Rail transit meteorological monitoring and early warning method and system based on multi-source data fusion

Abstract

The invention relates to a rail transit meteorological monitoring and early warning method and system based on multi-source data fusion, and the method comprises the steps: carrying out the multi-source data fusion of a satellite cloud picture, radar echoes, grid data, a national meteorological station, an intelligent meteorological station, and a water level monitoring station; customized, specialized and refined meteorological service product contents such as weather live monitoring, forecast data, early warning information and alarm information are provided in a visual form, 24-hour uninterrupted all-weather meteorological disaster monitoring service is carried out on a rail transit line, and efficient, rapid and straight-through transmission of meteorological information is realized. And subway operation safety and passenger travel safety are ensured.

Description

Technical Field

[0001] This application relates to the field of meteorological monitoring technology, and in particular to a method and system for meteorological monitoring and early warning of rail transit based on multi-source data fusion. Background Technology

[0002] As the backbone of urban public transportation, the safety and efficiency of rail transit directly affect the safe operation of rail transit, public safety, and socio-economic stability. In recent years, extreme weather events have occurred frequently around the world. Heavy rainfall, high temperatures, cold waves, and blizzards have led to frequent occurrences of water accumulation on rail transit tracks, resulting in backflow of rainwater, track icing, and snow accumulation. These phenomena not only reduce the wheel-rail adhesion coefficient, causing risks such as train slippage and increased braking distance, but may also cause secondary disasters such as signal equipment failure and catenary icing, seriously threatening train operation safety.

[0003] Currently, meteorological monitoring of rail transit mainly relies on the following technical means: deploying sensor equipment along the track to collect basic parameters such as track surface temperature, water film thickness, and the state of the covering; and using traditional algorithms such as weighted averaging and threshold judgment to fuse data from similar sensors, such as using temperature sensor thresholds to determine whether icing is possible and classifying water accumulation levels based on water film thickness sensor readings.

[0004] The aforementioned technologies focus on the simple fusion of single parameters or data from similar sensors, without fully considering the spatiotemporal correlation of multiple parameters such as temperature, water film thickness, and cover condition. This results in low data utilization and makes it difficult to reflect the comprehensive impact of meteorological conditions.

[0005] Based on this, this application provides a method and system for monitoring and early warning of meteorological conditions in rail transit based on multi-source data fusion. Summary of the Invention

[0006] To address the problem of low data utilization and difficulty in reflecting the comprehensive impact of meteorological conditions caused by focusing on simple fusion of single parameters or similar sensor data without fully considering the spatiotemporal correlation of multiple parameters such as temperature, water film thickness, and cover condition, this application provides a method and system for monitoring and early warning of rail transit meteorological conditions based on multi-source data fusion.

[0007] Firstly, this application provides a method for monitoring and early warning of meteorological conditions in rail transit based on multi-source data fusion, employing the following technical solution: including: Monitoring equipment is deployed along rail transit lines, stations, and key impact areas to collect meteorological parameters in real time. These meteorological parameters are then transmitted to edge computing nodes via a hybrid transmission network to obtain the raw monitoring dataset. Outlier removal and missing value imputation are performed on the original monitoring dataset, and the multi-source data are mapped to a three-dimensional time-space-parameter matrix by timestamps and orbital mileage markers to obtain a standardized feature dataset; Based on the standardized feature dataset, feature enhancement is performed using a hierarchical fusion algorithm to obtain the final fused feature set; Based on the final fused feature set, a risk prediction and graded early warning model is constructed to predict disasters and classify risk levels. The risk distribution along the track is displayed through a visualization platform, and early warning information is pushed out.

[0008] Preferably, the deployment of monitoring equipment along the rail transit line, at stations, and in key impact areas to collect meteorological parameters in real time, and the transmission of these meteorological parameters to edge computing nodes via a hybrid transmission network, yields a raw monitoring dataset, including: Monitoring equipment is deployed along the rail transit lines, at stations, and in key affected areas to collect the meteorological parameters in real time. The meteorological parameters include: temperature parameters, water film thickness parameters, road surface cover status, and macro-meteorological data. Temperature parameters are collected from the track surface and ambient temperature using contact sensors; water film thickness parameters are based on the principle of microwave reflection, using microwave sensors and amplitude-phase detectors to collect the amplitude ratio and phase difference of the reflection coefficient; road surface cover status is collected by sensitive elements to collect changes in the equivalent capacitance and resistance caused by the cover, outputting four status characteristics: dry, waterlogged, icy, and snow-covered; macro-meteorological data is obtained by connecting to urban meteorological stations to acquire wind speed and precipitation data, and by using an image recognition module to collect images of the track surface environment. The meteorological parameters are sent to the edge computing node through a hybrid transmission network to obtain the original monitoring dataset.

