Dangerous source diffusion prediction and early warning method and device

Abstract

The invention discloses a danger source diffusion prediction and early warning method and device, and relates to the technical field of pollution diffusion early warning, and the method employs advanced artificial intelligence algorithms, such as a neural network, logistic regression and a decision tree, and can predict the danger source diffusion condition and the danger degree at any position and at any time period in the future. Fixed or mobile equipment (such as an unmanned aerial vehicle, a robot and the like) is used for collecting data in real time, the collected data are more accurate, and through the data processing method, the model adapts to any collection time and point location. Deployment of data acquisition point locations is optimized through prediction results, model training is repeatedly performed on acquired real-time data to update a current model, and prediction accuracy and real-time performance are further improved.

Description

Technical Field

[0001] The present invention relates to the field of pollution diffusion early warning technology, and in particular to a method and device for predicting and warning the diffusion of hazardous sources based on artificial intelligence. Background Art

[0002] In the fields of industrial areas, urban management, and emergency response, hazardous sources such as chemical explosions, bacterial / viral outbreaks, and other emergencies occur. Over time, they spread to the surrounding areas through carriers such as air and organisms, causing great harm to the surrounding ecology and municipal equipment. These hazardous sources have the characteristics of rapid spread and wide impact range, posing a serious threat to public safety and environmental health. Traditional methods rely on fixed monitoring equipment and manual monitoring. These methods often have limitations in data collection, real-time and accuracy. The traditional gas diffusion model is often used to predict the spread range of hazardous sources, which fails to effectively achieve rapid prediction and timely warning of the spread of hazardous sources.

[0003] Patent document No.: 202111648544.5 records "A method and device for analyzing hazardous sources based on neural networks". This patent discloses a method for analyzing hazardous sources based on neural networks, which can achieve the effect of quickly locating hazardous sources when abnormal conditions occur in chemical parks. It includes the following steps: receiving gas data monitored by each hazardous gas probe at the front end and probe position information related to the hazardous gas probe; inputting the gas data into a convolutional neural network model for risk identification to determine the state of hazardous gases monitored at each location; if the state of the hazardous gas is abnormal, inputting the corresponding probe position information and gas concentration information into the position prediction model to perform a spatial position analysis of the hazardous source to determine the spatial position of the hazardous source; and displaying the spatial position of the hazardous source.

[0004] Disadvantages: 1. The patented technology can effectively locate the source of danger in space, but it does not cover the prediction and calculation of future trends in the status of hazardous gases; 2. The patented technology relies on probes to monitor hazardous gas data for analysis. If the source of danger spreads through non-gas carriers (such as biological carriers), the detection method is no longer applicable; 3. The patent fails to predict the degree of danger in various geographical locations within a certain period of time in the future and provide early warnings accordingly. 4. The position of the environmental monitoring probe in this patent is fixed. However, for monitoring points of emergencies, it is often necessary to use mobile devices such as drones and robots for dynamic monitoring. When the monitoring points change, such as the increase, decrease or movement of the monitoring locations, the neural network model cannot adapt to such changes due to its stringent requirements on the input data, which may cause the model to fail or reduce its accuracy.

[0005] The patent document with patent number: 202211235634.6 records "a major hazard source data center centralized safety monitoring and early warning system and method thereof". The patent discloses a major hazard source data center centralized safety monitoring and early warning system and method thereof, including a main control end module, the main control end module connection end is provided with a sub-regional module, the sub-regional module connection end is provided with a display module, the sub-regional module connection end is provided with a wireless interconnection module, the wireless interconnection module connection end is provided with an execution unit module, the execution unit module includes a communication device module, a controller module, a camera module and a sensor module, the main control end module connection end is provided with a major hazard monitoring and early warning module, the major hazard The risk monitoring and early warning module includes a major hazard source classification and identification module, a major hazard source real-time monitoring module, a hazard result simulation module and a hazard early warning module. The connection ends of the hazard early warning module are respectively provided with a hazard alarm module and a real-time risk early warning model module. The major hazard source classification and identification module includes a first-level hazard module, a second-level hazard module, a third-level hazard module and a fourth-level hazard module. The connection end of the major hazard source classification and identification module is provided with a sensor data identification module. The major hazard source classification and identification module is respectively electrically connected to the first-level hazard module, the second-level hazard module, the third-level hazard module and the fourth-level hazard module. The major hazard source classification and identification module is electrically connected to the sensor data identification module.

[0006] Disadvantages: 1. Although the patent proposes a complete set of safety monitoring and early warning systems and methods, it does not provide detailed descriptions of the specific implementation methods of some key technologies, such as the determination of the spread of dangerous sources, the prediction of duration, the setting of early warning mechanisms, and the assessment of the degree of danger. The clear treatment of these key technologies is crucial to ensure the timeliness and accuracy of dangerous sources; 2. The patent only classifies dangerous sources into different levels, that is, assesses their degree of danger, but does not further predict the degree of danger in various geographical locations within a certain period of time in the future, and provide corresponding early warnings accordingly. This means that it is limited in providing specific and targeted safety warnings for specific times and places.

[0007] The patent document with patent number: 202310579766.9 records the "hazardous chemical leakage and diffusion simulation method based on the influence of chemical park buildings". The purpose of the present invention is to provide a hazardous chemical leakage and diffusion simulation method based on the influence of chemical park buildings, including the following steps: collecting the leakage source's own parameters and geographical environment parameters, the leakage source's own parameters include gas leakage type, leakage hole shape; geographical environment parameters include leakage point longitude, latitude and altitude, wind speed, wind direction, and air pressure; calculating the hazardous gas leakage mass flow rate; constructing a hazardous gas leakage and diffusion model based on chemical park buildings; using a simulation system to numerically solve hazardous gas leakage accidents and evolve the concentration distribution law, and realize a visual display of the gas diffusion range.

[0008] Disadvantages: 1. The patent combines the leakage source's own parameters and geographical environment parameters to calculate the distribution law of hazardous gases, but it uses the traditional gas diffusion model. This model not only requires a high degree of professionalism, but also has a complex numerical solution process. Especially when multiple hazardous sources exist at the same time, modeling and solving will face greater difficulties; 2. The patented technology relies on the diffusion model to solve the distribution law of gas concentration. However, if the hazardous source diffuses and spreads through non-gaseous carriers, such as biological carriers, this method will no longer be applicable because it cannot effectively capture and analyze the diffusion dynamics of non-gaseous media; 3. The patent does not include the function of predicting the degree of danger of various geographical locations in a certain period of time in the future and proposing early warnings accordingly. This means that it is limited in providing specific and targeted safety warnings for specific times and places.

