A method for simulating and tracking the trajectory of wind-blown snow based on a numerical model of wind-blown snow

Abstract

The present application relates to the technical field of motion trajectory simulation tracking, in particular to a wind-blowing snow motion trajectory simulation tracking method based on a wind-blowing snow numerical model, in the present application, the spatial coupling relationship between local meteorological variables is extracted, a more discriminant meteorological regional division basis is constructed, the regional division is no longer dependent on single variable stratification, thereby the identification ability of local strong disturbance area is improved, the meteorological time series is analyzed by convolution neural network, the nonlinear time series evolution law between meteorological variables is captured, and the interactive mapping relationship between meteorological change and trajectory change is constructed, the response sensitivity of the prediction model to sudden weather patterns is enhanced, the random forest is used for weighted evaluation of the prediction results of multiple groups of models, the final prediction path is optimized according to the historical error distribution, the error source is tracked in a structured manner, and the adaptability and stability of the prediction result are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of motion trajectory simulation and tracking, and in particular to a method for simulating and tracking the motion trajectory of blowing snow based on a blowing snow numerical model. Background Art

[0002] The field of motion trajectory simulation and tracking technology uses computer simulation and digital algorithms to track, predict and analyze the motion trajectory of objects or phenomena. It predicts the motion trajectory based on physical models and uses sensors and computer systems to track and analyze the motion state in real time.

[0003] A method for simulating and tracking the trajectory of snowflakes in a blizzard based on a numerical model of blizzard motion is used to simulate and predict the trajectory of snow particles in blizzard phenomena, and to provide trajectory tracking and analysis. The purpose is to accurately simulate the dynamic trajectory of snow particles during blizzards through a numerical calculation model, and to assess the changes in their behavior under conditions such as wind speed, snow volume, and geographical environment, so as to effectively assess the severity and impact range of blizzard events and their potential threats to transportation, buildings, and other facilities.

[0004] The computing logic of existing technologies is centered on solving a fixed set of equations. It lacks a continuous response mechanism to changes in environmental conditions and the ability to extract local disturbance features. It is difficult to accurately restore the acceleration change process of snow particles under drastic changes in the wind field, resulting in prediction deviations in simulation results in spatially heterogeneous areas. Due to the lack of modeling capabilities for nonlinear interactions between meteorological variables, the prediction model finds it difficult to maintain temporal consistency in the output results during multiple rounds of meteorological changes. Furthermore, the computing load is limited by the performance of the central node, making it difficult to cope with application scenarios with large regional spans and multiple sources of disturbances. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a method for simulating and tracking the movement trajectory of wind-blown snow based on a wind-blown snow numerical model.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for simulating and tracking the trajectory of blowing snow based on a numerical model of blowing snow, comprising the following steps:

[0007] S1: Obtain real-time meteorological data through sensors, analyze meteorological parameters in different regions, identify periods with greater impact on wind and snow, and calculate the wind and snow environmental impact index;

[0008] S2: Based on the wind and snow environmental impact index, extract the local spatial features of meteorological data such as wind speed, temperature, and humidity through convolution operation, stratify the meteorological regions according to the influence of each feature, and generate a meteorological region feature layer;

[0009] S3: Based on the meteorological regional feature layer, a convolutional neural network is used to segment the time series data, train a model to compare changes between time nodes, analyze the correlation between meteorological parameters and trajectory changes, and generate a predicted value for the movement trajectory of the wind and snow;

[0010] S4: Based on the predicted value of the wind and snow movement trajectory, the meteorological data is distributed to the distributed computing platform, the task priority is dynamically adjusted, and resources are allocated to the areas with greater impact, thereby generating multi-region wind and snow trajectory simulation data;

[0011] S5: Based on the multi-regional snowstorm trajectory simulation data, random forest is used to perform weighted fusion of multiple model results, analyze the prediction error, judge the model adaptability, select the best model for tracking optimization, and generate the snowstorm trajectory optimization prediction result.

[0012] As a further solution of the present invention, the specific steps of generating the wind and snow environmental impact index are:

[0013] Real-time meteorological data is obtained through sensors, converted into meteorological parameters, and spatially divided by region. Wind speed fluctuations, temperature fluctuations, and humidity changes in each region are calculated and mapped to corresponding time periods to generate meteorological parameter fluctuation data;

[0014] Based on the meteorological parameter fluctuation data, the meteorological parameters of each region are gradually screened to analyze the time period related to the wind and snow phenomenon, identify the wind and snow impact period in each region, and generate wind and snow period impact data;

[0015] Based on the wind and snow period impact data, the variation range of meteorological parameters in each region is calculated, and the meteorological impact intensity in each region is combined with the time node to generate a wind and snow environmental impact index.

