A method for short-term forecasting of disaster-stricken areas in emergency Internet of Things (IoT)

Abstract

This invention relates to the field of meteorological forecasting technology, specifically disclosing a short-term forecasting method for post-disaster areas in an emergency Internet of Things (IoT). By establishing a two-level emergency IoT network and based on the convergent cross-mapping theory, the relationship between the sub-map and the overall map is explained by calculating the absolute value of the correlation coefficient between the post-disaster area and other areas. This provides short-term forecasting services for the post-disaster area, improving the accuracy of short-term forecasts and effectively alleviating the problem of decreased prediction performance caused by limited data uploads from gateway servers. It also solves the problem that traditional emergency IoT short-term forecasts can only be generated based on limited observation data within the region, making it difficult to match the accuracy of short-term forecasts generated by aggregating observation information from all sensor nodes, resulting in inconsistent short-term forecast quality.

Description

Technical Field

[0001] This application relates to the field of meteorological forecasting technology, and specifically discloses a short-term forecasting method for disaster-stricken areas in an emergency Internet of Things. Background Technology

[0002] Emergency communication networks, as networks that can provide reliable communication during crises, collect high-precision observation data from a large number of deployed sensors during operation, and obtain observation data collected by nearby sensors based on gateway servers. They can generate and broadcast short-term forecasts in a timely and reliable manner for disaster-stricken areas. In existing technologies, short-term forecasts refer to predictions of observable data within the next 12 hours, and are used for meteorological risk prevention and control and efficient crisis early warning.

[0003] However, limited emergency communication network bandwidth hinders gateway servers in non-disaster areas from uploading observational data, leading to the loss of inter-regional correlation data and ultimately reducing the accuracy of short-term forecasts. Simultaneously, limited data collection makes identifying potentially affected areas more difficult. Furthermore, when extreme weather events damage network infrastructure—fiber optic cables or base stations—network failures occur, causing communication interruptions. Such interruptions prevent numerous sensors from uploading their observational data to data centers (gateway servers), creating a critical gap between high-precision meteorological observations and short-term forecasts in post-disaster areas. This gap severely delays crisis response and management.

[0004] In existing technologies, in response to communication disruptions, an emergency Internet of Things (IoT) has been proposed that can ensure seamless access of distributed IoT to existing emergency communication networks. With its wide coverage and high robustness, it has become a key infrastructure for maintaining reliable communication and issuing early warning information to the public during crises.

[0005] The emergency IoT system employs a two-tier architecture. The bottom layer consists of several sensors deployed in different areas, each equipped with a gateway server. The upper layer comprises all gateway servers interconnected via a stable, low-bandwidth channel provided by an emergency communication network. During disasters such as extreme weather, this two-tier architecture ensures that sensors in the affected areas can promptly upload observational data to their respective gateway servers. These gateway servers then run a GNN-based machine learning model—the estimator Φ—to generate short-term forecasts.

[0006] While emergency IoT networks can provide short-term forecast services to disaster-stricken areas through multiple gateway servers when communication paths in traditional data centers are blocked, the fact remains that each sensor can only upload its observation data to its local gateway server means that short-term forecasts are generated based on limited observation data within its own region. This results in inconsistencies in the accuracy of short-term forecasts compared to those generated by aggregating observations from all sensor nodes.

[0007] This invention provides a short-term forecasting method for disaster-stricken areas in emergency Internet of Things (IoT) to solve the above-mentioned problems. Summary of the Invention

[0008] The purpose of this invention is to solve the problem that traditional emergency IoT short-term forecasts can only be generated based on limited observation data within the area, resulting in low accuracy and inconsistent short-term forecast quality.

[0009] To achieve the above objectives, the basic solution of this invention provides a short-term forecasting method for post-disaster areas in an emergency Internet of Things, comprising the following steps:

[0010] Step A1: Construct the overall graph with the sensors as vertices and the distance between every two sensors as edges;

[0011] Sensors are divided into regions, sub-maps are constructed for each region, and independent gateway servers collect observations from all sensors within their respective regions.

