Emergency management disaster dynamic early warning scheduling system based on multi-source data fusion

Abstract

The invention relates to an emergency management disaster dynamic early-warning scheduling system based on multi-source data fusion, which comprises a multi-source data acquisition module, a physical sensor-containing data acquisition sub-module, a network data acquisition sub-module, a multi-source data integration module, a multi-source data management module and a multi-source data management module, the system comprises a preprocessing sub-module, an annotation sub-module, a fusion processing sub-module, a dynamic early warning generation module and a precise resource scheduling module. Data are collected through a meteorological satellite, a ground monitoring station, an Internet of Things sensor and a network channel, and disaster situation sensing information is formed through cleaning, labeling and fusion processing. Based on a disaster evolution model of a physical and machine learning algorithm, multi-level time period early warning information is generated in real time in combination with historical data. According to the GIS and the emergency resource database, an improved algorithm is used for planning a resource allocation scheme, efficient early warning of disasters and accurate resource scheduling can be achieved, and the emergency management level is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of emergency management and disaster warning, and more particularly, to an emergency management disaster dynamic warning and dispatching system based on multi-source data fusion. Background Art

[0002] In today's society where disasters occur frequently, including natural disasters such as earthquakes, floods, typhoons, etc. and man-made disasters such as industrial accidents, fires, etc., traditional emergency management systems face many challenges. On the one hand, they usually rely on limited data sources and it is difficult to comprehensively and accurately grasp disaster information, with relatively large information blind spots. For example, relying solely on meteorological satellite data is difficult to accurately predict disasters in local areas, and the data from ground monitoring stations may also be unable to reflect the overall disaster situation due to locality and independence. On the other hand, traditional warning mechanisms are mostly static, unable to reflect the dynamic evolution process of disasters in real time, and lack scientificity and timeliness in resource scheduling, resulting in low efficiency of emergency rescue and difficulty in meeting the complex and changeable actual needs during disasters. Therefore, there is an urgent need to develop an innovative system that can fully integrate multi-source data and achieve dynamic disaster warning and resource scheduling. Summary of the Invention

[0003] The purpose of the present invention is to provide an emergency management disaster dynamic warning and dispatching system based on multi-source data fusion, which can solve the problems in the prior art that traditional warning mechanisms are mostly static, unable to reflect the dynamic evolution process of disasters in real time, and lack scientificity and timeliness in resource scheduling, resulting in low efficiency of emergency rescue and difficulty in meeting the complex and changeable actual needs during disasters.

[0004] To solve the above technical problems, the technical solution adopted in this application is as follows:

[0005] This application embodiment provides an emergency management disaster dynamic warning and dispatching system based on multi-source data fusion, including: a multi-source data acquisition module, the multi-source data acquisition module includes a physical sensor acquisition sub-module and a network acquisition sub-module, which are used to collect original data D related to disasters; a multi-source data integration module, which is used to preprocess, label, and fuse the original data D collected by the multi-source data acquisition module to generate disaster situation perception information I, providing a basis for warning; a dynamic warning generation module, which is used to generate warning information W from the disaster situation perception information I obtained by the multi-source data integration module; and a precise resource scheduling module, which is used to process the warning information W generated by the dynamic warning generation module and output resource allocation information S.

[0006] Multi-source data acquisition module: The multi-source data acquisition module is the basis of the emergency management disaster dynamic warning and dispatching system. It collects disaster-related data from multiple channels. The physical sensor acquisition submodule uses professional equipment, such as seismic sensors to capture crustal vibrations, and meteorological sensors to monitor meteorological elements such as wind speed and rainfall, to provide physical parameters for disaster monitoring. The network acquisition submodule uses the Internet to obtain data from social media and government platforms, including disaster site information shared by the public and disaster statistics and emergency plans released by the government. It summarizes two types of data D (hereinafter referred to as D) to provide raw data for subsequent processing.

[0007] Multi-source data integration module: This module receives the collected raw data, performs preprocessing, annotation and fusion, and generates disaster situation awareness information. The preprocessing uses algorithms such as sliding average and wavelet transform to remove noise, fill missing values, and improve data quality. The annotation is based on classification standards, adding labels such as disaster type, time, and location to enhance data analyzability. In the fusion stage, Kalman filtering is used to process time series data and mine time correlation; Kriging interpolation is used to fuse spatial data to grasp the spatial distribution of disasters, and finally I (hereinafter referred to as I) is generated to provide a basis for early warning.

[0008] Dynamic warning generation module: The dynamic warning generation module is one of the core of the system. It generates accurate warning information W (hereinafter referred to as W) based on the fused situation awareness information I and historical disaster analysis results A (hereinafter referred to as A) and constructs the disaster evolution model M. (·) (hereinafter referred to as M (·) ), LSTM is used to process time series data, and its gating mechanism is used to capture long-term dependencies and model disaster development trends; CNN is used to analyze spatial data, and key features are extracted through convolution, pooling, and fully connected layers. The model is optimized through large-scale data training and outputs W.

[0009] Precise resource scheduling module: The precise resource scheduling module plans the emergency resource allocation plan S (hereinafter referred to as S) according to the early warning information W, combined with GIS data G and emergency resource database information R (hereinafter referred to as R). The path planning adopts the improved Dijkstra algorithm, dynamically adjusts the path weight considering traffic conditions and weather, finds the optimal path, and sets objective functions such as minimizing transportation costs based on the linear programming resource allocation model. Combined with supply, demand, non-negative and other constraints, resources are reasonably allocated. The simulated annealing algorithm is introduced to accept new plans according to probability, iteratively optimize, and obtain the specific model formula of S to improve the efficiency of resource allocation.

[0010] In some embodiments of the present invention, the above-mentioned original data D of the multi-source data acquisition module is specifically:

[0011] D=D s +Dn In the formula, the multi-source data acquisition module is the basis of the system. The physical sensor acquisition sub-module collects sensor data such as seismic and meteorological data (D s ), for example, the seismic sensor captures the crustal vibrations, and the meteorological sensor monitors meteorological elements; the network acquisition sub-module obtains data (D n ) from social media and government platforms, such as disaster information shared by the public and the emergency plans released by the official, and aggregates the two parts of data to obtain D, providing the original data for subsequent processing.

[0012] In some embodiments of the present invention, the disaster situation perception information I of the above multi-source data integration module is specifically:

[0013] I = F(L(P(D)))

[0014] In the formula, this module undertakes the acquisition data and processes it through three steps: preprocessing, annotation, and fusion. In the preprocessing step, algorithms such as moving average and interpolation are used to clean and fill the data; in the annotation link, according to the classification standard, labels such as disaster type, time, and location are added; in the fusion step, Kalman filtering is used to process time series data, and Kriging interpolation is used to process spatial distribution data, and finally, the disaster situation perception information is generated.

[0015] In some embodiments of the present invention, the warning information W of the above dynamic warning generation module is specifically:

[0016] W = M(I, A)

[0017] In the formula, this module is one of the cores of the system. According to the fused disaster situation perception information I and the historical disaster analysis result A, the warning information W is generated. When constructing the disaster evolution model M(I, A), LSTM is used to process time series data to capture long-term dependencies; CNN is used to analyze spatial data to extract key features, and the model is trained and optimized with a large amount of data to obtain accurate warning information W.

[0018] In some embodiments of the present invention, the resource allocation information S of the above precise resource scheduling module is specifically:

[0019] S = 0(W, G, R)

[0020] In the formula, according to the warning information W, combined with the GIS data G and the information R of the emergency resource database, the emergency resource allocation plan S is planned. In the path planning, the improved Dijkstra algorithm is used, and the weights are adjusted considering traffic and weather to find the optimal path; based on the linear programming model, the objective function is set and combined with constraints such as supply, demand, and non-negativity to allocate resources; the simulated annealing algorithm is used to optimize the plan, that is, S.

[0021] The above is specifically as follows:

[0022]

[0023] Among them, the system collects data D from multiple sources, integrates it to generate I, then the early warning module generates W, and finally the resource scheduling module generates S. Each module collaborates to provide technical support for emergency management, reduce disaster losses, and ensure the safety of life and property.

[0024] In some embodiments of the present invention, the above physical sensor data acquisition sub-module includes meteorological satellites, ground monitoring stations, and Internet of Things sensors. Among them, meteorological satellites can provide information such as atmospheric temperature, humidity, cloud movement, and air pressure. Ground monitoring stations can collect data through sensors such as rain gauges, anemometers, wind vanes, seismographs, and seismometers. The Internet of Things sensors include water level sensors, seismic wave monitors, smoke sensors, temperature sensors, and combustible gas sensors, etc., which are used to monitor different disaster-related data.

[0025] In some embodiments of the present invention, the above network data acquisition sub-module uses web crawler technology to collect text, pictures, videos, and other information related to disasters from channels such as social media platforms, news websites, and government announcements.

[0026] In some embodiments of the present invention, the data preprocessing sub-module of the above multi-source data integration module uses algorithms such as moving average method, wavelet transform, linear interpolation method, and spline interpolation method to clean and process missing data, so as to achieve standardized conversion of data format, noise removal, outlier processing, and filling or marking of missing data.

[0027] The moving average method is as follows:

[0028]

[0029] Among them, the moving average method is used to smooth data and remove noise. Let the original data sequence be x1, x2,..., x n , the window size is k, and the moving average value is y i .