[0016] As a further embodiment of the present invention, the meteorological parameters include wind speed, temperature, humidity, air pressure, precipitation, radiation, cloud cover, visibility, dew point, meteorological phenomena, atmospheric stability, wind direction, wind frequency, evaporation, snow depth, snow grain size and snow water equivalent.

[0017] As a further solution of the present invention, the specific steps of generating the meteorological region characteristic layer are:

[0018] Based on the wind and snow environmental impact index, the wind speed, temperature, humidity and other data of each region are extracted, and the meteorological impact weight of each region is calculated by setting a threshold and weighting according to the impact degree to generate the regional meteorological data weight;

[0019] Based on the regional meteorological data weights, spatial distribution analysis is performed on the meteorological data in the region, key meteorological features of each region are extracted layer by layer, and local feature data of the meteorological region is generated;

[0020] Based on the local feature data of the meteorological region, the meteorological features of each region are integrated, multi-level feature mapping is performed, and a meteorological region feature layer is generated.

[0021] As a further solution of the present invention, the specific steps of generating the predicted value of the wind and snow movement trajectory are:

[0022] Based on the meteorological regional feature layer, a convolutional neural network is used to divide the data by time interval, extract the meteorological parameters of each time period and classify them into corresponding regions to generate a time series meteorological dataset;

[0023] Based on the time series meteorological dataset, meteorological parameters at different time points are compared, the correlation between changes in meteorological parameters and wind and snow trajectories is analyzed, and meteorological change and trajectory correlation data are generated through data difference detection;

[0024] Based on the meteorological change and trajectory correlation data, the meteorological parameters between each time node are compared with the change trend of the wind and snow trajectory to generate a wind and snow movement trajectory prediction value.

[0025] As a further solution of the present invention, the convolutional neural network is based on the formula:

[0026]

[0027] in: Indicates the Layer at time step The output features of Indicates that the previous layer is at time step The input features of Indicates the Tier convolution kernel weights, Indicates the The bias term of the layer, represents a nonlinear activation function, represents the time window width, Indicates that at time step The regulating factor on Indicates the Dynamic adjustment coefficient of the layer.

[0028] As a further solution of the present invention, the specific steps of generating the multi-region wind and snow trajectory simulation data are as follows:

[0029] Based on the predicted value of the wind and snow movement trajectory, the meteorological data is spatially allocated, and the data is allocated to each computing node according to regional importance to generate a regional meteorological data allocation result;

[0030] Based on the regional meteorological data allocation results, the priority of the computing tasks is adjusted according to the urgency of the impact of the wind and snow track, more computing resources are allocated to important areas, and regional task priority data is generated;

[0031] Based on the regional task priority data, the optimized computing tasks are processed in parallel on a distributed platform to generate multi-region wind and snow trajectory simulation data.

[0032] As a further solution of the present invention, the spatial allocation first evaluates the importance of each region based on the regional data in the predicted value of the wind and snow movement trajectory, combined with the meteorological influence range of each region, and based on the evaluation results, allocates the regional data to the corresponding computing nodes according to the influence to obtain the regional meteorological data allocation result.

[0033] As a further solution of the present invention, the specific steps of generating the wind and snow trajectory optimization prediction result are:

[0034] Based on the multi-region wind and snow trajectory simulation data, a random forest is used to weight the output of each model, adjust the impact of each region, integrate the predicted values ​​of each region, and generate a model weighted result;

[0035] Based on the weighted results of the model, the prediction error of each region is counted, the error value is calculated one by one, and the areas with larger errors are marked. By comparing with the set error threshold, regional error analysis data is generated;

[0036] Based on the regional error analysis data, the performance of each model in different regions is analyzed, and the most suitable model is selected by comparing the error sizes. Adjustments and optimizations are performed to generate optimized prediction results for wind and snow trajectories.

[0037] As a further solution of the present invention, the random forest is according to the formula:

[0038]

[0039] in: Represents the final weighted output result, Indicates the The weight coefficient of the region, Indicates the The predicted value of the region, Indicates the total number of regions, Indicates the Regional meteorological factor weights, Indicates the The terrain adaptability coefficient of the region, Indicates the The historical data adaptation coefficient of the region.

[0040] Compared with the prior art, the advantages and positive effects of the present invention are:

[0041] 1. This invention extracts the spatial coupling relationship between local meteorological variables to construct a more discriminative basis for meteorological regional division, eliminating the reliance on single-variable stratification for regional division and improving the ability to identify local areas of strong disturbance.

