Emergency resource intelligent scheduling optimization method and system based on large model analysis

Abstract

The invention provides an emergency resource intelligent scheduling optimization method and system based on large model analysis, and relates to the technical field of emergency resource scheduling. The method comprises the steps of collecting emergency event data and environment data; extracting emergency event features and environment features, constructing a prediction model based on a long short-term memory network, and obtaining a required emergency resource list; constructing an emergency resource evaluation model based on an analytic hierarchy process, and obtaining priority scores of the required emergency resources; constructing a multi-objective optimization model, and solving the multi-objective optimization model by using a particle swarm algorithm to generate a scheduling scheme; and constructing a decision function, and solving by using a genetic algorithm to obtain a scheduling decision. According to the method, through accurate prediction, scientific evaluation, multi-objective optimization and decision in combination with priority, the accuracy and efficiency of emergency resource scheduling and the resource utilization efficiency can be remarkably improved, powerful support is provided for emergency management work, and the method has important practical significance and application value.

Description

Technical Field

[0001] The present invention relates to the technical field of emergency resource scheduling, and in particular to an emergency resource intelligent scheduling optimization method based on large model analysis and an emergency resource intelligent scheduling optimization system based on large model analysis. Background Art

[0002] Emergency resource dispatch refers to the rational allocation and deployment of various emergency resources (such as personnel, supplies, equipment, and transportation) during emergencies to ensure a swift and efficient response to and handling of accidents or disasters. Its core goal is to optimize resource allocation, shorten response times, and minimize losses and risks.

[0003] Current scheduling methods focus primarily on the spatial layout, transportation routes, and time efficiency of resources, ignoring the varying priorities inherent in the critical role of resources in emergency response. This means that scheduling systems are unable to dynamically adjust priorities based on resource importance, instead mechanically allocating resources based on quantity and location. This lacks scientific assessment and prioritization of which resources are most critical and which can be postponed. This deficiency prevents limited transportation and scheduling capacity from being focused on securing the most critical resources in emergencies, reducing overall emergency response efficiency and effectiveness. Summary of the Invention

[0004] The present invention provides an emergency resource intelligent scheduling optimization method and system based on large model analysis, which are used to solve the defects in the prior art.

[0005] In one aspect, the present invention provides an intelligent scheduling optimization method for emergency resources based on large-scale model analysis, comprising: Emergency event data and environmental data are collected and preprocessed to obtain preprocessed data.

[0006] Emergency event characteristics and environmental characteristics are extracted from preprocessed data, and a prediction model based on long short-term memory network is constructed. Emergency event characteristics and environmental characteristics are input, and a list of required emergency resources is output.

[0007] An emergency resource assessment model based on the analytic hierarchy process was constructed to obtain the priority scores of the required emergency resources.

[0008] Collect emergency resource data for each resource in the required emergency resource list, build a multi-objective optimization model based on emergency resource data, emergency event data and environmental data, and use particle swarm optimization to solve the multi-objective optimization model and generate a scheduling plan.

[0009] According to the scheduling scheme, the decision function is constructed in combination with the priority score, and the genetic algorithm is used to solve it to obtain the scheduling decision.

[0010] According to the present invention, a method for intelligently scheduling and optimizing emergency resources based on large-scale model analysis includes emergency event data, including event type, event time, and event scope. Emergency resource data includes resource type, resource quantity, and resource distribution. Environmental data includes geographic information and meteorological information.

[0011] According to a method for intelligent scheduling optimization of emergency resources based on large-scale model analysis provided by the present invention, the process of extracting emergency event characteristics and environmental characteristics from preprocessed data includes: Use the equal-width discretization method to convert the continuous data in the preprocessed data into discrete category data. The conversion process includes: The continuous data in the preprocessed data is divided into k intervals, the interval width and the boundary of each interval are calculated, and discrete category data are obtained.

[0012] The information gain of each discrete category data for emergency event and environment classification is calculated using the feature selection method based on information gain.

[0013] Category data with information gain greater than the preset threshold are selected as the final emergency event features and environmental features.

[0014] According to the present invention, a method for intelligent scheduling optimization of emergency resources based on large model analysis is provided, and the process of constructing a prediction model based on a long short-term memory network includes: Collect historical emergency event data and historical environmental data.

[0015] Preprocess historical emergency event data and historical environmental data, extract historical emergency event characteristics and historical environmental characteristics, and mark the historical emergency resource list.

[0016] The long-short-term memory (LSTM) network structure is set to obtain a basic model, which consists of an input layer, a hidden layer, and an output layer. The input layer receives historical emergency event characteristics and historical environmental characteristics. The hidden layer contains multiple LSTM network units to capture the long-term dependencies between historical emergency event characteristics and historical environmental characteristics. The output layer outputs a list of predicted emergency resources.