[0030] The wavelet transform is as follows:

[0031]

[0032]

[0033] Among them, the continuous wavelet transform (CWT) x(t) is to convolve the signal with a wavelet function ψ(t). For the signal x(t), then W x (a, b) is the above mathematical relationship, where a is the scale parameter, which controls the stretching of the wavelet function; b is the translation parameter, which controls the translation of the wavelet function; ψ *Denotes the complex conjugate of ψ. The discrete wavelet transform (DWT) usually uses dyadic wavelets to decompose a signal into an approximation component and a detail component. Given a signal x[n], during the decomposition process, the approximation component cA j [n] and the detail component cD j [n] can be calculated through the above convolution formulas, where h[n] and g[n] are the coefficients of the low-pass filter and the high-pass filter respectively, and j represents the number of decomposition levels.

[0034] The linear interpolation method is as follows:

[0035]

[0036] In the formula, given two data points (x1, y1) and (x2, y2), for any x within the interval, the corresponding interpolation result can be calculated through the linear interpolation formula.

[0037] The spline interpolation method is as follows:

[0038] S(x) = a i + b i (x - x i ) + c i (x - x i ) 2 + d i (x - x i ) 3

[0039] Taking cubic spline interpolation as an example, assume that the known data points are (x i , y i ), i = 0, 1,..., n. On each sub-interval [x i , x i+1 , the cubic spline function can be expressed as the above formula.

[0040] In some embodiments of the present invention, the data annotation sub-module of the above multi-source data integration module adds labels to the data based on the disaster type classification system and geographic information system (GIS) data, and the annotation information includes data source, location, time, and related disaster characteristics, etc.

[0041] In some embodiments of the present invention, the data fusion processing sub-module of the above multi-source data integration module uses a data fusion algorithm based on Kalman filtering to fuse and predict time series data, uses Kriging interpolation to fuse and spatially interpolate spatially distributed data, applies the principal component analysis (PCA) algorithm to reduce the dimension of high-dimensional heterogeneous data, and uses a rule-based reasoning algorithm for consistency checking and deviation calibration.

[0042] The data fusion algorithm based on Kalman filtering is as follows:

[0043] State equation: X k = A k X k- + B k U k + W k

[0044] Observation equation: Z k = H k X k + V K

[0045] Predicted state:

[0046] Predicted covariance:

[0047] Kalman gain:

[0048] where X k is the state vector at time k, A k is the state transition matrix, U k is the control vector, B k is the control matrix, W k is the process noise, and W k ~ N(0, Q k ), N(0, Q k ) represents a Gaussian distribution with mean 0 and covariance Q k . In the observation equation, Z k is the observation vector at time k, H k is the observation matrix, V K is the observation noise, V K ~ N(0, R k ), where is the prediction of the state at time k based on the state at time k - 1, is the updated state estimate at time k, P k|k-1 and P k|k are the corresponding covariance matrices respectively, and I is the identity matrix.

[0049] The Kriging interpolation method is as follows:

[0050]

[0051] Let Z(x) be a regionalized variable with observed values Z(x i (1, 2…, n)) at locations x i . To estimate the value at location x0 where γ(x i , x j ) is the semivariogram, which describes the regionalized variable Z(x) at location xi and x j The degree of variation between them, and μ is the Lagrange multiplier.

[0052] Principal Component Analysis (PCA) algorithm:

[0053]

[0054] C = VΛV T

[0055]

[0056] Suppose there are m samples, and each sample has n-dimensional features, forming a data matrix X ∈ R m×n , first centralize the data to obtain Then calculate the covariance matrix C, and perform eigenvalue decomposition on the covariance matrix C to obtain C, where Λ is a diagonal matrix, and the diagonal elements are eigenvalues λ1 ≥ λ2 ≥ λ3 ≥ … ≥ λ n , and V is the eigenvector matrix. Select the eigenvectors corresponding to the first k largest eigenvalues, and project the original data onto these k eigenvectors to obtain the dimensionality-reduced data Y.

[0057] In some embodiments of the present invention, the disaster evolution modeling sub-module of the above-mentioned dynamic warning generation module constructs a disaster evolution model based on physical process modeling and machine learning algorithms, including a Long Short-Term Memory Network (LSTM) for processing time series data, a Convolutional Neural Network (CNN) for analyzing spatial data, and training and optimizing the model using a large amount of historical disaster case data and simulation data.

[0058] Long Short-Term Memory Network (LSTM) is as follows:

[0059] Forget gate: f t = σ(W f [h t-1 , x t +b f )

[0060]

[0061] Input gate: i t = σ(W i [h t-1 , x t +b i )

[0062]

[0063] Output gate: o t = σ(W o [h t-1 , xt ]+)

[0064] h t =o t ⊙tanh(C t )

[0065] The forget gate determines the cell state at the previous moment How much information needs to be forgotten? t is the output of the forget gate, which is a vector ranging from [0, 1]. Each element represents the degree to which the corresponding information is forgotten. σ is the sigmoid activation function. W f is the weight matrix of the forget gate, b f is the bias vector of the forget gate, [h t-1 , x t ] means to change the hidden state h of the previous moment t-1 and the current input x t splice together; input gate i t is the output of the sigmoid part of the input gate, W i is the weight matrix of the sigmoid part of the input gate, b i is the bias vector, is the candidate cell state, tanh is the hyperbolic tangent activation function, and w c is the weight matrix of the candidate cell state, b c is the bias vector, the output gate determines the cell state at the current moment; the output gate determines the cell state C at the current moment t How much information needs to be output to the hidden state h t , where o t is the output of the sigmoid part of the output gate, W o is the weight matrix of the output gate and b0 is the bias vector.

[0066] The convolutional neural network (CNN) is as follows:

[0067] Convolution operation:

[0068] Activation function: y i，j，c =max(0, Y i，j，c )

[0069] Loss function:

[0070] Optimization algorithm:

[0071] Where at position (i, j), the calculation of the cth channel of the output feature map Y is Y i，j，c , b cis the bias of the c-th channel, and the activation function uses the ReLu function y i，j，c , for the disaster evolution model, the commonly used loss function can be the mean squared error loss (MSE), where θ is the parameter of the model, is the predicted output of the model for the input x (i) , and the optimization algorithm uses stochastic gradient descent, where α is the learning rate. In actual training, mini-batch stochastic gradient descent (Mini-Batch SGD) is usually adopted, and a small batch of data is selected from the training dataset each time to calculate the gradient and update the parameters.

[0072] In some embodiments of the present invention, the dynamic warning information generation sub-module of the above-mentioned dynamic warning generation module triggers the warning mechanism through setting dynamic thresholds and gradient change detection algorithms according to the fusion data, the disaster evolution model, and the historical disaster big data analysis results, generates accurate warning information of different levels and different time periods, covering the expected impact range, intensity change, and possible secondary disasters of the disaster, and uses visualization technology to display it, and can customize the warning push method according to user needs and permissions.

[0073] In some embodiments of the present invention, the resource database and geographic information system integration sub-module of the above-mentioned precise resource scheduling module includes a geographic information system (GIS) that stores information such as terrain, traffic network, and population distribution, and an emergency resource distribution database that stores information such as the location, quantity, and performance of rescue teams, rescue equipment, and supplies. And it uses spatial indexing technology to optimize the data query and retrieval efficiency, matches the emergency resource location information with the GIS map through geocoding technology, and establishes a data update mechanism to ensure data accuracy and consistency.

[0074] In some embodiments of the present invention, the optimal resource allocation plan planning sub-module of the precise resource scheduling module uses an improved Dijkstra algorithm for path planning, considers traffic conditions and weather conditions, determines the resource allocation plan based on a linear programming-based resource allocation model, uses the simulated annealing algorithm to optimize the resource allocation plan, and establishes a real-time monitoring mechanism to track the resource transportation and allocation process,

[0075] The improved Dijkstra algorithm is as follows:

[0076] Let d[v] represent the current shortest distance estimate from the source node s to the node v. Initially, d[s]=0, and for v≠s, d[v]=∞,

[0077] Let S be the set of nodes for which the shortest path has been determined. Initially

[0078] The core steps of the algorithm are as follows:

[0079] When s ≠ V: Select a node u from V - S such that d[u] = min{d[v]: v ∈ V - S}, add u to S, and for the nodes adjacent to u (i.e., (u, v) ∈ E), update d[v]: d[v] = min(d[v], d[u] + w(u, v)).

[0080] When considering traffic conditions and weather conditions, the weight w(u, v) of the edge needs to be adjusted. Assume that the traffic condition is represented by t(u, v) and the weather condition is represented by w c (u, v). Then the new edge weight W(u, v) = w(u, v) × t(u, v) × w c (u, v).

[0081] The resource allocation model based on linear programming is as follows:

[0082] Let R be the set of resource types, L be the set of demand locations, S be the set of supply locations. Let x r，s，l represent the quantity of resource r allocated from supply location s to demand location l. The objective function: Assume the goal of resource allocation is to minimize the total cost. Let c r，s，l be the cost of allocating a unit of resource r from supply location s to demand location l. Then the objective function is

[0083] Constraints: Supply constraint: The total amount of resources provided by each supply location s cannot exceed its available quantity. Let α r，s be the quantity of resource r owned by supply location s. Then for all r ∈ R and s ∈ S;

[0084] Demand constraint: The total amount of resources obtained by each demand location l should meet its demand. Let b r，l be the demand for resource r at demand location l. Then for all r ∈ R and l ∈ L;

[0085] Non - negativity constraint: x r，s，l ≥ 0, for all r ∈ R, s ∈ S and l ∈ L.