[0012] Step A2: The area with abnormal observations at the current moment is divided into the post-disaster area, and the gateway server of the area broadcasts the historical observations of the area up to the current moment;

[0013] Step A3: Each gateway server in a non-disaster area generates a short-term forecast for the post-disaster area based on the historical observations of its own area and the historical observations of the post-disaster area at the same time. It then filters and obtains a set of valid short-term forecasts and sends them back to the gateway server in the post-disaster area.

[0014] Step A4: The gateway server in the post-disaster area, with the optimization objective of minimizing the sum of the products of the absolute value ratio of the correlation coefficient between each non-disaster area and the post-disaster area and the expected normalized mean absolute error of the corresponding short-term forecast for the non-disaster area, alternately executes the following iterative optimization process:

[0015] Calculate the absolute values ​​of the correlation coefficients between each non-disaster area and the post-disaster area based on the current short-term forecasts;

[0016] The new short-term forecast is obtained by weighting and fusing all current short-term forecasts using the absolute value of this coefficient as the weight.

[0017] The process continues until the decrease in the objective function value between the current and previous iterations is less than a preset threshold, at which point the final short-term forecast and correlation coefficient set are obtained.

[0018] Furthermore, in step A1, the distance between two sensors in the same area is calculated using the following formula:

[0019] ;

[0020] In the formula, This indicates that it is located in the m-th region. The i-th sensor, This indicates that it is located in the m-th region. The j-th sensor, Indicates the total number of regions. Indicates the first One sensor With the One sensor The distance between them.

[0021] Furthermore, the observations from each sensor are represented in vector form.

[0022] Furthermore, in step A2, the historical observations broadcast in the post-disaster area are output in the form of a sequence list.

[0023] Furthermore, in step A3, the non-disaster area The gateway server is equipped with a GNN-based machine learning model and predicts post-disaster areas using the following formula. Meteorological conditions as short-term forecasts :

[0024] ;

[0025] In the formula, Indicates the post-disaster area Historical observation sequence, Indicates non-disaster areas The historical observation sequence, and k≠m;

[0026] In the formula, Its elements follow a Gaussian distribution. Indicates at time Lower disaster area The mean of all sensor observations. Indicates at time Lower non-disaster areas The variance of all sensor observations.

[0027] Furthermore, in step A3, the absolute error with the actual observed values ​​of the post-disaster area at the next time step is used to determine the confidence level. Below, the accuracy threshold is met. The required short-term forecast is an effective short-term forecast.

[0028] Furthermore, in step A4, the following optimization objective function is established:

[0029] ;

[0030] In the formula, Indicates non-disaster areas and post-disaster areas At any moment The correlation coefficient, Indicates non-disaster areas and post-disaster areas At any moment absolute value of the correlation coefficient This represents the sum of the absolute values ​​of the correlation coefficients for all non-disaster areas;

[0031] In the formula, Indicates the difference between predicted and actual values The expectation of the norm;

[0032] In the formula, This indicates the number of sensors in the area. Indicates the precision threshold. This indicates the total number of regions.

[0033] Furthermore, in step A4, after establishing the optimization objective function, based on the current short-term forecast, a conservative strategy is adopted to calculate the upper bound of the probability when the short-term forecast of each non-disaster area does not meet the accuracy threshold. Then, through the regularization function and the Lagrange multiplier method, the absolute value of the correlation coefficient between each non-disaster area and the post-disaster area is obtained.

[0034] Furthermore, post-disaster areas and non-disaster areas At any moment absolute value of correlation coefficient Calculate using the following formula:

[0035] ;

[0036] In the formula, , This represents a dummy variable that iterates through all non-disaster regions. This represents the historical observation sequence of the corresponding non-disaster area.

[0037] Furthermore, in obtaining post-disaster areas and non-disaster areas At any moment absolute value of correlation coefficient Subsequently, the short-term forecasts from multiple gateway servers are weighted and fused using the following formula to obtain a new short-term forecast:

[0038] .