[0009] It can be seen that the main disadvantages of the prior art are as follows:

[0010] 1. The prediction of hazardous source diffusion is solved by establishing a mathematical model based on the principle of gas diffusion. This model not only requires a high degree of professionalism, but also has a complex numerical solution process. Especially when multiple hazardous sources exist at the same time, modeling and solving will face greater difficulties.

[0011] 2. Currently, the prediction of hazardous source diffusion relies on diffusion models to solve the distribution law of gas concentration. However, if the hazardous source diffuses through non-gaseous carriers (such as biological carriers), this method will no longer be applicable because it cannot effectively capture and analyze the diffusion dynamics of non-gaseous media.

[0012] 3. When determining the degree of hazard of a dangerous source, only the location of the center of the dangerous source is considered, and it does not include the function of predicting the degree of hazard at various geographical locations in a certain period of time in the future and issuing warnings accordingly. This means that it is limited in providing specific and targeted safety warnings for specific times and locations.

[0013] 4. In the monitoring of the spread of hazardous sources, fixed-position monitoring equipment is usually used. However, for monitoring points of emergencies, it is often necessary to use mobile devices such as drones and robots for irregular dynamic monitoring. When the monitoring points change, such as the increase or decrease or movement of monitoring locations, the neural network model cannot adapt to such changes due to its strict requirements on input data (if it has not been specially processed), which may cause the model to fail or reduce accuracy;

[0014] 5. The model is not updated in a timely manner, and the real-time and accuracy of the predicted data are not high. Summary of the invention

[0015] In view of the above problems, the present invention provides a method and device for predicting and warning the spread of dangerous sources, which are used to overcome the above problems or at least partially solve the above problems.

[0016] The present invention provides the following scheme:

[0017] A method for predicting and warning the spread of dangerous sources, comprising:

[0018] Obtain multi-source heterogeneous data of all existing monitoring points by means of fixed and / or mobile devices at preset time intervals, wherein the multi-source heterogeneous data at least includes the hazard source parameters, environmental parameters and social impact data corresponding to each of the existing monitoring points;

[0019] Preprocessing the multi-source heterogeneous data and extracting features; the features include label features and hazard source diffusion features extracted based on the position information between the two existing monitoring points; the hazard source diffusion features include the spatiotemporal position features between the two existing monitoring points, the hazard source characteristic features corresponding to each of the existing monitoring points, and the meteorological features corresponding to each of the existing monitoring points;

[0020] Determine label data, the label data is the feature to be predicted; use the spatiotemporal location feature, the hazard source characteristic feature and the meteorological feature as input of a hazard source diffusion model, use the label data as output of the hazard source diffusion model to train the hazard source diffusion model to obtain a target hazard source diffusion model;

[0021] Determine the center position and delay time of the point to be predicted, and calculate the spatiotemporal position characteristics to be predicted based on the center position and the delay time;

[0022] According to the spatiotemporal position characteristics to be predicted, a number of valid monitoring points are obtained by screening from all the existing monitoring points, wherein the valid monitoring points are the existing monitoring points within a preset spatial range and time window; the spatial range is a range defined with the central position as the center;

[0023] Input the spatiotemporal location characteristics, the hazard source characteristic characteristics and the meteorological characteristics corresponding to the respective effective monitoring points into the hazard source diffusion model for prediction to obtain a number of initial prediction results;

[0024] A plurality of the initial prediction results are fused to obtain the target prediction result of the point to be predicted.

[0025] Preferably: the hazard source is determined to be based on air transmission, and the concentration of hazardous gases is extracted as the label feature; the hazard source is determined to be a biological epidemic, and indicators reflecting the impact of the epidemic on ecology are extracted as the label feature.

[0026] Preferably: the spatiotemporal position features include the monitoring time interval between the two existing monitoring points, the spherical distance between the two existing monitoring points, the vertical height between the two existing monitoring points, and the relative orientation between the two existing monitoring points;

[0027] The hazard source characteristic features are features effective for diffusion prediction that are extracted based on the hazard source characteristics.

[0028] Preferably: the hazard source diffusion model is obtained by constructing a machine learning algorithm or a neural network.

[0029] Preferably: the spatial range is a range specified by a user with the central position as the center;

[0030] Alternatively, the spatial range is centered at the central position, and the range within the prediction area is dynamically determined based on the target monitoring value.

[0031] Preferably: the fusion method includes an averaging method or a weighted averaging method; the weighted averaging method includes assigning different weights to different prediction results according to the distance between the existing monitoring point and the point to be predicted and the wind direction factor, and calculating the weighted average value.

[0032] Preferably: based on the target prediction results, in the predicted key diffusion area or the key area of ​​the expected change trend, determine to adjust and increase the deployment density of the collection equipment to optimize the monitoring points.

[0033] Preferably: based on the actual meaning of the value of the tag feature, combined with the actual situation or international standards, the thresholds of different hazard levels are determined to divide the hazard levels;

[0034] Based on the target prediction result, a convex hull algorithm or a kernel density estimation algorithm is used to calculate and obtain boundary lines of different danger levels;

[0035] In chronological order, the danger range level maps after different delay times are combined together to construct a danger source diffusion situation map.

[0036] Preferably: determine the danger level of the warning according to the actual situation, and trigger the warning information in combination with social information such as personnel, ecology, facilities, etc. within the range;

[0037] After the warning is triggered, warning information will be released to the public and relevant departments through multiple channels. At the same time, personalized risk avoidance advice and emergency response guidelines will be provided in combination with road traffic and personnel distribution information.

