Meteorological service center data visualization and interaction platform

Abstract

The invention discloses a meteorological service center data visualization and interaction platform, and particularly relates to the field of meteorological data visualization, which comprises a multi-source heterogeneous data acquisition and preprocessing module, a space-time fusion and data enhancement module, a distributed storage and intelligent index module, a dynamic visualization module, an intelligent decision and early warning module and an interaction module. According to the invention, multi-source meteorological data are collected through an API and a protocol, and wavelet transform noise reduction processing is carried out by using an edge computing node; carrying out space-time alignment and fusion, and filling missing data with LSTM; dynamic visualization and interaction are realized by using WebGL (Web Graphics Library) and ARCore / ARKit; the WRF and the Transform are integrated to generate early warning information, and decision suggestions are pushed through a natural language interface; according to the method, the WRF numerical model and the Transform algorithm are combined, accurate meteorological early warning is provided for the user, meanwhile, the geographic space index is constructed based on the GeoHash algorithm, efficient geographic space query and data retrieval are achieved, a real-time feedback mechanism and a multi-dimensional visual view are provided, and the operation process and experience of the user are optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological data visualization, and more specifically, to a data visualization and interaction platform for a meteorological service center. Background Art

[0002] In modern society, people's demand for meteorological services is no longer limited to simple weather information broadcasts, but focuses more on personalized, refined, and visual service experience. The public hopes to obtain meteorological information that is closely related to their lives, such as travel weather recommendations, suitability for outdoor activities, etc., and can understand meteorological data and forecast results in an intuitive and easy-to-understand way to make better decisions.

[0003] Meteorological data is an objective record of the state of the atmosphere and its changing patterns, and is of great significance for weather forecasting, meteorological disaster warning, and climate change research. With the improvement of observation technology and computing power, the types and quantities of meteorological data are increasing rapidly.

[0004] However, traditional meteorological services still have some shortcomings in actual use. For example, traditional meteorological services often rely on static reports and simplified data presentation methods, which makes it difficult to meet the needs of the public and professional fields for timely, accurate and in-depth meteorological information. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a data visualization and interaction platform for a meteorological service center to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions: Multi-source heterogeneous data acquisition and preprocessing module: used to collect meteorological data in real time from meteorological satellites, ground sensor networks, meteorological radars and third-party APIs, obtain raw meteorological data, and perform noise reduction on the raw meteorological data through edge computing nodes; Spatiotemporal fusion and data enhancement module: Based on the Kriging interpolation algorithm and Bayesian fusion model, the original meteorological data is spatiotemporally aligned and fused to generate a comprehensive meteorological data set, and the missing data is filled in through the LSTM neural network; Distributed storage and intelligent indexing module: Use the time series database InfluxDB to store real-time comprehensive meteorological data sets, use MongoDB to store unstructured radar images, and build geospatial indexes based on the GeoHash algorithm; Dynamic visualization module: used to render dynamic maps and 3D weather fields based on WebGL technology, and use ARCore / ARKit virtual-reality fusion visualization; Intelligent decision-making and warning module: Integrates the WRF numerical model and Transformer prediction algorithm to generate meteorological warning information, and pushes decision-making suggestions to users through a natural language interface; Interaction module: used to receive user input and support users to interact with visualized data through at least one of timeline, geographic area selection, and data filtering conditions.

[0007] Preferably, in the multi-source heterogeneous data acquisition and preprocessing module, global meteorological satellite data is accessed through the API interface, satellite cloud images, infrared radiation data and atmospheric temperature vertical profile data are downloaded in real time, and historical satellite data is regularly obtained using the FTP protocol; for the ground sensor network, meteorological station data is accessed through the MQTT protocol, and Doppler radar data is accessed through the TCP / IP protocol to obtain precipitation intensity, storm path and wind field information in real time, and integrate third-party meteorological service APIs to supplement the data coverage of the platform.

[0008] Preferably, in the spatiotemporal fusion and data enhancement module, the meteorological data with inconsistent spatial resolution are interpolated by the Kriging interpolation algorithm, and the discrete point data of the ground meteorological station are interpolated into high-resolution grid data to generate a continuous spatial distribution map; at the same time, a linear interpolation algorithm is used to align the data with inconsistent temporal resolution, and the ground sensor data collected every hour is time-aligned with the satellite data collected every 15 minutes to generate a meteorological data set with a unified time series.

[0009] Based on the time-space alignment, the Bayesian fusion model is used to perform weighted fusion of multi-source meteorological data. By dynamically adjusting the weights of data sources, the module generates a high-confidence comprehensive meteorological data set. Among them, the comprehensive meteorological dataset is a dataset obtained by integrating meteorological data from different sources, after spatiotemporal alignment, weighted fusion and dynamic weight adjustment; When there are conflicts in the spatiotemporal dimensions of multi-source data, the module uses a neural network model based on the attention mechanism to generate confidence-weighted results to ensure the reliability of the fused data.

[0010] The isolation forest algorithm is then used to detect outliers in the comprehensive meteorological data set, and abnormal data points are marked through a visual interface. For the detected abnormal data, the module uses a variational autoencoder to reconstruct and generate alternative values.