[0017] The basic model is trained by taking historical emergency event characteristics and historical environmental characteristics as input and the corresponding historical emergency resource list as output.

[0018] The stochastic gradient descent algorithm is used to update the model parameters, and the mean square error is used as the loss function. The model parameters that meet the preset accuracy are retained to obtain the prediction model.

[0019] According to a method for intelligent scheduling optimization of emergency resources based on large-scale model analysis provided by the present invention, the process of obtaining priority scores of required emergency resources includes: The emergency resource assessment problem is broken down into the goal layer, the criteria layer, and the solution layer. The goal layer is used to prioritize the required emergency resources. The criteria layer includes three criteria, corresponding to the importance, availability, and timeliness of the required emergency resources. The solution layer represents each emergency resource in the required emergency resource list.

[0020] For each criterion in the criterion layer, the relative importance of various resources in the solution layer is compared to construct a judgment matrix.

[0021] The eigenvalue method is used to calculate the maximum eigenvalue of the judgment matrix and its corresponding eigenvector, and the eigenvalue vector is normalized to obtain the weight vector.

[0022] Multiply the weight of each criterion by the weight of the resource in the corresponding solution layer, and sum them up to obtain the comprehensive priority score of each resource.

[0023] According to the present invention, a method for intelligent scheduling optimization of emergency resources based on large model analysis is provided, and the process of constructing a multi-objective optimization model includes: Set the objective function, which includes minimizing resource transportation costs, minimizing resource response time, and maximizing resource supply and demand matching.

[0024] Constraints are determined to obtain a multi-objective optimization model. Constraints include supply constraints, demand constraints, and non-negativity constraints. Supply constraints require that the transport quantity of each resource from each supply point must not exceed its supply quantity. Demand constraints require that the quantity of each resource received by each demand point must meet its basic demand. Non-negativity constraints require that the transport quantity cannot be negative.

[0025] According to the present invention, an emergency resource intelligent scheduling optimization method based on large model analysis is provided. The process of using a particle swarm algorithm to solve a multi-objective optimization model includes: A group of particles are randomly initialized, each particle represents a possible scheduling solution.

[0026] For each particle, its fitness value is calculated according to the objective function of the multi-objective optimization model.

[0027] Update the particle position and velocity, terminate the algorithm when the preset maximum number of iterations is reached, and output the scheduling plan corresponding to the global optimal position as the approximate optimal solution of the multi-objective optimization model.

[0028] According to the present invention, a method for intelligent scheduling optimization of emergency resources based on large model analysis is provided, and the process of constructing a decision function in combination with priority scoring includes: Define the priority weight coefficient, introduce the priority weight to adjust the objective function according to the multi-objective optimization model, and construct the decision function, resource supply cost decision function, resource scheduling time decision function and resource demand satisfaction degree decision function.

[0029] Genetic algorithm is used to solve the decision function and obtain the optimized scheduling decision.

[0030] According to the present invention, a method for intelligent scheduling optimization of emergency resources based on large model analysis is provided, wherein the process of solving the decision function using a genetic algorithm includes: Real number coding is used, and the scheduling quantity of each resource in the scheduling plan is used as the gene of the chromosome. A certain number of individuals are randomly generated in the feasible solution space to form the initial population. Each individual represents a possible solution to the decision function.

[0031] The weighted summation method is used to transform multiple decision functions into a single-objective fitness function.

[0032] Randomly select k individuals from the current population to form a tournament group, compare the fitness values ​​of the k individuals, and select the individual whose fitness reaches the preset threshold as the parent.

[0033] Randomly pair two of the selected parent individuals. For each pair of parent individuals, randomly select a crossover point, exchange the genes of the two parent individuals after the crossover point, and generate two offspring individuals.

[0034] Set the crossover probability, and perform a crossover operation with the crossover probability for each pair of parent individuals.

[0035] For each individual in the offspring population, the preset mutation probability is used to decide whether to mutate. If the individual needs to mutate, one or more genes on the chromosome are randomly selected for mutation. For the selected genes, a random mutation amount is generated according to the Gaussian distribution and the gene value is updated.

[0036] For each individual in the mutated offspring population, the fitness value is recalculated, the offspring population and the parent population are merged, and individuals are selected in descending order of fitness value to form the next generation population, keeping the population size unchanged.

[0037] When the maximum number of iterations is reached, the iteration is stopped and the individual with the highest fitness in the population is output, which is the optimal scheduling solution.

[0038] On the other hand, the present invention also provides an emergency resource intelligent scheduling optimization system based on large model analysis, comprising: The data acquisition module is used to collect emergency event data and environmental data and perform preprocessing to obtain preprocessed data.