[0038] A device for predicting and warning the spread of dangerous sources, used to execute the above-mentioned method for predicting and warning the spread of dangerous sources, comprising:

[0039] A training data acquisition unit is used to collect multi-source heterogeneous data of all existing monitoring points at preset time intervals through fixed and / or mobile devices, wherein the multi-source heterogeneous data at least includes the hazard source parameters, environmental parameters and social impact data collected corresponding to each of the existing monitoring points;

[0040] A feature extraction unit, used for preprocessing the multi-source heterogeneous data and extracting features; the features include label features and hazard source diffusion features extracted based on the position information between the two existing monitoring points; the hazard source diffusion features include the spatiotemporal position features between the two existing monitoring points, the hazard source characteristic features corresponding to each of the existing monitoring points, and the meteorological features corresponding to each of the existing monitoring points;

[0041] A model training unit is used to determine label data, wherein the label data is a feature to be predicted; the spatiotemporal location feature, the hazard source characteristic feature and the meteorological feature are used as inputs of a hazard source diffusion model, and the label data is used as outputs of the hazard source diffusion model to train the hazard source diffusion model to obtain a target hazard source diffusion model; the hazard source diffusion model is obtained by using a machine learning algorithm or a neural network construction;

[0042] The data acquisition unit of the point to be predicted is used to determine the center position and delay time of the point to be predicted, and calculate the spatiotemporal position characteristics to be predicted according to the center position and the delay time;

[0043] An effective monitoring point determination unit is used to select and obtain a number of effective monitoring points from all the existing monitoring points according to the spatiotemporal position characteristics to be predicted, wherein the effective monitoring points are the existing monitoring points within a preset spatial range and time window; the spatial range is a range defined with the central position as the center;

[0044] An initial prediction result output unit, used to input the spatiotemporal location characteristics, the hazard source characteristic characteristics and the meteorological characteristics corresponding to the respective effective monitoring points into the hazard source diffusion model for prediction to obtain a number of initial prediction results;

[0045] The target prediction result calculation unit is used to fuse several of the initial prediction results to obtain the target prediction result of the point to be predicted.

[0046] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0047] The embodiment of the present application provides a method and device for predicting and warning the spread of dangerous sources, which adopts advanced artificial intelligence algorithms such as neural networks, logistic regression and decision trees, and can predict the spread and degree of danger of dangerous sources at any location and any time period in the future. Using fixed or mobile devices (such as drones, robots, etc.) to collect data in real time, the collected data is more accurate. After the data processing method of the present invention, the model adapts to any collection time and point. The deployment of data collection points is optimized by predicting the results, and the current model is repeatedly trained and updated for the collected real-time data, which further improves the accuracy and real-time performance of the prediction. Combining multi-source data such as dangerous source parameters, environmental parameters and social impact, a comprehensive data processing and analysis system is constructed to achieve a comprehensive and multi-dimensional analysis of the spread of dangerous sources. Through real-time analysis and prediction by the data center, the spread trend of dangerous sources, the danger level and the timely release of early warning information can be quickly discovered, and at the same time, combined with social data analysis, risk avoidance suggestions and emergency response guidelines can be quickly specified. It can be widely applicable to various dangerous sources such as chemical accidents, nuclear leaks, biological epidemics and different diffusion carriers (air, water, food chain, etc.), and meet specific needs in different scenarios through modular design and flexible configuration.

[0048] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0050] Figure 1 It is a flow chart of a method for predicting and warning the spread of dangerous sources provided by an embodiment of the present invention;

[0051] Figure 2 It is an overall flow chart provided by an embodiment of the present invention;

[0052] Figure 3 is a schematic diagram of optimized positions of monitoring points provided by an embodiment of the present invention;

[0053] Figure 4 is a schematic diagram of optimized positions of monitoring points provided by an embodiment of the present invention;

[0054] Figure 5 is a schematic diagram of the danger range level provided by an embodiment of the present invention;

[0055] Figure 6It is a schematic diagram of a hazard source diffusion prediction and early warning device provided by an embodiment of the present invention;

[0056] Figure 7 It is a schematic diagram of a hazard source diffusion prediction and early warning device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all of the embodiments. Based on the embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.

[0058] See also Figure 1 , is a method for predicting and warning the spread of dangerous sources provided by an embodiment of the present invention, such as Figure 1 As shown, the method may include:

[0059] S101: acquiring multi-source heterogeneous data of all existing monitoring points at preset time intervals through fixed and / or mobile equipment, wherein the multi-source heterogeneous data at least includes the hazard source parameters, environmental parameters and social impact data corresponding to each of the existing monitoring points;

[0060] S102: Preprocessing the multi-source heterogeneous data and extracting features; the features include label features and hazard source diffusion features extracted based on the position information between the two existing monitoring points; the hazard source diffusion features include the spatiotemporal position features between the two existing monitoring points, the hazard source characteristic features corresponding to each of the existing monitoring points, and the meteorological features corresponding to each of the existing monitoring points;

[0061] In specific implementation, the embodiments of the present application may provide that the spatiotemporal position features include the monitoring time interval between the two existing monitoring points, the spherical distance between the two existing monitoring points, the vertical height between the two existing monitoring points, and the relative orientation between the two existing monitoring points;

[0062] The hazard source characteristic features are features effective for diffusion prediction that are extracted based on the hazard source characteristics.

[0063] S103: Determine label data, where the label data is the feature that needs to be predicted; use the spatiotemporal location features, the hazard source characteristic features and the meteorological features as inputs to a hazard source diffusion model, and use the label data as outputs of the hazard source diffusion model to train the hazard source diffusion model to obtain a target hazard source diffusion model; the hazard source diffusion model is constructed based on a machine learning algorithm or a neural network.

[0064] The method provided in the embodiment of the present application is applicable to both the prediction of the spread of pollution sources based on air transmission and the prediction of the spread of biological epidemics. In specific implementation, the embodiment of the present application can provide the determination of the danger source based on air transmission, extracting the concentration of hazardous gas as the label feature; determining that the danger source is a biological epidemic, extracting indicators reflecting the ecological impact caused by the epidemic as the label feature.

[0065] S104: Determine the center position and delay time of the point to be predicted, and calculate the spatiotemporal position features to be predicted according to the center position and the delay time;

[0066] S105: Filtering and obtaining a number of valid monitoring points from all the existing monitoring points according to the spatiotemporal position characteristics to be predicted, wherein the valid monitoring points are the existing monitoring points within a preset spatial range and time window; the spatial range is a range defined with the central position as the center; the spatial range is a range specified by a user with the central position as the center;

[0067] Alternatively, the spatial range is centered at the central position, and the range within the prediction area is dynamically determined based on the target monitoring value.