[0011] Preferably, in the distributed storage and intelligent indexing module, the time series database InfluxDB is used to store the real-time comprehensive meteorological data set, and InfluxDB organizes data using measurement values, tags, fields, and timestamps; each data point contains a measurement value, a set of tags, one or more fields, and a timestamp; the comprehensive meteorological data set is written to the time series database in real time through the API or client library of InfluxDB; the written data is sorted by timestamp and stored in a specific data bucket; when querying, the query language of InfluxDB is used to filter and aggregate operations by time range, tag, or field; MongoDB uses binary JSON format to store documents, and each document contains data of any structure. Radar image data is stored as binary data with metadata, including timestamp, resolution, and format. The raw weather image data is written to the database through MongoDB's API or client library; each document contains image data and its related metadata. When querying, use MongoDB's query language to filter by time range, location or other metadata.

[0012] Preferably, in the dynamic visualization module, WebGL technology is based on the Web graphics library of OpenGL ES, 3D graphics are directly rendered in the browser, and the GPU is called by JavaScript to perform graphics calculation and 3D rendering; the rendering pipeline of WebGL includes vertex shader and fragment shader, the vertex shader processes vertex data, the fragment shader calculates pixel color, and maps satellite images or meteorological grid data to the 3D model through texture mapping; Among them, vertex data includes map coordinates and meteorological data points; Rendering dynamic maps and 3D meteorological fields based on WebGL technology also includes converting comprehensive meteorological data sets into a format processed by WebGL. When the data is updated, WebGL dynamically adjusts the input of vertex and fragment shaders and renders in real time. In the rendering of 3D meteorological fields, the `Points` object of Three.js is used to create a particle group to simulate dynamic effects such as wind fields and clouds. The position and color of each particle are dynamically updated according to the comprehensive meteorological data set. At the same time, the volume cloud effect is realized through the GLSL shader, and the dynamic changes of the cloud layer are simulated by combining the noise algorithm, and the lighting effect is calculated using ray tracing technology. ARCore and ARKit are mobile augmented reality frameworks, including motion tracking, environmental understanding and anchor management. Motion tracking tracks the device position through the camera and IMU sensor, environmental understanding detects planes and estimates lighting conditions, and anchor management fixes virtual objects at specific locations in the real world. In the superposition of virtual and real data, the 3D original meteorological model generated by WebGL is output to the GLB format and loaded into the AR scene. The position, rotation and scale of the 3D original meteorological model are dynamically adjusted according to the meteorological data to drive the state change of the AR model.

[0013] Preferably, in the intelligent decision-making and warning module, the WRF numerical model performs high-resolution simulations of meteorological phenomena such as typhoon paths and precipitation distribution based on the principles of atmospheric dynamics, and generates minute-level meteorological forecast results, including core parameters such as wind speed, temperature, air pressure, and precipitation; the Transformer prediction algorithm uses historical meteorological data to capture long-term dependencies through a self-attention mechanism to predict future meteorological trends, especially the probability of occurrence of extreme weather events (such as heavy rains and typhoons) and short-term fluctuations in meteorological parameters. In order to improve the prediction accuracy, the system performs a weighted fusion of the WRF numerical model output and the Transformer prediction results to generate a comprehensive prediction result, and divides it into three levels of warnings: red, orange, and yellow according to the severity of the comprehensive prediction result; For example, when the typhoon path prediction error is less than 50 kilometers and the wind speed exceeds level 12, the system will trigger a red warning.

[0014] After the warning information is generated, the BERT model converts structured data into natural language descriptions and generates decision recommendations for different users.

[0015] Preferably, in the interaction module, the user is allowed to interact with the data visualization platform of the meteorological service center by integrating the timeline, geographic area selection and data screening conditions; the timeline interaction enables the user to select a time period by sliding a bar or clicking, dynamically requesting the backend API to obtain corresponding data according to the time range selected by the user, and updating the visualization chart in real time; The geographic area selection function is based on the map library. Users can select the target area using the rectangle, circle or polygon tool. The system will automatically parse the longitude and latitude boundaries of the selected area, obtain the meteorological data within the longitude and latitude boundaries of the selected area from the backend, and update the visualization view. The data filtering conditions provide UI components such as drop-down menus, sliders, and multiple-selection boxes, which support users to filter data by a combination of temperature, precipitation, and wind speed parameters. For example, data points with a temperature higher than 30°C and a precipitation less than 5 mm can be filtered. The system converts the conditions into API query parameters, filters out the data sets that meet the conditions from the database, and dynamically renders them to the front-end chart or map.