[0042] 2. This invention uses a convolutional neural network to segment and analyze meteorological time series, capturing the nonlinear temporal evolution patterns between meteorological variables and establishing an interactive mapping relationship between meteorological changes and trajectory changes, thereby enhancing the prediction model's sensitivity to sudden weather events.

[0043] 3. This invention uses a distributed computing architecture to prioritize multi-region simulation tasks based on their impact and dynamically allocate resources, thus avoiding wasted computing resources and improving computational accuracy in key areas.

[0044] 4. In this invention, a random forest algorithm is used to perform weighted evaluation of prediction results from multiple models, optimizing the final prediction path based on historical error distribution. This achieves structured tracking of error sources and improves the adaptability and stability of prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic diagram of the main steps of the present invention;

[0046] Figure 2 This is a schematic diagram of the refinement of S1 of the present invention;

[0047] Figure 3 This is a schematic diagram of the refinement of S2 of the present invention;

[0048] Figure 4 This is a schematic diagram of the refinement of S3 of the present invention;

[0049] Figure 5 This is a schematic diagram of the refinement of S4 of the present invention;

[0050] Figure 6 This is a schematic diagram of the refinement of S5 of the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0052] Example 1

[0053] See also Figure 1The application provides a technical solution: a wind-blowing snow movement trajectory simulation tracking method based on a wind-blowing snow numerical model, comprising the following steps:

[0054] S1: obtaining real-time meteorological data through a sensor, identifying a time period with a greater influence on wind-blowing snow by analyzing meteorological parameters in different regions, and calculating a wind-blowing snow environmental influence index;

[0055] S2: based on the wind-blowing snow environmental influence index, extracting local spatial features of meteorological data such as wind speed, air temperature, and humidity through convolution operation, layering meteorological regions according to the influence of each feature, and generating meteorological region feature layers;

[0056] S3: based on the meteorological region feature layer, using a convolutional neural network to cut time series data, training the model to compare changes between time nodes, analyzing the correlation between meteorological parameters and trajectory changes, and generating wind-blowing snow movement trajectory prediction values;

[0057] S4: based on the wind-blowing snow movement trajectory prediction value, distributing meteorological data to a distributed computing platform, dynamically adjusting task priorities, allocating resources to regions with greater influence, and generating multi-region wind-blowing snow trajectory simulation data;

[0058] S5: based on the multi-region wind-blowing snow trajectory simulation data, using a random forest to weight and fuse multiple model results, analyze prediction errors, judge model adaptability, select the best model for tracking optimization, and generate wind-blowing snow trajectory optimization prediction results.

[0059] Please refer to Figure 2 The specific steps for generating the wind-blowing snow environmental influence index are as follows:

[0060] Obtain real-time meteorological data through a sensor, convert it into meteorological parameters, and divide the space by region. Calculate the wind speed fluctuation, air temperature fluctuation, and humidity change of each region, and map them to the corresponding time period to generate meteorological parameter fluctuation data;

[0061] Based on the meteorological parameter fluctuation data, gradually filter the meteorological parameters of each region, analyze the time period related to wind-blowing snow phenomenon, identify the wind-blowing snow influence period in each region, and generate wind-blowing snow period influence data;

[0062] Based on the wind-blowing snow period influence data, calculate the change amplitude of the meteorological parameters of each region, and combine the meteorological influence intensity in each region with the time node to generate the wind-blowing snow environmental influence index;

[0063] Based on the original meteorological observation data output by the sensor, the wavelet transform method is used to process the time series of three meteorological signals: wind speed, temperature, and humidity. The wavelet basis selected is the Daubechies 4th-order wavelet, and the decomposition level is 5. The wavelet multi-scale decomposition operation is performed on each meteorological parameter in sequence. The local volatility is calculated by extracting the approximate coefficient and detail coefficient of each layer. The analysis window width is set to 15 minutes and the sliding step is set to 1 minute. The decomposed signal is reconstructed along the time axis to construct the meteorological parameter change trend sequence. The K-means++ clustering method is then used for spatial partitioning. The input sample is the average value of the reconstructed wind speed, temperature, and humidity time series in the specified area grid. The cluster center is initialized by the default mechanism of the K-means++ algorithm, and the number of clusters is set to 8. The Euclidean distance is used to measure the meteorological differences between each grid to complete the attribution classification of meteorological data in the spatial grid, and finally generate meteorological parameter fluctuation data.

