeds the preset maximum threshold, it is determined as a coordinate deviation;

[0056] If more than 50% of the reference points have coordinate deviations, readjust the kernel function parameters and perform interpolation; if the deviation ratio is less than 50%, only perform local correction on the deviation area;

[0057] S3015. Match and align the verified high-resolution grid data with the meteorological data point by point according to the spatial coordinates. If there are uncovered areas, use the nearest neighbor interpolation method to supplement the missing values to ensure the integrity and consistency of the fused grid data.

[0058] Furthermore, it is set that in S3012, adaptively adjusting the smoothness coefficient in combination with the spatial distribution density of the meteorological data is specifically as follows: If the density of the meteorological data shows a sudden change in a local area, adjust the smoothness coefficient according to the density gradient. When the density gradient is greater than the preset threshold, the smoothness coefficient is reduced to 70% of the original value to suppress over-smoothing of the interpolation. When the density gradient is less than the preset threshold, the smoothness coefficient is increased to 130% of the original value to eliminate the interference of data noise.

[0059] In the above-mentioned S3013, when performing the fine-grained layer interpolation operation, if the change trends of the meteorological elements in adjacent grids are consistent, use linear interpolation. If the change trends are significantly different, switch to the non-linear interpolation algorithm to avoid interpolation distortion.

[0060] Furthermore, it is set that S400 is specifically as follows:

[0061] S401. Conduct a traceability analysis on the marked grid points exceeding the preset threshold, extract the corresponding data source types, spatio-temporal resolution characteristics, and abnormal fluctuation records of the comprehensive weight matrix, and generate an abnormal traceability report.

[0062] S402. Based on the abnormal traceability report, dynamically adjust the empirical parameter weights for the meteorological element model parameters associated with the marked grid points in the numerical weather prediction model according to the deviation direction and amplitude, including the following sub-steps:

[0063] S4021. According to the type of meteorological element of the marked grid point, extract the corresponding set of empirical parameters from the numerical weather prediction model, and associate the predicted value calculated in S3021 with the meteorological element value after fusion in S200 to establish a mapping relationship table between the parameters and the deviation.

[0064] S4022. Based on the direct difference between the predicted value and the meteorological element value of the same grid point in S3021, determine the deviation direction. Specifically:

[0065] S4021.1. Obtain the direct difference between the predicted value and the meteorological element value of the same grid point in S3021, and detect the sign consistency of the difference in the continuous time series based on the preset time window. If more than 80% of the differences are positive within the same window, trigger a positive deviation determination. If more than 80% of the differences are negative, trigger a negative deviation determination.

[0066] S4021.2. If the positive deviation determination is triggered, confirm that the predicted value of the numerical weather prediction model is continuously higher than the fused meteorological element value, and generate a command to reduce the weight of the gain type parameter; if the negative deviation determination is triggered, confirm that the predicted value is continuously lower than the meteorological element value, and generate a command to increase the weight of the compensation type parameter.

[0067] S4021.3. According to the confirmation result of the deviation direction in S4021.2, add a dynamic mark to the associated meteorological element model parameter in the model parameter correction coefficient table. The mark types include a weight decay factor or a weight enhancement factor, and the weight decay factor and the weight enhancement factor are in a proportional relationship with the average value of all grid point differences calculated in S3022.

[0068] S4021.4. Perform regional association expansion. If the same meteorological element triggers the same deviation direction determination at 3 or more adjacent grid points, enhance the regionalization of the associated parameter weight adjustment command, expand the single-point adjustment amount to the regional range, and the adjustment amplitude decays with the grid distance.

[0069] S4023. Based on the average value of all grid point differences calculated in S3022, combined with the preset error reference value, calculate the parameter weight adjustment amount:

[0070] If the predicted value is higher than the meteorological element value, assign a negative direction coefficient and proportionally reduce the corresponding parameter weight; if the predicted value is lower than the meteorological element value, assign a positive direction coefficient and proportionally increase the corresponding parameter weight.

[0071] Divide the average value by the preset error reference value to obtain the normalized adjustment ratio.

[0072] Multiply the direction coefficient by the normalized adjustment ratio to obtain the final adjustment amount of the parameter weight, ensuring that the adjustment amplitude is linearly related to the deviation degree.

[0073] S4024. Perform smoothing constraint to limit the single-time parameter weight adjustment amplitude not to exceed 10% of the original weight. If the calculated value exceeds the range, perform threshold truncation processing to ensure the stability of the model.

[0074] S403. Use historical assimilation data to train a lightweight neural network, input the adjusted difference in S3023 into the network, predict the short-term meteorological element change trend, and feedback the prediction result back to the weight allocation module in S200 to update the comprehensive weight matrix.

[0075] Further set that the specific S403 is as follows:

[0076] S4031. Extract the adjusted difference, the generated comprehensive weight matrix, and the corresponding timestamps from the historical assimilation data, and generate labeled time-series training samples according to the sliding time-window mechanism. The label is the actual change value of meteorological elements in the next hour.

[0077] S4032. Construct a lightweight neural network, including an input layer, two convolutional layers, a single long short-term memory unit, and a fully connected output layer. The convolution kernel size is set to 3×3, and the stride matches the meteorological grid resolution.

[0078] S4033. Adopt an adaptive learning rate mechanism. The initial learning rate is set to 0.001. If the loss of the validation set does not decrease for three consecutive training cycles, the learning rate decays to 50% of the original value. At the same time, introduce an early stopping mechanism to prevent overfitting.