[0016] Technical effects and advantages of the present invention: The present invention acquires diversified meteorological information by collecting data sources from meteorological satellites, ground sensors and meteorological radars, and using API interfaces, MQTT and TCP / IP protocols. The original meteorological data is subjected to denoising and compression processing through edge computing nodes, and multi-source data is integrated and a comprehensive meteorological data set is generated through Kriging interpolation and Bayesian fusion methods. The isolation forest algorithm is then used to detect outliers, and the data is reconstructed in combination with a variational autoencoder. The 3D meteorological field rendering and augmented reality functions realized based on WebGL technology enable users to analyze meteorological data in an intuitive environment. Users can flexibly interact with visual charts through timelines, geographic area selections and data screening methods, and update the displayed content in real time. The present invention provides users with accurate meteorological warnings by combining the WRF numerical model with the Transformer algorithm, and at the same time builds a geographic spatial index based on the GeoHash algorithm to achieve efficient geographic spatial query and data retrieval, and provides a real-time feedback mechanism and multi-dimensional visual views to optimize user operation processes and experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of module connection of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] See also Figure 1 As shown, the present invention provides a data visualization and interaction platform for a meteorological service center, which includes a multi-source heterogeneous data acquisition and preprocessing module, a spatiotemporal fusion and data enhancement module, a distributed storage and intelligent indexing module, a dynamic visualization module, an intelligent decision-making and early warning module, and an interaction module.

[0020] Multi-source heterogeneous data acquisition and preprocessing module: used to collect meteorological data in real time from meteorological satellites, ground sensor networks, meteorological radars and third-party APIs, obtain raw meteorological data, and perform noise reduction on the raw meteorological data through edge computing nodes; In the multi-source heterogeneous data acquisition and preprocessing module, global meteorological satellite data is accessed through the API interface, and satellite cloud images, infrared radiation data and atmospheric temperature vertical profile data are downloaded in real time. At the same time, the FTP protocol is used to regularly obtain historical satellite data; for the ground sensor network, the widely distributed meteorological station data is accessed through the MQTT protocol, and the Doppler radar data is accessed through the TCP / IP protocol to obtain precipitation intensity, storm path and wind field information in real time, and the OpenWeatherMap third-party meteorological service API is integrated to supplement the data coverage of the platform.

[0021] After data collection is completed, the raw data is preprocessed using lightweight computing nodes deployed at the edge of the sensor network. The edge computing nodes are rapidly deployed and expanded based on Docker container technology, and run wavelet transform denoising and compression programs. The specific method of using wavelet transform algorithm to reduce noise of original meteorological data is as follows: Step A1: Select wavelet basis function and decomposition level; Step A2: performing wavelet decomposition on the original meteorological data signal to obtain approximate coefficients and detail coefficients; The calculation method of wavelet decomposition is as follows: ,in, Represented as the original meteorological data signal, It is represented as the approximate coefficient of the b-th layer and is represented as the low-frequency component of the signal; It is represented as the detail coefficient of the bth layer, indicating the high-frequency component of the signal, C is represented as the number of decomposition layers, and b is represented as the layer index of the wavelet decomposition, indicating the scale of the current decomposition; Step A3: Threshold processing is performed on the detail coefficients. The specific calculation method is as follows: ,in, It is represented as the detail coefficient of the bth layer after threshold processing, is represented as the threshold, It is represented as the detail coefficient of the b-th layer, and sign is represented as the sign function; The detail coefficient below the threshold is regarded as useless signal, and the detail coefficient is set to 0; the detail coefficient above the threshold is regarded as useful signal, and the detail coefficient is reduced; The threshold calculation method is as follows: ,in, is represented as the threshold, Expressed as the noise standard deviation, Expressed as signal length; Among them, the signal length is the number of time steps of the original meteorological data; the signal length is used to adjust the size of the threshold to adapt to signals of different lengths.

[0022] The calculation method of noise standard deviation is as follows: ,in, Expressed as the noise standard deviation, Expressed as the median, Expressed as the detail coefficient of the b-th layer; Step A4: reconstruct the noise-reduced signal by inverse wavelet transforming the approximate coefficients and the detail coefficients after threshold processing. The specific calculation method is: ,in, Represented as the original meteorological data signal after noise reduction, It is represented as the approximate coefficient of the b-th layer, it is represented as the detail coefficient of the b-th layer after threshold processing, C is represented as the number of decomposition layers, b is represented as the layer index of wavelet decomposition, and it represents the scale of the current decomposition; The above steps can remove the noise in the original meteorological data.

[0023] Spatiotemporal fusion and data enhancement module: Based on the Kriging interpolation algorithm and Bayesian fusion model, the original meteorological data is spatiotemporally aligned and fused to generate a comprehensive meteorological data set, and the missing data is filled in through the LSTM neural network; In the spatiotemporal fusion and data enhancement module, the meteorological data with inconsistent spatial resolution are interpolated by the Kriging interpolation algorithm, and the discrete point data of the ground meteorological station are interpolated into high-resolution grid data to generate a continuous spatial distribution map; at the same time, the linear interpolation algorithm is used to align the data with inconsistent temporal resolution, and the ground sensor data collected every hour and the satellite data collected every 15 minutes are time-aligned to generate a meteorological data set with a unified time series.

[0024] Based on the time-space alignment, the Bayesian fusion model is used to perform weighted fusion of multi-source meteorological data. By dynamically adjusting the weights of data sources, a high-confidence comprehensive meteorological data set is generated. For example, in typhoon path prediction, the module integrates satellite cloud images, ground wind speed data and radar echo data to generate more accurate typhoon path prediction results; When there are conflicts in the spatiotemporal dimensions of multi-source data, a neural network model based on the attention mechanism is used to generate confidence-weighted results to ensure the reliability of the fused data.