[0009] Preferably, the original monitoring dataset undergoes outlier removal and missing value imputation, and multi-source data is mapped to a three-dimensional time-space-parameter matrix using timestamps and orbital odometers to obtain a standardized feature dataset, including: Outlier identification and removal are performed on continuous parameters such as temperature and water film thickness in the original monitoring dataset using statistical criteria. Extreme values ​​that deviate from the normal range are identified by calculating the data distribution characteristics. Data samples that conform to the orbital meteorological monitoring scenario are retained, and abnormal readings caused by sensor noise and environmental interference are eliminated. For the periods of data loss in the original monitoring dataset caused by sensor disconnection, communication delay, or harsh environment, spatiotemporal interpolation is used to fill in the missing values. Multi-source heterogeneous data are aligned according to timestamps to synchronize the collection time of the meteorological parameters. Spatial location matching is achieved through orbital odometers, and the data of each sensor is mapped to the corresponding orbital segment or station coordinates. Based on the aligned time, space and parameter dimension data, a three-dimensional matrix of time-space-parameter is constructed, where the matrix row dimension is the time series, the column dimension is the orbital spatial location, and the depth dimension is the meteorological parameter type, resulting in an initial feature dataset containing time labels, spatial coordinates and the corresponding meteorological parameters. Based on the initial feature dataset, the meteorological parameters in the three-dimensional matrix are standardized, converting parameters of different dimensions into feature values ​​of a unified dimension. This ensures the consistency of multi-source data processing in subsequent hierarchical fusion algorithms and outputs the standardized feature dataset that can be directly used for fusion modeling.

[0010] Preferably, the step of filling in missing values ​​in the original monitoring dataset for periods of data loss due to sensor disconnection, communication delay, or adverse environmental conditions using spatiotemporal interpolation includes: Interpolation is performed on the effective observations before and after the missing data points in the time dimension, and spatial supplementation is performed by combining the spatial correlation data of adjacent monitoring stations to generate complete time series data covering the entire monitoring period.

[0011] Preferably, the step of performing feature enhancement based on the standardized feature dataset using a hierarchical fusion algorithm to obtain the final fused feature set includes: The data from similar sensors in the standardized feature dataset are fused using a weighted average method. The weights are dynamically allocated based on the historical measurement accuracy of each sensor to eliminate systematic biases caused by individual hardware differences and output a set of basic parameters after data-level fusion. Based on the aforementioned set of basic parameters, a dual-model parallel processing architecture is constructed: a GA-BP neural network is used, with temperature, salinity, and original water film thickness as input features. The initial weights and thresholds of the neural network are optimized through a genetic algorithm, and the water film thickness features after compensating for environmental interference are output; a PSO-BP neural network is used, with attenuation ratio, phase difference, and temperature as input features. The network structure parameters are optimized through a particle swarm optimization algorithm, and the classification results of four types of road surface cover states—dry, waterlogged, icy, and snow-covered—are output. The water film thickness characteristics and the classification results of the covering materials are combined with macro-meteorological data and equipment operating status parameters. Multi-source information is fused through Bayesian estimation, and the output bias of the middle layer model is dynamically corrected to generate the final fused feature set containing the compensated water film thickness, covering material status label, and temperature gradient.

[0012] Preferably, the step of constructing a risk prediction and graded early warning model based on the final fused feature set, and performing disaster prediction and risk level classification, includes: Using the time-series data of temperature, water film thickness, and wind speed in the final fused feature set as input, a disaster prediction is performed using the SO-CNN-GRU neural network model: the size of the convolutional kernel and the number of hidden layer nodes of the gated recurrent unit are optimized by the snake optimization algorithm, and the weight of key meteorological features is enhanced by the SE attention mechanism, and the road icing time in the future preset period is output. Based on the final fused feature set and the time of road icing, a multi-parameter risk assessment system is constructed: water accumulation risk level, icing risk level, and cover risk level. Among them, the water accumulation risk level is based on the compensated water film thickness as the core indicator, and is corrected by the ambient temperature, and is divided into four levels of warning; the freezing risk level is based on the predicted freezing time, real-time temperature and water film thickness as joint indicators, and is divided into four levels of warning; the covering risk level is based on the covering status label and combined with the temperature gradient parameter, and outputs the risk level corresponding to dryness, water accumulation, freezing and snow accumulation.

[0013] Preferably, after constructing the multi-parameter risk assessment system based on the final fused feature set and the road surface icing occurrence time, the method further includes: The risk entropy weighting method is used to dynamically allocate the weights of each risk level indicator and calculate the comprehensive risk value: the objective weights of water accumulation, icing and covering risks are determined by information entropy theory, and subjective weights are adjusted by experts. The multi-dimensional risk levels are integrated into a comprehensive early warning level, realizing the improvement from single-parameter early warning to multi-parameter collaborative early warning. Based on historical accident data and real-time monitoring data, the thresholds for each risk level are dynamically updated using the sliding window method: when extreme weather conditions are detected, an emergency threshold correction mechanism is triggered to shorten the early warning response time and output the final early warning result, which includes the comprehensive early warning level, risk location, and duration.

[0014] Secondly, this application discloses a rail transit meteorological monitoring and early warning device based on multi-source data fusion, which adopts the following technical solution, including: The multi-source data module is used to deploy monitoring equipment along rail transit lines, stations and key impact areas to collect meteorological parameters in real time, and send the meteorological parameters to edge computing nodes through a hybrid transmission network to obtain the raw monitoring dataset. The data processing module is used to remove outliers and fill in missing values ​​in the original monitoring dataset, and to map the multi-source data to a three-dimensional time-space-parameter matrix through timestamps and track mileage markers to obtain a standardized feature dataset. The feature fusion module is used to enhance features based on the standardized feature dataset using a hierarchical fusion algorithm to obtain the final fused feature set; The risk prediction module is used to predict disasters and classify risk levels. It also displays the monitoring and forecasting distribution range of major meteorological elements along the rail transit line through a visualization platform and pushes targeted meteorological disaster early warning information.