[0086] Furthermore, the emergency management disaster dynamic early - warning scheduling system includes: a multi - source data collection module, a multi - source data integration module, a dynamic early - warning generation module, and a precise resource scheduling module.

[0087] The above - mentioned multi - source data collection module includes:

[0088] The physical sensor data acquisition sub-module constructs a multi-channel physical sensor data acquisition network that includes meteorological satellites, ground monitoring stations, and Internet of Things sensors. The meteorological satellite continuously sends large-scale meteorological data, such as atmospheric temperature, humidity, cloud movement, air pressure, etc., to the ground receiving station according to the predetermined orbit and time interval. Ground monitoring stations are deployed with various high-precision sensors at different geographical locations. For example, rain gauges accurately measure rainfall, anemometers and wind vanes monitor wind speed and direction in real time, seismographs and seismometers keenly capture crustal movement information, etc. These monitoring stations transmit the collected data to the data center at a fixed frequency through wired or wireless communication methods. The Internet of Things sensor network is widely distributed in areas where disasters may occur. Water level sensors are installed in rivers, lakes, reservoirs, and coastal areas to be able to monitor the subtle changes in water level in real time; seismic wave monitors are deployed in seismic active zones and surrounding areas to accurately detect the propagation and intensity of seismic waves; smoke sensors, temperature sensors, and combustible gas sensors are set in areas prone to fires such as forests and factories to timely detect fire hazards.

[0089] The network data acquisition sub-module uses web crawler technology to collect text, pictures, videos, etc. related to disasters from channels such as social media platforms, news websites, and government announcements through the network data acquisition interface. These network data will be transmitted to the data center through network protocols to provide more dimensional data support for the comprehensive assessment of disasters.

[0090] The above multi-source data integration module includes:

[0091] The data preprocessing sub-module, after the data received from the multi-source data acquisition module arrives at the data center, first enters the data preprocessing process. For data from different sensors and network channels, standardization conversion is carried out according to their data types and formats. For example, convert the binary data format of meteorological satellites to the common CSV format, convert the analog signals collected by sensors to digital signals, and unify the timestamp format of the data. Then, use data cleaning algorithms to remove noise data, outliers, and duplicate data. For the noise in time series data, use the moving average method or wavelet transform for smoothing; for outliers that deviate significantly from the normal range, identify and correct or delete them according to the statistical characteristics and physical meaning of the data. Next, deal with missing data. For data with a certain time pattern, use linear interpolation or spline interpolation to fill it; for data without an obvious pattern, make a reasonable estimate according to the context and relevant factors of the data or mark it as a missing state.

[0092] Data annotation sub-module: The preprocessed data enters the data annotation sub-module. Based on the disaster type classification system and Geographic Information System (GIS) data, labels are added to the data. For example, for water level data, label the name of the river or water area it belongs to, the longitude and latitude information of the monitoring location, and the time label corresponding to the data; for seismic wave data, label the approximate area where the earthquake occurred, the magnitude range (preliminary estimate), and the monitoring time, etc. At the same time, combining historical disaster data and expert knowledge, key labelings are carried out on data features that may be related to the occurrence of disasters. For example, in rainy weather, data with rainfall exceeding a certain threshold and a long duration are labeled as high-risk flood warning-related data. Through data annotation, the identifiability and usability of the data are improved, facilitating subsequent data fusion and analysis.

[0093] Data fusion processing sub-module: The Kalman filter-based data fusion algorithm is used to fuse and predict the collected time series data (such as meteorological data continuously monitored by meteorological satellites, water level change data of water level sensors, etc.), improving the accuracy and reliability of the data. For spatially distributed data (such as the geographical distribution data of ground monitoring stations), Kriging interpolation method is used for data fusion and spatial interpolation to fill in the data blank areas and optimize the data distribution. The Principal Component Analysis (PCA) algorithm is used to reduce the dimension of high-dimensional heterogeneous data, extract key information, and reduce the complexity of data processing. At the same time, a rule-based reasoning algorithm is used to conduct consistency checks and deviation calibrations on data from different sources but with similar monitoring purposes to ensure data consistency. For example, for temperature data of the same area measured by different meteorological stations, data calibration is carried out by setting rules such as temperature difference thresholds and geographical distance weights.

[0094] The above dynamic warning generation module includes:

[0095] Disaster evolution modeling sub-module: Based on physical process modeling and machine learning algorithms (such as Long Short-Term Memory network LSTM for processing time series data, Convolutional Neural Network CNN for analyzing spatial data), a disaster evolution model covering various disasters is constructed. For flood disasters, combining the hydrodynamic equation and the LSTM network, considering the impact of factors such as rainfall, topography, and river hydrological characteristics on flood evolution; for earthquake disasters, using the elastic wave propagation theory and the CNN model, combining geological structures, seismic wave propagation characteristics, etc. to establish corresponding disaster evolution models. During the model construction process, a large amount of historical disaster case data and simulation data are collected to train and optimize the model, continuously adjusting the parameters and structure of the model to improve the prediction accuracy and generalization ability of the model;

[0096] At the same time, the system integrates a large amount of historical disaster case big data, uses cluster analysis algorithms to classify and extract features, and explores the commonalities and characteristics of the development of different types of disasters as an important reference for the model. The association rule mining algorithm is used to analyze the correlation between disaster factors and further improve the disaster evolution model. For example, through analysis, it is found that the probability of fires in surrounding areas will increase within a period of time after an earthquake. This correlation is incorporated into the earthquake disaster evolution model so that the secondary disasters that may be caused can be considered at the same time when warning;

[0097] The dynamic warning information generation submodule can track the development trend of disasters in real time based on the fused data, combined with the above-mentioned disaster evolution model and the historical disaster big data analysis results. The parameters of the disaster evolution model are updated and corrected using real-time data so that it can accurately reflect the real-time status of the disaster. By setting dynamic thresholds and gradient change detection algorithms, when disaster indicators (such as the rate of flood level rise, earthquake magnitude changes, etc.) exceed the threshold or show abnormal changes, the warning mechanism is triggered.

[0098] According to the real-time disaster status, the system dynamically generates accurate early warning information of different levels (such as general, severe, serious, etc.) and different time periods. The early warning information not only includes the expected impact range and intensity changes of the disaster, but also predicts in detail the secondary disasters that may be caused. For example, for typhoon disasters, the typhoon path, intensity changes, and possible secondary disasters such as heavy rain, floods, and landslides will be predicted; for fires, the possible expansion of the affected area will be predicted based on the speed of fire spread and wind direction. Visualization technology is used to display early warning information to users in intuitive forms such as maps and charts to improve the readability and comprehensibility of the information. In the process of generating early warning information, different early warning push methods will be customized according to user needs and permissions, such as pushing detailed professional reports to government emergency management departments and sending concise and clear text message notifications to the public.

[0099] The above-mentioned precise resource scheduling module includes:

[0100] Resource database and geographic information system integration submodule, the system includes geographic information system (GIS) and preset emergency resource distribution database. GIS system stores detailed geospatial information, including topography, transportation network, population distribution, etc., and continuously updates and improves geographic information through satellite remote sensing images and geographic mapping data. Emergency resource distribution database covers detailed information of various emergency resources, such as the location, number, professional capabilities of rescue teams, and the storage location, quantity, performance of rescue equipment and materials. Spatial indexing technology (such as R-tree index) is used to optimize GIS data and resource database to improve data query and retrieval efficiency.

[0101] The two are interrelated, providing comprehensive information support for resource scheduling. By using geocoding technology to match the location information of emergency resources with the GIS map, the visual management of resources is realized. In the system initialization stage, data calibration and verification are carried out on the resource database and the GIS system to ensure the accuracy and consistency of the data. At the same time, a data update mechanism is established. When new resources are added or the resource status changes, the database information is updated in a timely manner to ensure the effectiveness of resource scheduling.

[0102] The optimal resource allocation plan planning sub-module, after receiving the early warning information, the system uses the improved Dijkstra algorithm for path planning based on the location, scope and intensity of the disaster, combined with the GIS and emergency resource database information, to find the optimal transportation route from the emergency resource storage point to the disaster area. During the path planning process, real-time factors such as traffic conditions and weather conditions affecting transportation are considered. For example, in heavy rain weather, waterlogged road sections are marked and the path planning strategy is adjusted. For the resource allocation problem, a resource allocation model based on linear programming is adopted, with the objective function of minimizing rescue costs and maximizing rescue effects, considering constraints such as the balance between resource supply and demand and transportation capacity, to determine the allocation plan of various rescue resources.