[0064] Based on the meteorological parameter fluctuation data, a double-threshold mutation detection method is used to perform regional fluctuation identification processing on the wind speed, temperature, and humidity time series. Upper and lower thresholds are set for the three types of meteorological variables in each grid, and the upper and lower limits of mutations are set based on the historical sliding average superimposed with 2 times the standard deviation. Meteorological data points that exceed the range are labeled as outliers. Subsequently, a sliding window matching algorithm is used for time period identification. The window size is set to 3 hours and the sliding step is 15 minutes. The total number of outliers in each window is counted. If the number of outliers in each of three consecutive windows exceeds 60% of the window capacity, the time is marked as a period of possible wind and snow impact. Windows with adjacent time periods of no more than 1 hour are merged to obtain continuous time intervals. After processing, the regional number and time period information are combined to generate wind and snow period impact data.

[0065] Based on the impact data of wind and snow periods, the Z-Score standardization method is used to normalize the change amplitudes of wind speed, temperature, and humidity in the corresponding period. The standardization process standardizes the meteorological data based on the historical 30-day sliding mean and standard deviation in the region, and outputs the normalized amplitude values ​​of the three parameters respectively. Subsequently, the exponentially weighted moving average method is used to perform a weighted operation on the normalized data. The weighting factor is set to 0.4 to give recent changes a higher weight. The weighted data series of wind speed, temperature, and humidity are output respectively, and then fused based on the set linear weights. The wind speed weight is set to 0.5, the temperature is set to 0.3, and the humidity is set to 0.2. The fused value is bound to the regional number and the corresponding time node and encoded into a ternary data set to finally generate the wind and snow environmental impact index.

[0066] Meteorological parameters include wind speed, temperature, humidity, air pressure, precipitation, radiation, cloud cover, visibility, dew point, meteorological phenomena, atmospheric stability, wind direction, wind frequency, evaporation, snow depth, snow grain size and snow water equivalent.

[0067] Referring to Figure 3 The specific steps for generating the meteorological regional feature layer are as follows:

[0068] Based on the wind-blowing snow environmental impact index, the wind speed, air temperature, humidity and other data of each region are extracted, and by setting a threshold value, the meteorological impact weight of each region is calculated by weighting according to the influence degree, and the regional meteorological data weight is generated;

[0069] Based on the regional meteorological data weight, the spatial distribution of the meteorological data in the region is analyzed, and the key meteorological features of each region are extracted layer by layer to generate the meteorological regional local feature data;

[0070] Based on the meteorological regional local feature data, the meteorological features of each region are fused, and multi-level feature mapping is performed to generate the meteorological regional feature layer;

[0071] Based on the wind-blowing snow environmental impact index, the fuzzy comprehensive evaluation method is used to weight process the wind speed, air temperature, humidity data of each region. First, set the fuzzy level set to five equal interval partitions, which correspond to five influence levels of extremely low, low, medium, high and extremely high respectively, and set the weight vector to wind speed 0.4, air temperature 0.35 and humidity 0.25. For each input data, the fuzzy matrix is constructed by using the maximum membership degree principle, and then the fuzzy synthesis operation is performed. The synthesis method selects the weighted average type operation structure, and the calculation process outputs the comprehensive evaluation value of each region as the regional meteorological weight index. Finally, all regional indicators are mapped to the corresponding geographical grid number, and the regional meteorological data weight is output;

[0072] Based on the regional meteorological data weight, the inverse distance weighted interpolation method is used to analyze the spatial distribution of the meteorological data in the region. The interpolation weight function is set to a power function structure, the power parameter is set to 2, the nearest 10 meteorological station data around each grid cell is taken as the reference point, the interpolation weight coefficient is calculated according to the inverse distance relationship between the interpolation center point and the reference point, and the interpolation filling operation of wind speed, air temperature and humidity in the regional space is completed. Then, the threshold extraction method based on hierarchical clustering is used to decompose the spatial data layer by layer after filling. The similarity measurement method is set to Manhattan distance, and the merging strategy is the minimum distance principle. For each type of meteorological factor, the grid cells with a fluctuation amplitude higher than 20% of the median value are extracted and classified, and the merged results are collected according to the region number, and the meteorological regional local feature data is output;

[0073] Based on the local feature data of meteorological regions, the autoencoder neural network method is used to fuse the meteorological characteristics of each region. The input layer dimension is set to the type of meteorological factor multiplied by the number of regions, the activation function is ReLU, the encoding layer dimension is compressed layer by layer to one-fourth of the input dimension, the L2 regularization constraint is adopted, the decoding layer reconstructs the compressed representation, and the mean square error is used as the loss function. The Adam optimizer is used in the training process, the learning rate is set to 0.001, the batch size is 32, and the number of iterations is set to 100. The principal component transformation is performed on the output feature vector of the fused intermediate layer, and the feature dimension of the top 80% cumulative contribution rate is extracted as the high-dimensional representation of the regional meteorology, and finally the meteorological regional feature layer is generated.