[0068] S106: inputting the spatiotemporal location characteristics, the hazard source characteristics and the meteorological characteristics corresponding to the effective monitoring points into the hazard source diffusion model for prediction to obtain a number of initial prediction results;

[0069] S107: Fusing several of the initial prediction results to obtain the target prediction result of the point to be predicted. The fusion method includes an average method or a weighted average method; the weighted average method includes assigning different weights to different prediction results according to the distance between the existing monitoring point and the point to be predicted and the wind direction factor, and calculating the weighted average.

[0070] The sources of danger in the embodiments of the present application mainly refer to substances and states that, after emergencies such as chemical explosions in industrial parks, water pollution, and biological epidemic outbreaks, continue to spread to the surrounding areas over time through carriers such as air and organisms, causing great harm to the surrounding ecology, municipal equipment, etc.

[0071] The hazardous source spread prediction and early warning method provided in the embodiment of the present application utilizes fixed monitoring stations and mobile sensor equipment (such as drones, ground robots, etc.) to achieve comprehensive and dynamic collection of direct parameters of hazardous sources (such as gas concentration, radiation intensity), environmental influencing factors (geographic location, wind direction and speed, temperature and humidity, etc.), and social impact data (casualties, traffic conditions, etc.), and upload them to the data center.

[0072] The hazard source diffusion model is trained using data obtained from several existing monitoring points. After determining the location of the predicted point and the delay time required for the prediction, a prediction spatial range and a time window for effective prediction can be determined based on the center position of the predicted point. The existing monitoring points used in the training are searched within the spatial range and the data collection time is within the time window as effective monitoring points. These effective monitoring points are close to the predicted point in space, and the data collection time is close to the start time of the current prediction, which can better characterize the pollution source diffusion at the predicted point in the future.

[0073] Then, the relevant data corresponding to these effective monitoring points and the relevant data of the points to be predicted are used as the input of the trained hazard source diffusion model. Finally, the hazard source diffusion model outputs multiple initial prediction results. Finally, the multiple initial prediction results are fused to obtain the final target prediction result.

[0074] Based on data analysis, we extract key features and labels that are closely related to the spread of hazardous sources, and use machine learning (such as deep learning, ensemble learning) and artificial intelligence algorithms to build a prediction model that can accurately simulate the spread of hazardous sources. The model input features are decoupled from time and location, and can be applied to any monitoring location and any collection time point;

[0075] By using the trained model, the real-time incoming data can be quickly processed and analyzed to achieve instant prediction of the spread of hazardous sources and risk level assessment of various regions in the future, providing a scientific basis for decision-making.

[0076] In order to further optimize the deployment of collection points and dynamically update the prediction model, the embodiment of the present application can also provide the key areas of predicted diffusion areas or expected change trends based on the target prediction results, determine to adjust and increase the deployment density of collection equipment to optimize the monitoring points. Based on the prediction results of the diffusion model, the deployment of collection points is dynamically optimized to improve the timeliness and accuracy of the collected data, and the prediction model is dynamically updated to improve the reliability of the prediction results.

[0077] In order to further achieve timely warning and emergency response, the embodiment of the present application can also provide the actual meaning of the value of the tag feature, and determine the thresholds of different hazard levels based on the actual situation or international standards to divide the hazard level;

[0078] Based on the target prediction result, a convex hull algorithm or a kernel density estimation algorithm is used to calculate and obtain boundary lines of different danger levels;

[0079] In chronological order, the danger range level maps after different delay times are combined together to construct a danger source diffusion situation map.

[0080] Determine the danger level of the warning based on the actual situation, and trigger the warning information based on social information such as personnel, ecology, and facilities within the scope;

[0081] After the warning is triggered, warning information will be released to the public and relevant departments through multiple channels. At the same time, personalized risk avoidance advice and emergency response guidelines will be provided in combination with road traffic and personnel distribution information.

[0082] By constructing a hazard source change situation map, the early warning mechanism is automatically triggered according to the prediction results, and early warning information is released to the public and relevant departments through multiple channels (such as SMS, APP push, social media, broadcasting, etc.), and personalized risk avoidance suggestions and emergency response guidelines are provided to minimize potential hazards.

[0083] The following is a detailed introduction to the hazard source spread prediction and early warning method provided by this application.

[0084] The present invention provides a method and device for predicting and warning the spread of hazardous sources based on artificial intelligence, aiming to solve the technical problems of predicting and warning the spread of hazardous sources through multi-source heterogeneous data collection, intelligent fusion, real-time prediction and intelligent warning.

[0085] like Figure 2 As shown, the basic steps are as follows:

[0086] Step 1: Multi-source heterogeneous data collection.

[0087] Fixed and mobile equipment are used to collect data on hazard source parameters, environmental parameters and social impacts at irregular intervals, including the following data:

[0088] The collection of parameters of the hazard source itself includes but is not limited to the following characteristics: concentration of hazardous gases (hazard sources are transmitted via air); intuitive variables reflecting the harm caused by the hazard source, such as the number of infected people, the number of dead poultry, etc. (hazard sources are transmitted via non-air carriers); monitoring point time; monitoring point location.

[0089] Meteorological data feature extraction, including but not limited to the following features: temperature, air pressure, humidity, wind direction, wind speed, etc.

[0090] The collection of social information data is mainly used for the dispatch and optimization of emergency response resources after early warning, including but not limited to the following features: distribution data of people / organisms in the area, terrain data such as road traffic, etc.

[0091] The frequency and location of monitoring should be determined according to the diffusion characteristics of the hazard source. For example, for a hazard source that spreads rapidly, the monitoring frequency can be set to every few minutes or hours, and the monitoring points can be set within a range of tens to hundreds of meters; while for a hazard source such as a biological epidemic that spreads slowly, the monitoring frequency can be carried out daily, and the distance between monitoring points can be appropriately increased.

[0092] Step 2: Label feature extraction.

[0093] The label features are extracted based on the actual situation of the hazard source. If it is based on air transmission, the concentration of the hazardous gas is extracted. (Concentration at monitoring location i at time t);

[0094] If it is a biological epidemic, then extract indicators reflecting the ecological impact of the epidemic, such as the number of casualties / number of biological infections, etc., and record them as represents the number of casualties at monitoring location i at time t).

[0095] Step 3: Extraction of hazard source diffusion characteristics.