[0103] The system will accurately match the rescue forces and materials to the disaster area to ensure that resources can reach the most needed places at the fastest speed and the shortest path. For example, for an earthquake disaster in the mountains, the system will give priority to deploying nearby mountain rescue teams and helicopters and other resources suitable for mountain rescue; for urban waterlogging, pumping equipment and sandbags and other materials with sufficient drainage capacity will be arranged at the nearest location. The simulated annealing algorithm is used to optimize and adjust the resource allocation plan to improve the feasibility and effectiveness of the plan. During the resource allocation process, a real-time monitoring mechanism is established to track and feedback the transportation and allocation process of resources, and problems that may occur, such as transportation vehicle failures and material shortages, are processed in a timely manner to ensure the smooth progress of the rescue operation.

[0104] Secondly, the present invention provides the working steps of the emergency management disaster dynamic early warning and scheduling system with multi-source data fusion, and the steps include: multi-source data collection step, multi-source data integration step, dynamic early warning generation step and precise resource scheduling step.

[0105] The above multi-source data collection step includes:

[0106] After the system is started, the physical sensor data collection sub-module and the network data collection sub-module are started simultaneously.

[0107] In the physical sensor data collection sub-module, meteorological satellites, ground monitoring stations, and Internet of Things sensors start to work, collect data at a predetermined frequency and transmit it to the data center.

[0108] The network data collection sub-module collects data from network channels such as social media through web crawler technology and transmits it to the data center.

[0109] The above multi-source data integration steps include:

[0110] The data received from the multi-source data collection module enters the data preprocessing sub-module. First, it performs a standardized conversion of the data format to make it conform to a unified data format specification. Then, corresponding data cleaning algorithms are used for different types of data to remove noise, outliers, and duplicate data. Next, the missing data is processed, and appropriate interpolation methods or estimation methods are selected according to the data characteristics for filling or marking.

[0111] The preprocessed data enters the data annotation sub-module, and based on the disaster type classification system and GIS data, labels are added to the data to annotate information such as the source, location, time, and related disaster characteristics of the data.

[0112] The annotated data enters the data fusion and processing sub-module. The Kalman filter algorithm is applied to fuse and predict time series data, and the Kriging interpolation method is used for spatial distribution data for fusion and interpolation. Then, the principal component analysis algorithm is used to reduce the dimension of high-dimensional heterogeneous data and extract key information. Finally, the rule-based reasoning algorithm performs consistency checking and deviation calibration on data with similar monitoring purposes, completes the data fusion and integration process, generates disaster situation awareness information, and stores it in the database for subsequent module use.

[0113] The above dynamic warning generation steps include:

[0114] The dynamic warning generation module reads the fused disaster situation awareness data from the database and inputs it into a pre-constructed disaster evolution model. The model calculates and analyzes the data, and at the same time combines the results of cluster analysis and association rule mining of big data of historical disaster cases to track the development trend of disasters in real time.

[0115] The system continuously monitors the changes in disaster indicators through the set dynamic threshold and gradient change detection algorithm. When the disaster indicators exceed the threshold or show abnormal changes, the warning mechanism is triggered. The system generates accurate warning information of different levels and different time periods according to the current disaster status and model prediction results, including the expected impact range of the disaster, intensity changes, and possible secondary disasters, etc., and publishes the warning information to relevant departments and the public through multiple communication channels (such as text messages, broadcasts, network platforms, etc.). When publishing the warning information, customized push is performed according to the needs and permissions of users.

[0116] The above precise resource scheduling steps include:

[0117] When receiving the early warning information, the precise resource scheduling module immediately extracts relevant information from the GIS system and the emergency resource distribution database, including the location, scope, intensity of the disaster, and the location, quantity, performance, etc. of the emergency resources.

[0118] Use the improved Dijkstra algorithm for path planning, calculate the optimal transportation route from the emergency resource storage point to the disaster area, and consider factors such as real-time traffic conditions and weather conditions during the planning process. At the same time, based on the resource allocation model of linear programming, considering constraints such as resource supply-demand balance and transportation capacity, determine the allocation plan of various rescue resources. Use the simulated annealing algorithm to optimize and adjust the resource allocation plan to ensure the feasibility and effectiveness of the plan.

[0119] Organize and coordinate the transportation and allocation of emergency resources according to the allocation plan, track the transportation status of resources in real time, establish a monitoring mechanism, and promptly handle problems that occur during the transportation process to ensure that resources can reach the disaster area in a timely and accurate manner.

[0120] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0121] 1. The multi-source data acquisition module of the present invention, which includes a physical sensor data acquisition sub-module and a network data acquisition sub-module, can comprehensively cover various data sources, providing rich information for disaster assessment from different dimensions. The physical sensor data acquisition sub-module can collect data from meteorological satellites, ground monitoring stations, and Internet of Things sensors in real time and accurately. These data cover key information in multiple aspects such as the atmosphere, geology, hydrology, and fire. The network data acquisition sub-module can obtain text, image, and video information related to disasters from channels such as social media and news websites, realizing all-round information monitoring of disasters and avoiding decision-making mistakes caused by a single information source.

[0122] 2. The present invention can utilize the large-scale meteorological data provided by meteorological satellites, which helps to detect the potential risks of meteorological disasters in advance; various sensors of ground monitoring stations can finely monitor local environmental parameters and provide accurate data for disaster assessment; water level sensors, seismic wave monitors, etc. in Internet of Things sensors are specifically targeted at different disasters, can conduct targeted monitoring of disasters, improve the perception ability of various disasters, provide sufficient basic data for subsequent analysis and decision-making, and can more accurately grasp the early signs of disasters.

[0123] 3. The present invention cleans and processes missing data for data through algorithms such as the moving average method, wavelet transform, linear interpolation method, and spline interpolation method, which can effectively improve data quality, remove noise, outliers, and duplicate data, fill in missing data, ensure the accuracy of subsequent data processing and analysis, avoid model prediction deviations and decision-making errors caused by data problems, and lay a foundation for accurate disaster assessment and early warning.

[0124] 4. The present invention adds labels to data based on the disaster type classification system and GIS data, which can clarify the source, location, time, and disaster characteristics of the data, making the data more identifiable and usable, facilitating subsequent data management and utilization, improving data processing efficiency, facilitating the transfer and operation of data in different systems and analysis processes, and providing a clearer guidance for subsequent fusion and analysis.

[0125] 5. The data fusion processing sub-module of the multi-source data integration module of the present invention uses Kalman filtering to fuse and predict time series data, ensuring the accuracy and reliability of time series data; Kriging interpolation method for the fusion and interpolation of spatially distributed data can optimize the distribution of spatial data and fill in data gaps; principal component analysis algorithm for dimensionality reduction of high-dimensional heterogeneous data to reduce data processing complexity; rule-based reasoning algorithm for consistency checking and deviation calibration to ensure the consistency of similar data from different sources, thereby forming more accurate and comprehensive disaster situation awareness information and providing better data support for disaster analysis.

[0126] 6. The disaster evolution modeling sub-module of the present invention constructs a disaster evolution model based on physical process modeling and machine learning algorithms (such as LSTM and CNN), and uses a large amount of historical disaster case data and simulation data for training and optimization, which can more accurately simulate the development process of disasters, considering time series and spatial data characteristics, improving the prediction accuracy of disaster evolution, and can predict the development trend of disasters in advance, providing a scientific basis for deploying rescue and prevention measures in advance.

[0127] 7. The dynamic warning information generation sub-module of the dynamic warning generation module of the present invention triggers the warning mechanism by setting dynamic thresholds and gradient change detection algorithms, and can dynamically generate accurate warning information of different levels and different time periods according to real-time data, covering the expected impact range, intensity change, and possible secondary disasters of the disaster, using visualization technology to display and customize the push method, realizing dynamic and accurate warning, enabling different users to receive appropriate information according to their own needs, and helping to take timely response measures to reduce disaster losses.

[0128] 8. In the present invention, the topographic and geomorphic information, transportation network information, and population distribution information stored in GIS are correlated with the resource information stored in the emergency resource distribution database. The spatial indexing technology is used to optimize the data query and retrieval efficiency, and the geographical coding technology is used to match the resource location information, ensuring the accuracy and consistency of the data, providing a comprehensive and accurate information basis for resource allocation, and facilitating the rapid allocation of resources and decision-making.

[0129] 9. The present invention uses an improved Dijkstra algorithm for path planning and takes into account real-time traffic conditions and weather conditions, which can plan the optimal transportation route; a resource allocation model based on linear programming is used to determine the resource allocation plan, and the simulated annealing algorithm is used to optimize the resource allocation. At the same time, a real-time monitoring mechanism is established to ensure that resources can reach the disaster area at the fastest speed and the shortest path, realizing the efficient allocation of resources, improving the efficiency and effect of emergency rescue, and minimizing the damage to people and property caused by disasters to the greatest extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0130] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0131] Figure 1 is the principle block diagram of the present invention;

[0132] Figure 2 is the flowchart of the working steps of the emergency management disaster dynamic early warning and dispatching system of the present invention;

[0133] Figure 3 is the detailed flowchart of the working steps of the emergency management disaster dynamic early warning and dispatching system of the present invention;

[0134] Figure 4 is the structural block diagram of an electronic device provided by an embodiment of the present invention.

[0135] Icons: 101 - Memory; 102 - Processor; 103 - Communication Interface. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0136] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Usually, the components of the embodiments of the present application described and illustrated in the drawings here can be arranged and designed in various different configurations.

[0137] Accordingly, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed present application, but merely represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the scope of protection of the present application.