[0039] The principle and effect of this basic scheme are as follows:

[0040] Compared with existing technologies, this invention establishes a two-level emergency IoT network. Based on the convergent cross-mapping theory, it interprets the relationship between the subgraph and the overall graph by calculating the absolute value of the correlation coefficient between the post-disaster area and the non-disaster area, thereby providing short-term forecast services for the post-disaster area, improving the accuracy of short-term forecasts, and effectively alleviating the problem of decreased prediction performance caused by limited data uploads from the gateway server. At the same time, this invention also judges the intensity of the impact of the post-disaster area on the non-disaster area by judging the absolute value of the correlation coefficient between the post-disaster area and the non-disaster area. This solves the problem that the short-term forecasts of traditional emergency IoT can only be generated based on limited observation data within the area, and the accuracy is difficult to match that of short-term forecasts generated by aggregating the observation information of all sensor nodes, resulting in inconsistent short-term forecast quality. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 A flowchart of a short-term forecasting method for disaster-stricken areas in an emergency Internet of Things (IoT) proposed in this application is shown. Detailed Implementation

[0043] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0044] A short-term forecasting method for disaster-stricken areas in an emergency Internet of Things (IoT) system, implementing, for example... Figure 1 As shown, it includes the following steps:

[0045] Step A1: Construct the overall graph with the sensors as vertices and the distance between every two sensors as edges;

[0046] Sensors are divided into regions, and sub-maps are constructed for each region. An independent gateway server collects the observations of all sensors within its assigned region.

[0047] Specifically, the expression for the site plan is as follows:

[0048] G=(V,A);

[0049] In the formula, V represents a set containing N vertices, each vertex corresponding to a sensor, N = |V|, and A represents the weighted adjacency matrix. ,in, Indicates the first One sensor With the One sensor The distance between them.

[0050] In this embodiment, a total of 9 sensors are deployed: Among them, sensors With sensors The distance between them is 10 kilometers, and the sensors With sensors The distance between them is 15 kilometers. .

[0051] Then, the local map is mapped to its corresponding region to divide it into sub-maps. The area covered by all sensors is divided into M sub-regions, and the calculation of the distance between two sensors in the same region is optimized using the following formula:

[0052] ;

[0053] In the formula, This indicates that it is located in the m-th region. The i-th sensor, This indicates that it is located in the m-th region. The j-th sensor.

[0054] In this embodiment, each region Each system is equipped with an independent gateway server to aggregate observations from all sensors within its designated area. Furthermore, in this embodiment, only historical observations from the disaster-stricken area are uploaded to the emergency communication network, along with short-term forecasts transmitted from non-disaster areas. Local historical observations from non-disaster areas are processed entirely within their local gateway servers.

[0055] Step A2: The area with abnormal observations at the current moment is divided into the post-disaster area, and the gateway server of the area broadcasts the historical observations of the area up to the current moment.

[0056] Specifically, the constructed graph model includes a master graph and various subgraphs divided by region. For the current time t, the master graph... The observed values ​​at each vertex, i.e., the sensor's observed values, are represented by the following vector formula:

[0057] ;

[0058] For the region at the current time t, Corresponding subgraph The observed values ​​at each vertex, i.e., the sensor's observed values, are represented by the following vector formula:

[0059] ;

[0060] In the formula, , where n represents the number of sensors in the area. Indicates the region The observation value of the nth sensor at time t.

[0061] In this embodiment, the region It contains three sensors: , , At the current time t, the wind speed observed by the sensor is: The unit is meters per second.

[0062] Furthermore, when the region At any moment When extreme weather events occur, the region If the observed value exceeds a preset state threshold, the observed value is determined to be an abnormal state, and the region is then... The area was designated as a post-disaster zone, and was located within the post-disaster zone. The gateway server began broadcasting to the region from the time before the disaster. Up to the current moment The historical observations are broadcast as a sequence list, expressed as follows:

[0063] ;

[0064] In the formula, In the moments before the disaster Lower disaster area The observations from all sensors.