[0025] The isolation forest algorithm is then used to detect outliers in the comprehensive meteorological data set, and abnormal data points are marked through a visual interface. For the detected abnormal data, the module uses a variational autoencoder to reconstruct and generate alternative values; The calculation method for detecting outliers in meteorological data using the isolation forest algorithm is as follows: Step B1: randomly select a feature c from the feature matrix E; the data in the feature matrix is ​​a comprehensive meteorological data set; use a density-based segmentation point selection method to select a segmentation point d in the data sparse area; The calculation method of the feature matrix E is as follows: , where E represents the feature matrix and h represents the number of samples. Expressed as the number of features; Step B2: Build an isolation tree, divide the comprehensive meteorological data set into two subsets according to the split point d, and recursively repeat step B1 for each subset until there is only one sample in the subset or the tree reaches the maximum depth; the maximum depth is ; Step B3: Repeat steps B1 to B2 to construct f isolated trees to form an isolation forest. Step B4: Calculate the path length, for each sample , traverse each isolated tree and calculate the path length from the root node to the leaf node; then take the average of the path lengths of all isolated trees to get the sample The average path length of Step B5: Calculate the outlier value. The specific method for calculating the outlier value is: ,in, Represented as a sample The anomaly score ranges from between, is expressed as a normalization factor, It is represented as the average path length, and h is the number of samples; The calculation method of the normalization factor is as follows: ,in, is expressed as a normalization factor, Expressed as a harmonic number, , h represents the sample size; The calculation method of harmonic number is as follows: ,in, It is expressed as a harmonic number; Step B6: Determine the outlier value. The calculated outlier value range is ,but is an outlier. If the calculated outlier is within the range ,but is the normal value; Outliers in the comprehensive meteorological data set are removed through steps B1 to B6.

[0026] The method of using variational autoencoders to reconstruct outliers in a comprehensive meteorological dataset is as follows: Step D1: Use the isolation forest algorithm to detect outliers in the comprehensive meteorological data set and mark the abnormal data points; standardize the comprehensive meteorological data set to make its mean 0 and standard deviation 1; then use the normal data as a training set to train the VAE model; Step D2: Define the encoder and decoder using a deep learning framework; then train the VAE model using normal data to optimize the reconstruction loss and KL divergence; Step D3: Input the abnormal data into the encoder to obtain the mean and variance of the latent space, and then , the variance is Sample latent variable m from the normal distribution of ; input the latent variable m into the decoder to generate reconstructed data ,Then the reconstructed data is de-normalized to restore the range of the comprehensive meteorological dataset; Step D4: Plot the abnormal data and the reconstructed data in the comprehensive meteorological data set in the same chart to visually compare the recovery effect, and finally calculate the error between the reconstructed data and the true value; Through steps D1 to D4, outliers in the comprehensive meteorological data set can be reconstructed so that the outliers are reconstructed into normal values ​​and substitute values ​​are generated; For example, when the temperature data of a ground weather station is abnormal, the module reconstructs the temperature value of the station based on the data of neighboring stations; Use LSTM neural network to predict and fill in the missing comprehensive meteorological data set; The specific method of LSTM neural network for comprehensive meteorological data set prediction and filling is as follows: Step C1: Divide the comprehensive meteorological dataset into a training set and a test set, and then format the comprehensive meteorological dataset into the form of LSTM input; Step C2: Use the deep learning framework to build an LSTM model and configure the network structure and hyperparameters. The model includes an input layer, an LSTM layer, and a fully connected layer. Among them, the input layer is used to receive sequence data; the LSTM layer is used to define the number and parameters of LSTM units; the fully connected layer is after the LSTM layer and is used for output prediction; Step C3: Input the training set into the LSTM model for training, update the weights through the back-propagation algorithm until the loss function converges, and use the validation set to monitor overfitting during the training process; Step C4: After the LSTM model training is completed, the performance of the LSTM model is evaluated using the test set and the evaluation index is calculated; Step C5: Input the time series containing missing values ​​into the trained LSTM model, generate predicted missing values ​​through the LSTM model, and finally fill the predicted missing values ​​back into the comprehensive meteorological dataset to form the final dataset; Among them, the time series is the daily meteorological data, which is the observation record from the set start time to the end time; Among them, the evaluation indicators include root mean square error and determination coefficient.

[0027] Distributed storage and intelligent indexing module: Use the time series database InfluxDB to store real-time comprehensive meteorological data sets, use MongoDB to store unstructured radar images, and build geospatial indexes based on the GeoHash algorithm; In the distributed storage and intelligent indexing module, the time series database InfluxDB is used to store real-time comprehensive meteorological data sets. InfluxDB uses measurement values, tags, fields, and timestamps to organize data. Each data point contains a measurement value, a set of tags, one or more fields, and a timestamp. The comprehensive meteorological data set is written to the time series database in real time through the API or client library of InfluxDB. The written data is sorted by timestamp and stored in a specific data bucket. When querying, the query language of InfluxDB is used to filter and aggregate operations by time range, tag, or field. For example, you can query the temperature trend of a weather station in the past hour, or calculate the average humidity value of a certain area; MongoDB uses binary JSON format to store documents, and each document contains data of any structure. Radar image data is stored as binary data with metadata, including timestamp, resolution, and format. The raw weather image data is written to the database through MongoDB's API or client library; each document contains image data and its related metadata; when querying, MongoDB's query language is used to filter by time range, location or other metadata; Unstructured radar images refer to data obtained from radar systems that are not organized or labeled according to a fixed data model; radar systems record targets by emitting electromagnetic waves and receiving reflected signals; radar waves are reflected when passing through different objects to form image data.