[0138] The following will describe in detail some embodiments of the present application in conjunction with the accompanying drawings. Without conflict, the following various embodiments and the various features in the embodiments can be combined with each other.

[0139] Embodiment 1

[0140] Please refer to Figures 1-3 , Figure 1 , which is a block diagram of the principle of the present invention; Figure 2 , which is a flowchart of the working steps of the emergency management disaster dynamic early warning and dispatching system in the present invention; Figure 3 , which is a detailed flowchart of the working steps of the emergency management disaster dynamic early warning and dispatching system in the present invention.

[0141] This embodiment provides an emergency management disaster dynamic early warning and dispatching system based on multi-source data fusion, including: a multi-source data acquisition module, which includes a physical sensor acquisition sub-module and a network acquisition sub-module, and is used to collect original disaster-related data D; a multi-source data integration module, which is used to preprocess, label, and fuse the original data D collected by the multi-source data acquisition module to generate disaster situation perception information I as a basis for early warning; a dynamic early warning generation module, which is used to generate early warning information W from the disaster situation perception information I obtained by the multi-source data integration module; and a precise resource dispatching module, which is used to process the early warning information W generated by the dynamic early warning generation module and output resource allocation information S.

[0142] Multi-source data acquisition module: The multi-source data acquisition module is the basis of the emergency management disaster dynamic early warning and dispatching system, collecting disaster-related data from multiple channels. The physical sensor acquisition sub-module uses professional equipment, such as seismic sensors to capture crustal vibrations and meteorological sensors to monitor meteorological elements such as wind speed and rainfall, providing physical parameters for disaster monitoring. The network acquisition sub-module uses the Internet to obtain data from social media and government platforms, including disaster scene information shared by the public and disaster statistics and emergency plans released by the official, and summarizes the two types of data D (hereinafter simply referred to as D) to provide raw materials for subsequent processing.

[0143] Multi-source Data Integration Module: This module receives the collected raw data, performs preprocessing, annotation, and fusion to generate disaster situation awareness information. In preprocessing, algorithms such as moving average and wavelet transform are used to remove noise and fill in missing values to improve data quality. For annotation, according to the classification standard, labels such as disaster type, time, and location are added to enhance data analyzability. In the fusion stage, Kalman filtering is used to process time series data to mine time correlations; Kriging interpolation is used to fuse spatial data to grasp the spatial distribution of disasters. Finally, I (abbreviated as I hereinafter) is generated, providing a basis for early warning.

[0144] Dynamic Early Warning Generation Module: The dynamic early warning generation module is one of the cores of the system. Based on the fused situation awareness information I and the historical disaster analysis result A (abbreviated as A hereinafter), it generates accurate early warning information W (abbreviated as W hereinafter) and constructs a disaster evolution model M (·) (abbreviated as M hereinafter (·) ) When using LSTM to process time series data, its gating mechanism is used to capture long-term dependencies and model the development trend of disasters; CNN is used to analyze spatial data, and key features are extracted through convolutional, pooling, and fully connected layers. The model is optimized through a large amount of data training, and W is output.

[0145] Precise Resource Scheduling Module: The precise resource scheduling module plans the emergency resource allocation plan S (abbreviated as S hereinafter) according to the early warning information W, in combination with the GIS data G and the emergency resource database information R (abbreviated as R hereinafter). The path planning uses an improved Dijkstra algorithm, considering traffic conditions and weather to dynamically adjust the path weights and find the optimal path. Based on the resource allocation model of linear programming, objective functions such as minimizing transportation costs are set, combined with constraints such as supply, demand, and non-negativity, to reasonably allocate resources. The simulated annealing algorithm is introduced to accept new solutions according to probability and iteratively optimize to obtain the specific model formula of S, improving the efficiency of resource allocation.

[0146] In some embodiments of the present invention, the raw data D of the above multi-source data acquisition module is specifically:

[0147] D = D s + D n

[0148] In the formula, the multi-source data acquisition module is the basis of the system. The physical sensor acquisition sub-module collects sensor data such as seismic and meteorological sensors (D s ), such as seismic sensors capturing crustal vibrations and meteorological sensors monitoring meteorological elements; the network acquisition sub-module obtains data (D n ) from social media and government platforms, such as disaster information shared by the public and official emergency plans. The two parts of data are summarized to obtain D, providing the raw data for subsequent processing.

[0149] In some embodiments of the present invention, the disaster situation awareness information I of the above multi-source data integration module is specifically as follows:

[0150] I = F(L(P(D)))

[0151] In the formula, this module undertakes the acquisition of data, and processes it through three steps: preprocessing, annotation, and fusion. In the preprocessing step, algorithms such as moving average and interpolation are used to clean and fill the data; in the annotation link, according to the classification standard, labels such as disaster type, time, and location are added; in the fusion step, Kalman filtering is used to process time series data, and Kriging interpolation is used to process spatial distribution data, and finally the disaster situation awareness information is generated.

[0152] In some embodiments of the present invention, the warning information W of the above dynamic warning generation module is specifically as follows:

[0153] W = M(I, A)

[0154] In the formula, this module is one of the cores of the system. According to the fused disaster situation awareness information I and the historical disaster analysis result A, the warning information W is generated. When constructing the disaster evolution model M(I, A), LSTM is used to process time series data to capture long-term dependencies; CNN is used to analyze spatial data to extract key features, and the model is trained and optimized with a large amount of data to obtain accurate warning information W.

[0155] In some embodiments of the present invention, the resource allocation information S of the above precise resource scheduling module is specifically as follows:

[0156] S = 0(W, G, R)

[0157] In the formula, according to the warning information W, combined with the GIS data G and the emergency resource database information R, the emergency resource allocation plan S is planned. In the path planning, the improved Dijkstra algorithm is used, and the weights are adjusted considering traffic and weather to find the optimal path; based on the linear programming model, the objective function is set and combined with constraints such as supply, demand, and non-negativity to allocate resources; the simulated annealing algorithm is used to optimize the plan, that is, S.

[0158] The above is specifically as follows:

[0159]

[0160] Among them, the system collects data D from multiple sources, generates I through integration, then generates W by the warning module, and finally generates S by the resource scheduling module. Each module cooperates to provide technical support for emergency management, reduce disaster losses, and ensure the safety of life and property.

[0161] In some embodiments of the present invention, the above physical sensor data acquisition sub-module includes meteorological satellites, ground monitoring stations, and Internet of Things sensors. Among them, meteorological satellites can provide information such as atmospheric temperature, humidity, cloud movement, and air pressure. Ground monitoring stations can collect data through sensors such as rain gauges, anemometers, wind vanes, seismographs, and seismometers. Internet of Things sensors include water level sensors, seismic wave monitors, smoke sensors, temperature sensors, and combustible gas sensors, etc., which are used to monitor different disaster-related data.

[0162] In some embodiments of the present invention, the above network data acquisition sub-module uses web crawler technology to collect text, pictures, videos, and other information related to disasters from channels such as social media platforms, news websites, and government announcements.

[0163] In some embodiments of the present invention, the data preprocessing sub-module of the above multi-source data integration module uses algorithms such as moving average method, wavelet transform, linear interpolation method, and spline interpolation method to clean and process missing data of the data, so as to achieve standardized conversion of data format, noise removal, outlier processing, and filling or marking of missing data.

[0164] The moving average method is as follows:

[0165]

[0166] Among them, the moving average method is used to smooth data and remove noise. Let the original data sequence be x1, x2,..., x n , the window size is k, and the moving average value is y i .

[0167] The wavelet transform is as follows:

[0168]

[0169]

[0170] Among them, the continuous wavelet transform (CWT) x(t) convolves the signal with a wavelet function ψ(t). For the signal x(t), then W x (a, b) is the above mathematical relationship. Among them, a is the scale parameter, which controls the stretching of the wavelet function; b is the translation parameter, which controls the translation of the wavelet function; ψ * represents the complex conjugate of ψ. The discrete wavelet transform (DWT) usually uses binary wavelets to decompose the signal into an approximation component and a detail component. Let the signal be x[n]. During the decomposition process, the approximation component cA j [n] and the detail component cD j [n] can be calculated through the above convolution formula. Among them, h[n] and g[n] are the coefficients of the low-pass filter and the high-pass filter respectively, and j represents the number of decomposition layers.

[0171] The linear interpolation method is as follows:

[0172]

[0173] In the formula, given two data points (x1, y1) and (x2, y2), for any x within the interval, the corresponding interpolation result can be calculated through the linear interpolation formula.

[0174] The spline interpolation method is as follows:

[0175] S(x) = a i + b i (x - x i ) + c i (x - x i ) 2 + d i (x - x i ) 3

[0176] Among them, taking cubic spline interpolation as an example, assuming that the known data points are (x i , y i ), i = 0, 1,..., n, on each sub-interval [x i , x i+1 , the cubic spline function can be expressed as the above formula.

[0177] In some embodiments of the present invention, the data annotation sub-module of the above multi-source data integration module adds labels to the data based on the disaster type classification system and Geographic Information System (GIS) data, and the annotation information includes data source, location, time, and related disaster characteristics, etc.