[0039] The emergency resource prediction module is used to extract emergency event characteristics and environmental characteristics from preprocessed data, build a prediction model based on long short-term memory network, input emergency event characteristics and environmental characteristics, and output a list of required emergency resources.

[0040] The resource priority assessment module is used to build an emergency resource assessment model based on the hierarchical analysis method to obtain the priority score of the required emergency resources.

[0041] The scheduling plan generation module is used to collect the emergency resource data of each resource in the required emergency resource list, build a multi-objective optimization model based on the emergency resource data, emergency event data and environmental data, and use the particle swarm algorithm to solve the multi-objective optimization model to generate a scheduling plan.

[0042] The scheduling decision generation module is used to construct a decision function based on the scheduling plan and priority score, and use the genetic algorithm to solve it to obtain the scheduling decision.

[0043] The present invention provides an intelligent scheduling optimization method and system for emergency resources based on large model analysis. Through the sequence data processing capability of the long short-term memory network, it can capture the complex patterns of emergency events and environmental factors changing over time, and accurately predict the required emergency resource list. Avoid excessive or insufficient allocation of resources, and provide a reliable foundation for subsequent scheduling work. Constructing an emergency resource evaluation model based on the hierarchical analysis method can comprehensively and scientifically determine the priority scores of required emergency resources, ensure that when resources are limited, the most important resources are allocated first, and improve the effectiveness of emergency response. By constructing a multi-objective optimization model and solving it using the particle swarm algorithm, the optimal solution can be quickly found among many possible scheduling schemes. It can not only reduce the cost of emergency resource scheduling, but also shorten the scheduling time and improve the efficiency of resource utilization. After generating the scheduling scheme, a decision function is constructed according to the priority score and solved using a genetic algorithm, so that the final scheduling decision can be more in line with actual needs and ensure that important resources can be given priority. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 This is a flow chart of an emergency resource intelligent scheduling optimization method based on large model analysis provided by an embodiment of the present invention; Figure 2 This is a structural diagram of an emergency resource intelligent scheduling optimization system based on large model analysis provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0046] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0047] The following combination Figure 1-Figure 2 The present invention describes an emergency resource intelligent scheduling optimization method and system based on large model analysis.

[0048] Figure 1 This is a flow chart of an emergency resource intelligent scheduling optimization method based on large model analysis provided by an embodiment of the present invention.

[0049] like Figure 1 As shown, an embodiment of the present invention provides an emergency resource intelligent scheduling optimization method and system based on large model analysis. The execution subject can be an emergency resource intelligent scheduling optimization method based on large model analysis, and the method includes: Emergency event data and environmental data are collected and preprocessed to obtain preprocessed data.

[0050] Emergency event data includes event type, event time, and event scope. Emergency resource data includes resource type, resource quantity, and resource distribution.

[0051] Collect emergency and environmental data from a wide range of sources. Emergency data may come from reports from emergency management departments, feedback from on-site rescue personnel, and monitoring equipment records; environmental data can be obtained from meteorological departments, geographic information systems, and transportation departments.

[0052] Integrate the collected data from multiple sources and store them in a unified data warehouse or database. Since data formats and standards may vary from data source to data source, data format conversion and unification are required to ensure data consistency and compatibility.

[0053] Check whether there are missing values ​​in the data. Different methods can be used to handle missing values. If there are few missing values, you can directly delete the records containing missing values. If there are many missing values, you can use statistics such as mean, median, mode, etc. to fill in the missing values, or use interpolation to estimate them.

[0054] Identify outliers in the data, which may be caused by data entry errors, sensor failures, etc. Outliers can be deleted, corrected, or smoothed.

[0055] Check whether there are duplicate records in the data. For duplicate records, keep only one to avoid data redundancy from interfering with subsequent analysis.

[0056] Convert the data to a standard normal distribution with a mean of 0 and a standard deviation of 1. Standardization can eliminate the dimensional differences between different features and make different features comparable.

[0057] Emergency event characteristics and environmental characteristics are extracted from preprocessed data, and a prediction model based on long short-term memory network is constructed. Emergency event characteristics and environmental characteristics are input, and a list of required emergency resources is output.

[0058] Extract features related to emergency resource scheduling from preprocessed data. For example, extract features such as road congestion level and average vehicle speed from traffic data.

[0059] The process of extracting emergency event features and environmental features from preprocessed data includes: Use the equal-width discretization method to convert the continuous data in the preprocessed data into discrete categorical data. Continuous data can be the specific time of an event, or the temperature and wind speed in meteorological information. The conversion process includes: Assume that the continuous data in the preprocessed data is , divide it into k intervals, calculate the interval width, the formula is expressed as:

[0060] Calculate the boundaries of each interval to obtain discrete category data. The formula is expressed as:

[0061] Where, Indicates the maximum value in continuous data. Represents the minimum value in continuous data, i=0,1,…,k.