[0065] Step A3: Each gateway server in a non-disaster area generates a short-term forecast for the disaster-stricken area based on the historical observations of its own area and the historical observations of the disaster-stricken area at the same time. It then filters and obtains a set of valid short-term forecasts and sends them back to the gateway server in the disaster-stricken area.

[0066] Specifically, in the post-disaster area Complete the historical observation sequence Following the broadcast, non-disaster areas The gateway server in the disaster area at the same time Historical observation series The observed values ​​in the data, and the historical observation series of this region. The observed values ​​are taken as input, and the GNN-based machine learning model, estimator Φ, running on the gateway server, predicts the post-disaster area using the following formula. Meteorological conditions as short-term forecasts :

[0067] ;

[0068] In the formula, Indicates non-disaster areas Historical observation sequence, And k≠m;

[0069] In the formula, Its elements follow a Gaussian distribution. Indicates at time Lower disaster area The mean of all sensor observations. Indicates at time Lower non-disaster areas The variance of all sensor observations.

[0070] In this embodiment, the GNN-based machine learning model running on each gateway server, the estimator Φ, is a commonly used GNN-based machine learning model applied to the gateway server in the prior art.

[0071] Ultimately, the gateway servers deployed in all regions were obtained for the disaster-stricken areas. The generated short-term forecast set .

[0072] In this embodiment, a metric is established. The quality of the predicted short-term forecasts is evaluated and screened, among which, δ represents the accuracy threshold, and δ represents the confidence level replenishment. Each non-disaster region... In obtaining the next moment Post-disaster areas Short-term forecast Subsequently, its short-term forecast With the next moment Post-disaster areas The actual observed value The absolute error should be within the confidence level. Below, the accuracy threshold is met. The requirement is expressed as follows:

[0073] ;

[0074] In this embodiment, with confidence level Below, the accuracy threshold is met. The required short-term forecast results are valid short-term forecast results; otherwise, they are invalid short-term forecast results, and invalid short-term forecast results are discarded. Specifically, if the region... If an extreme weather event occurs at 10:00 AM, the region will be affected. The area is divided into post-disaster zones, and a precision threshold is set in the established metric. When non-disaster areas Targeting the disaster-stricken areas The generated short-term wind speed forecast for 10:30 AM is as follows: At that time, corresponding to 10:30 AM in the post-disaster area The actual wind speed was [55, 60, 62]T. The calculated average absolute error between the short-term wind speed forecast and the actual wind speed was 7, which exceeded the accuracy threshold. Therefore, this short-term forecast did not meet the required accuracy and was classified as an invalid short-term forecast and removed.

[0075] Furthermore, if it is not a disaster area Internal sensor at all times The obtained meteorological conditions and post-disaster areas If the obtained meteorological conditions are almost opposite, it indicates a non-disaster area. At the same time, another extreme weather event occurred.

[0076] Step A4: The gateway server in the post-disaster area, with the optimization objective of minimizing the sum of the products of the absolute value ratio of the correlation coefficient between each non-disaster area and the post-disaster area and the expected normalized mean absolute error of the corresponding short-term forecast for the non-disaster area, alternately executes the following iterative optimization process:

[0077] Calculate the absolute values ​​of the correlation coefficients between each non-disaster area and the post-disaster area based on the current short-term forecasts;

[0078] The new short-term forecast is obtained by weighting and fusing all current short-term forecasts using the absolute value of this coefficient as the weight.

[0079] The process continues until the decrease in the objective function value between the current and previous iterations is less than a preset threshold, at which point the final short-term forecast and correlation coefficient set are obtained.

[0080] Furthermore, in the first iteration, the arithmetic mean of all valid short-term forecasts is used as the initial current short-term forecast. Also, the sum of the absolute values ​​of the correlation coefficients is the absolute value of the total correlation coefficient.