[0178] In some embodiments of the present invention, the data fusion processing sub-module of the above multi-source data integration module uses a data fusion algorithm based on Kalman filtering to fuse and predict time series data, uses Kriging interpolation method to fuse and spatially interpolate spatially distributed data, uses Principal Component Analysis (PCA) algorithm to reduce the dimension of high-dimensional heterogeneous data, and uses a rule-based reasoning algorithm for consistency checking and deviation calibration.

[0179] The data fusion algorithm based on Kalman filtering is as follows:

[0180] State equation: X k = A k X k- + B k U k + W k

[0181] Observation equation: Z k = H k X k + VK

[0182] Prediction state:

[0183] Prediction covariance:

[0184] Kalman gain:

[0185] where X k is the state vector at time k, A k is the state transition matrix, U k is the control vector, B k is the control matrix, W k is the process noise, and W k ~ N(0, Q k ), N(0, Q k ) represents a Gaussian distribution with mean 0 and covariance Q k . The observation equation where Z k is the observation vector at time k, H k is the observation matrix, V K is the observation noise, V K ~ N(0, R k ), where is the prediction of the state at time k based on the state at time k - 1, is the updated state estimate at time k, P k|k-1 and P k|k are the corresponding covariance matrices respectively, and I is the identity matrix.

[0186] The Kriging interpolation method is as follows:

[0187]

[0188] where let Z(x) be a regionalized variable, with observed values Z(x i (1, 2…, n)) at positions x i . To estimate the value at position x0 where γ(x i , x j ) is the semivariogram, which describes the degree of variation of the regionalized variable Z(x) between positions x i and x j , and μ is the Lagrange multiplier.

[0189] Principal Component Analysis (PCA) algorithm:

[0190]

[0191] C = VΛV T

[0192]

[0193] Suppose there are m samples, and each sample has n-dimensional features, forming a data matrix X ∈ R m×n , first centralize the data to obtain Then calculate the covariance matrix C, and perform eigenvalue decomposition on the covariance matrix C to obtain C, where Λ is a diagonal matrix, and the diagonal elements are eigenvalues λ1 ≥ λ2 ≥ λ3 ≥ … ≥ λ n , and V is the eigenvector matrix. Select the eigenvectors corresponding to the first k largest eigenvalues, and project the original data onto these k eigenvectors to obtain the dimensionality-reduced data Y.

[0194] In some embodiments of the present invention, the disaster evolution modeling sub-module of the above dynamic warning generation module constructs a disaster evolution model based on physical process modeling and machine learning algorithms, including a long short-term memory network (LSTM) for processing time series data, a convolutional neural network (CNN) for analyzing spatial data, and uses a large amount of historical disaster case data and simulation data to train and optimize the model.

[0195] The long short-term memory network (LSTM) is as follows:

[0196] Forget gate: f t = σ(W f [h t-1 , x t +b f )

[0197]

[0198] Input gate: i t = σ(W i [h t-1 , x t +b i )

[0199]

[0200] Output gate: o t = σ(W o [h t-1 , x t +)

[0201] h t = o t ⊙tanh(C t )

[0202] Among them, the forget gate determines how much information of the previous cell state needs to be forgotten, f tis the output of the forget gate, which is a vector with values in the range [0, 1]. Each element represents the degree to which the corresponding information is forgotten. σ is the sigmoid activation function, and W f is the weight matrix of the forget gate, and b f is the bias vector of the forget gate. [h t-1 , x t means concatenating the hidden state h t-1 at the previous time step and the input x t at the current time step; in the input gate, i t is the output of the sigmoid part of the input gate, W i is the weight matrix of the sigmoid part of the input gate, and b i is the bias vector. is the candidate cell state, tanh is the hyperbolic tangent activation function, and w c is the weight matrix of the candidate cell state, and b c is the bias vector. The output gate determines the cell state at the current time step; the output gate determines the cell state C t of how much information needs to be output to the hidden state h t , where o t is the output of the sigmoid part of the output gate, W o is the weight matrix of the output gate, and b0 is the bias vector.

[0203] The convolutional neural network (CNN) is as follows:

[0204] Convolution operation:

[0205] Activation function: y i，j，c = max(0, Y i，j，c )

[0206] Loss function

[0207] Optimization algorithm:

[0208] Among them, at the position (i, j), the calculation of the c-th channel of the output feature map Y is Y i，j，c , b c is the bias of the c-th channel. The activation function uses the ReLu function y i，j，c . For the disaster evolution model, the commonly used loss function can be the mean squared error loss (MSE), where θ is the parameter of the model, is the output of the model for the input x (i)The predicted output, and the optimization algorithm uses stochastic gradient descent, where α is the learning rate. In actual training, mini-batch stochastic gradient descent (Mini-Batch SGD) is usually adopted, and a small batch of data is selected from the training dataset each time to calculate the gradient and update the parameters.

[0209] In some embodiments of the present invention, the dynamic warning information generation sub-module of the above-mentioned dynamic warning generation module triggers the warning mechanism through setting dynamic thresholds and gradient change detection algorithms according to the fusion data, the disaster evolution model, and the historical disaster big data analysis results, generates accurate warning information of different levels and different time periods, covering the expected impact range of the disaster, intensity changes, and possible secondary disasters, and uses visualization technology to display it, and can customize the warning push method according to user needs and permissions.

[0210] In some embodiments of the present invention, the resource database and the geographic information system integration sub-module of the above-mentioned precise resource scheduling module include a geographic information system (GIS) that stores information such as terrain and landforms, transportation networks, and population distributions, and an emergency resource distribution database that stores information such as the locations, quantities, and performances of rescue teams, rescue equipment, and materials. And the spatial index technology is used to optimize the data query and retrieval efficiency, the emergency resource location information is matched with the GIS map through geocoding technology, and a data update mechanism is established to ensure the accuracy and consistency of the data.

[0211] In some embodiments of the present invention, the optimal resource allocation plan planning sub-module of the precise resource scheduling module uses an improved Dijkstra algorithm for path planning, considers traffic conditions and weather conditions, determines the resource allocation plan based on a linear programming-based resource allocation model, optimizes the resource allocation plan using the simulated annealing algorithm, and establishes a real-time monitoring mechanism to track the resource transportation and allocation process.

[0212] The improved Dijkstra algorithm is as follows:

[0213] Let d[v] represent the current shortest distance estimate from the source node s to the node v. Initially, d[s]=0, and for v≠s, d[v]=∞.

[0214] Let S be the set of nodes for which the shortest path has been determined. Initially

[0215] The core steps of the algorithm are as follows:

[0216] When s≠V: Select the node u from V-S such that d[u]=min{d[v]: v∈V-S}, add u to S, and for the nodes adjacent to u (i.e., (u, v)∈E), update d[v]: d[v]=min(d[v], d[u]+w(u, v)).

[0217] When considering traffic conditions and weather conditions, the weight w(u, v) of an edge needs to be adjusted. Assume that the traffic condition is represented by t(u, v) and the weather condition is represented by w c (u, v). Then the new edge weight W(u, v) = w(u, v) × t(u, v) × w c (u, v).

[0218] The resource allocation model based on linear programming is as follows:

[0219] Let R be the set of resource types, L be the set of demand locations, S be the set of supply locations. Let x r，s，l represent the quantity of resource r allocated from supply location s to demand location l. The objective function: Assume the goal of resource allocation is to minimize the total cost. Let c r，s，l be the cost of allocating a unit of resource r from supply location s to demand location l. Then the objective function is

[0220] Constraint conditions: Supply constraint: The total amount of resources provided by each supply location s cannot exceed its available amount. Let α r，s be the quantity of resource r owned by supply location s. Then for all r ∈ R and s ∈ S;

[0221] Demand constraint: The total amount of resources obtained by each demand location l should meet its demand. Let b r，l be the demand for resource r at demand location l. Then for all r ∈ R and l ∈ L;

[0222] Non - negativity constraint: x r，s，l ≥ 0, for all r ∈ R, s ∈ S and l ∈ L.

[0223] Embodiment 2

[0224] Based on Embodiment 1, this Embodiment 2 is used for a specific implementation scenario, specifically:

[0225] Flood disaster emergency management, system setup and data collection:

[0226] In the multi - source data collection module, the meteorological satellite of the physical sensor data collection sub - module continuously monitors the atmospheric state, obtaining information such as rainfall, cloud thickness, wind speed, etc.; rain gauges and water level sensors of the ground monitoring stations are deployed around rivers and reservoirs to monitor rainfall and water level changes in real - time; water level sensors in the Internet of Things sensor network are spread throughout key water areas. At the same time, the network data collection sub - module collects news, reports and public feedback about floods from social media platforms and local government websites through web crawler technology.

[0227] The data collection frequency is set to once every 15 minutes.

[0228] Data integration and analysis:

[0229] Data enters the multi-source data integration module. After passing through the data preprocessing sub-module, data in different formats is converted into unified structured data. The noise in rainfall and water level data is removed by the moving average method, and the missing data caused by equipment failures is filled using linear interpolation. The data annotation sub-module annotates the water area and location information for water level and rainfall data based on Geographic Information System (GIS) data. In the data fusion processing sub-module, the Kalman filter is used to fuse the water level and rainfall data in the time series, the Kriging interpolation method is adopted to improve the spatial distribution data, the principal component analysis algorithm is used to reduce the dimensionality of high-dimensional data, and the rule-based reasoning algorithm is used to calibrate the deviations of different monitoring points.