[0062] The information gain-based feature selection method is used to calculate the information gain of each discrete category data for emergency events and environmental classification. The formula is expressed as:

[0063] Where H(S) is the information entropy of the data set S, and the formula is expressed as:

[0064] Where, represents the probability of category i appearing in the data set S, m represents the number of categories, Representing discrete categorical data The value is a subset of v, Representation subset The number of samples, Indicates the number of samples in the dataset S.

[0065] By calculating the information gain of each category data, the category data with information gain greater than the preset threshold is selected as the final emergency event characteristics and environmental characteristics.

[0066] For discrete categorical features, such as event types, resource types, and geographic region categories, use one-hot encoding to convert them into numeric vectors. Assume that a categorical feature has n distinct values, and represent each value as an n-dimensional vector, where only the corresponding position is set to 1, and all other positions are set to 0.

[0067] The process of building a prediction model based on LSTM networks includes: Collect historical emergency event data and historical environmental data.

[0068] Preprocess historical emergency event data and historical environmental data, extract historical emergency event characteristics and historical environmental characteristics, and mark the historical emergency resource list.

[0069] The long-short-term memory (LSTM) network structure is set to obtain a basic model, which consists of an input layer, a hidden layer, and an output layer. The input layer receives historical emergency event characteristics and historical environmental characteristics. The hidden layer contains multiple LSTM network units to capture the long-term dependencies between historical emergency event characteristics and historical environmental characteristics. The output layer outputs a list of predicted emergency resources.

[0070] The basic model is trained by taking historical emergency event characteristics and historical environmental characteristics as input and the corresponding historical emergency resource list as output.

[0071] The stochastic gradient descent algorithm is used to update the model parameters, and the mean square error is used as the loss function. The model parameters that meet the preset accuracy are retained to obtain the prediction model. The mean square error formula is expressed as:

[0072] Where, Represents a list of predicted emergency resources, Represents the historical emergency resource list, and n represents the number of samples.

[0073] Taking flood events as an example, records of similar flood events that occurred in the past 10 years were obtained, including event type (flood), scale (affected area, water level), occurrence time, duration, and affected population.

[0074] Collect historical environmental data: Collect relevant environmental factors, including rainfall, temperature, humidity, topographic data (such as river slope and city elevation), and infrastructure information (such as the condition of drainage systems).

[0075] For each historical event, a list of resources actually used is annotated as the output label of supervised learning.

[0076] Before model training, the emergency management department compiled data from 20 historical flood events. Each event contained 10 dimensions of emergency event characteristics and 8 dimensions of environmental characteristics, as well as a labeled list of four types of resources (tents, food, medical teams, and rescue boats) to form a training dataset.

[0077] The collected raw data was cleaned (e.g., handling missing values: filling missing rainfall data with the mean) and normalized (e.g., scaling values ​​such as affected area and population to the range [0, 1] to ensure data comparability). Textual data (e.g., event type) was converted to numerical encoding.

[0078] Key features are extracted from the preprocessed data, including event scale (affected area, affected population), severity (peak water level, duration of the event), and dynamic changes (time series data of the event, such as hourly changes in water level).

[0079] Extract environmental factors that affect resource demand, such as meteorological conditions (average rainfall, temperature changes), geographical conditions (terrain complexity, altitude), and external factors such as seasonal information.

[0080] The extracted historical features include emergency event characteristics (affected area, affected population, peak water level, duration) and environmental characteristics (average rainfall, temperature, terrain complexity index). These features are combined into an input vector, which is then labeled and output as a resource list.

[0081] Set up an LSTM basic model, including: Input layer: Receives historical emergency event characteristics and historical environmental characteristics (input dimension is 18, including event scale, environmental factors, etc.).

[0082] Hidden layer: Contains 128 LSTM units (stacked in two layers) to capture long-term dependencies between features (such as the lagged impact of flood duration on resource demand).

[0083] Output layer: Outputs a predicted list of emergency resources (a 4-dimensional vector corresponding to the number of tents, food, medical teams, and rescue boats).

[0084] Input feature vectors of 20 historical events (each event corresponds to a sequence of time steps, such as 24 hours before the event to the end of the event).

[0085] Output a marked list of historical emergency resources.

[0086] The stochastic gradient descent optimizer is used with a learning rate of 0.001. The mean square error is used as the loss function to calculate the difference between the predicted resource list and the actual resource list.

[0087] The training was iterated for 1000 rounds, and the model parameters with MSE lower than 0.05 (preset accuracy) were retained to finally obtain the trained prediction model.

[0088] After training the model on historical data, it learned a pattern: when rainfall exceeds a certain threshold and affects more than a certain number of people, the demand for medical teams increases significantly (because flooding is prone to causing illness). Leveraging the time series processing capabilities of LSTM, the model captures the dynamic changes in resource demand as events unfold (for example, food demand rises sharply 24 hours after a water level peak).