[0074] See also Figure 4 ,The specific steps to generate the predicted value of the wind and snow movement trajectory are:

[0075] Based on the meteorological regional feature layer, a convolutional neural network is used to divide the data by time interval, extract the meteorological parameters of each time period and classify them into corresponding regions to generate a time series meteorological dataset;

[0076] Based on the time series meteorological data set, the meteorological parameters at different time nodes are compared, the correlation between the changes in meteorological parameters and the trajectory of wind and snow is analyzed, and the correlation data between meteorological changes and trajectories is generated through data difference detection;

[0077] Based on the correlation data of meteorological changes and trajectories, the changing trends of meteorological parameters and wind and snow trajectories at each time node are compared to generate the predicted value of wind and snow movement trajectory;

[0078] Based on the meteorological regional feature layer, the convolutional neural network method is used to divide the meteorological data into time intervals. The time interval is set to one hour, and a three-dimensional tensor structure is used as the input format. The input dimension is set to the number of channels multiplied by the number of regions multiplied by the number of time steps. The number of channels corresponds to three parameters: wind speed, temperature, and humidity. The convolution kernel size is set to 3×3, the step size is set to 1, the padding method is same, and the activation function is ReLU. The dimension of the first convolution output feature map is 64, the dimension of the second convolution layer is 128, the maximum pooling layer pooling kernel size is 2×2, and the pooling step size is 2. After the convolution operation is completed, the wind speed, temperature, and humidity in each time period are extracted as time dimension vectors, and the vectors are classified into corresponding spatial regions according to the regional index. Finally, a mapping table of the regional number and the time period joint index is constructed and output as a time series meteorological dataset.

[0079] Based on a time series meteorological dataset, a Laplace distance-based difference detection algorithm is used to compare and analyze meteorological parameters in different time periods. The input is the meteorological vector of the same area in adjacent time periods. The algorithm constructs an adjacency graph and calculates the Laplace eigenvalue of each node in the graph to determine the changing trend of wind speed, temperature, and humidity between two time points. The parameter change is marked when the eigenvalue difference is greater than the set threshold. The threshold is set to the median of the 7-day moving standard deviation of each parameter. The marked change points are then compared with historical wind and snow trajectory records. The vector cosine similarity method is used to evaluate the relative consistency between the change points and the trajectory morphology. When the similarity is greater than 0.85, the associated data is retained. Finally, the regional index and time period pairing information and their similarity results are output to generate meteorological change and trajectory association data.

[0080] Based on the data associated with meteorological changes and trajectories, a gated recurrent unit network method is used to model and analyze the meteorological parameters and trajectory change trends at different time nodes. The network structure contains a single-layer GRU unit, the input sequence length is set to 12, and the time step corresponds to the hourly unit. Each input sequence contains three parameters: wind speed, temperature, and humidity, as well as the displacement increments of the wind and snow trajectory in the x and y directions. The hidden layer dimension is set to 256, the activation function is tanh, the update gate and reset gate are controlled by the sigmoid function respectively, and the mean square error loss is calculated between the output sequence and the actual trajectory change. The Adam optimizer is used in the training process, the learning rate is set to 0.0005, the batch size is 64, and the training rounds are 200. The predicted output is encoded as a prediction index according to the region and time node, and finally the predicted value of the wind and snow movement trajectory is generated.

[0081] Convolutional neural network, according to the formula:

[0082]

[0083] in: Indicates the Layer at time step The output features of Indicates that the previous layer is at time step The input features of Indicates the Tier convolution kernel weights, Indicates the The bias term of the layer, represents a nonlinear activation function, represents the time window width, Indicates that at time step The regulating factor on Indicates the Dynamic adjustment coefficient of the layer;

[0084] Execution process: First, through each time step Calculate the input features of the previous layer through convolution operation With the The convolution kernel weights of the layer The weighted sum of , get the convolution result, and then add the bias term to the result , and after a nonlinear activation function After processing, the adjustment factor is introduced , used for the time step The response is dynamically adjusted to make adaptive corrections to the trajectory of the wind and snow in different time periods, and finally the coefficient is dynamically adjusted. It is used to adjust the sensitivity of the time step t according to the historical movement pattern of the wind and snow, so as to accurately simulate the trajectory changes of the wind and snow in each time period. The movement trajectory of the wind and snow can be more accurately simulated and tracked in the numerical model.