[0096] 1. Extraction of associated features of monitoring locations

[0097] Since the monitoring location and monitoring time are not fixed, the collected data is a sparse matrix as shown in Table 1 below:

[0098] Table 1 Monitoring time and location

[0099]

[0100] Among them, record 1 indicates data collection, and no record indicates no data collection. In actual situations, since mobile devices dynamically collect data at irregular intervals, the two-dimensional array composed of monitoring time and monitoring location is a large and sparse matrix. The following data processing is now performed:

[0101] Location data is expressed in the form of latitude, longitude and altitude, such as in Indicates the latitude of the monitoring location at time t, represents the longitude of monitoring location i at time t, represents the altitude of monitoring location i at time t.

[0102] The monitoring time interval between two monitoring locations (accuracy can be selected based on actual conditions, such as minutes, hours, or days) is calculated as follows:

[0103] T i,j =T i -T j , where T i,j ≥0

[0104] Calculate the relative distance and orientation of two monitoring locations. The calculation process is as follows:

[0105] A. First, convert the longitude and latitude into Geohash codes. The longitude and latitude of the center of the Geohash are represented as the monitoring position of the area. The accuracy of the monitoring distance (i.e., the length of the Geohash code) is determined according to the characteristics of the hazard source. The purpose of this is to reduce the amount of calculation when predicting the spread range of the hazard source.

[0106] B. Calculate the spherical distance between the longitude and latitude of the centers of two Geohash blocks. The formula is as follows:

[0107]

[0108] Where: d i,j is the spherical distance between the centers of two Geohash blocks. R is the radius of the Earth. and are the latitudes of the first and second Geohash block centers, respectively, expressed in radians. λ1 and λ2 are the longitudes of the first and second Geohash block centers, respectively, expressed in radians. is the difference between two latitudes.

[0109] Δλ=λ1-λ2 is the difference between the two longitudes.

[0110] c. Calculate the relative height between the center longitude and latitude of two Geohash blocks. The formula is as follows:

[0111] v i,j =asl i -asl j

[0112] Where: v i,j is the relative height between the centers of two Geohash blocks. i is the altitude of monitoring location i. j is the altitude of monitoring location j.

[0113] d. Calculate the relative position between the longitude and latitude of the centers of the two Geohash blocks, and record a i,j , represents the direction of monitoring position i at monitoring position j. If monitoring position i is due north of monitoring position j, then a i,j =0; if monitoring location i is due south of monitoring location j, then a i,j =180. i,j The value range of is [0,360).

[0114] Based on the monitoring point information, the extracted features are summarized in Table 2:

[0115] Table 2 Correlation characteristics between two monitoring locations

[0116] variable meaning <![CDATA[T i,j ]]> Monitoring time interval between two monitoring locations <![CDATA[d i,j ]]> Spherical distance between two monitoring positions <![CDATA[v i,j ]]> Vertical height between two monitoring locations <![CDATA[a i,j ]]> The relative position between the two monitoring locations

[0117] To calculate the characteristic values ​​required for the model input, first determine two parameters based on the characteristics of the hazard source, namely the β parameter that determines the size of the spatial range and the γ parameter that determines the time range. The location association characteristic calculation method is as follows:

[0118] First, calculate the number of other monitoring locations within the spherical range of the β distance for each monitoring location. Assuming that monitoring location i has M monitoring locations within the spherical range of radius β, of which the monitoring records (monitoring time) of R monitoring locations are earlier than the monitoring time of monitoring location i and within the time interval γ, then monitoring location i will generate R model features. The position association features are shown in Table 3:

[0119] Table 3 Location correlation characteristics between different monitoring locations

[0120] Location Information Time interval Relative distance Relative height Relative position <![CDATA[L i,1 ]]> <![CDATA[T i,1 ]]> <![CDATA[d i,1 ]]> <![CDATA[v i,1 ]]> <![CDATA[a i,1 ]]> …… …… …… …… …… <![CDATA[L i,j ]]> <![CDATA[T i,j ]]> <![CDATA[d i,j ]]> <![CDATA[v i,j ]]> <![CDATA[a i,j ]]> <![CDATA[L i,R ]]> <![CDATA[T i,R ]]> <![CDATA[d i,R ]]> <![CDATA[v i,R ]]> <![CDATA[a i,R ]]>

[0121] The above monitoring location association features are also called spatiotemporal location features.

[0122] Expressed in Q, Q=(T i,j ,d i,j ,v i,j ,a i,j )

[0123] 2. Extraction of hazard source characteristics. Hazard source characteristics refer to the extraction of features that are effective for diffusion prediction based on the characteristics of the hazard source, such as gas concentration (expressed by F1), etc. n Indicates the nth characteristic feature of the hazard source. It indicates the concentration of a hazardous gas at monitoring location j at time t. Since time is expressed as time difference in the characteristic, this characteristic can be expressed as (F 1,j ,…,F n,j ) indicates. The characteristics of the hazard source and its temporal and spatial location characteristics are shown in Table 4.

[0124] Table 4 Hazard source characteristics and temporal and spatial location characteristics

[0125] Location Information Spatiotemporal location feature Q Hazard source characteristics F <![CDATA[L i,j ]]> <![CDATA[(T i,j ,d i,j ,v i,j ,a i,j )]]> <![CDATA[(F 1,j ,…,F n,j )]]>

[0126] 3. Extraction of meteorological data features. Meteorological features are represented by W, which mainly include temperature, air pressure, humidity, wind speed and wind direction, etc., then W = (W 1,j ,…,W k,j ).

[0127] The model features and label data extracted based on location information are shown in Table 5. (The label data is the feature to be predicted, which can be 1 or more, that is, k ≥ 1):

[0128] Table 5 Model features and label data

[0129]

[0130] Step 4: Hazard source diffusion model training. Use machine learning algorithms (such as decision trees, polynomial regression, etc.) or neural networks, and use Q, F, and W as model inputs. o As model output, a multi-objective regression model is trained.

[0131] Step 5: Predict the spread of hazardous sources.

[0132] Input the location and delay time of the point to be predicted, and output the target data of the location after the specified delay time.