[0015] Thirdly, this application also provides a control device, the device comprising: It includes a memory and a processor, wherein the memory stores a computer program that can be loaded and executed by the processor, such as the above-described method for monitoring and early warning of rail transit meteorology based on multi-source data fusion.

[0016] Fourthly, this application also provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above regarding the rail transit meteorological monitoring and early warning method based on multi-source data fusion.

[0017] In summary, this application first deploys monitoring equipment along the rail transit line, at stations, and in key impact areas to collect meteorological parameters such as temperature and water film thickness in real time. This data is then transmitted to edge computing nodes via a hybrid transmission network to form a raw monitoring dataset. Subsequently, outlier removal and missing value imputation are performed on the raw data. Multi-source data are then mapped to a three-dimensional time-space-parameter matrix using timestamps and track mileage markers, constructing a standardized feature dataset containing time labels, spatial coordinates, and corresponding meteorological parameters. Based on this standardized feature dataset, feature enhancement is performed using a hierarchical fusion algorithm, specifically including low-level data-level fusion (dynamically allocating weights using a weighted average method for data from similar sensors). The system employs a multi-level fusion approach, involving three stages: eliminating hardware discrepancies, mid-level feature fusion (using GA-BP neural networks to compensate for water film thickness and PSO-BP neural networks to classify cover status), and high-level decision-level fusion (combining macro-meteorological data with Bayesian estimation to correct model biases). This yields a final fused feature set containing compensated water film thickness and cover status labels. Based on this final fused feature set, a risk prediction and graded early warning model is constructed. An SO-CNN-GRU neural network is used to predict icing time. Multi-dimensional risk level classification and risk entropy weighting are used for comprehensive evaluation, and early warning thresholds are dynamically adjusted. Finally, a visualization platform displays the risk distribution along the track and pushes early warning information. Through deep fusion of multi-source data and intelligent algorithms, the system effectively improves the accuracy of meteorological parameter monitoring and risk warning, providing precise decision support for rail transit operation and scheduling, and significantly ensuring train safety under extreme weather conditions. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a method for monitoring and issuing early warnings for rail transit meteorological conditions based on multi-source data fusion.

[0019] Figure 2 This is a structural block diagram of a rail transit meteorological monitoring and early warning device based on multi-source data fusion. Detailed Implementation

[0020] The following combination Figure 1 - Figure 2 This application will be described in further detail.

[0021] The implementing entity of this application is a meteorological monitoring system, whose functional modules are generally divided into 13 major areas: setting function area, login function area, real-time monitoring function area, weather forecast function area, early warning information release function area, alarm information release function area, forecast monitoring and alarm function area, video monitoring function area, data overlay function area for stations along the line, single-station meteorological service function area, decision service function area, historical query function area, and meteorological element threshold setting function area. The system connects to automatic weather stations, water accumulation monitoring stations, video surveillance, and sensors for temperature, humidity, precipitation, wind direction and speed, air pressure, water level, and visibility.

[0022] Reference Figure 1 The embodiments of this application include at least steps S10 to S40.

[0023] S10 deploys monitoring equipment along rail transit lines, stations, and key impact areas to collect meteorological parameters in real time. The meteorological parameters are then sent to edge computing nodes via a hybrid transmission network to obtain the raw monitoring dataset.

[0024] S20 performs outlier removal and missing value imputation on the original monitoring dataset, and maps the multi-source data to a three-dimensional time-space-parameter matrix using timestamps and track mileage markers to obtain a standardized feature dataset.

[0025] S30, based on a standardized feature dataset, uses a hierarchical fusion algorithm to enhance features and obtain the final fused feature set.

[0026] S40, based on the final fused feature set, constructs a risk prediction and graded early warning model to predict disasters and classify risk levels. It also displays the risk distribution along the track through a visualization platform and pushes early warning information.

[0027] Specifically, the process begins by deploying monitoring equipment along the rail transit line to collect meteorological parameters, which are then transmitted to edge nodes via a hybrid transmission network to form a raw monitoring dataset. Next, outlier removal and missing value imputation are performed on the raw data, and it is mapped to a time-space-parameter three-dimensional matrix using timestamps and mileage markers to obtain a standardized feature dataset. Subsequently, a hierarchical fusion algorithm (low-level data, mid-level feature, and high-level decision) is used to enhance the features, generating a final fused feature set. Finally, a risk prediction and graded early warning model is constructed based on this feature set to predict disasters and classify risk levels. The model is then used to display the risk distribution along the rail line and push early warning information through a visualization platform.

[0028] In some embodiments, step S10 specifically includes the following steps: deploying monitoring equipment along the rail transit line, at stations, and in key impact areas to collect meteorological parameters in real time. The meteorological parameters include: temperature parameters, water film thickness parameters, road surface cover status, and macro-meteorological data. The temperature parameters are collected by contact sensors to measure the temperature of the track surface and the environment. The water film thickness parameter is based on the principle of microwave reflection, and the amplitude ratio and phase difference of the reflection coefficient are collected by microwave sensors and amplitude-phase detectors. The road surface cover status is obtained by collecting the changes in equivalent capacitance and resistance caused by the cover through sensitive elements, and outputting four types of status characteristics: dry, water accumulation, icing, and snow accumulation. The macro-meteorological data is obtained by connecting to urban meteorological stations to obtain wind speed and precipitation data, and by collecting images of the track surface environment through an image recognition module. The meteorological parameters are sent to edge computing nodes through a hybrid transmission network to obtain the original monitoring dataset.