[0089] When a flood occurs in City B, current emergency event data (affected area, affected population, peak water level) and environmental data (average rainfall, temperature, terrain complexity) are collected.

[0090] The new data is subjected to the same preprocessing (cleaning and normalization) and feature vectors are extracted (for example, the input sequence includes hourly data from the start of the event to the current time).

[0091] The new feature vector is input into the trained LSTM prediction model. The model uses the LSTM units in the hidden layer to analyze the long-term dependencies between features (such as the relationship between heavy rainfall and subsequent medical needs) and outputs a predicted list of emergency resources.

[0092] For the flood event in City B, the model outputs a list of required emergency resources: P4 tents, P3 food parcels, P1 medical team, and P2 rescue boats. Higher water levels and population impacts increase the need for tents and medical resources, while continued rainfall increases the need for rescue boats (due to the complex terrain).

[0093] Construct an emergency resource assessment model based on the analytic hierarchy process to obtain the priority scores of required emergency resources. The process includes: The emergency resource assessment problem is broken down into the goal layer, the criteria layer, and the solution layer. The goal layer is used to prioritize the required emergency resources. The criteria layer includes three criteria, corresponding to the importance, availability, and timeliness of the required emergency resources. The solution layer represents each emergency resource in the required emergency resource list.

[0094] For each criterion in the criterion layer, the relative importance of various resources in the solution layer is compared to construct a judgment matrix. The formula is:

[0095] Where, It indicates the importance of resource i relative to resource j, z indicates the type of resource, and its value ranges from 1 to 9 and the reciprocal.

[0096] Use the eigenvalue method to calculate the maximum eigenvalue of the judgment matrix A and its corresponding eigenvector, and normalize the eigenvalue vector to obtain the weight vector. The formula is expressed as follows:

[0097] Where, and denote the i-th and j-th eigenvectors respectively.

[0098] Multiply the weight of each criterion by the weight of the resource in the corresponding solution layer, and sum them up to get the comprehensive priority score of each resource. Let the weight vector of the criterion layer be , the weight matrix of each criterion at the solution level is , then the comprehensive priority scoring formula for each resource is expressed as:

[0099] Where, represents the priority score of the i-th resource, represents the weight vector of the kth criterion layer, represents the weight matrix of the i-th criterion.

[0100] After the prediction model outputs a list of required emergency resources, the emergency center needs to prioritize the four types of resources. The goal is to prioritize the dispatch of critical resources when resources are limited (such as insufficient transportation capacity).

[0101] Key contradiction: If all resources are transported at the same time, the capacity of the currently available transport vehicles will exceed (for example, only 70% of the resources can be transported), and trade-offs must be made based on priority.

[0102] Priority scoring process: based on the analytic hierarchy process.

[0103] It is completed in four steps, including establishing a hierarchy, constructing a judgment matrix, calculating weights, and synthesizing priority scores.

[0104] Establish a hierarchy, including: Target layer, determines the resource scheduling priority (highest target).

[0105] At the criteria level, there are three evaluation criteria: Importance (C1): the direct impact of resources on the survival / safety of disaster victims (such as medical teams saving lives); Availability (C2): the current inventory and difficulty of obtaining them (such as whether local warehouses have reserves); Timeliness (C3): whether resources need to be delivered within the golden time (such as rescue boats need to be in place within 6 hours).

[0106] At the solution level, there are four resources to be evaluated: R1: medical team (P1), R2: rescue boat (P2), R3: food package (P3), and R4: tent (P4).

[0107] A judgment matrix was constructed, and the emergency expert team compared the criteria and resources pairwise and assigned values ​​on a scale of 1-9 ( Table 1 ).

[0108] Table 1: Meaning of scales 1-9 for chromatography analysis:

[0109] Criteria layer comparison (target: resource priority): Importance (C1) has the most direct impact on survival (e.g., lack of medical teams will lead to death) and is significantly more important than timeliness (C3) (scale 5); Timeliness (C3) is slightly more important than availability (C2) (scale 3); Generate the criterion layer judgment matrix A:

[0110] Among them, the main diagonal = 1 (self-comparison), C1vsC2 = 5 (importance is obviously more important than availability), and C3vsC2 = 3 (timeliness is slightly more important than availability).

[0111] Resource layer comparison (for each criterion), a. Criterion: Importance (C1): Medical teams (R1) are most critical to saving lives and are significantly more important than rescue boats (R2) (scale 5); medical teams (R1) are strongly more important than food (R3) (scale 7); and rescue boats (R2) are slightly more important than tents (R4) (scale 3).