[0133] 1. Model prediction input feature calculation. The user enters the center latitude and longitude (latitude, longitude) and altitude (altitude, asl) of the Geohash code corresponding to the position to be predicted, and specifies a delay time Δt. Then, according to the feature extraction method detailed in step 3, we will calculate the spatiotemporal position feature Q based on these input information (center position and delay time). Within a specific spatial range (within the β distance) and time window (within the γ time range), if there are H historical records of monitoring points, then these records will be used to generate H corresponding feature data, each of which contains the feature description of the corresponding monitoring point under the specified spatiotemporal conditions.

[0134] 2. Model prediction: Before making model predictions, you first need to clarify the prediction range. Here are two flexible ways to determine the prediction area:

[0135] (a) Specify the prediction area: Users can directly specify a clear area range, and the model will only predict the spread of hazardous sources within the area. This method is suitable for scenarios where users already have a preliminary understanding of potential hazardous areas or specific concerns. By accurately defining the prediction area, computing resources can be used more efficiently and prediction results for the required area can be quickly obtained.

[0136] (b) Dynamically determine the prediction area based on the target monitoring value: This method can be used when the user is uncertain about the boundaries of the potential danger zone. The model starts from the location to be predicted, gradually expands the prediction range outward, and predicts the spread of hazardous sources for each newly expanded area. During the prediction process, the predicted hazardous source concentration, impact range or other target values ​​will be continuously monitored. Once the predicted value drops below the threshold set by the user, it is considered that the impact of the hazardous source diffusion at that location has been weakened to an acceptable level, and further range diffusion prediction for that location is stopped. This method can dynamically adapt to the characteristics and environmental conditions of different hazardous sources to ensure the accuracy and practicality of the prediction results.

[0137] Secondly, it is necessary to clarify the time granularity in the prediction process: clarify the delay time of the location to be predicted, which refers to the time interval from the current moment to a certain point in the future that we need to predict. According to the specific characteristics and propagation speed of the hazard source, we need to determine an appropriate delay time granularity. This granularity can be different time units such as 1 minute, 30 minutes, 1 hour, 24 hours, etc. Its selection should be based on the rate of hazard source diffusion, the need for prediction accuracy, and the limitation of computing resources. The determination of the delay time granularity will directly affect the timeliness and accuracy of the prediction results.

[0138] After determining the delay time granularity, the target detection value after the time granularity can be predicted. The "target detection value" here may refer to specific indicators such as the concentration of hazardous sources, the scope of impact, and the potential hazard level, which can quantitatively reflect the spread of hazardous sources at a specific time point.

[0139] According to the determined spatial range and time window, effective monitoring points suitable for the predicted point can be selected from all existing monitoring points. The corresponding data of the effective monitoring points are input into the hazard source diffusion model for prediction to obtain several initial prediction results.

[0140] 3. Fusion of multi-point prediction results. Since the initial prediction results of multiple valid monitoring points after the delay time Δt are obtained through the above method, that is, H corresponding feature data will produce H initial prediction results, these results may vary due to factors such as the location of the monitoring point and environmental conditions. In order to obtain the comprehensive prediction result of the position to be predicted after the delay time Δt, these results need to be fused.

[0141] The fusion method can be selected according to specific needs, including but not limited to the following fusion methods:

[0142] Simple average method: directly calculate the average of all prediction results as the final prediction result.

[0143] Weighted average method: Different weights are assigned to different prediction results according to factors such as the distance between the monitoring point and the location to be predicted, and the wind direction, and then the weighted average value is calculated.

[0144] Figure 3 Indicates that when H = 7, L i Seven prediction values will be generated, and through the fusion of the prediction results, a final prediction value is obtained. Fusion formula:

[0145] L i = w1·L1 + … + w7·L7.

[0146] Step 6: Optimize the detection points based on the prediction results.

[0147] Given the dynamic characteristics of the spread of the hazard source, real-time and effective dynamic monitoring data is crucial for improving the timeliness and accuracy of model prediction. By integrating this dynamic data, the trend and changes of the spread of the hazard source can be captured more accurately. As Figure 4 shown, based on this prediction result, the deployment density of the acquisition equipment can be flexibly adjusted and increased in the key spread areas predicted or the key areas of the expected change trend, and the monitoring points can be optimized to ensure the comprehensiveness and accuracy of the monitoring, so as to more effectively respond to potential risks.

[0148] Step 7: Determine the hazard range level.

[0149] Based on the actual meaning of the label feature values and combined with the actual situation or international standards, the threshold values of different hazard degrees are determined to divide the hazard levels, such as the thresholds th1, th2, …, th5, and th1 < th2 < … < th5. They respectively correspond to IV mild hazard, III moderate hazard, II high hazard, I extreme hazard, etc.

[0150] Based on the prediction result, as Figure 5 shown, calculate the boundary lines of different hazard levels, and convex hull algorithm, kernel density estimation algorithm, etc. can be used to determine the boundary lines of different hazard degrees.

[0151] Step 8: Construct a graph of the change situation of the hazard source.

[0152] Arrange the hazard range level graphs after different delay times in chronological order and combine them to construct a graph of the spread situation of the hazard source.

[0153] Step 9: Hazard status warning.

[0154] Determine the warning hazard level according to the actual situation (such as the predicted range in a certain future period is in "III moderate hazard"), and at the same time combine the social situation information such as the personnel, ecology, and facilities within this range to trigger a warning message.

[0155] Step 10: Warning release and response.

[0156] After the warning is triggered, the warning information is released to the public and relevant departments through multiple channels (such as SMS, APP push, social media, broadcast, etc.), and personalized risk avoidance suggestions and emergency response guidelines are provided in combination with road traffic, personnel distribution and other information.

[0157] In short, the danger source diffusion prediction and early warning method provided by the present application adopts advanced artificial intelligence algorithms, such as neural networks, logistic regression and decision trees, etc., which can predict the danger source diffusion and danger level at any location and any time period in the future. Using fixed or mobile equipment (such as drones, robots, etc.) to collect data in real time, the collected data is more accurate. After the data processing method of the present invention, the model adapts to any collection time and point. The deployment of data collection points is optimized by predicting the results, and the current model is repeatedly trained and updated for the collected real-time data, which further improves the accuracy and real-time performance of the prediction. Combining multi-source data such as the danger source's own parameters, environmental parameters and social impact, a comprehensive data processing and analysis system is constructed to achieve a comprehensive and multi-dimensional analysis of the danger source diffusion. Through real-time analysis and prediction by the data center, the danger source diffusion trend, danger level and early warning information can be quickly discovered, and the social situation data analysis can be combined to quickly specify risk avoidance suggestions and emergency response guidelines. It can be widely applicable to various danger sources such as chemical accidents, nuclear leaks, biological epidemics and different diffusion carriers (air, water, food chain, etc.), and meet specific needs in different scenarios through modular design and flexible configuration.