[0029] Specifically, monitoring equipment is deployed along the rail transit lines, at stations, and in key affected areas to collect meteorological parameters in real time, including temperature, water film thickness, road surface cover status, and macro-meteorological data. Temperature parameters are collected from the track surface and ambient temperature using contact sensors. Water film thickness parameters are collected based on the microwave reflection principle using microwave sensors and amplitude-phase detectors to collect the amplitude ratio and phase difference of the reflection coefficient. Road surface cover status is collected by sensitive elements to collect changes in the equivalent capacitance and resistance caused by the cover, outputting four status characteristics: dry, waterlogged, icy, and snow-covered. Macro-meteorological data is obtained by connecting to urban meteorological stations to acquire wind speed and precipitation data, and by using image recognition modules to collect images of the track surface environment. These meteorological parameters are then sent to edge computing nodes through a hybrid transmission network to obtain the raw monitoring dataset.

[0030] In some embodiments, step S20 specifically includes the following steps: outlier identification and removal for continuous parameters such as temperature and water film thickness in the original monitoring dataset using statistical criteria; identifying extreme values ​​deviating from the normal range by calculating data distribution characteristics; retaining data samples consistent with the orbital meteorological monitoring scenario; and eliminating abnormal readings caused by sensor noise and environmental interference. For data loss periods in the original monitoring dataset due to sensor disconnection, communication delays, or adverse environments, spatiotemporal interpolation is used to fill in missing values. Multi-source heterogeneous data are aligned according to timestamps to synchronize the acquisition time of meteorological parameters, and spatial positioning is achieved through orbital mileage markers. The matching process maps sensor data to corresponding track segments or station coordinates. Based on the aligned time, space, and parameter dimensions, a three-dimensional time-space-parameter matrix is ​​constructed, where the matrix's row dimension represents the time series, column dimension represents the track spatial location, and depth dimension represents the meteorological parameter type. This yields an initial feature dataset containing time labels, spatial coordinates, and corresponding meteorological parameters. Based on this initial feature dataset, the meteorological parameters in the three-dimensional matrix are standardized, converting parameters with different dimensions into feature values ​​with a unified dimension. This ensures consistent processing of multi-source data by subsequent hierarchical fusion algorithms, outputting a standardized feature dataset that can be directly used for fusion modeling.

[0031] Furthermore, time-dimensional interpolation is performed based on the effective observations before and after the missing data points, and spatial supplementation is performed by combining the spatial correlation data of adjacent monitoring stations to generate complete time series data covering the entire monitoring period.

[0032] Specifically, for continuous parameters such as temperature and water film thickness, statistical criteria are used to identify and remove extreme values ​​that deviate from the normal range, eliminating abnormal readings caused by sensor noise and environmental interference. For data loss caused by sensor disconnection, spatiotemporal interpolation is used, combining effective observations before and after the missing point to perform time dimension interpolation and supplementing spatial correlation data of adjacent stations, generating a complete time series covering the entire time period. Then, multi-source heterogeneous data are aligned to the time dimension by timestamps and matched to spatial locations by track mileage markers, mapping them to the corresponding track segments or station coordinates, constructing a three-dimensional time-space-parameter matrix of rows (time)-columns (space)-depth (parameters), obtaining an initial feature dataset containing time labels, spatial coordinates, and meteorological parameters. Finally, meteorological parameters of different dimensions in the three-dimensional matrix are standardized and converted into uniform dimension feature values, outputting a standardized feature dataset that can be directly used for fusion modeling.

[0033] In some embodiments, step S30 specifically includes the following steps: weighted averaging is used to fuse sensor data of the same type in the standardized feature dataset; weights are dynamically allocated based on the historical measurement accuracy of each sensor to eliminate systematic biases caused by individual hardware differences, and a data-level fused basic parameter set is output; based on the basic parameter set, a dual-model parallel processing architecture is constructed: a GA-BP neural network is used, with temperature, salinity, and original water film thickness as input features; the initial weights and thresholds of the neural network are optimized using a genetic algorithm to output water film thickness features after compensating for environmental interference; a PSO-BP neural network is used, with attenuation ratio, phase difference, and temperature as input features; the network structure parameters are optimized using a particle swarm optimization algorithm to output classification results for four types of road cover states: dry, waterlogged, icy, and snowy; the water film thickness features and cover classification results are combined with macro-meteorological data and equipment operating status parameters, and multi-source information is fused using Bayesian estimation to dynamically correct the output bias of the intermediate model, generating a final fused feature set containing compensated water film thickness, cover state labels, and temperature gradients.

[0034] Specifically, a weighted average method is first applied to data from similar sensors, dynamically allocating weights based on historical measurement accuracy to eliminate hardware discrepancies and obtain a basic parameter set after data-level fusion. Next, a dual-model parallel processing architecture is constructed. A GA-BP neural network is used with temperature, salinity, and original water film thickness as inputs. A genetic algorithm optimizes the weight thresholds to output water film thickness features after compensating for environmental interference. Simultaneously, a PSO-BP neural network is used with attenuation ratio, phase difference, and temperature as inputs. A particle swarm optimization algorithm optimizes the network parameters to classify and output four types of road cover states: dry, waterlogged, icy, and snowy. Finally, the water film thickness features and cover classification results are combined with macro-meteorological data and equipment operating status parameters. Bayesian estimation is used to fuse multi-source information to dynamically correct the output bias of the intermediate model, generating a final fused feature set containing compensated water film thickness, cover state labels, and temperature gradients.