[0079] S4034. Input the adjusted difference of S3023 at the current moment into the trained network, output the predicted value of the change trend of meteorological elements in the next hour, and combine this predicted value with the real-time stability index of the data source in S200 to generate a weight correction factor.

[0080] S4035. Dynamically adjust the comprehensive weight matrix based on the weight correction factor generated in S4034. Specifically: increase the weight of the data source with high consistency between the predicted trend and the real-time data, and decrease the weight of the data source with low consistency. The updated weight matrix is directly applied to the next data fusion.

[0081] In summary, the present invention has the following beneficial effects:

[0082] The present invention can accurately identify abnormal data and perform targeted repair, and dynamically adjust the weights of data sources through weighting, effectively avoiding the interference of low-quality data on the fusion result. Then, by improving the assimilation step of the numerical weather prediction model, the final prediction accuracy is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 It is a schematic flowchart of the embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0084] The present invention will be further described in detail below with reference to the accompanying drawings.

[0085] As shown in the Figure 1 accompanying drawings;

[0086] This embodiment discloses a multi-source heterogeneous meteorological data assimilation and fusion method, including the following steps:

[0087] S100. Synchronously obtain meteorological data from multiple data sources through a distributed data interface and perform preprocessing. The data sources include satellite remote sensing data, ground meteorological station data, and meteorological radar data.

[0088] S200, after completing the preprocessing of the meteorological data from multiple data sources in S100, based on the weighted least squares method, respectively fuse the data from the multiple data sources into meteorological element values ​​of each meteorological element;

[0089] S300, assimilating the meteorological element values ​​obtained by fusion in S200 with the numerical weather forecast model.

[0090] S400, in conjunction with S100, S200 and S300, can dynamically optimize numerical weather forecast model parameters and generate executable meteorological warning instructions.

[0091] Among them, the pre-processing process in S100 is:

[0092] S101. Construct a meteorological element threshold rule base, which includes the minimum and maximum values ​​of each meteorological element to preliminarily screen out abnormal data; define the effective range of temperature as -90℃ to 60℃, which covers common climate conditions from extreme cold to tropical high temperatures; set the effective range of air pressure to 500hPa to 1100hPa to adapt to changes in air pressure at different altitudes; define the effective range of relative humidity to be 0% to 100%, representing humidity conditions from completely dry to saturated; set the effective range of wind speed to 0m / s to 75m / s to adapt to regular weather conditions and special meteorological events, such as typhoons or tornadoes.

[0093] S102, performing quality inspection on the data points in each meteorological element, and further screening out abnormal data;

[0094] S103, repairing the screened abnormal data;

[0095] S104, normalizing each meteorological element one by one to map different meteorological elements to the same numerical range; the temperature is normalized by Min-Max, the air pressure is logarithmically transformed and then normalized by Min-Max, the humidity is normalized by linear scaling, and the wind speed is first de-skewed and then normalized by Min-Max.

[0096] Among them, S102 is specifically:

[0097] A sliding time window mechanism is adopted to dynamically detect the data points in each meteorological element, and based on the window length and step size corresponding to each meteorological element, the corresponding local outlier factor is calculated according to all the data points in each window. If the local outlier factor is greater than the set threshold of 0.5, all data points in the window are judged as abnormal data.

[0098] Among them, the sliding time window mechanism can automatically adjust the window length and step size according to the volatility of each meteorological element; specifically:

[0099] Calculate the standard deviation and mean of each data point within the current window for each meteorological element. Then, divide the calculated standard deviation by the mean to obtain the coefficient of variation. When the coefficient of variation is higher than the set maximum threshold, it indicates that the data changes violently, and the window length is shortened; when the coefficient of variation is lower than the set minimum threshold, it indicates that the data changes gently, and the window length is extended.

[0100] The formula for calculating the standard deviation is:

[0101] ;

[0102] where σ t is the standard deviation, x i is the value of the i-th data point within the current sliding window, μ is the mean of each data point within the current window, and N represents the number of data points contained within the current window;

[0103] The formula for calculating the coefficient of variation is:

[0104] CV t =σ t / μ;

[0105] where CV t represents the coefficient of variation;

[0106] The formula for dynamically adjusting the window length is defined as:

[0107] T t+1 =max(T min ,min(T max ,T t +α(β - CV t )));

[0108] where T t+1 is the new window length, T t is the current window length, T min and T max are respectively the minimum value of the window length, 15 minutes, and the maximum value, 60 minutes, α is the adjustment rate, and β is the expected value;

[0109] After calculating the standard deviation and coefficient of variation of each data point within the current window, compare the coefficient of variation with the set maximum threshold of 0.1 and minimum threshold of 0.05;

[0110] If CV t >0.1, the window length is reduced to T t+1 =T t -α(CV t-0.1);

[0111] If CV t < 0.05, the window length is increased to T t+1 = T t + α(0.05 - CV t ).

[0112] Extract the values of the data points of each meteorological element at the current time and the previous time, and divide them by the time interval between the current time and the previous time to obtain the change rate at the current time. When the change rate at the current time is greater than the set threshold, shorten the step size to more accurately capture the rapidly changing trend; when the change rate at the current time is less than the set threshold, extend the step size to reduce the computational amount.