[0081] Specifically, in this embodiment, the following optimization objective function is established:

[0082] ;

[0083] In the formula, Indicates non-disaster areas and post-disaster areas At any moment The correlation coefficient, Indicates non-disaster areas and post-disaster areas At any moment absolute value of the correlation coefficient This represents the sum of the absolute values ​​of the correlation coefficients for all non-disaster areas;

[0084] In the formula, Indicates the difference between predicted and actual values The expected value of the norm (mean absolute error).

[0085] After establishing the optimization objective function, based on the current short-term forecast... Using a conservative strategy, the short-term forecasts for each non-disaster area did not meet the accuracy threshold, i.e. The upper bound of the probability at time t is finally obtained by the following expression:

[0086] ;

[0087] Then, using a regularization function: Using the Lagrange multiplier method, the post-disaster area can be solved. and non-disaster areas At any moment absolute value of correlation coefficient The expression is as follows:

[0088] ;

[0089] In the formula, To ensure that regional observations highly correlated with the post-disaster area occupy a prominent position in short-term forecasts, This represents a dummy variable that iterates through all non-disaster regions. This represents the historical observation sequence of the corresponding non-disaster area.

[0090] The process of the Lagrange multiplier method is as follows:

[0091] Based on the objective optimization function, variables are introduced. And convert it into a Lagrange function of the following form:

[0092] ;

[0093] In the formula, It represents the Lagrange multiplier.

[0094] Furthermore, regarding Find the partial derivative and set it to zero, then consider the constraints. We obtain the following formula:

[0095] ;

[0096] ;

[0097] Final integration of the post-disaster area and non-disaster areas At any moment absolute value of correlation coefficient The expression.

[0098] As the number of participating servers increases, a weighted combination strategy for aggregating short-term forecasts from each server yields better accuracy compared to the output of a single server. Therefore, in obtaining post-disaster regional... and non-disaster areas At any moment absolute value of correlation coefficient Subsequently, the short-term forecasts from multiple gateway servers are weighted and fused using the following formula to obtain a new short-term forecast:

[0099] ;

[0100] In this embodiment, the disaster-stricken area and non-disaster areas At any moment absolute value of correlation coefficient Reflecting the post-disaster area and non-disaster areas The correlation.

[0101] Furthermore, based on the absolute value of the correlation coefficient between the disaster-stricken area and other areas at the next moment, the absolute value of the correlation coefficient is used as the magnitude of the impact. When the magnitude of the impact exceeds the alarm threshold, an emergency alarm is issued to the other area. For example, if area If an extreme weather event occurs at 10:00 AM, the region Areas designated as post-disaster zones, or non-disaster zones The estimator mounted on the gateway server predicted that at 10:30 AM, the non-disaster area... and post-disaster areas absolute value of correlation coefficient If the alarm threshold of 0.7 is exceeded, then the area will be... Issue an emergency alarm.

[0102] Finally, when the decrease in the objective function value between the current and previous iterations is less than a preset threshold, the final short-term forecast is obtained. and correlation coefficient set .

[0103] This invention establishes a two-level emergency IoT network. Based on the convergent cross-mapping theory, it interprets the relationship between the subgraph and the overall graph by calculating the absolute value of the correlation coefficient between the post-disaster area and the non-disaster area, thereby providing short-term forecast services for the post-disaster area, improving the accuracy of short-term forecasts, and effectively alleviating the problem of decreased prediction performance caused by limited data uploads from the gateway server. At the same time, this invention also assesses the intensity of the impact of the post-disaster area on the non-disaster area by judging the absolute value of the correlation coefficient between the post-disaster area and the non-disaster area.