[0028] For example, you can query radar image data for a certain area within a certain time period, or find satellite cloud images at a specific resolution; The GeoHash algorithm regards the earth's surface as a two-dimensional plane based on longitude and latitude coordinates, and divides the plane into smaller sub-blocks through recursive binary division. Each binary division uses 0 or 1 to represent the interval to which the target point belongs according to its location. The binary codes of latitude and longitude are interleaved and merged, and then Base32 encoded to finally obtain the GeoHash string. When building a geospatial index, the corresponding GeoHash code of the points in the geographic space is calculated according to their latitude and longitude coordinates, and the code and point information are stored in the database or index structure; when retrieving points near a certain point, the GeoHash code of the target point is used to find codes with the same or similar prefixes, so that the adjacent points can be quickly located; For example, when you need to query other stations near a certain weather station, you can quickly locate the adjacent geographical area by matching the first few bits of the GeoHash code. Assuming the GeoHash code of the target weather station is wx4g0, you can find nearby weather stations by querying all codes with the prefix wx4g; GeoHash is an algorithm that encodes geographic coordinates into strings. It is suitable for building geospatial indexes and supports efficient geographic location queries. Through GeoHash encoding, the geographic location information of meteorological data can be converted into strings and indexes can be built based on these strings.

[0029] Dynamic visualization module: used to render dynamic maps and 3D weather fields based on WebGL technology, and use ARCore / ARKit virtual-reality fusion visualization; In the dynamic visualization module, WebGL technology is based on the Web graphics library of OpenGL ES, which directly renders 3D graphics in the browser, and calls the GPU to perform graphics calculations and 3D rendering through JavaScript; the rendering pipeline of WebGL includes vertex shaders and fragment shaders, the vertex shader processes vertex data, the fragment shader calculates pixel color, and maps satellite images or meteorological grid data onto the 3D model through texture mapping; Among them, vertex data includes map coordinates and meteorological data points; satellite image data is obtained through satellite remote sensing technology. The remote sensors on the satellite capture the electromagnetic wave signals reflected or emitted from the earth's surface. After processing and analysis, high-resolution satellite images are formed; meteorological raster data is obtained through meteorological observation stations, satellite remote sensing, radar and other means. After processing and gridding, the data is represented in the form of a two-dimensional grid, and each grid unit contains the meteorological element value of the area; Rendering dynamic maps and 3D meteorological fields based on WebGL technology also includes converting comprehensive meteorological data sets into a format that can be processed by WebGL. When the data is updated, WebGL dynamically adjusts the input of vertex and fragment shaders and renders in real time. In 3D meteorological field rendering, the `Points` object of Three.js is used to create a particle group to simulate dynamic effects such as wind fields and clouds. The position and color of each particle are dynamically updated according to the original meteorological data. At the same time, the volume cloud effect is realized through the GLSL shader, and the dynamic changes of the cloud layer are simulated by combining the noise algorithm, and the lighting effect is calculated using ray tracing technology. ARCore and ARKit are mobile augmented reality frameworks, including motion tracking, environmental understanding and anchor management. Motion tracking tracks the device position through the camera and IMU sensor, environmental understanding detects planes and estimates lighting conditions, and anchor management fixes virtual objects at specific locations in the real world. In the superposition of virtual and real data, the 3D original meteorological model generated by WebGL is output to the GLB format and loaded into the AR scene. The position, rotation and scale of the 3D original meteorological model are dynamically adjusted according to the meteorological data to drive the state change of the AR model. Users can customize charts and select meteorological parameters through drop-down menus, filter data for specific time and area through date selectors and map area selection tools, support multiple chart types such as line charts, heat maps, wind rose charts, etc., dynamically generate charts based on the data source and filtering conditions selected by the user, and realize the linkage between charts and 3D maps or AR scenes; the system reduces memory usage by loading data in blocks, loads meteorological data in blocks by area or time, uses multi-level of detail technology to adjust the details of 3D models according to the viewing distance, and distributes data parsing and calculation tasks to background threads through Web Worker multi-threading; WebGL downgrades old browsers to 2D Canvas rendering through Feature Detection. The AR adaptation solution includes using the WebXR standard on the web to support the latest versions of Chrome and Firefox, and exporting iOS / Android native applications on the mobile side through Unity.