[0085] See also Figure 5 ,The specific steps to generate multi-region wind and snow trajectory simulation data are:

[0086] Based on the predicted value of the wind and snow movement trajectory, the meteorological data is spatially allocated and the data is allocated to each computing node according to the regional importance to generate the regional meteorological data allocation result;

[0087] Based on the regional meteorological data allocation results, the priority of computing tasks is adjusted according to the urgency of the impact of wind and snow tracks, more computing resources are allocated to important areas, and regional task priority data is generated;

[0088] Based on regional task priority data, the optimized computing tasks are processed in parallel on a distributed platform to generate multi-region wind and snow trajectory simulation data;

[0089] Based on the predicted value of the wind and snow movement trajectory, a data space segmentation method based on KD Tree is used to perform spatial allocation operations on meteorological data, and a three-dimensional spatial index structure is constructed. The node coordinates use longitude, latitude and wind speed as the three-axis coordinates. An index sequence is constructed for all samples in the input meteorological dataset in a Z-shaped scanning order. The maximum capacity of each node is set to 64 records. The split threshold is the median of the number of samples in the current layer. The tree structure is recursively constructed to a depth of no less than 12 layers. During the allocation process, the region number and spatial coordinates in the trajectory prediction value are used as query parameters. Each meteorological data is assigned to the target region number according to the spatial neighborhood returned by the K nearest neighbor query. The region number is then hashed to the specified computing node, and the regional meteorological data allocation result is finally output.

[0090] Based on the results of regional meteorological data allocation, the TOPSIS ranking method is used to set the regional task priority. First, an evaluation matrix is ​​constructed, with columns including trajectory displacement amplitude, trajectory slope change rate, historical trajectory overlap, and meteorological disturbance frequency. Vector normalization is performed on the data in each column of the matrix, and weight vectors are set to 0.4, 0.3, 0.2, and 0.1, respectively. The Euclidean distance between each region and the ideal optimal solution and the worst solution is calculated, and the regions are sorted according to relative proximity. The sorting index is mapped to the task queue, and the highest priority region number corresponds to the highest weighted task tag. Then, a task scheduling dictionary is constructed, and the sorting results are encoded as a key-value pair structure and input into the scheduling engine, outputting the regional task priority data.

[0091] Based on regional task priority data, the RDD parallel computing mechanism based on Apache Spark is used to execute optimization computing tasks on a distributed platform. First, each regional task object is encapsulated into a ResilientDistributed Dataset in key-value pair format. Each record has the region number as the key and the task execution script as the value. The SparkContext object is called and the number of task partitions is set to 64. The scheduling strategy is set to prioritize the key marked as high. The mapPartitions function is used to batch process each task, and the maximum number of processing records per partition is set to 100. During task execution, broadcast variables are used to uniformly synchronize the wind and snow physical constraint parameters. Finally, after each node executes, the output data is merged using the reduceByKey function to generate multi-region wind and snow trajectory simulation data.

[0092] For spatial allocation, we first evaluate the importance of each region based on the regional data in the predicted value of the wind and snow movement trajectory and the meteorological impact range of each region. Based on the evaluation results, we allocate the regional data to the corresponding computing nodes according to their influence to obtain the regional meteorological data allocation results.

[0093] See also Figure 6 ,The specific steps to generate the optimized prediction results of wind and snow trajectory are:

[0094] Based on multi-regional wind and snow trajectory simulation data, random forests were used to weight the output of each model, adjust the impact of each region, and integrate the predicted values ​​of each region to generate a model weighted result.

[0095] Based on the model weighted results, the prediction error of each region is counted, the error value is calculated one by one, and the areas with larger errors are marked. By comparing with the set error threshold, regional error analysis data is generated;

[0096] Based on regional error analysis data, we analyze the performance of each model in different regions, select the most suitable model by comparing the error size, make adjustments and optimizations, and generate the optimized prediction results of wind and snow trajectory;

[0097] Based on the simulated data of wind and snow trajectories in multiple regions, the random forest regression algorithm is used to perform weighted processing on the output results of each model. When building the model, the number of base learners is set to 100 decision trees, the maximum depth of each tree is set to 15, the minimum number of sample splits is 4, the feature selection method is automatic, and the Bootstrap sampling strategy is used to perform resampling operations on the input data. The regional number is used as the grouping index. After each group of data is input into the forest model and the prediction operation is performed, the predicted output value of each tree is recorded. Subsequently, the regional influence factor is set according to the standard deviation of wind speed fluctuation in each region. All prediction results of each region are weighted according to the regional influence factor. Finally, the prediction results of each region are merged and the model weighted result is output.