[0158] See also Figure 6 , the embodiment of the present application can also provide a hazard source diffusion prediction and early warning device, such as Figure 6 As shown, the device for executing the above-mentioned method for predicting and warning the spread of dangerous sources may include:

[0159] The training data acquisition unit 601 is used to collect multi-source heterogeneous data of all existing monitoring points at preset time intervals through fixed and / or mobile devices, wherein the multi-source heterogeneous data at least includes the hazard source parameters, environmental parameters and social impact data collected corresponding to each of the existing monitoring points;

[0160] A feature extraction unit 602 is used to preprocess the multi-source heterogeneous data and extract features; the features include label features and hazard source diffusion features extracted based on the position information between the two existing monitoring points; the hazard source diffusion features include the spatiotemporal position features between the two existing monitoring points, the hazard source characteristic features corresponding to each of the existing monitoring points, and the meteorological features corresponding to each of the existing monitoring points;

[0161] The model training unit 603 is used to determine label data, where the label data is the feature to be predicted; the spatiotemporal location feature, the hazard source characteristic feature and the meteorological feature are used as inputs of a hazard source diffusion model, and the label data is used as outputs of the hazard source diffusion model to train the hazard source diffusion model to obtain a target hazard source diffusion model; the hazard source diffusion model is obtained by using a machine learning algorithm or a neural network construction;

[0162] The data acquisition unit 604 of the point to be predicted is used to determine the center position and delay time of the point to be predicted, and calculate the spatiotemporal position characteristics to be predicted according to the center position and the delay time;

[0163] The effective monitoring point determination unit 605 is used to select a number of effective monitoring points from all the existing monitoring points according to the spatiotemporal position characteristics to be predicted, wherein the effective monitoring points are the existing monitoring points within a preset spatial range and time window; the spatial range is a range defined with the central position as the center;

[0164] An initial prediction result output unit 606 is used to input the spatiotemporal location characteristics, the hazard source characteristics and the meteorological characteristics corresponding to the respective effective monitoring points into the hazard source diffusion model for prediction to obtain a number of initial prediction results;

[0165] The target prediction result calculation unit 607 is used to fuse the several initial prediction results to obtain the target prediction result of the point to be predicted.

[0166] The embodiment of the present application may also provide a hazard source diffusion prediction and early warning device, the device comprising a processor and a memory:

[0167] The memory is used to store program code and transmit the program code to the processor;

[0168] The processor is used to execute the steps of the above-mentioned hazard source diffusion prediction and early warning method according to the instructions in the program code.

[0169] like Figure 7 As shown, a hazard source diffusion prediction and warning device provided in an embodiment of the present application may include: a processor 10, a memory 11, a communication interface 12 and a communication bus 13. The processor 10, the memory 11 and the communication interface 12 all communicate with each other through the communication bus 13.

[0170] In the embodiment of the present application, the processor 10 may be a central processing unit (CPU), an application specific integrated circuit, a digital signal processor, a field programmable gate array or other programmable logic devices, etc.

[0171] The processor 10 can call the program stored in the memory 11. Specifically, the processor 10 can execute the operations in the embodiment of the hazard source diffusion prediction and early warning method.

[0172] The memory 11 is used to store one or more programs, which may include program codes, and the program codes include computer operation instructions. In the embodiment of the present application, the memory 11 at least stores programs for implementing the following functions:

[0173] Obtain multi-source heterogeneous data of all existing monitoring points by means of fixed and / or mobile devices at preset time intervals, wherein the multi-source heterogeneous data at least includes the hazard source parameters, environmental parameters and social impact data corresponding to each of the existing monitoring points;

[0174] Preprocessing the multi-source heterogeneous data and extracting features; the features include label features and hazard source diffusion features extracted based on the position information between the two existing monitoring points; the hazard source diffusion features include the spatiotemporal position features between the two existing monitoring points, the hazard source characteristic features corresponding to each of the existing monitoring points, and the meteorological features corresponding to each of the existing monitoring points;

[0175] Determine label data, the label data is the feature to be predicted; use the spatiotemporal location feature, the hazard source characteristic feature and the meteorological feature as input of a hazard source diffusion model, use the label data as output of the hazard source diffusion model to train the hazard source diffusion model to obtain a target hazard source diffusion model; the hazard source diffusion model is obtained by using a machine learning algorithm or a neural network construction;

[0176] Determine the center position and delay time of the point to be predicted, and calculate the spatiotemporal position characteristics to be predicted based on the center position and the delay time;

[0177] According to the spatiotemporal position characteristics to be predicted, a number of valid monitoring points are obtained by screening from all the existing monitoring points, wherein the valid monitoring points are the existing monitoring points within a preset spatial range and time window; the spatial range is a range defined with the central position as the center;

[0178] Input the spatiotemporal location characteristics, the hazard source characteristic characteristics and the meteorological characteristics corresponding to the respective effective monitoring points into the hazard source diffusion model for prediction to obtain a number of initial prediction results;

[0179] A plurality of the initial prediction results are fused to obtain the target prediction result of the point to be predicted.

[0180] In one possible implementation, the memory 11 may include a program storage area and a data storage area, wherein the program storage area can store an operating system and application programs required for at least one function (such as a file creation function, a data reading and writing function), etc.; the data storage area can store data created during use, such as initialization data, etc.

[0181] In addition, the memory 11 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device or other volatile solid-state storage device.

[0182] The communication interface 12 may be an interface of a communication module, and is used to connect to other devices or systems.

[0183] Of course, it should be noted that Figure 7 The structure shown does not constitute a limitation on the hazard source diffusion prediction and early warning device in the embodiment of the present application. In actual applications, the hazard source diffusion prediction and early warning device may include Figure 7 More or fewer components than shown, or combinations of certain components.

[0184] The embodiment of the present application may also provide a computer-readable storage medium, which is used to store program code, and the program code is used to execute the steps of the above-mentioned hazard source diffusion prediction and early warning method.