[0030] Intelligent decision-making and warning module: Integrates the WRF numerical model and Transformer prediction algorithm to generate meteorological warning information, and pushes decision-making suggestions to users through a natural language interface; In the intelligent decision-making and warning module, the WRF numerical model simulates typhoon paths, precipitation distribution and other meteorological phenomena with high resolution based on the principles of atmospheric dynamics, and generates minute-level meteorological forecast results, including core parameters such as wind speed, temperature, air pressure and precipitation. The Transformer prediction algorithm uses historical meteorological data to capture long-term dependencies through a self-attention mechanism to predict future meteorological trends, especially the probability of occurrence of extreme weather events (such as heavy rains and typhoons) and short-term fluctuations in meteorological parameters. In order to improve the prediction accuracy, the output of the WRF numerical model is weightedly fused with the prediction results of the Transformer to generate a comprehensive prediction result, which is then divided into three levels of warning: red, orange and yellow according to the severity of the comprehensive prediction result. For example, when the typhoon path prediction error is less than 50 kilometers and the wind speed exceeds level 12, the system will trigger a red warning.

[0031] After the warning information is generated, the BERT model converts structured data into natural language descriptions and generates decision recommendations for different users; The specific method of using the Transformer prediction algorithm based on historical meteorological data is as follows: Assume that historical meteorological data is a time series ,in, represents the meteorological parameter vector at time t, where t represents the time step, , R represents a real number set, n represents the feature dimension, and T represents the length of the time series; the relationship between time t and time series T is: ; Map the time series X to a high-dimensional space to get the input embedding ,in, Represented as the embedding vector at time t, ; Represented as model dimensions; Among them, the model dimension is the dimension of the embedding vector, which represents the representation of the input data at each time step in the high-dimensional feature space; The calculation method of the embedding vector is as follows: ,in, Represented as an embedding vector, Expressed as an embedding weight matrix, Represented as the meteorological parameter vector at time t, Represented as a bias vector in the embedding vector; Since Transformer has no explicit time information, position encoding is added ,in, It is expressed as the position encoding of time step t, and the calculation method is as follows: ,in, is represented as the sth dimension of the position encoding, t is represented as the time step, Represented as model dimensions; Add the input embedding and position encoding to get the input of the Transformer ; Then perform a linear transformation on the Transformer input W to obtain the query matrix, key matrix and value matrix; the specific calculation method of the query matrix is: , where I represents the query matrix and W represents the input of Transformer. Represented as a learnable weight matrix; The calculation method of the key matrix is ​​as follows: , where P represents the key matrix and W represents the input of Transformer. Represented as a learnable weight matrix; The calculation method of the value matrix is ​​as follows: , where U represents the value matrix and W represents the input of Transformer. Represented as a learnable weight matrix; Calculate the dot product of the query and the key, and normalize it through the Sofamax function to get the attention score matrix. The calculation method is as follows: , where Y represents the attention score matrix, I represents the query matrix, and P represents the key matrix. Represented as transposed, It is represented as the dimensions of query matrix and key matrix; Use the attention score to perform weighted summation on the value matrix to obtain the self-attention output. The specific calculation method is: , where y represents the self-attention output, Y represents the attention score matrix, and Y1 represents the attention score pair matrix; In order to capture the features of different subspaces, Transformer uses a multi-head attention mechanism to split the query matrix, key matrix, and value matrix into v heads respectively. Each head calculates the attention output independently. The specific calculation method is: , where K represents the single-head attention output, Represented as the query matrix of the vth head, is represented as the key matrix of the vth head, Represented as the value matrix of the vth head, Represented as a single-head attention calculation function; The outputs of all heads are concatenated and linearly transformed to obtain the final multi-head attention output. The calculation method is as follows: ,in, Represented as multi-head attention output, Concat is represented as an operation used to concatenate multiple vectors or matrices according to the specified dimension; It is represented as the output weight matrix, and z is the number of heads; After the self-attention mechanism, Transformer uses a feedforward neural network to further extract features; the multi-head attention output is subjected to two linear transformations and nonlinear activations. The specific calculation method is: ,in, Represented as the matrix input to the feedforward neural network, and Expressed as a weight matrix, , Represented as a bias vector; The long-term dependencies of historical meteorological data are captured through the self-attention mechanism, and features are extracted by combining multi-head attention and feedforward neural networks to obtain the final decoding output, which is then linearly transformed to obtain the final prediction result.

[0032] Interaction module: used to receive user input and support users to interact with visualized data through at least one of timeline, geographic area selection, and data filtering conditions; In the interactive module, by integrating the timeline, geographic area selection and data screening conditions, users can interact with the meteorological service center data visualization platform in an intuitive and flexible way; the timeline interaction allows users to select a time period by sliding a bar or clicking, dynamically requesting the backend API to obtain corresponding data according to the time range selected by the user, and updating the visualization chart in real time; For example, if the user drags the timeline to the typhoon active period, the typhoon path and precipitation heat map on the map will simultaneously display the meteorological changes during that period; The geographic area selection function is based on the map library. Users can select the target area using the rectangle, circle or polygon tool. The system will automatically parse the longitude and latitude boundaries of the selected area, obtain the meteorological data within the longitude and latitude boundaries of the selected area from the backend, and update the visualization view. For example, after selecting a coastal area, the system only displays the real-time wind speed and storm surge forecast information for that area; The data filtering conditions provide UI components such as drop-down menus, sliders, and multiple-selection boxes, which support users to filter data by temperature, precipitation, and wind speed parameter combinations. For example, the system can filter data points with a temperature higher than 30°C and a precipitation less than 5 mm. The system converts the conditions into API query parameters, filters out the data sets that meet the conditions from the database, and dynamically renders them to the front-end chart or map. The three interaction methods work together through a unified interface. Users can simultaneously select the combination of conditions of "Summer 2023", "Yangtze River Basin", and "Daily precipitation exceeds 100 mm". The system automatically aggregates the filtering logic of time, space, and attribute dimensions to generate a compound query request. After the back-end returns the data, the front-end synchronously updates the multi-dimensional visualization view, such as a 3D precipitation distribution map superimposed on a time series line chart. To optimize the user experience, the interactive module also provides a real-time feedback mechanism, such as displaying a progress bar or animated prompts when data is loading, to ensure the consistency of user operations and the efficiency of system responses. Meteorological researchers quickly locate historical extreme weather events through the timeline, analyze the spatial impact range in combination with geographic area selection, and then use data screening to compare the correlation between different meteorological parameters.