[0113] The formula for calculating the change rate is:

[0114] ;

[0115] where Δx t represents the change rate at the current time t, x t represents the value of the data point at the current time t, x t-1 represents the value of the data point at the previous time t - 1, and Δt represents the time interval;

[0116] The formula for dynamically adjusting the step size is defined as:

[0117] Δt t+1 = max(Δt min , min(Δt max , Δt t + γ(δ - Δx t )));

[0118] where Δt t+1 is the step size for the next calculation, Δt min and Δt max are the minimum step size of 1 minute and the maximum step size of 15 minutes respectively, γ is the adjustment rate, and δ is the desired change rate threshold;

[0119] If Δx t > 0.1, the step size is shortened to Δt t+1 = Δt t - γ(Δx t - 0.1);

[0120] If Δx t < 0.05, the step size is extended to Δt t+1 = Δt t + γ(0.05 - Δx t ).

[0121] Among them, S103 is specifically as follows:

[0122] If the abnormal data is ground meteorological station data; use the Delaunay triangulation algorithm to perform spatial triangulation on the ground stations. Each station is connected to other stations to form triangular sides, and then determine the adjacency relationship between the stations of the abnormal ground meteorological station and the stations of other ground meteorological stations. Then select the 5 stations of the ground meteorological stations that are closest to the station of the abnormal ground meteorological station; for the selected 5 stations of the ground meteorological stations, use the weighted average method to replace the abnormal data, and determine the weight according to the distance between these stations of the ground meteorological stations and the station of the abnormal ground meteorological station. The weight is inversely proportional to the distance; calculate the weighted average value according to the weights of the selected 5 stations of the ground meteorological stations to replace the abnormal data.

[0123] The specific process is shown in the following formula:

[0124] ;

[0125] Among them, is the interpolated abnormal data, y j is the data of the jth ground meteorological station, w j is the weight of the jth ground meteorological station.

[0126] If the abnormal data is satellite remote sensing data or meteorological radar data, then use the data points within 3 time instants before and after in the time dimension for time interpolation repair.

[0127] Assume that the current time is t, then select the data points at times t - 3, t - 2, t - 1, t + 1, t + 2, t + 3.

[0128] Among them, S200 is specifically as follows:

[0129] S201. Since the spatio-temporal resolution and error characteristics of satellite remote sensing, ground meteorological station, and meteorological radar data are significantly different, it is necessary to assign dynamic weights to each data source and meteorological element to balance the contributions;

[0130] S202. Multiply the weights of the three aspects of data source type, spatio-temporal proximity, and data quality item by item to generate a comprehensive weight matrix;

[0131] The comprehensive weight matrix W k The formula is:

[0132] W k =W source ⊙W time ⊙Q quality；

[0133] Among them, W source represents the data source type weight matrix, W timeRepresents the spatio-temporal proximity weight matrix, W space Represents the data quality weight matrix, and ⊙ represents element-wise multiplication.

[0134] S203. Based on S201 and S202, each meteorological element data will correspond to a comprehensive weight matrix, and the fused meteorological element value is obtained based on the weighted least squares method;

[0135] The process is shown as follows:

[0136] ;

[0137] Where is the fused meteorological element value, a i is the data of the i-th data source, is the weight of the i-th data source.

[0138] Among them, S201 is specifically as follows:

[0139] Allocate weights based on the data source type;

[0140] For the surface meteorological station data, the highest basic weight of 0.6 is assigned at the corresponding geographical location;

[0141] For satellite remote sensing data, the weight of the wide-area coverage area is increased to 0.5, and the weight of the dense area of ground stations is reduced to 0.3 to avoid conflict with the surface meteorological station data;

[0142] For meteorological radar data, the weight is dynamically adjusted according to the scene change; for example, in the case of precipitation and sudden change of wind speed during short-term severe weather changes, the weight is increased to 0.7.

[0143] Allocate weights based on spatio-temporal proximity; for the target fusion time, the weight is dynamically adjusted according to the time difference; the weight of the data within ±15 minutes of the target fusion time is 1.0, and it decays to 0.3 if it exceeds 1 hour;

[0144] Allocate weights based on data quality; for the abnormal data repaired in S103, the weight is reduced proportionally according to the confidence of the repair algorithm; the confidence of the Delaunay triangulation algorithm is 0.8, and the confidence of time interpolation is 0.6. If a data point is repeatedly determined as abnormal data, specifically manifested as the local outlier factor > 0.5 in S102 more than 3 times, the weight is directly assigned to 0 to avoid contaminating the fusion result.

[0145] Among them, S300 is specifically as follows:

[0146] S301. The low-resolution data of the numerical weather prediction model is upsampled to the same resolution as the meteorological data through bicubic interpolation, so that the meteorological data and the data of the numerical weather prediction model can be compared on the same spatial grid;

[0147] S302. Calculate the difference between the predicted value of the numerical weather prediction model and the meteorological element value.

[0148] Specifically, S302 is as follows:

[0149] S3021. For the same meteorological element at each grid point, calculate the direct difference between the predicted value and the meteorological element value;

[0150] ;

[0151] Among them, ΔE(x,y,t) is the deviation between the predicted value and the meteorological element value, and E NWP is the predicted value of the numerical weather prediction model, is the specific value of the meteorological element value at the spatial coordinates (x,y) and time t;

[0152] Through the above formula, the prediction deviation of the local area can be directly reflected. For example, the temperature in a certain area is overestimated by 2°C.