[0035] In some embodiments, step S40 specifically includes the following steps: using the time series data of temperature, water film thickness, and wind speed in the final fused feature set as input, a disaster prediction is performed using an SO-CNN-GRU neural network model: the size of the convolutional kernel and the number of hidden layer nodes of the gated recurrent unit are optimized by the snake optimization algorithm, and the weights of key meteorological features are enhanced by combining the SE attention mechanism to output the road icing occurrence time within a preset future period; based on the final fused feature set and the road icing occurrence time, a multi-parameter risk assessment system is constructed: water accumulation risk level, icing risk level, and cover risk level; Among them, the water accumulation risk level is based on the compensated water film thickness as the core indicator, and is corrected by the ambient temperature, and is divided into four levels of warning; the freezing risk level is based on the predicted freezing time, real-time temperature and water film thickness as joint indicators, and is divided into four levels of warning; the covering risk level is based on the covering status label and combined with the temperature gradient parameter, and outputs the risk level corresponding to dryness, water accumulation, freezing and snow accumulation.

[0036] Furthermore, the risk entropy weighting method is used to dynamically allocate the weights of each risk level indicator and calculate the comprehensive risk value: the objective weights of water accumulation, icing, and covering risks are determined by information entropy theory, and subjective weights are adjusted by experts to integrate the multi-dimensional risk levels into a comprehensive early warning level, thereby improving the early warning from single-parameter early warning to multi-parameter collaborative early warning; based on historical accident data and real-time monitoring data, the thresholds of each risk level are dynamically updated by the sliding window method: when extreme weather conditions are detected, an emergency threshold correction mechanism is triggered to shorten the early warning response time and output the final early warning result including the comprehensive early warning level, risk location, and duration.

[0037] Specifically, the system first uses time-series data on temperature, water film thickness, and wind speed as input. Then, it uses an SO-CNN-GRU neural network (which optimizes the number of convolutional kernels and GRU hidden layer nodes using a snake optimization algorithm, and enhances key features using an SE attention mechanism) to predict the road icing time for a preset period. Next, it constructs a multi-parameter risk assessment system. Water accumulation risk is categorized into four levels based on compensated water film thickness combined with temperature correction. Icing risk is categorized into four levels based on predicted icing time, real-time temperature, and water film thickness. Cover risk is output based on state labels combined with temperature gradients. Finally, it employs a risk entropy weighting method (using information entropy to determine objective weights combined with expert subjective weights) to integrate multi-dimensional risk levels into a comprehensive warning level. The threshold is dynamically updated using a sliding window method, and emergency corrections are triggered by extreme weather to shorten the response time. The final output includes the comprehensive warning level, risk location, and duration, and the warning information is displayed and pushed through a visualization platform.

[0038] The implementation principle of a rail transit meteorological monitoring and early warning method based on multi-source data fusion in this application embodiment is as follows: First, monitoring equipment is deployed along the rail transit line, at stations, and in key impact areas to collect meteorological parameters such as temperature and water film thickness in real time and send them to edge computing nodes through a hybrid transmission network to form a raw monitoring dataset; then, outlier removal and missing value imputation are performed on the raw data, and the multi-source data is mapped to a time-space-parameter three-dimensional matrix through timestamps and track mileage markers to construct a standardized feature dataset containing time labels, spatial coordinates, and corresponding meteorological parameters; based on the standardized feature dataset, feature enhancement is performed through a hierarchical fusion algorithm, specifically including low-level data-level fusion (for sensors of the same type). The data is processed using a weighted average method to dynamically allocate weights to eliminate hardware differences, a mid-level feature-level fusion (using a GA-BP neural network to compensate for water film thickness and a PSO-BP neural network to classify cover status), and a high-level decision-level fusion (combining macro-meteorological data with Bayesian estimation to correct model bias), resulting in a final fused feature set containing compensated water film thickness, cover status labels, etc. Finally, a risk prediction and graded early warning model is constructed based on this final fused feature set. An SO-CNN-GRU neural network is used to predict icing time, and a multi-dimensional risk level classification and risk entropy weighting method are used for comprehensive evaluation. Early warning thresholds are dynamically adjusted, and the risk distribution along the track is displayed through a visualization platform, with early warning information pushed out. Thus, through deep fusion of multi-source data and intelligent algorithms, the accuracy of meteorological parameter monitoring and risk early warning is effectively improved, providing precise decision support for rail transit operation scheduling and significantly ensuring train safety under extreme weather conditions.

[0039] Figure 1 This is a flowchart illustrating a rail transit meteorological monitoring and early warning method based on multi-source data fusion in one embodiment. It should be understood that, although... Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows; unless explicitly stated otherwise, there is no strict order requirement for the execution of these steps, and they can be executed in other orders; and Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0040] Based on the same technical concept, referring to Figure 2 This application also provides a rail transit meteorological monitoring and early warning device based on multi-source data fusion, which adopts the following technical solution. The device includes: The multi-source data module is used to deploy monitoring equipment along rail transit lines, stations and key impact areas to collect meteorological parameters in real time and send the meteorological parameters to edge computing nodes through a hybrid transmission network to obtain the raw monitoring dataset. The data processing module is used to remove outliers and fill in missing values ​​in the original monitoring dataset, and to map multi-source data to a three-dimensional time-space-parameter matrix through timestamps and track mileage markers to obtain a standardized feature dataset. The feature fusion module is used to enhance features based on a standardized feature dataset using a hierarchical fusion algorithm to obtain the final fused feature set. The risk prediction module is used to build a risk prediction and graded early warning model based on the final fused feature set, to predict disasters and classify risk levels, and to display the risk distribution along the track through a visualization platform and push early warning information.