[0098] Based on the model weighted results, a point-by-point error statistics method is used to calculate the prediction error of each region. The input parameters are the comparison pairs of the predicted trajectory values ​​and the actual observed values ​​of each region. The predicted trajectory and the observed trajectory of each region are first aligned along the time axis, and the trajectory data are synchronized using an interpolation step of 5 minutes. Then, the absolute difference between the predicted value and the observed value is calculated at each time point. The error threshold is set to 5 meters, and the error values ​​are accumulated by time period and the average is calculated as the regional average error. After sorting the errors of all regions, the region numbers with errors higher than the threshold are marked. A three-tuple data structure containing the region index, error value, and threshold status is constructed to output the regional error analysis data.

[0099] Based on the regional error analysis data, a model accuracy optimization algorithm is used to compare and analyze the performance of multiple candidate models in each region. The input data is the error matrix between the prediction results of multiple models and the true values ​​in each region. The rows of the matrix represent the model numbers and the columns represent the region numbers. The cumulative error values ​​of each model in different regions are counted, and the model number with the smallest error in each region is selected as the optimal model for the region. A mapping table between the region number and the optimal model number is established, and the parameters of all optimal models are extracted and reconstructed. The sampling parameters, tree depth parameters and splitting strategies are inherited and optimized. Finally, all regional models are retrained based on the optimized parameters, and the final merged prediction results are output to generate the optimized prediction results of wind and snow trajectories.

[0100] Random forest, according to the formula:

[0101]

[0102] in: Represents the final weighted output result, Indicates the The weight coefficient of the region, Indicates the The predicted value of the region, Indicates the total number of regions, Indicates the Regional meteorological factor weights, Indicates the The terrain adaptability coefficient of the region, Indicates the The historical data adaptation coefficient of the region;

[0103] Implementation process: First, the predicted value of each region From the independent wind and snow numerical model, the movement trajectory of wind and snow in each area is simulated, and then the weight coefficient is introduced The predicted values ​​are preliminarily weighted according to the degree of influence of each region, taking into account the importance of different regions in the simulation of wind and snow trajectories, and then the meteorological factor weights are introduced , adjust the meteorological conditions of each area so that the impact of meteorological conditions on the simulation results can be accurately reflected, and then introduce the terrain adaptability coefficient , used to adjust the performance of wind and snow under different terrain conditions, affect the trajectory of wind and snow, and finally the historical data adaptation coefficient was introduced to adjust the impact of the stability of regional historical data on the simulation of wind and snow trajectory, enhance the adaptability of the model to unstable areas, and the final weighted output results It is a combination of all regional forecast values ​​and can more accurately simulate and track the movement trajectory of wind and snow.

[0104] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for simulating and tracking the trajectory of blowing snow based on a numerical model of blowing snow, characterized in that: The following steps are involved: S1: Real-time meteorological data is obtained through sensors and spatially divided by region. By analyzing the meteorological parameters of different regions, the wind and snow impact period in each region is identified, and the change range of the meteorological parameters in each region is calculated. The meteorological impact intensity in each region is combined with the time node to obtain the wind and snow environmental impact index; S2: Based on the wind and snow environmental impact index, extract the local spatial features of the wind speed, temperature, and humidity meteorological data through convolution operations, stratify the meteorological regions according to the influence of each feature, and generate a meteorological region feature layer; S3: Based on the meteorological regional feature layer, a convolutional neural network is used to segment the time series data, and a gated recurrent unit network model is trained to compare changes between time nodes, analyze the correlation between meteorological parameters and trajectory changes, and generate a prediction value of the wind and snow movement trajectory encoded as a prediction index by region and time node; S4: Based on the predicted value of the wind and snow movement trajectory, the meteorological data is distributed to the distributed computing platform, the task priority is dynamically adjusted, and resources are allocated to important areas to generate multi-region wind and snow trajectory simulation data; S5: Based on the multi-region wind and snow trajectory simulation data, a random forest is used to perform weighted fusion of multiple model results, wherein the multiple models include a gated recurrent unit network model, the prediction error is analyzed, the model adaptability is judged, and the best model is selected for tracking optimization to generate a wind and snow trajectory optimization prediction result; The random forest is based on the formula: ; in: Represents the final weighted output result, Indicates the The weight coefficient of the region, Indicates the The predicted value of the region, Indicates the total number of regions, Indicates the Regional meteorological factor weights, Indicates the The terrain adaptability coefficient of the region, Indicates the The historical data adaptation coefficient of the region.