[0153] S3022. Calculate the average value of the differences of all grid points to evaluate the systematic deviation of the numerical weather prediction model; then calculate the root mean square error to measure the dispersion degree between the predicted value and the meteorological element value, and reflect the uncertainty of the numerical weather prediction model;

[0154] The process is shown in the following formula:

[0155] ;

[0156] Among them, represents the average value of the differences of all grid points, and N represents the total number of all spatial grid points;

[0157] ;

[0158] Among them, RMSE represents the root mean square error, and N represents the total number of all spatial grid points.

[0159] S3023. Combine the comprehensive weight matrix generated in S202 to adjust the difference of each grid point according to the weight of the data source, and avoid the interference of low-quality data on the difference calculation.

[0160] ;

[0161] Among them, W k is the comprehensive weight matrix generated in S202, corresponding to the spatial coordinates (x,y) and time t;

[0162] Specifically, after S3023, the following steps are also available:

[0163] S3024. Visualize the difference result in the form of a spatial heat map. Red represents the positive deviation where the predicted value is higher than the meteorological element value, and blue represents the negative deviation where the predicted value is lower than the meteorological element value, which is convenient for intuitively locating the concentrated area of deviation and marking the grid points that exceed the preset threshold.

[0164] Among them, S301 specifically is:

[0165] S3011. Obtain the original grid resolution of the numerical weather prediction model and the actual grid resolution of the meteorological data, and calculate the proportionality coefficient between the two:

[0166] Based on the original row resolution and original column resolution of the numerical weather prediction model, and the actual row resolution and actual column resolution of the meteorological data, calculate the resolution proportionality coefficients in the row direction and column direction respectively; the row direction proportionality coefficient is equal to the meteorological data row resolution divided by the original model row resolution, and the column direction proportionality coefficient is equal to the meteorological data column resolution divided by the original model column resolution.

[0167] S3012. Dynamically adjust the kernel function parameters of the bicubic interpolation algorithm based on the proportionality coefficient:

[0168] According to the larger value among the row direction proportionality coefficient and the column direction proportionality coefficient, set the sampling radius of the kernel function to 1.5 times of the larger value, and adaptively adjust the smoothness coefficient in combination with the spatial distribution density of the meteorological data; when the meteorological data density is higher than the preset threshold, the smoothness coefficient is reduced to retain the detailed features; when the meteorological data density is lower than the preset threshold, the smoothness coefficient is increased to enhance the data smoothness.

[0169] S3013. Perform layer-by-layer interpolation processing on the low-resolution data of the numerical weather prediction model:

[0170] Adopt a hierarchical interpolation strategy to divide the original low-resolution data into a coarse-grained layer and a fine-grained layer:

[0171] The operation for the coarse-grained layer interpolation is to perform preliminary interpolation on the original grid based on the bicubic interpolation algorithm to generate intermediate-resolution data;

[0172] The operation for the fine-grained layer interpolation is to dynamically adjust the interpolation step size according to the local gradient change of the meteorological data on the basis of the intermediate-resolution data, and finally generate high-resolution grid data with the same resolution as the meteorological data;

[0173] S3014. Verify the spatial coordinate consistency between the interpolated high-resolution grid data and the meteorological data:

[0174] Randomly select a preset number of key grid points from the meteorological data as reference points;

[0175] Calculate the coordinate offset between the interpolated data and the reference points. If the offset exceeds the preset maximum threshold, it is determined as a coordinate deviation.

[0176] If coordinate deviations exist in more than 50% of the reference points, re-adjust the kernel function parameters and perform interpolation. If the deviation ratio is lower than 50%, only perform local correction on the deviation area.

[0177] S3015. Match and align the verified high-resolution grid data and meteorological data point by point according to the spatial coordinates. If there are uncovered areas, use the nearest neighbor interpolation method to supplement the missing values to ensure the integrity and consistency of the fused grid data.

[0178] Among them, in S3012, adaptively adjusting the smoothness coefficient in combination with the spatial distribution density of meteorological data is specifically as follows: If the density of meteorological data shows a sudden change in a local area, adjust the smoothness coefficient according to the density gradient. When the density gradient is greater than the preset threshold, the smoothness coefficient is reduced to 70% of the original value to suppress over-smoothing of interpolation. When the density gradient is less than the preset threshold, the smoothness coefficient is increased to 130% of the original value to eliminate data noise interference.

[0179] In the aforementioned S3013, when performing the fine-grained layer interpolation operation, if the change trends of meteorological elements in adjacent grids are consistent, use linear interpolation. If the change trends are significantly different, switch to a non-linear interpolation algorithm to avoid interpolation distortion.

[0180] Among them, S400 is specifically as follows:

[0181] S401. Conduct a traceability analysis on the marked grid points exceeding the preset threshold, extract the corresponding data source types, spatio-temporal resolution characteristics, and records of abnormal fluctuations in the comprehensive weight matrix, and generate an abnormal traceability report.

[0182] S402. Based on the abnormal traceability report, dynamically adjust the empirical parameter weights for the meteorological element model parameters associated with the marked grid points in the numerical weather prediction model according to the deviation direction and amplitude, including the following sub-steps:

[0183] S4021. According to the type of meteorological element of the marked grid point, extract the corresponding set of empirical parameters from the numerical weather prediction model, and associate the predicted value calculated in S3021 with the meteorological element value after fusion in S200 to establish a mapping relationship table between the parameters and the deviation.