[0112] Generate matrix B1:

[0113] b. Criterion: Availability (C2), Data Support: There are sufficient food packages in the local warehouse (high availability, small value), and the rescue boat needs to be transferred from other places (low availability, large value).

[0114] Combined with data judgment: Food (R3) has the highest availability and is significantly more important than medical teams (R1) (scale 5). Tents (R4) are slightly more important than rescue boats (R2) (scale 3), generating matrix B2:

[0115] c. Criterion: Timeliness (C3). Current flood situation: Water levels are still rising, and trapped people must be evacuated within 6 hours. Rescue boats (R2) must be immediately available and are slightly more important than medical teams (R1) (Scale 3). Food (R3) can be delayed for 12 hours and is equally important as tents (R4) (Scale 1).

[0116] Generate matrix B3:

[0117] Calculate the weights of each layer and use the eigenvalue method to solve the weight vector.

[0118] Criterion layer weight W c , calculate the maximum eigenvalue λ of matrix A max ≈3.085, corresponding to the normalized eigenvector:

[0119] Importance (63.7%) > Timeliness (25.8%) > Availability (10.5%).

[0120] Resource layer weights (for each criterion), importance criterion (C1) weight W1:

[0121] That is, the medical team (60%) was the top priority, followed by the rescue boat (21.7%).

[0122] Usability criterion (C2) weight W2:

[0123] That is, food (48.9%) is the easiest to obtain, and rescue boats (13.2%) are the most difficult to transport.

[0124] Timeliness criterion (C3) weight W3:

[0125] That is, the rescue boat (51.1%) needs to be delivered as quickly as possible, while the tent (12.8%) can be delayed.

[0126] Calculate the comprehensive priority score of each resource according to the priority score synthesis formula.

[0127] Table 2: Comprehensive priority rating of emergency resources:

[0128] The final score ranking is: medical team (0.468) > rescue boat (0.284) > food package (0.145) > tent (0.102).

[0129] Collect emergency resource data for each resource in the required emergency resource list, build a multi-objective optimization model based on emergency resource data, emergency event data and environmental data, and use particle swarm optimization to solve the multi-objective optimization model and generate a scheduling plan.

[0130] Environmental data includes geographic information and meteorological information.

[0131] The process of building a multi-objective optimization model includes: Set the objective function, which includes minimizing resource transportation costs, minimizing resource response time, and maximizing resource supply and demand matching. Minimizing resource transportation costs includes: Assume that the transportation cost of emergency resource i from supply point j to demand point k is , the transport quantity is , then the resource transportation cost objective function is expressed as:

[0132] Minimizing resource response time includes: Assume that the transportation time of emergency resource i from supply point j to demand point k is , then the resource response time objective function is expressed as:

[0133] Maximizing the matching of resource supply and demand includes: Assume that the demand quantity of emergency resource i at demand point k is , the supply quantity of emergency resource i from supply point j is , then the resource supply and demand matching objective function is expressed as:

[0134] Determine the constraints, including supply constraints, demand constraints, and non-negativity constraints. The supply constraint is expressed as:

[0135] This means that the amount of each resource transported from each supply point does not exceed its supply.

[0136] The demand constraint is expressed as:

[0137] It means that the amount of each resource obtained by each demand point must meet its basic needs.

[0138] The non-negativity constraint is expressed as:

[0139] Indicates that the shipping quantity cannot be negative.

[0140] In this embodiment, city B is divided into a severely affected area, a moderately affected area, and a slightly affected area according to the severity of the flood. Resource requirements are matched according to the specific disaster situation, for example: D1: Severely affected area (60% of resources required), including 15 medical teams, 36 rescue boats, 9,000 food packages, and 1,800 tents. D2: Moderately affected area (30% of resources required), including 7 medical teams, 18 rescue boats, 4,500 food packages, and 900 tents. D3: Slightly affected area (10% of resources required), including 3 medical teams, 6 rescue boats, 1,500 food packages, and 300 tents.

[0141] Obtain information on supply points that can supply the required emergency resources, including: S1: Local warehouse (limited supply), including 10 medical teams, 20 rescue boats, 5,000 food parcels, and 1,000 tents. S2: Neighboring city warehouse (replenishable), including 20 medical teams, 40 rescue boats, 10,000 food parcels, and 2,000 tents.

[0142] The process of using particle swarm optimization to solve a multi-objective optimization model includes: A group of particles are randomly initialized, each particle represents a possible scheduling scheme, and the position vector of the particle is set represents the position of the i-th particle in the d-dimensional search space, where Represents the value of the i-th particle in the j-th dimension, corresponding to the decision variable in the multi-objective optimization model , initialize the particle's velocity vector .

[0143] For each particle, its fitness value is calculated according to the objective function of the multi-objective optimization model.