[0230] The obtained fusion data is input into the dynamic warning generation module. In the disaster evolution modeling sub-module, a flood disaster evolution model based on the hydrodynamic equation and Long Short-Term Memory Network (LSTM) simulates the flood evolution according to rainfall, topography, and river hydrological characteristics. Through the dynamic threshold and gradient change detection algorithm, a warning is triggered when the rising speed of the water level exceeds the set threshold.

[0231] Warning and resource scheduling:

[0232] The dynamic warning information generation sub-module generates warning information including the flood inundation range, intensity, and secondary disasters such as possible landslides, and pushes it to relevant departments and the public in the form of maps and text messages.

[0233] The precise resource scheduling module plans the best rescue route according to the GIS and emergency resource database information, arranges the deployment of rescue materials such as sandbags and inflatable boats and rescue personnel from nearby storage points, and optimizes the deployment plan through the simulated annealing algorithm to ensure that resources reach the disaster area within 1 hour, achieving efficient flood control and rescue.

[0234] Comparative example 1

[0235] A flood disaster management system that only relies on meteorological satellite data.

[0236] Only using meteorological satellite data, it monitors rainfall and cloud information, lacking data from ground monitoring stations, Internet of Things sensors, and network information.

[0237] Due to the lack of local detailed information and on-site monitoring, it is impossible to accurately judge the local water level changes of floods and the actual ground conditions, and the prediction of the flood development trend is inaccurate. For example, when floods break out in small watersheds, it is impossible to detect the rapid rise of local water levels in a timely manner, resulting in delayed warnings and untimely resource allocation. Eventually, some residents in the affected areas fail to evacuate in time, the time for rescue resources to reach the affected locations is prolonged, and the losses caused by floods increase by more than 30% compared with the system of the present invention.

[0238] Embodiment 3

[0239] Based on Embodiment 1, this Embodiment 3 is used for specific implementation scenarios, specifically: earthquake disaster emergency management system, setting and data collection:

[0240] The seismograph of the physical sensor data collection sub-module is deployed in the earthquake zone, and the seismograph and ground monitoring stations monitor crustal movements and ground vibrations; the network data collection sub-module collects news reports and social media information related to earthquakes.

[0241] The data collection period is set for real-time transmission, and data is transmitted once abnormal fluctuations are detected.

[0242] Data integration and analysis:

[0243] In the multi-source data integration module, wavelet transform is used to process the noise in the seismic wave data, and abnormal data is corrected according to historical data and physical laws. The data annotation sub-module marks the source area and magnitude range of the seismic wave data, uses the principal component analysis algorithm to process the multi-dimensional crustal movement data, and the rule-based reasoning algorithm ensures the consistency of data from different monitoring stations. The data fusion processing sub-module fuses data such as seismic waves and ground vibrations, and uses a disaster evolution model combining convolutional neural network (CNN) and elastic wave propagation theory to analyze the development of earthquake disasters.

[0244] An alarm is triggered when the magnitude change exceeds the set gradient.

[0245] Alarm and resource scheduling:

[0246] The dynamic alarm information generation sub-module generates alarm information including magnitude, focal depth, aftershock prediction, and possible range of casualties, and sends it to the emergency management department and medical institutions in the form of broadcasts and professional reports.

[0247] The precise resource scheduling module allocates resources such as earthquake rescue teams and life detectors, plans the optimal transportation route according to GIS information and the improved Dijkstra algorithm, considers road conditions and weather impacts, and enables the resources to reach the epicenter area in the shortest time, improving the timeliness and effectiveness of earthquake rescue.

[0248] Comparative Example 2

[0249] Traditional static earthquake disaster early warning system.

[0250] Based only on fixed earthquake monitoring equipment and simple threshold judgment, without combining multi-source data and complex disaster evolution models.

[0251] When an earthquake occurs, it is unable to accurately predict aftershock situations and the impacts on surrounding areas, and cannot adjust the resource allocation plan according to real-time traffic and weather information. In an earthquake disaster, due to the lack of accurate prediction of aftershocks, rescue workers faced risks of secondary disasters during subsequent rescue operations, and the resource allocation did not consider road damage conditions, resulting in low rescue efficiency. Compared with the system of the present invention, the rescue operation was delayed by about 2 hours, increasing the risks of casualties and property losses.

[0252] Example 4

[0253] Based on Example 1, this Example 4 is used for a specific implementation scenario, specifically: fire disaster emergency management, system setup and data collection:

[0254] The temperature sensors and smoke sensors of the physical sensor data collection sub-module are deployed in easily flammable areas such as factories and forests, and the network data collection sub-module collects network information on fire hazards.

[0255] The data collection frequency is once per minute.

[0256] Data integration and analysis:

[0257] The multi-source data integration module performs standardized conversion and outlier processing on the sensor data. The data annotation sub-module marks the location and time information of the temperature and smoke data. The data fusion processing sub-module uses the Kalman filter and rule-based reasoning algorithms to fuse different sensor data. The disaster evolution modeling sub-module predicts the spread of the fire based on the physical model of fire propagation and machine learning algorithms.

[0258] An early warning is triggered when the temperature and smoke indicators reach dangerous thresholds.

[0259] Early warning and resource scheduling:

[0260] The dynamic early warning information generation sub-module generates early warning information on the development, spread direction, and possible affected areas of the fire, and notifies the surrounding residents and the fire department through alarms and electronic displays.

[0261] The precise resource scheduling module calls fire trucks and fire extinguishing equipment, and allocates resources through a linear programming-based resource allocation model to ensure that the fire is extinguished within the shortest time and the fire losses are reduced.

[0262] Comparative Example 3

[0263] Single-source fire monitoring system.

[0264] Only temperature sensors are installed in the factory, and network data and other sensors are not used.

[0265] When a fire occurs, the overall situation of the fire cannot be fully grasped, and the judgment of the fire spread speed and possible affected range is inaccurate. For example, due to the lack of smoke sensor data, the severity and development trend of the fire cannot be judged, resulting in unreasonable resource allocation. The fire spread range is 20% larger than the prediction of the system of the present invention, causing greater property losses, and the secondary disaster assessment of the surrounding environment is insufficient, affecting the effectiveness of the entire emergency management.

[0266] Through the comparison between the above embodiments and the comparative examples, it can be seen that the emergency management disaster dynamic early warning and dispatching system based on multi-source data fusion of the present invention has significant advantages in disaster monitoring, early warning, and resource dispatching, and can provide more comprehensive, accurate, timely, and effective emergency management services for different types of disasters, improving the overall ability to respond to disasters.

[0267] Please refer to Figure 4 , which is a structural block diagram of an electronic device provided by an embodiment of the present invention. An embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by the processor 102, the system described above is implemented. If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in various embodiments of the present application. The foregoing storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical disks, etc., which can store program codes.

[0268] The above is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0269] It will be apparent to those skilled in the art that the present application is not limited to the details of the exemplary embodiments described above, and that the present application can be implemented in other specific forms without departing from the spirit or essential features of the present application. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the present application is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims be included in the present application. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. An emergency management disaster dynamic early warning and dispatching system based on multi-source data fusion, characterized in that, Including: A multi-source data acquisition module, which includes a physical sensor acquisition sub-module and a network acquisition sub-module, and is used to collect original data D related to disasters; A multi-source data integration module, which is used to preprocess, annotate and fuse the original data D collected by the multi-source data acquisition module, generate disaster situation awareness information I, and provide a basis for early warning; A dynamic early warning generation module, which is used to generate early warning information W from the disaster situation awareness information I obtained by the multi-source data integration module; A precise resource scheduling module, which is used to process the early warning information W generated by the dynamic early warning generation module and output resource allocation information S.

2. The emergency management disaster dynamic early warning and dispatching system based on multi-source data fusion according to claim 1, wherein The original data D of the multi-source data acquisition module is specifically: D = D s + D n Where D s is the data collected by the physical sensor acquisition sub-module; D n is the data collected by the network acquisition sub-module.

3. The emergency management disaster dynamic early warning and dispatching system based on multi-source data fusion according to claim 1, characterized in that The disaster situation awareness information I of the multi-source data integration module is specifically: I = F(L(P(D))) In the formula, after three steps of preprocessing, annotation and fusion, the preprocessing uses a moving average and interpolation algorithm to clean and fill data; in the annotation link, according to the classification standard, disaster type, time and location labels are added; for the fusion, the Kalman filter is used to process time series data, and Kriging interpolation is used to process spatial distribution data, and finally the disaster situation awareness information I is generated.

4. The emergency management disaster dynamic early warning and dispatching system based on multi-source data fusion according to claim 1, characterized in that, The early warning information W of the dynamic early warning generation module is specifically: W = M(I, A) In the formula, the early warning information W is generated based on the fused disaster situation awareness information I and the historical disaster analysis result A. When constructing the disaster evolution model M(I, A), LSTM is used to process time series data to capture long-term dependencies; CNN is used to analyze spatial data to extract key features, and the model is optimized through a large amount of data training to obtain the early warning information W.