[0184] S4022. Based on the direct difference between the predicted value and the meteorological element value of the same grid point in S3021, determine the deviation direction. Specifically:

[0185] S4021.1. Obtain the direct difference between the predicted value and the meteorological element value at the same grid point in S3021, and detect the sign consistency of the difference in the continuous time series based on a preset time window; if more than 80% of the differences are positive within the same window, trigger a positive deviation determination; if more than 80% of the differences are negative, trigger a negative deviation determination;

[0186] S4021.2. If a positive deviation determination is triggered, confirm that the predicted value of the numerical weather prediction model is continuously higher than the fused meteorological element value, and generate an instruction to reduce the weight of the gain type parameter; if a negative deviation determination is triggered, confirm that the predicted value is continuously lower than the meteorological element value, and generate an instruction to increase the weight of the compensation type parameter;

[0187] S4021.3. According to the confirmation result of the deviation direction in S4021.2, add a dynamic mark to the associated meteorological element model parameter in the model parameter correction coefficient table. The mark types include a weight decay factor or a weight enhancement factor, and the weight decay factor and the weight enhancement factor are in a proportional relationship with the average value of all grid point differences calculated in S3022;

[0188] S4021.4. Perform regional association expansion. If the same meteorological element triggers the same deviation direction determination at 3 or more adjacent grid points, enhance the regionalization of the associated parameter weight adjustment instruction, expand the single-point adjustment amount to the regional range, and the adjustment amplitude decays with the grid distance;

[0189] S4023. Based on the average value of all grid point differences calculated in S3022, combined with a preset error reference value, calculate the parameter weight adjustment amount:

[0190] If the predicted value is higher than the meteorological element value, assign a negative direction coefficient and reduce the corresponding parameter weight proportionally; if the predicted value is lower than the meteorological element value, assign a positive direction coefficient and increase the corresponding parameter weight proportionally;

[0191] Divide the average value by the preset error reference value to obtain the normalized adjustment ratio;

[0192] Multiply the direction coefficient by the normalized adjustment ratio to obtain the final adjustment amount of the parameter weight, ensuring that the adjustment amplitude is linearly related to the deviation degree;

[0193] S4024. Perform smoothing constraints to limit the single-time parameter weight adjustment amplitude not to exceed 10% of the original weight. If the calculated value exceeds the range, perform threshold truncation processing to ensure the stability of the model;

[0194] S403. Train a lightweight neural network using historical assimilation data, input the adjusted difference in S3023 into the network, predict the short-term meteorological element change trend, and feedback the prediction result back to the weight allocation module in S200 to update the comprehensive weight matrix in S202.

[0195] Among them, S403 is specifically as follows:

[0196] S4031. Extract the adjusted difference in S3023, the comprehensive weight matrix generated by S202, and the corresponding timestamps from the historical assimilation data, and generate labeled time series training samples according to the sliding time window mechanism. The label is the actual change value of the meteorological element in the next hour.

[0197] S4032. Construct a lightweight neural network, including an input layer, two convolutional layers, a single long short-term memory unit, and a fully connected output layer. The convolution kernel size is set to 3×3, and the stride matches the meteorological grid resolution.

[0198] S4033. Adopt an adaptive learning rate mechanism. The initial learning rate is set to 0.001. If the validation set loss does not decrease for three consecutive training cycles, the learning rate is decayed to 50% of the original value. At the same time, introduce an early stopping mechanism to prevent overfitting.

[0199] S4034. Input the adjusted difference in S3023 at the current moment into the trained network, output the predicted value of the meteorological element change trend in the next hour, and combine this predicted value with the real-time stability index of the data source in S200 to generate a weight correction factor.

[0200] S4035. Dynamically adjust the comprehensive weight matrix in S202 based on the weight correction factor generated in S4034. Specifically: increase the weight of the data source with high consistency between the predicted trend and the real-time data, and decrease the weight of the data source with low consistency. The updated weight matrix is directly applied to the next data fusion.

[0201] Application Example 1

[0202] Generate a virtual typhoon based on historical typhoon parameters, with the maximum wind speed at the center simulated as 60 m / s and the path simulated as moving towards the southeast coast of China; use WRF to generate simulated satellite, ground station, and radar data.

[0203] Inject abnormal data into S100 to simulate real-world interference. Specifically:

[0204] The instantaneous wind speed at a certain ground meteorological station reaches 80 m / s.

[0205] Inject a wind speed of 70 m / s in the outer region of the typhoon. This value is within the effective range of wind speed from 0 m / s to 75 m / s for constructing the meteorological element threshold rule base, but this value significantly deviates from the average wind speed of 50 m / s at surrounding stations in the local window.

[0206] The wind speed of a certain surface meteorological station suddenly increases from 50 m / s to 70 m / s within 10 minutes, and the change rate , is greater than the threshold of 0.1 to trigger dynamic step adjustment.

[0207] In S101, the instantaneous wind speed of a certain surface meteorological station reaches 80 m / s, which significantly exceeds the threshold of 75 m / s and is determined as abnormal data; the wind speed data of 70 m / s does not exceed the threshold of 75 m / s, but needs to be further detected through S102.

[0208] In S102, the coefficient of variation CV t = 0.15, which is greater than the highest threshold of 0.1, and the window length is shortened from 30 minutes to 15 minutes; within the shortened window, the local outlier factor LOF of the wind speed of 70 m / s is 0.62, which is greater than the set threshold of 0.5, and is determined as abnormal data.