[0144] Update the particle position and velocity. The velocity update formula is expressed as:

[0145] Where, Indicates the current speed. represents the speed after speed update, R represents the inertia weight, and represents the acceleration constant, and is a random number between [0,1], represents the individual optimal position of the i-th particle, represents the global optimal position.

[0146] The position update formula is expressed as:

[0147] Where, represents the current position of particle i, Represents the updated position of particle i.

[0148] The particle position and velocity are continuously updated, and the algorithm is terminated when the preset maximum number of iterations is reached. The scheduling plan corresponding to the global optimal position is output as the approximate optimal solution of the multi-objective optimization model.

[0149] Based on the emergency resource demand during a flood event in City B and the information about available supply points, a multi-objective optimization model was constructed and solved using a particle swarm optimization algorithm. The resulting scheduling plan is as follows: Medical Team: S1 to D1, transporting 8 medical teams, at a cost of 1,600 yuan, and taking 3 hours. S2 to D1, transporting 7 medical teams, at a cost of 2,800 yuan, and taking 6 hours. S1 to D2, transporting 2 medical teams, at a cost of 300 yuan, and taking 2 hours. S2 to D2, transporting 5 medical teams, at a cost of 1,500 yuan, and taking 4 hours. S2 to D3, transporting 3 medical teams, at a cost of 600 yuan, and taking 3 hours.

[0150] Rescue Boat: S1 to D1, 15 boats, cost 7,500 yuan, time required 3 hours. S2 to D1, 21 boats, cost 16,800 yuan, time required 6 hours. S2 to D2, 18 boats, cost 10,800 yuan, time required 4 hours. S2 to D3, 6 boats, cost 2,400 yuan, time required 3 hours.

[0151] Food packages: S1 to D2, 4,000 meals, 32,000 yuan, 2 hours. S2 to D1, 9,000 meals, 135,000 yuan, 6 hours. S2 to D3, 1,500 meals, 12,000 yuan, 3 hours.

[0152] Tents: From S1 to D1, the transport quantity is 1,000 tents, the transport cost is 100,000 yuan, and the required time is 3 hours. From S2 to D3, the transport quantity is 300 tents, the transport cost is 24,000 yuan, and the required time is 3 hours.

[0153] According to the scheduling scheme, the decision function is constructed in combination with the priority score, and the genetic algorithm is used to solve it to obtain the scheduling decision.

[0154] The process of building a decision function based on priority scores includes: Define the priority weight coefficient, the formula is expressed as:

[0155] In the formula, z represents the type of resource, Represents the priority score of the j-th resource in the scheduling scheme.

[0156] According to the multi-objective optimization model, priority weights are introduced to adjust the objective function and construct a decision function, including: Resource supply cost decision function, the formula is expressed as:

[0157] The weighted coefficient for tent transportation costs is only 0.102, reducing its cost focus. The weighted coefficient for medical team costs is 0.468, strengthening its cost control.

[0158] Resource scheduling time decision function, the formula is expressed as:

[0159] A medical team delay of 1 hour = 0.468 × penalty value. A tent delay of 1 hour = 0.102 × penalty value, with a higher tolerance.

[0160] The decision function of the degree to which resources meet demand is expressed as follows:

[0161] min(⋅) ensures that over-delivery does not increase the score, focusing on the satisfaction rate of basic needs. The medical team satisfaction rate is 95%, contributing 0.468×0.95; the tent satisfaction rate is 30%, contributing only 0.102×0.30.

[0162] Where, represents the transportation cost of emergency resource i from supply point j to demand point k, represents the scheduling quantity of the i-th resource in the scheduling scheme, represents the optimized resource supply cost decision function, represents the optimized resource scheduling time decision function, represents the transportation time of emergency resource i from supply point j to demand point k, represents the demand quantity of emergency resource i at demand point k, Represents the decision function of the degree to which the optimized resources meet the demand, where Indicates taking and The minimum value in .

[0163] Genetic algorithm is used to solve the decision function and obtain the optimized scheduling decision.

[0164] The process of solving the decision function using genetic algorithms includes: Real number coding is used, and the scheduling quantity of each resource in the scheduling plan is used as the gene of the chromosome. A certain number of individuals are randomly generated in the feasible solution space to form the initial population. Each individual represents a possible solution to the decision function.

[0165] The weighted summation method is used to transform multiple decision functions into a single-objective fitness function, which can be expressed as follows:

[0166] Where, 、 、 is the weight coefficient.

[0167] Randomly select k individuals from the current population to form a tournament group, compare the fitness values ​​of the k individuals, and select the individual whose fitness reaches the preset threshold as the parent.

[0168] Randomly pair two of the selected parent individuals. For each pair of parent individuals, randomly select a crossover point, exchange the genes of the two parent individuals after the crossover point, and generate two offspring individuals.

[0169] Set the crossover probability, and perform a crossover operation with the crossover probability for each pair of parent individuals.