[0185] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0186] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.

[0187] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.

[0188] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A method for predicting and warning the spread of dangerous sources, characterized in that: include: Obtain multi-source heterogeneous data of all existing monitoring points by means of fixed and / or mobile devices at preset time intervals, wherein the multi-source heterogeneous data at least includes the hazard source parameters, environmental parameters and social impact data corresponding to each of the existing monitoring points; Preprocessing the multi-source heterogeneous data and extracting features; The features include label features and danger source diffusion features extracted based on the position information between the two existing monitoring points; the danger source diffusion features include the spatiotemporal position features between the two existing monitoring points, the danger source characteristic features corresponding to each of the existing monitoring points, and the meteorological features corresponding to each of the existing monitoring points; Determine label data, the label data is the feature to be predicted; use the spatiotemporal location feature, the hazard source characteristic feature and the meteorological feature as input of a hazard source diffusion model, use the label data as output of the hazard source diffusion model to train the hazard source diffusion model to obtain a target hazard source diffusion model; Determine the center position and delay time of the point to be predicted, and calculate the spatiotemporal position characteristics to be predicted based on the center position and the delay time; According to the spatiotemporal position characteristics to be predicted, a number of valid monitoring points are obtained by screening from all the existing monitoring points, wherein the valid monitoring points are the existing monitoring points within a preset spatial range and time window; the spatial range is a range defined with the central position as the center; Input the spatiotemporal location characteristics, the hazard source characteristic characteristics and the meteorological characteristics corresponding to the respective effective monitoring points into the hazard source diffusion model for prediction to obtain a number of initial prediction results; A plurality of the initial prediction results are fused to obtain the target prediction result of the point to be predicted.

2. The method for predicting and warning the spread of dangerous sources according to claim 1, characterized in that: The hazard source is determined to be based on air transmission, and the concentration of hazardous gases is extracted as the label feature; the hazard source is determined to be a biological epidemic, and indicators reflecting the impact of the epidemic on ecology and other aspects are extracted as the label feature.

3. The method for predicting and warning the spread of dangerous sources according to claim 1, characterized in that: The spatiotemporal position features include the monitoring time interval between the two existing monitoring points, the spherical distance between the two existing monitoring points, the vertical height between the two existing monitoring points, and the relative orientation between the two existing monitoring points; The hazard source characteristic features are features effective for diffusion prediction that are extracted based on the hazard source characteristics.

4. The method for predicting and warning the spread of dangerous sources according to claim 1, characterized in that: The hazard source diffusion model is obtained by constructing a machine learning algorithm or a neural network.

5. The method for predicting and warning the spread of dangerous sources according to claim 1, characterized in that: The spatial range is a range specified by a user with the central position as the center; Alternatively, the spatial range is centered at the central position and the range within the prediction area is dynamically determined based on the target monitoring value.

6. The method for predicting and warning the spread of dangerous sources according to claim 1, characterized in that: The fusion method includes an averaging method or a weighted averaging method; the weighted averaging method includes assigning different weights to different prediction results according to the distance between the existing monitoring point and the point to be predicted and the wind direction factor, and calculating the weighted average value.

7. The method for predicting and warning the spread of dangerous sources according to claim 1, characterized in that: Based on the target prediction results, in the predicted key diffusion areas or key areas of expected change trends, determine to adjust and increase the deployment density of the collection equipment to optimize the monitoring points.

8. The method for predicting and warning the spread of dangerous sources according to claim 1, characterized in that: Based on the actual meaning of the value of the label feature, combined with the actual situation or international standards, determine the thresholds of different hazard levels to divide the hazard levels; Based on the target prediction result, a convex hull algorithm or a kernel density estimation algorithm is used to calculate and obtain boundary lines of different danger levels; In chronological order, the danger range level maps after different delay times are combined together to construct a danger source diffusion situation map.

9. The method for predicting and warning the spread of dangerous sources according to claim 1, characterized in that: Determine the danger level of the warning based on the actual situation, and trigger the warning information based on social information such as personnel, ecology, and facilities within the scope; After the warning is triggered, warning information will be released to the public and relevant departments through multiple channels. At the same time, personalized risk avoidance advice and emergency response guidelines will be provided in combination with road traffic and personnel distribution information.

10. A device for predicting and warning the spread of dangerous sources, characterized in that: Used to execute the hazard source diffusion prediction and early warning method according to any one of claims 1 to 9, the device comprises: A training data acquisition unit is used to collect multi-source heterogeneous data of all existing monitoring points at preset time intervals through fixed and / or mobile devices, wherein the multi-source heterogeneous data at least includes the hazard source parameters, environmental parameters and social impact data collected corresponding to each of the existing monitoring points; A feature extraction unit, used for preprocessing the multi-source heterogeneous data and extracting features; the features include label features and hazard source diffusion features extracted based on the position information between the two existing monitoring points; the hazard source diffusion features include the spatiotemporal position features between the two existing monitoring points, the hazard source characteristic features corresponding to each of the existing monitoring points, and the meteorological features corresponding to each of the existing monitoring points; A model training unit is used to determine label data, wherein the label data is a feature to be predicted; the spatiotemporal location feature, the hazard source characteristic feature and the meteorological feature are used as inputs of a hazard source diffusion model, and the label data is used as outputs of the hazard source diffusion model to train the hazard source diffusion model to obtain a target hazard source diffusion model; the hazard source diffusion model is obtained by using a machine learning algorithm or a neural network construction; The data acquisition unit of the point to be predicted is used to determine the center position and delay time of the point to be predicted, and calculate the spatiotemporal position characteristics to be predicted according to the center position and the delay time; An effective monitoring point determination unit is used to select and obtain a number of effective monitoring points from all the existing monitoring points according to the spatiotemporal position characteristics to be predicted, wherein the effective monitoring points are the existing monitoring points within a preset spatial range and time window; the spatial range is a range defined with the central position as the center; An initial prediction result output unit, used to input the spatiotemporal location characteristics, the hazard source characteristic characteristics and the meteorological characteristics corresponding to the respective effective monitoring points into the hazard source diffusion model for prediction to obtain a number of initial prediction results; The target prediction result calculation unit is used to fuse several of the initial prediction results to obtain the target prediction result of the point to be predicted.

Images