[0041] In some embodiments, the multi-source data module is specifically used to deploy monitoring equipment along rail transit lines, stations, and key impact areas to collect meteorological parameters in real time. The meteorological parameters include: temperature parameters, water film thickness parameters, road surface cover status, and macro-meteorological data. Temperature parameters are collected from the track surface and ambient temperature using contact sensors; water film thickness parameters are based on the principle of microwave reflection, using microwave sensors and amplitude-phase detectors to collect the amplitude ratio and phase difference of the reflection coefficient; road surface cover status is collected by sensitive elements to collect changes in the equivalent capacitance and resistance caused by the cover, outputting four status characteristics: dry, waterlogged, icy, and snow-covered; macro-meteorological data is obtained by connecting to urban meteorological stations to acquire wind speed and precipitation data, and by using an image recognition module to collect images of the track surface environment. Meteorological parameters are sent to edge computing nodes via a hybrid transmission network to obtain the raw monitoring dataset.

[0042] In some embodiments, the data processing module is specifically used to identify and remove outliers from continuous parameters such as temperature and water film thickness in the original monitoring dataset using statistical criteria, identify extreme values ​​that deviate from the normal range by calculating data distribution characteristics, retain data samples that conform to the orbital meteorological monitoring scenario, and eliminate abnormal readings caused by sensor noise and environmental interference. For periods of data loss in the original monitoring dataset due to sensor disconnection, communication delay, or adverse environment, spatiotemporal interpolation is used to fill in the missing values. Multi-source heterogeneous data are aligned according to timestamps to synchronize the collection time of meteorological parameters. Spatial location matching is achieved through orbital odometers, and the data of each sensor is mapped to the corresponding orbital segment or station coordinates. Based on the aligned time, space and parameter dimension data, a three-dimensional matrix of time-space-parameter is constructed, where the matrix row dimension is the time series, the column dimension is the orbital spatial location, and the depth dimension is the meteorological parameter type, resulting in an initial feature dataset containing time labels, spatial coordinates and corresponding meteorological parameters. Based on the initial feature dataset, the meteorological parameters in the three-dimensional matrix are standardized, and the parameters with different dimensions are converted into feature values ​​with a unified dimension. This ensures that the subsequent hierarchical fusion algorithm processes the multi-source data in a consistent manner, and outputs a standardized feature dataset that can be directly used for fusion modeling.

[0043] In some embodiments, the data processing module is specifically used to perform time-dimensional interpolation based on the effective observations before and after the missing data points, and to perform spatial supplementation by combining the spatial correlation data of adjacent monitoring stations, thereby generating complete time-series data covering the entire monitoring period.

[0044] In some embodiments, the feature fusion module is specifically used to perform fusion processing on the same type of sensor data in the standardized feature dataset using a weighted average method, dynamically assigning weights according to the historical measurement accuracy of each sensor, eliminating systematic biases caused by individual hardware differences, and outputting a set of basic parameters after data-level fusion. Based on the basic parameter set, a dual-model parallel processing architecture is constructed: a GA-BP neural network is used, with temperature, salinity and original water film thickness as input features. The initial weights and thresholds of the neural network are optimized through a genetic algorithm, and the water film thickness features after compensating for environmental interference are output; a PSO-BP neural network is used, with attenuation ratio, phase difference and temperature as input features. The network structure parameters are optimized through a particle swarm optimization algorithm, and the classification results of four types of road cover states: dry, waterlogged, icy and snowy are output. By combining water film thickness characteristics and cover classification results with macro-meteorological data and equipment operating status parameters, multi-source information fusion is performed through Bayesian estimation. The output bias of the mid-level model is dynamically corrected to generate a final fused feature set containing compensated water film thickness, cover status labels, and temperature gradient.

[0045] In some embodiments, the risk warning module is specifically used to take the temperature, water film thickness, and wind speed time series data in the final fused feature set as input, and use the SO-CNN-GRU neural network model to predict disasters: the convolution kernel size and the number of hidden layer nodes of the gated recurrent unit of the convolutional neural network are optimized by the snake optimization algorithm, and the weight of key meteorological features is enhanced by the SE attention mechanism, and the road icing time in the future preset period is output. Based on the final fused feature set and the time of road icing, a multi-parameter risk assessment system is constructed: water accumulation risk level, icing risk level, and cover risk level; Among them, the water accumulation risk level is based on the compensated water film thickness as the core indicator, and is corrected by the ambient temperature, and is divided into four levels of warning; the freezing risk level is based on the predicted freezing time, real-time temperature and water film thickness as joint indicators, and is divided into four levels of warning; the covering risk level is based on the covering status label and combined with the temperature gradient parameter, and outputs the risk level corresponding to dryness, water accumulation, freezing and snow accumulation.

[0046] In some embodiments, the risk warning module is also used to dynamically allocate the weights of each risk level indicator using the risk entropy weight method to calculate the comprehensive risk value: the objective weights of water accumulation, icing, and covering risks are determined by the information entropy theory, and the subjective weights are adjusted by experts to integrate the multi-dimensional risk levels into a comprehensive warning level, thereby achieving an improvement from single-parameter warning to multi-parameter collaborative warning. Based on historical accident data and real-time monitoring data, the thresholds for each risk level are dynamically updated using the sliding window method: when extreme weather conditions are detected, an emergency threshold correction mechanism is triggered to shorten the early warning response time and output the final early warning result, which includes the comprehensive early warning level, risk location, and duration.