[0209] In S103, for the abnormal data in the surface meteorological station, 5 adjacent stations are selected through the Delaunay triangulation algorithm, and the weighted average repair value is 53 m / s. The error is reduced from 70 m / s to 53 m / s. Compared with the simulated true value of 55 m / s, the error < 4%.

[0210] In S200, the weight distribution is as follows: in the typhoon core area, that is, within the simulated radius of 100 km, the radar weight is increased to 0.7, and the satellite weight is reduced to 0.3. Compared with the true radar weight of 0.72 and the true satellite weight of 0.28 in the simulation, the error < 5%.

[0211] The comparison of fusion accuracy is as follows:

[0212] The root mean square error of wind speed for single radar data is 6.8 m / s, and the root mean square error of the wind speed of the fused meteorological element values is 3.2 m / s; the accuracy of wind speed prediction has been significantly improved.

[0213] After assimilation in S300, the 24-hour path error of the typhoon is reduced from 65 km to 28 km. By assimilating the fused meteorological element values into the numerical weather prediction model, the accuracy of the model is significantly improved and the prediction error is reduced.

[0214] Application Example 2

[0215] Simulate a cold wave process, covering the mid-latitude region, with the average temperature dropping suddenly by 15°C within 24 hours, local wind speed reaching 20 m / s, and accompanied by heavy snow over a large area; use WRF to generate simulation data and inject data anomalies.

[0216] In S100, a ground meteorological station reports an extreme low temperature due to a sensor failure at a site, exceeding the actual value by 12, triggering the dynamic detection mechanism; in addition, the temperature at a certain ground meteorological station drops suddenly within 10 minutes, and the change rate exceeds the threshold of 0.1, triggering dynamic step adjustment; in a meteorological radar, a certain radar reflectivity factor increases abnormally, exceeding the actual value by 20 dBZ; in satellite remote sensing, the surface temperature data is missing in a local area due to cloud cover.

[0217] In S102, the anomalies of the temperature element are determined by the coefficient of variation CV t = 0.077 and the local outlier factor LOF = 0.62 to identify the abnormal ground meteorological station sites, and they are repaired based on the Delaunay triangulation algorithm, with an error ≤ 0.5°C; the radar data is repaired using time interpolation, and the error is reduced to within 3 dBZ after repair; the satellite data is repaired by filling the missing area using time interpolation, and the root mean square error ≤ 1.2°C.

[0218] In S200, the basic weight of the ground meteorological station data for the temperature element is 0.6, the weight of the meteorological radar data is 0.378, and the weight of the satellite remote sensing data is 0.28, and the temperature fusion error is reduced from 5°C of a single data source to 0.7°C.

[0219] In S300, the resolution of the numerical weather prediction model is increased from 10 km to 1 km to achieve grid-level alignment; after bias correction, the root mean square error of temperature prediction is reduced from 4.2°C to 1.8°C, an improvement of 57%.

[0220] At the same time, the heat map shows a positive deviation in the coastal area, 2°C higher than the prediction, and it is corrected to within ±0.5°C after assimilation.

[0221] In the second application embodiment of this invention, the temperature accuracy of the cold wave process can be improved by more than 50%, and the ability to capture extreme weather is significantly enhanced, verifying the effectiveness of multi-source data dynamic fusion and assimilation.

[0222] This specific embodiment is only an explanation of the present invention, and it is not a limitation of the present invention. Those skilled in the art can make modifications without creative contributions to this embodiment according to needs after reading this specification, but as long as it is within the scope of the claims of the present invention, it is protected by the patent law.

Claims

1. A multi-source heterogeneous meteorological data assimilation and fusion method, characterized in that The following steps are included: S100, synchronously acquiring meteorological data from multiple data sources through a distributed data interface and performing preprocessing, the data sources including satellite remote sensing data, ground meteorological station data, and meteorological radar data; S200, after completing the preprocessing of the meteorological data from multiple data sources in S100, based on the weighted least squares method, respectively fuse the data from the multiple data sources into meteorological element values ​​of each meteorological element; S300, assimilating the meteorological element values ​​obtained by fusion in S200 with the numerical weather forecast model; S400, in conjunction with S100, S200 and S300, can dynamically optimize numerical weather forecast model parameters and generate executable weather warning instructions; Among them, S200 is specifically: Multiply the weights of data source type, spatiotemporal proximity, and data quality item by item to generate a comprehensive weight matrix; The data of each meteorological element will correspond to a comprehensive weight matrix, and the fused meteorological element value will be obtained based on the weighted least squares method; S300 is specifically: S301, upgrading the low-resolution data of the numerical weather forecast model to the same resolution as the meteorological data by bicubic interpolation, so that the meteorological data and the data of the numerical weather forecast model can be compared on the same spatial grid; S302, calculating the difference between the predicted value of the numerical weather forecast model and the meteorological element value; specifically: S3021. For the same meteorological element at each grid point, calculate the direct difference between the predicted value and the meteorological element value; S3022, calculating the average value of all grid point difference values ​​to evaluate the systematic deviation of the numerical weather forecast model; then calculating the root mean square error to measure the degree of dispersion between the predicted value and the meteorological element value, reflecting the uncertainty of the numerical weather forecast model; S3023. In combination with the generated comprehensive weight matrix, the difference value of each grid point is adjusted according to the weight of the data source to avoid interference of low-quality data on the difference calculation; S3024. Visualize the difference results in the form of a spatial heat map, where red indicates a positive deviation of the predicted value that is higher than the meteorological element value, and blue indicates a negative deviation of the predicted value that is lower than the meteorological element value, so as to facilitate intuitive positioning of the deviation concentration area and mark the grid points that exceed the preset threshold.