5. The emergency management disaster dynamic early warning and dispatching system based on multi-source data fusion according to claim 1, wherein, The resource allocation information S of the precise resource scheduling module is specifically: S = O(W, G, R) In the formula, according to the early warning information W, combined with GIS data G and emergency resource database information R, an emergency resource allocation plan S is planned. The path planning uses an improved Dijkstra algorithm, considering traffic and weather to adjust weights to find the optimal path. Based on a linear programming model, the objective function is set and combined with supply, demand and non-negative constraint conditions to allocate resources, and the simulated annealing algorithm is used to optimize the plan to obtain the resource allocation information S.

6. The emergency management disaster dynamic early warning and dispatching system based on multi-source data fusion according to claim 3, characterized in that, The preprocessing of the multi-source data integration module uses a moving average method, wavelet transform, linear interpolation method and spline interpolation method to clean and process missing data, so as to realize the standard conversion of data format, noise removal, outlier processing and filling or marking of missing data. The moving average method is as follows: In the formula, let the original data sequence be \(x_1, x_2, \ldots, x\) n , the window size be \(k\), and the moving average be \(y\) i ; The wavelet transform is as follows: In the formula, a is the scale parameter; b is the translation parameter; ψ * denotes the complex conjugate of ψ; Given the signal x[n], during the decomposition process, the approximation component cA j [n] and the detail component cD j [n] can be calculated by the convolution formula; h[n] and g[n] are the coefficients of the low-pass filter and high-pass filter respectively; j represents the number of decomposition levels; The linear interpolation method is as follows: In the formula, given two data points (x1, y1) and (x2, y2), for any within the interval, its corresponding interpolation result can be calculated through the linear interpolation formula; The spline interpolation method is as follows: S(x) = a i + b i (x - x i ) + c i (x - x i ) 2 + d i (x - x i ) 3 where, taking cubic spline interpolation as an example, it is assumed that the known data points are (x i , y i ), i = 0, 1, …, n. On each sub-interval [x i , x i+1 , the cubic spline function can be expressed as the above formula.

7. The emergency management disaster dynamic early warning and dispatching system based on multi-source data fusion according to claim 3, wherein The fusion of the multi-source data integration module uses a Kalman filter-based data fusion algorithm to fuse and predict time series data, uses Kriging interpolation to fuse and spatially interpolate spatially distributed data, uses the principal component analysis (PCA) algorithm to reduce the dimension of high-dimensional heterogeneous data, and uses a rule-based reasoning algorithm for consistency checking and deviation calibration. The Kalman filter-based data fusion algorithm is as follows: State equation: X k = A k X k- + B k U k + W k Observation equation: Z k = H k X k + V K Prediction status: Predicted covariance: Kalman gain: In the state equation, X k is the state vector at time k, A k is the state transition matrix, U k is the control vector, B k is the control matrix, W k is the process noise, and W k ~ N(0, Q K ), N(0, Q k ) represents a Gaussian distribution with a mean of 0 and a covariance of Q k . In the observation equation, Z k is the observation vector at time k, H k is the observation matrix, V K is the observation noise, V K ~ N(0, R K ), where is the prediction of the state at time k based on the state at time k - 1, is the updated state estimate at time k, P k|k-1 and P k|k are the corresponding covariance matrices respectively, and I is the identity matrix; The Kriging interpolation method is as follows: In the formula, let Z(x) be a regionalized variable, at position x i (1, 2, …, n) there are observed values Z(x i ), and it is necessary to estimate the value at position x0 where γ(x i , x j ) is the semivariogram, which describes the degree of variation of the regionalized variable Z(x) between positions x i and x j , and μ is the Lagrange multiplier; The principal component analysis (PCA) algorithm is: C = V Λ V T where, assuming there are m samples, each sample has n-dimensional features, forming a data matrix X ∈ R m×n , first centralize the data to obtain Then calculate the covariance matrix C, and perform eigenvalue decomposition on the covariance matrix C to obtain C, where Λ is a diagonal matrix and the diagonal elements are eigenvalues λ1≥λ2≥λ3≥…≥λ n , where V is the eigenvector matrix. Select the eigenvectors corresponding to the top k largest eigenvalues and project the original data onto these k eigenvectors to obtain the dimensionality-reduced data Y.

8. The emergency management disaster dynamic early warning and dispatching system based on multi-source data fusion according to claim 4, characterized in that, The disaster evolution modeling of the dynamic early warning generation module constructs a disaster evolution model based on physical process modeling and machine learning algorithms, including a long short-term memory network (LSTM) for processing time series data, a convolutional neural network (CNN) for analyzing spatial data, and uses a large amount of historical disaster case data and simulation data to train and optimize the model. The long short-term memory network (LSTM) is as follows: Forgotten Gate: f t = σ(W f [h t-1 , x t + b f ) Input gate: i t = σ(W i [h t-1 , x t + b i ) Input gate: o t = σ(W o [h t-1 , t +) h t = o t ⊙tanh(C t ) Wherein, the forget gate determines the cell state at the previous moment how much information needs to be forgotten, f t is the output of the forget gate, which is a vector with a value range between [0,1]. Each element represents the degree to which the corresponding information is forgotten. σ is the sigmoid activation function, and W f is the weight matrix of the forget gate, and b f is the bias vector of the forget gate. [h t-1 , x t means concatenating the hidden state h t-1 at the previous moment and the input x t at the current moment; in the input gate, i t is the output of the sigmoid part of the input gate. W i is the weight matrix of the sigmoid part of the input gate, and b i is the bias vector. is the candidate cell state. Tanh is the hyperbolic tangent activation function, and w c is the weight matrix of the candidate cell state, and b c is the bias vector. The output gate determines the cell state at the current moment; the output gate determines the cell state C t how much information needs to be output to the hidden state h t , where o t is the output of the sigmoid part of the output gate. W o is the weight matrix of the output gate, and b0 is the bias vector. The convolutional neural network (CNN) is as follows: Convolution operation: Activation function: y i,j,c = max(0, Y i,j，c ) Loss function: Optimization algorithm: wherein, at the position (i, j), the calculation of the c-th channel of the output feature map Y is Y i,j,c , b c is the bias of the c-th channel, and the activation function uses the ReLu function y i,j,c . For the disaster evolution model, the commonly used loss function can be the mean square error loss (MSE), where θ is the parameter of the model, is the predicted output of the model for the input x (i) , and the optimization algorithm uses stochastic gradient descent, where α is the learning rate. In actual training, mini-batch stochastic gradient descent (Mini-Batch SGD) is adopted, and a small batch of data is selected from the training dataset each time to calculate the gradient and update the parameters.

9. The emergency management disaster dynamic early warning and dispatching system based on multi-source data fusion according to claim 1, characterized in that, The precise resource scheduling module includes a resource database and a geographic information system integration sub-module. The resource database and geographic information system integration sub-module includes a geographic information system (GIS) for storing topographic features, transportation networks, and population distribution information, and an emergency resource distribution database for storing the location, quantity, and performance information of rescue teams, rescue equipment, and supplies. It uses spatial indexing technology to optimize data query and retrieval efficiency, matches the emergency resource location information with the GIS map through geocoding technology, and at the same time establishes a data update mechanism to ensure data accuracy and consistency.

10. The emergency management disaster dynamic early warning and dispatching system based on multi-source data fusion according to claim 1, characterized in that, The precise resource scheduling module also includes an optimal resource allocation plan planning sub-module. The optimal resource allocation plan planning sub-module uses an improved Dijkstra algorithm for path planning, considers traffic conditions and weather conditions, determines the resource allocation plan based on a linear programming-based resource allocation model, uses the simulated annealing algorithm to optimize the resource allocation plan, and establishes a real-time monitoring mechanism to track the resource transportation and allocation process. The improved Dijkstra algorithm is as follows: Let d[v] represent the current shortest distance estimate from the source node s to node v. Initially, d[s]=0, and for v≠s, d[v]=∞. Let S be the set of nodes for which the shortest path has been determined. Initially, The core steps of the algorithm are as follows: When s≠V: Select a node u from V-S such that d[u]=min{d[v]: v∈V-S}, add u to S, and for the nodes adjacent to u (i.e., (u, v)∈E), update d[v]: d[v]=min(d[v], d[u]+w(u, v)); When considering traffic conditions and weather conditions, the weight w(u, v) of the edge needs to be adjusted. Assuming that the traffic condition is represented by t(u, v) and the weather condition is represented by w c (u, v), then the new edge weight W(u, v)=w(u, v)×t(u, v)×wc(u, v); The linear programming-based resource allocation model is as follows: Let \(R\) be the set of resource types, \(L\) be the set of demand locations, \(S\) be the set of supply locations, and let \(x\) r，s，1 represent the quantity of resource \(r\) allocated from supply location \(s\) to demand location \(l\). The objective function: Assuming the goal of resource allocation is to minimize the total cost, let \(c\) r，s，1 be the cost of allocating one unit of resource \(r\) from supply location \(s\) to demand location \(l\). Then the objective function is Constraints: Supply constraint: The total amount of resources provided by each supply location s cannot exceed its available amount. Let α r，s be the quantity of resource r owned by supply location s. Then for all r ∈ R and s ∈ S; Demand constraint: The total amount of resources obtained by each demand location l should meet its demand. Let b r，1 be the demand of demand location l for resource r, then for all r ∈ R and l ∈ L; Non - negativity constraint: x r，s，l ≥ 0, for all r ∈ R, s ∈ S and l ∈ L.

Images