[0047] This application also discloses a control device.

[0048] Specifically, the control device includes a memory and a processor. The memory stores a computer program that can be loaded and executed by the processor to implement the aforementioned rail transit meteorological monitoring and early warning method based on multi-source data fusion.

[0049] This application also discloses a computer-readable storage medium.

[0050] Specifically, the computer-readable storage medium stores a computer program that can be loaded and executed by a processor, such as the above-mentioned rail transit meteorological monitoring and early warning method based on multi-source data fusion. The computer-readable storage medium includes, for example, various media that can store program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0051] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A rail transit meteorological monitoring and early warning method based on multi-source data fusion, characterized in that, The application relates to a meteorological disaster prediction and early warning method for rail transit, and a device thereof. The method comprises the following steps: deploying monitoring equipment along a rail transit line, stations and key influence areas, collecting meteorological parameters in real time, sending the meteorological parameters to an edge computing node through an Internet of Things wireless transmission network to obtain an original monitoring data set, performing outlier rejection and missing value filling on the original monitoring data set, mapping multi-source data to a three-dimensional matrix of time, space and parameters through a time stamp and a rail mileage mark to obtain a standardized feature data set, performing feature enhancement on the standardized feature data set through a hierarchical fusion algorithm to obtain a final fusion feature set, constructing a risk prediction and grading early warning model based on the final fusion feature set, performing disaster prediction and risk level division, and displaying the monitoring and prediction distribution range of each main meteorological element along the rail transit line through a visualization platform and pushing targeted meteorological disaster early warning information. The method comprises the following steps: deploying monitoring equipment along a rail transit line, stations and key influence areas, collecting meteorological parameters in real time, sending the meteorological parameters to an edge computing node through an Internet of Things wireless transmission network to obtain an original monitoring data set, performing outlier rejection and missing value filling on the original monitoring data set, mapping multi-source data to a three-dimensional matrix of time, space and parameters through a time stamp and a rail mileage mark to obtain a standardized feature data set, performing feature enhancement on the standardized feature data set through a hierarchical fusion algorithm to obtain a final fusion feature set, constructing a risk prediction and grading early warning model based on the final fusion feature set, performing disaster prediction and risk level division, and displaying the monitoring and prediction distribution range of each main meteorological element along the rail transit line through a visualization platform and pushing targeted meteorological disaster early warning information. The method comprises the following steps: deploying monitoring equipment along a rail transit line, stations and key influence areas, collecting meteorological parameters in real time, sending the meteorological parameters to an edge computing node through an Internet of Things wireless transmission network to obtain an original monitoring data set, performing outlier rejection and missing value filling on the original monitoring data set, mapping multi-source data to a three-dimensional matrix of time, space and parameters through a time stamp and a rail mileage mark to obtain a standardized feature data set, performing feature enhancement on the standardized feature data set through a hierarchical fusion algorithm to obtain a final fusion feature set, constructing a risk prediction and grading early warning model based on the final fusion feature set, performing disaster prediction and risk level division, and displaying the monitoring and prediction distribution range of each main meteorological element along the rail transit line through a visualization platform and pushing targeted meteorological disaster early warning information. The method comprises the following steps: deploying monitoring equipment along a rail transit line, stations and key influence areas, collecting meteorological parameters in real time, sending the meteorological parameters to an edge computing node through an Internet of Things wireless transmission network to obtain an original monitoring data set, performing outlier rejection and missing value filling on the original monitoring data set, mapping multi-source data to a three-dimensional matrix of time, space and parameters through a time stamp and a rail mileage mark to obtain a standardized feature data set, performing feature enhancement on the standardized feature data set through a hierarchical fusion algorithm to obtain a final fusion feature set, constructing a risk prediction and grading early warning model based on the final fusion feature set, performing disaster prediction and risk level division, and displaying the monitoring and prediction distribution range of each main meteorological element along the rail transit line through a visualization platform and pushing targeted meteorological disaster early warning information. 2.The rail transit meteorological monitoring and early warning method based on multi-source data fusion of claim 1, wherein, The method comprises the following steps: deploying monitoring equipment along a rail transit line, stations and key influence areas, collecting meteorological parameters in real time, sending the meteorological parameters to an edge computing node through an Internet of Things wireless transmission network to obtain an original monitoring data set, performing outlier rejection and missing value filling on the original monitoring data set, mapping multi-source data to a three-dimensional matrix of time, space and parameters through a time stamp and a rail mileage mark to obtain a standardized feature data set, performing feature enhancement on the standardized feature data set through a hierarchical fusion algorithm to obtain a final fusion feature set, constructing a risk prediction and grading early warning model based on the final fusion feature set, performing disaster prediction and risk level division, and displaying the monitoring and prediction distribution range of each main meteorological element along the rail transit line through a visualization platform and pushing targeted meteorological disaster early warning information. ​ ​ ​ 3.The rail transit meteorological monitoring and early warning method based on multi-source data fusion of claim 2, characterized in that, ​ ​ ​ The multi-source heterogeneous data is aligned in time dimension according to timestamps, the collection time of the meteorological parameters is synchronized, the spatial position matching is realized through the track mileage mark, the sensor data is mapped to the corresponding track section or site coordinate, and based on the aligned time, space and parameter dimension data, a three-dimensional matrix of time-space-parameter is constructed, wherein the row dimension of the matrix is the time sequence, the column dimension is the track space position, and the depth dimension is the meteorological parameter type, so as to obtain an initial feature data set containing time labels, space coordinates and corresponding meteorological parameters; Based on the initial feature data set, the meteorological parameters in the three-dimensional matrix are standardized, different dimension parameters are converted into feature values of a unified dimension, the consistency of the subsequent hierarchical fusion algorithm for multi-source data is ensured, and the standardized feature data set which can be directly used for fusion modeling is output. 4.The rail transit meteorological monitoring and early warning method based on multi-source data fusion of claim 3, characterized in that, For the data loss period caused by sensor disconnection, communication delay or harsh environment in the original monitoring data set, a time-space interpolation method is used for missing value filling, including: Based on the valid observation values before and after the missing data points, time dimension interpolation is performed, and spatial correlation data of adjacent monitoring sites are combined for spatial supplement to generate complete time series data covering the entire monitoring period.