2. The multi-source heterogeneous meteorological data assimilation and fusion method according to claim 1, wherein: The preprocessing process in S100 is as follows: S101, constructing a meteorological element threshold rule base, wherein the meteorological element threshold rule base includes the minimum and maximum values ​​of each meteorological element to preliminarily filter out abnormal data; S102, performing quality inspection on the data points in each meteorological element, and further screening out abnormal data; specifically, adopting a sliding time window mechanism, dynamically inspecting the data points in each meteorological element, and calculating the corresponding local outlier factor according to all the data points in each window based on the window length and step length corresponding to each meteorological element, and if the local outlier factor is greater than a set threshold, all the data points in the window are determined to be abnormal data; Among them, the sliding time window mechanism can automatically adjust the window length and step size according to the volatility of each meteorological element. Specifically: calculate the standard deviation and mean of each data point within the current window for each meteorological element, and then divide the calculated standard deviation by the mean to obtain the coefficient of variation. When the coefficient of variation is higher than the set maximum threshold, indicating that the data changes violently, the window length is shortened; when the coefficient of variation is lower than the set minimum threshold, indicating that the data changes gently, the window length is extended; Extract the values of the data points at the current moment and the previous moment for each meteorological element, and divide them by the time interval between the current moment and the previous moment to obtain the change rate at the current moment. When the change rate at the current moment is greater than the set threshold, the step size is shortened to more accurately capture the rapidly changing trend; when the change rate at the current moment is less than the set threshold, the step size is extended to reduce the computational amount; S103. Repair the screened abnormal data; S104. Perform normalization processing on each meteorological element one by one to map different meteorological elements to the same numerical range.

3. A multi-source heterogeneous meteorological data assimilation and fusion method according to claim 2, characterized in that: S103 is specifically as follows: If the abnormal data is ground meteorological station data; use the Delaunay triangulation algorithm to perform spatial triangulation on the ground stations. Each station is connected to other stations to form triangle sides, and then determine the adjacency relationship between the station of the abnormal ground meteorological station and the stations of other ground meteorological stations. Then select the 5 stations of the ground meteorological stations that are closest to the station of the abnormal ground meteorological station; for the selected 5 stations of the ground meteorological stations, use the weighted average method to replace the abnormal data, and determine the weight according to the distance between these ground meteorological station stations and the station of the abnormal ground meteorological station. The weight is inversely proportional to the distance; calculate the weighted average value according to the weights of the selected 5 stations of the ground meteorological stations to replace the abnormal data; If the abnormal data is satellite remote sensing data or meteorological radar data, then use the data points within 3 moments before and after in the time dimension for time interpolation repair.

4. A multi-source heterogeneous meteorological data assimilation and fusion method according to claim 1, characterized in that: S301 is specifically as follows: S3011. Obtain the original grid resolution of the numerical weather prediction model and the actual grid resolution of the meteorological data, and calculate the proportionality coefficient between the two: Based on the original row resolution and original column resolution of the numerical weather prediction model, as well as the actual row resolution and actual column resolution of the meteorological data, calculate the resolution proportionality coefficients in the row direction and column direction respectively; The row direction proportionality coefficient is equal to the meteorological data row resolution divided by the original model row resolution, and the column direction proportionality coefficient is equal to the meteorological data column resolution divided by the original model column resolution; S3012. Dynamically adjust the kernel function parameters of the bicubic interpolation algorithm based on the proportionality coefficient: According to the larger value of the row direction proportionality coefficient and the column direction proportionality coefficient, set the sampling radius of the kernel function to 1.5 times of the larger value, and adaptively adjust the smoothness coefficient in combination with the spatial distribution density of the meteorological data; when the meteorological data density is higher than the preset threshold, the smoothness coefficient is reduced to retain the detail features; when the meteorological data density is lower than the preset threshold, the smoothness coefficient is increased to enhance the data smoothness; S3013. Perform layer-by-layer interpolation on the low-resolution data of the numerical weather prediction model: Adopt a hierarchical interpolation strategy to divide the original low-resolution data into a coarse-grained layer and a fine-grained layer: The operation for interpolating the coarse-grained layer is to perform preliminary interpolation on the original grid based on the bicubic interpolation algorithm to generate intermediate-resolution data; The operation for interpolating the fine-grained layer is to dynamically adjust the interpolation step size according to the local gradient change of the meteorological data on the basis of the intermediate-resolution data, and finally generate high-resolution grid data with the same resolution as the meteorological data; S3014. Verify the spatial coordinate consistency between the interpolated high-resolution grid data and the meteorological data: Randomly select a preset number of key grid points from the meteorological data as reference points; Calculate the coordinate offset between the interpolated data and the reference points. If the offset exceeds the preset maximum threshold, it is determined as a coordinate deviation; If more than 50% of the reference points have coordinate deviations, readjust the kernel function parameters and perform interpolation again; if the deviation ratio is less than 50%, only perform local correction on the deviation area; S3015. Match and align the verified high-resolution grid data and the meteorological data point by point according to the spatial coordinates; if there are uncovered areas, use the nearest neighbor interpolation method to supplement the missing values to ensure the integrity and consistency of the fused grid data.