[0033] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A data visualization and interaction platform for a meteorological service center, characterized in that: include: Multi-source heterogeneous data acquisition and preprocessing module: used to collect meteorological data in real time from meteorological satellites, ground sensor networks, meteorological radars and third-party APIs, obtain raw meteorological data, and perform noise reduction on the raw meteorological data through edge computing nodes; Spatiotemporal fusion and data enhancement module: Based on the Kriging interpolation algorithm and Bayesian fusion model, the original meteorological data is spatiotemporally aligned and fused to generate a comprehensive meteorological data set, and the missing data is filled in through the LSTM neural network; Distributed storage and intelligent indexing module: Use the time series database InfluxDB to store real-time comprehensive meteorological data sets, use MongoDB to store unstructured radar images, and build geospatial indexes based on the GeoHash algorithm; Dynamic visualization module: used to render dynamic maps and 3D weather fields based on WebGL technology, and use ARCore / ARKit virtual-reality fusion visualization; Intelligent decision-making and warning module: Integrates the WRF numerical model and Transformer prediction algorithm to generate meteorological warning information, and pushes decision-making suggestions to users through a natural language interface; Interaction module: used to receive user input and support users to interact with visualized data through at least one of timeline, geographic area selection, and data filtering conditions.

2. A meteorological service center data visualization and interaction platform according to claim 1, characterized in that: In the multi-source heterogeneous data acquisition and preprocessing module, global meteorological satellite data is accessed through the API interface, satellite cloud images, infrared radiation data and atmospheric temperature vertical profile data are downloaded in real time, and historical satellite data is regularly obtained using the FTP protocol; For the ground sensor network, the weather station data is accessed through the MQTT protocol, and the Doppler radar data is also accessed through the TCP / IP protocol to obtain precipitation intensity, storm path and wind field information in real time, and integrate third-party meteorological service APIs to supplement the platform's data coverage; After data collection is completed, the raw data is preprocessed using lightweight computing nodes deployed at the edge of the sensor network; The edge computing nodes run wavelet transform denoising and compression programs based on Docker container technology and extensions.

3. A meteorological service center data visualization and interaction platform according to claim 1, characterized in that: In the spatiotemporal fusion and data enhancement module, the meteorological data with inconsistent spatial resolutions are interpolated by the Kriging interpolation algorithm, and the discrete point data of the ground meteorological station are interpolated into high-resolution grid data to generate a continuous spatial distribution map; at the same time, the linear interpolation algorithm is used to align the data with inconsistent temporal resolutions, and the ground sensor data collected every hour is aligned with the satellite data collected every 15 minutes to generate a unified time series meteorological data set; Based on the time-space alignment, the Bayesian fusion model is used to perform weighted fusion of multi-source meteorological data; Generate a high-confidence comprehensive meteorological dataset by dynamically adjusting the weights of data sources; When there are conflicts in multi-source data in the temporal and spatial dimensions, a neural network model based on the attention mechanism is used to generate confidence-weighted results. The isolation forest algorithm is then used to detect outliers in the comprehensive meteorological data set, and abnormal data points are marked through a visual interface. For the detected abnormal data, a variational autoencoder is used to reconstruct it and generate alternative values.

4. A meteorological service center data visualization and interaction platform according to claim 3, characterized in that: The calculation method for detecting outliers in a comprehensive meteorological dataset using the isolation forest algorithm is as follows: Step B1: randomly select a feature c from the feature matrix E; the data in the feature matrix is ​​a comprehensive meteorological data set; use a density-based segmentation point selection method to select a segmentation point d in the data sparse area; The calculation method of the feature matrix E is as follows: , where E represents the feature matrix and h represents the number of samples. Expressed as the number of features; Step B2: Build an isolation tree, divide the comprehensive meteorological data set into two subsets according to the split point d, and recursively repeat step B1 for each subset until there is only one sample in the subset or the tree reaches the maximum depth; the maximum depth is ; Step B3: Repeat steps B1 to B2 to construct f isolated trees to form an isolation forest. Step B4: Calculate the path length, for each sample , traverse each isolated tree and calculate the path length from the root node to the leaf node; then take the average of the path lengths of all isolated trees to get the sample The average path length of Step B5: Calculate the outlier value. The specific method for calculating the outlier value is: ,in, Represented as a sample The anomaly score ranges from between, Expressed as a normalization factor, It is represented as the average path length, and h is the number of samples; The calculation method of the normalization factor is as follows: ,in, Expressed as a normalization factor, Expressed as a harmonic number, , h represents the sample size; The calculation method of harmonic number is as follows: ,in, It is expressed as a harmonic number; Step B6: Determine the outlier value. The calculated outlier value range is ,but is an outlier. If the calculated outlier is within the range ,but is the normal value; Outliers in the comprehensive meteorological data set are removed through steps B1 to B6.