5. The rail transit meteorological monitoring and early warning method based on multi-source data fusion according to claim 4, characterized in that, Based on the standardized feature data set, feature enhancement is performed through a hierarchical fusion algorithm to obtain a final fusion feature set, including: The same type of sensor data in the standardized feature data set is fused by using a weighted average method, the historical measurement accuracy of each sensor is dynamically allocated a weight, the systematic deviation caused by individual differences of hardware is eliminated, and a basic parameter set after data level fusion is output; Based on the basic parameter set, a double-model parallel processing architecture is constructed: a GA-BP neural network is used, temperature, salinity and original water film thickness are used as input features, the initial weight and threshold of the neural network are optimized by a genetic algorithm, and a water film thickness feature after compensation of environmental interference is output; a PSO-BP neural network is used, attenuation ratio, phase difference and temperature are used as input features, network structure parameters are optimized by a particle swarm algorithm, and classification results of four types of road coverings, i.e., dry, water accumulation, icing and snow, are output; The water film thickness feature and the covering classification result are combined with macro-weather data and equipment operation state parameters, multi-source information fusion is performed through Bayesian estimation, output deviation of the middle layer model is dynamically corrected, and the final fusion feature set containing compensated water film thickness, covering state label and temperature gradient is generated. 6.The rail transit meteorological monitoring and early warning method based on multi-source data fusion of claim 1, wherein, Based on the final fusion feature set, a risk prediction and grading early warning model is constructed to perform disaster prediction and risk level division, including: Taking the temperature, water film thickness and wind speed time series data in the final fusion feature set as input, a SO-CNN-GRU neural network model is used for disaster prediction: the convolution kernel size of the convolutional neural network and the number of hidden layer nodes of the gated recurrent unit are optimized by a snake optimization algorithm, the weight of the key meteorological features is enhanced by combining an SE attention mechanism, and the road icing occurrence time in a future preset period is output. Based on the final fusion feature set and the road surface icing occurrence time, a multi-parameter risk assessment system is constructed: water accumulation risk level, icing risk level and cover risk level; Among them, the water accumulation risk level is divided into four levels of warning by taking the compensated water film thickness as the core index and combining with the environmental temperature; the icing risk level is divided into four levels of warning by taking the predicted icing time, real-time temperature and water film thickness as joint indexes; the cover risk level is based on the cover state label and combined with the temperature gradient parameter to output the risk level corresponding to dry, water accumulation, icing and snow.

7. The rail transit meteorological monitoring and early warning method based on multi-source data fusion according to claim 6, characterized in that, After the multi-parameter risk assessment system is constructed based on the final fusion feature set and the road surface icing occurrence time, the following steps are further included: The risk entropy weight method is used to dynamically allocate the weight of each risk level index, and the comprehensive risk value is calculated: the objective weights of water accumulation, icing and cover risk are determined through information entropy theory, and the subjective weights are adjusted combined with the expert to fuse the multi-dimensional risk level into the comprehensive warning level, realizing the promotion from single parameter warning to multi-parameter collaborative warning; Based on historical accident data and real-time monitoring data, the threshold values of each risk level are dynamically updated through the sliding window method: when extreme weather conditions are monitored, the threshold emergency correction mechanism is triggered to shorten the warning response time, and the final warning result including the comprehensive warning level, risk position and duration is output.

8. A rail transit meteorological monitoring and early warning device based on multi-source data fusion, characterized in that, The device comprises: A multi-source data module is used to deploy monitoring equipment along the rail transit line, stations and key influence areas, to collect meteorological parameters in real time, and to send the meteorological parameters to edge computing nodes through a hybrid transmission network to obtain an original monitoring data set; A data processing module is used to perform outlier rejection and missing value filling on the original monitoring data set, and to map multi-source data to a three-dimensional matrix of time-space-parameters through time stamp and rail mileage mark to obtain a standardized feature data set; A feature fusion module is used to perform feature enhancement based on the standardized feature data set through a hierarchical fusion algorithm to obtain a final fusion feature set; A risk prediction module is used to perform disaster prediction and risk level division, and to display the monitoring and forecasting distribution range of each main meteorological element along the rail transit line through a visualization platform and to push targeted meteorological disaster warning information.

9. A control device, characterized by The device comprises: A memory and a processor, wherein the memory stores a computer program capable of being loaded and executed by the processor to perform the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program capable of being loaded and executed by the processor to perform the method of any one of claims 1 to 7 is stored.

Images