[0104] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any indirect modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A short-term forecasting method for post-disaster areas in an emergency Internet of Things (IoT) system, characterized in that, Includes the following steps: Step A1: Construct the overall graph with the sensors as vertices and the distance between every two sensors as edges; Sensors are divided into regions, sub-maps are constructed for each region, and independent gateway servers collect observations from all sensors within their respective regions. The observations from each sensor are expressed in vector form; Step A2: The area with abnormal observations at the current moment is divided into the post-disaster area, and the gateway server of the area broadcasts the historical observations of the area up to the current moment; Step A3: Each gateway server in a non-disaster area generates a short-term forecast for the post-disaster area based on the historical observations of its own area and the historical observations of the post-disaster area at the same time. It then filters and obtains a set of valid short-term forecasts and sends them back to the gateway server in the post-disaster area. In step A3, the non-disaster area The gateway server is equipped with a GNN-based machine learning model and predicts post-disaster areas using the following formula. Meteorological conditions as short-term forecasts : ； In the formula, Indicates the post-disaster area Historical observation sequence, Indicates non-disaster areas The historical observation sequence, k≠m; In the formula, Its elements follow a Gaussian distribution. Indicates at time Lower disaster area The mean of all sensor observations. Indicates at time Lower non-disaster areas Variance of all sensor observations; Step A4: The gateway server in the post-disaster area, with the optimization objective of minimizing the sum of the products of the absolute value ratio of the correlation coefficient between each non-disaster area and the post-disaster area and the expected normalized mean absolute error of the corresponding short-term forecast for the non-disaster area, alternately executes the following iterative optimization process: Calculate the absolute values ​​of the correlation coefficients between each non-disaster area and the post-disaster area based on the current short-term forecasts; The new short-term forecast is obtained by weighting and fusing all current short-term forecasts using the absolute value of this coefficient as the weight. The process continues until the decrease in the objective function value between the current and previous iterations is less than a preset threshold, at which point the final short-term forecast and correlation coefficient set are obtained. In step A4, the following optimization objective function is established: ； In the formula, Indicates non-disaster areas and post-disaster areas At any moment The correlation coefficient, Indicates non-disaster areas and post-disaster areas At any moment absolute value of the correlation coefficient This represents the sum of the absolute values ​​of the correlation coefficients for all non-disaster areas; In the formula, Indicates the difference between predicted and actual values The expectation of the norm; In the formula, This indicates the number of sensors in the area. Indicates the precision threshold. Indicates the total number of regions; In step A4, after establishing the optimization objective function, based on the current short-term forecast, a conservative strategy is adopted to calculate the upper probability bound when the short-term forecast of each non-disaster area does not meet the accuracy threshold. Then, through the regularization function and the Lagrange multiplier method, the absolute value of the correlation coefficient between each non-disaster area and the post-disaster area is obtained.

2. The method for short-term forecasting of post-disaster areas in an emergency Internet of Things according to claim 1, characterized in that, In step A1, the distance between two sensors in the same area is calculated using the following formula: ； In the formula, This indicates that it is located in the m-th region. The i-th sensor, This indicates that it is located in the m-th region. The j-th sensor, Indicates the total number of regions. Indicates the first One sensor With the One sensor The distance between them.

3. The method for short-term forecasting of post-disaster areas in an emergency Internet of Things according to claim 1, characterized in that, In step A2, the historical observations broadcast in the post-disaster area are output as a sequence list.

4. A short-term forecasting method for post-disaster areas in an emergency Internet of Things according to claim 1, characterized in that, In step A3, the absolute error with the actual observed value of the post-disaster area at the next time step is used to satisfy the accuracy threshold at a confidence level of 1-δ. The required short-term forecast is an effective short-term forecast, where δ represents the confidence level replenishment.

5. A short-term forecasting method for post-disaster areas in an emergency Internet of Things according to claim 1, characterized in that, Post-disaster areas and non-disaster areas At any moment absolute value of correlation coefficient Calculate using the following formula: ； In the formula, , This represents a dummy variable that iterates through all non-disaster regions. This represents the historical observation sequence of the corresponding non-disaster area.

6. A short-term forecasting method for post-disaster areas in an emergency Internet of Things according to claim 1, characterized in that, In obtaining the disaster-stricken area and non-disaster areas At any moment absolute value of correlation coefficient Subsequently, the short-term forecasts from multiple gateway servers are weighted and fused using the following formula to obtain a new short-term forecast: 。

Images