5. The data visualization and interaction platform of a meteorological service center according to claim 1, characterized in that: In the distributed storage and intelligent indexing module, the time series database InfluxDB is used to store real-time comprehensive meteorological data sets. InfluxDB organizes data using measurement values, tags, fields, and timestamps. Each data point contains a measurement value, a set of tags, one or more fields, and a timestamp. The comprehensive meteorological data set is written to the time series database in real time through the API or client library of InfluxDB. The written data is sorted by timestamp and stored in a specific data bucket. When querying, use InfluxDB's query language to filter and aggregate by time range, tag, or field.

6. A meteorological service center data visualization and interaction platform according to claim 1, characterized in that: The specific method of building a geospatial index based on the GeoHash algorithm is: The GeoHash algorithm regards the earth's surface as a two-dimensional plane based on longitude and latitude coordinates, and divides the plane into smaller sub-blocks through recursive binary division. Each binary division uses 0 or 1 to represent the interval to which the target point belongs according to its location. The binary codes of latitude and longitude are interleaved and merged, and then Base32 encoded to finally obtain the GeoHash string. When constructing a geospatial index, the corresponding GeoHash code of the points in the geographic space is calculated according to their latitude and longitude coordinates, and the code and point information are stored in the database or index structure; When retrieving points near a certain point, you can quickly locate nearby points by searching for codes with the same prefix based on the GeoHash code of the target point.

7. The data visualization and interaction platform of a meteorological service center according to claim 1, characterized in that: In the dynamic visualization module, WebGL technology is based on the Web graphics library of OpenGLES, which directly renders 3D graphics in the browser, and calls the GPU for graphics calculation and 3D rendering through JavaScript; the rendering pipeline of WebGL includes vertex shaders and fragment shaders, the vertex shader processes vertex data, the fragment shader calculates pixel color, and maps satellite images or meteorological grid data to the 3D model through texture mapping.

8. A data visualization and interaction platform for a meteorological service center according to claim 6, characterized in that: in, Rendering dynamic maps and 3D meteorological fields based on WebGL technology also includes converting comprehensive meteorological data sets into a format processed by WebGL. When the data is updated, WebGL dynamically adjusts the input of vertex and fragment shaders and renders in real time. In the rendering of 3D meteorological fields, the `Points` object of Three.js is used to create a particle group to simulate the dynamic effects of wind fields and clouds. The position and color of each particle are dynamically updated according to the comprehensive meteorological data set. At the same time, the volume cloud effect is realized through the GLSL shader, and the dynamic changes of the cloud layer are simulated by combining the noise algorithm, and the lighting effect is calculated using ray tracing technology. ARCore and ARKit are mobile augmented reality frameworks, including motion tracking, environmental understanding and anchor management. Motion tracking tracks the device position through the camera and IMU sensor, environmental understanding detects planes and estimates lighting conditions, and anchor management fixes virtual objects at specific locations in the real world. In the superposition of virtual and real data, the 3D original meteorological model generated by WebGL is output to the GLB format and loaded into the AR scene. The position, rotation and scale of the 3D original meteorological model are dynamically adjusted according to the meteorological data to drive the state change of the AR model.

9. A meteorological service center data visualization and interaction platform according to claim 1, characterized in that: In the intelligent decision-making and warning module, the WRF numerical model simulates the meteorological phenomena of typhoon paths and precipitation distribution based on the principles of atmospheric dynamics to generate minute-level meteorological forecast results; the Transformer prediction algorithm uses historical meteorological data to capture long-term dependencies through a self-attention mechanism to predict future meteorological trends; the WRF numerical model output is weightedly fused with the Transformer prediction results to generate a comprehensive prediction result, which is then divided into three levels of warning: red, orange, and yellow according to the severity of the comprehensive prediction results.

10. The data visualization and interaction platform of a meteorological service center according to claim 1, characterized in that: In the interactive module, the user is allowed to interact with the data visualization platform of the meteorological service center by integrating the timeline, geographic area selection and data screening conditions; the timeline interaction allows the user to select a time period by sliding the bar or clicking, dynamically requesting the backend API to obtain the corresponding data according to the time range selected by the user, and updating the visualization chart in real time; The geographic area selection function is based on the map library. Users can select the target area using the rectangular, circular or polygonal tools. The system automatically parses the longitude and latitude boundaries of the selected area, obtains the meteorological data within the longitude and latitude boundaries of the selected area from the backend, and updates the visualization view.

Images