2. The method for simulating and tracking the trajectory of blowing snow based on a blowing snow numerical model according to claim 1, characterized in that: The specific steps of generating the wind and snow environmental impact index are as follows: Real-time meteorological data is obtained through sensors, converted into meteorological parameters, and wind speed fluctuations, temperature fluctuations, and humidity changes in each area are calculated and mapped to corresponding time periods to generate meteorological parameter fluctuation data; Based on the meteorological parameter fluctuation data, the meteorological parameters of each region are gradually screened to analyze the time period related to the wind and snow phenomenon, identify the wind and snow impact period in each region, and generate wind and snow period impact data; Based on the wind and snow period impact data, a wind and snow environmental impact index is generated.

3. The method for simulating and tracking the trajectory of blowing snow based on a blowing snow numerical model according to claim 1, characterized in that: The meteorological parameters include wind speed, temperature, humidity, air pressure, precipitation, radiation, cloud cover, visibility, dew point, meteorological phenomena, atmospheric stability, wind direction, wind frequency, evaporation, snow depth, snow grain size and snow water equivalent.

4. The method for simulating and tracking the trajectory of blowing snow based on a blowing snow numerical model according to claim 1, characterized in that: The specific steps of generating the meteorological region characteristic layer are: Based on the wind and snow environmental impact index, wind speed, temperature, and humidity data of each region are extracted, and by setting a threshold and weighting them according to the degree of influence, the meteorological impact weight of each region is calculated to generate the regional meteorological data weight; Based on the regional meteorological data weights, spatial distribution analysis is performed on the meteorological data in the region, key meteorological features of each region are extracted layer by layer, and local feature data of the meteorological region is generated; Based on the local feature data of the meteorological region, the meteorological features of each region are integrated, multi-level feature mapping is performed, and a meteorological region feature layer is generated.

5. The method for simulating and tracking the trajectory of snowstorms based on a snowstorm numerical model according to claim 1, characterized in that: The specific steps of generating the predicted value of the wind and snow motion trajectory are as follows: Based on the meteorological regional feature layer, a convolutional neural network is used to divide the data by time interval, extract the meteorological parameters of each time period and classify them into corresponding regions to generate a time series meteorological dataset; Based on the time series meteorological data set, meteorological parameters at different time nodes are compared, the correlation between changes in meteorological parameters and wind and snow trajectories is analyzed, and meteorological change and trajectory correlation data are generated through data difference detection; Based on the meteorological change and trajectory correlation data, the meteorological parameters between each time node are compared with the change trend of the wind and snow trajectory to generate a wind and snow movement trajectory prediction value.

6. The method for simulating and tracking the trajectory of blowing snow based on a blowing snow numerical model according to claim 1, characterized in that: The convolutional neural network is based on the formula: ; in: Indicates the Layer at time step The output features of Indicates that the previous layer is at time step The input features of Indicates the Tier convolution kernel weights, Indicates the The bias term of the layer, represents a nonlinear activation function, represents the time window width, Indicates that at time step The regulating factor on Indicates the Dynamic adjustment coefficient of the layer.

7. The method for simulating and tracking the trajectory of blowing snow based on a blowing snow numerical model according to claim 1, characterized in that: The specific steps for generating the multi-region wind and snow trajectory simulation data are as follows: Based on the predicted value of the wind and snow movement trajectory, the meteorological data is spatially allocated, and the data is allocated to each computing node according to regional importance to generate a regional meteorological data allocation result; Based on the regional meteorological data allocation results, the priority of the computing tasks is adjusted according to the urgency of the impact of the wind and snow track, more computing resources are allocated to important areas, and regional task priority data is generated; Based on the regional task priority data, the optimized computing tasks are processed in parallel on a distributed platform to generate multi-region wind and snow trajectory simulation data.

8. The method for simulating and tracking the trajectory of blowing snow based on a blowing snow numerical model according to claim 7, characterized in that: The spatial allocation first evaluates the importance of each region based on the regional data in the predicted value of the wind and snow movement trajectory, combined with the meteorological impact range of each region, and based on the evaluation results, allocates the regional data to the corresponding computing nodes according to the influence to obtain the regional meteorological data allocation result.

9. The method for simulating and tracking the trajectory of blowing snow based on a blowing snow numerical model according to claim 1, characterized in that: The specific steps for generating the wind and snow trajectory optimization prediction result are as follows: Based on the multi-region wind and snow trajectory simulation data, a random forest is used to weight the output results of each model, adjust according to the degree of influence of each region, integrate the predicted values ​​of each region, and generate a model weighted result; Based on the weighted results of the model, the prediction error of each region is counted, the error value is calculated one by one, and the number of the region with error higher than the threshold is marked. By comparing with the set error threshold, regional error analysis data is generated; Based on the regional error analysis data, the performance of each model in different regions is analyzed, and the most suitable model is selected by comparing the error sizes. Adjustments and optimizations are performed to generate optimized prediction results for wind and snow trajectories.

Images