5. A multi-source heterogeneous meteorological data assimilation and fusion method according to claim 4, characterized in that: In S3012, the specific method of adaptively adjusting the smoothness coefficient in combination with the spatial distribution density of the meteorological data is as follows: if the density of the meteorological data shows a sudden change in a local area, adjust the smoothness coefficient according to the density gradient; when the density gradient is greater than the preset threshold, the smoothness coefficient is reduced to 70% of the original value to inhibit excessive smoothing of the interpolation; when the density gradient is less than the preset threshold, the smoothness coefficient is increased to 130% of the original value to eliminate the interference of data noise; In the aforementioned S3013, when performing the interpolation operation of the fine-grained layer, if the change trends of the meteorological elements of adjacent grids are consistent, linear interpolation is adopted; if the change trends are significantly different, switch to the non-linear interpolation algorithm to avoid interpolation distortion.

6. A multi-source heterogeneous meteorological data assimilation and fusion method according to claim 1, characterized in that: S400 is specifically as follows: S401. Conduct a traceability analysis on the marked grid points exceeding the preset threshold, extract the corresponding data source types, spatio-temporal resolution characteristics, and abnormal fluctuation records of the comprehensive weight matrix, and generate an abnormal traceability report; S402. Based on the abnormal traceability report, dynamically adjust the empirical parameter weights for the meteorological element model parameters associated with the marked grid points in the numerical weather prediction model according to the deviation direction and amplitude, including the following sub-steps: S4021. According to the meteorological element type of the marked grid points, extract the corresponding set of empirical parameters from the numerical weather prediction model, and associate the predicted value calculated in S3021 with the meteorological element value after fusion in S200 to establish a mapping relationship table between the parameters and the deviation; S4022. Based on the direct difference between the predicted value and the meteorological element value of the same grid point in S3021, determine the deviation direction; specifically: S4021.

1. Obtain the direct difference between the predicted value and the meteorological element value at the same grid point in S3021, and detect the sign consistency of the difference in the continuous time series based on a preset time window; if more than 80% of the differences are positive within the same window, trigger a positive deviation determination; if more than 80% of the differences are negative, trigger a negative deviation determination; S4021.

2. If a positive deviation determination is triggered, confirm that the predicted value of the numerical weather prediction model is continuously higher than the fused meteorological element value, and generate an instruction to reduce the weight of the gain type parameter; if a negative deviation determination is triggered, confirm that the predicted value is continuously lower than the meteorological element value, and generate an instruction to increase the weight of the compensation type parameter; S4021.

3. According to the confirmation result of the deviation direction in S4021.2, add a dynamic mark to the associated meteorological element model parameter in the model parameter correction coefficient table. The mark types include a weight decay factor or a weight enhancement factor, and the weight decay factor and the weight enhancement factor are in a proportional relationship with the average value of all grid point differences calculated in S3022; S4021.

4. Perform regional association expansion. If the same meteorological element triggers the same deviation direction determination at 3 or more adjacent grid points, enhance the regionalization of the associated parameter weight adjustment instruction, extend the single-point adjustment amount to the regional scope, and the adjustment amplitude decays with the grid distance; S4023. Calculate the parameter weight adjustment amount based on the average value of all grid point differences calculated in S3022, combined with a preset error reference value: If the predicted value is higher than the meteorological element value, assign a negative direction coefficient and proportionally reduce the corresponding parameter weight; If the predicted value is lower than the meteorological element value, assign a positive direction coefficient and proportionally increase the corresponding parameter weight; Divide the average value by the preset error reference value to obtain the normalized adjustment ratio; Multiply the direction coefficient by the normalized adjustment ratio to obtain the final adjustment amount of the parameter weight, ensuring that the adjustment amplitude is linearly related to the deviation degree; S4024. Perform smoothing constraints to limit the single-time parameter weight adjustment amplitude not to exceed 10% of the original weight. If the calculated value exceeds the range, perform threshold truncation processing to ensure the stability of the model; S403. Use historical assimilation data to train a lightweight neural network, input the adjusted difference in S3023 into the network, predict the short-term meteorological element change trend, and feedback the prediction result back to the weight allocation module in S200 to update the comprehensive weight matrix.

7. A multi-source heterogeneous meteorological data assimilation and fusion method according to claim 6, characterized in that: Specifically, S403 is as follows: S4031. Extract the adjusted difference in S3023, the generated comprehensive weight matrix, and the corresponding time stamps from the historical assimilation data, and generate labeled time series training samples according to the sliding time window mechanism. The label is the actual change value of the meteorological element in the next 1 hour; S4032. Construct a lightweight neural network, including an input layer, two convolutional layers, a single long short-term memory unit, and a fully connected output layer. The convolution kernel size is set to 3×3, and the stride matches the meteorological grid resolution; S4033. Adopt an adaptive learning rate mechanism. Set the initial learning rate to 0.

001. If the loss of the validation set does not decrease for three consecutive training cycles, the learning rate decays to 50% of the original value. At the same time, introduce an early stopping mechanism to prevent overfitting; S4034. Input the adjusted difference at the current moment S3023 into the trained network, output the predicted value of the meteorological element change trend in the next hour, and combine this predicted value with the real-time stability index of the data source in S200 to generate a weight correction factor; S4035. Based on the weight correction factor generated in S4034, dynamically adjust the comprehensive weight matrix. Specifically: increase the weight of the data source with high consistency between the predicted trend and the real-time data, and decrease the weight of the data source with low consistency. The updated weight matrix is directly applied to the next data fusion.

Images