[0170] For each individual in the offspring population, the preset mutation probability is used to decide whether to mutate. If the individual needs to mutate, one or more genes on the chromosome are randomly selected for mutation. For the selected genes, a random mutation amount is generated according to the Gaussian distribution and the gene value is updated.

[0171] For each individual in the mutated offspring population, the fitness value is recalculated, the offspring population and the parent population are merged, and individuals are selected in descending order of fitness value to form the next generation population, keeping the population size unchanged.

[0172] When the maximum number of iterations is reached, the iteration is stopped and the individual with the highest fitness in the population is output, which is the optimal scheduling solution.

[0173] In this example, a decision function is constructed by combining priority scores and solved using a genetic algorithm. The final dispatch decision is: Medical Team: The original 8 medical teams from S1 to D1 will be increased to 10, preserving local resources in S1, reducing costs and improving timeliness. The original 7 medical teams from S2 to D1 will be reduced to 5, freeing up capacity for rescue boats. Rescue Boats: Eight new rescue boats will be added from S1 to D2 to ensure timeliness in the affected areas. Food Packs: The number of food packs from S2 to D3 will be reduced from 1,500 to 1,000. Tents: The number of tents from S1 to D1 will be reduced from 1,000 to 0. All other items remain unchanged.

[0174] Thanks to optimized dispatch decisions, the average medical team arrival time was reduced from 4.2 hours to 3.1 hours, a 27% reduction. Total transportation costs were also reduced from 225,400 yuan to 201,500 yuan, an 11% decrease.

[0175] In summary, this embodiment provides a method for intelligently optimizing emergency resource scheduling based on large-scale model analysis. This method, using a prediction model based on an LSTM network, breaks through the limitations of static experience. By analyzing disaster evolution patterns (such as the nonlinear relationship between flood level rise rate and medical resource demand), environmental coupling factors (the potential impact of geological conditions and rescue equipment loss coefficients), and historical response data, it constructs a multidimensional spatiotemporal mapping of resource demand. Compared to traditional manual estimation, this method significantly reduces resource waste caused by miscalculation (such as excessive allocation of tents, which squeezes out life-saving equipment transportation capacity).

[0176] Through the hierarchical analysis method, the importance of resources is converted into a computable priority scoring system, and a hierarchical decision-making scale is constructed under complex constraints, forcing the scheduling model to prioritize lifeline resources and avoid strategic mistakes caused by human emotional bias.

[0177] The particle swarm optimization algorithm, within a multi-dimensional search space of objectives (cost, time, and matching), discovered Pareto-optimal solutions that are unattainable with traditional scheduling. When the genetic algorithm incorporates resource weighting, the system automatically reduces low-priority resources in critical capacity shortage scenarios, ensuring a 100% delivery rate for medical equipment. The priority weighting factor incorporates the results of hierarchical analysis into the cost function, ensuring that high-value resources receive excess capacity.

[0178] The plan's priority scoring system and objective function conversion process establish a traceable, verifiable, and explainable decision-making chain. Audit departments can verify the scientific nature of the dispatch plan by replaying the genetic algorithm iteration process, significantly enhancing the credibility of emergency management.

[0179] Based on the same general inventive concept, the present invention also protects an emergency resource intelligent scheduling optimization system based on large model analysis. The emergency resource intelligent scheduling optimization system based on large model analysis provided by the present invention is described below. The emergency resource intelligent scheduling optimization system based on large model analysis described below and the emergency resource intelligent scheduling optimization method based on large model analysis described above can be referenced to each other.

[0180] Figure 2 This is a structural diagram of an emergency resource intelligent scheduling optimization system based on large model analysis provided by an embodiment of the present invention.

[0181] like Figure 2 As shown, an emergency resource intelligent scheduling optimization system based on large model analysis includes a memory, a processor and a computer program stored in the memory and runnable on the processor. The processor includes a data acquisition module, an emergency resource prediction module, a resource priority evaluation module, a scheduling scheme generation module and a scheduling decision generation module.

[0182] The data acquisition module is used to collect emergency event data and environmental data and perform preprocessing to obtain preprocessed data.

[0183] The emergency resource prediction module is used to extract emergency event characteristics and environmental characteristics from preprocessed data, build a prediction model based on long short-term memory network, input emergency event characteristics and environmental characteristics, and output a list of required emergency resources.

[0184] The resource priority assessment module is used to construct an emergency resource assessment model based on the hierarchical analysis method to obtain the priority scores of the required emergency resources.

[0185] The scheduling scheme generation module is used to collect the emergency resource data of each resource in the required emergency resource list, build a multi-objective optimization model based on the emergency resource data, emergency event data and environmental data, and use the particle swarm algorithm to solve the multi-objective optimization model to generate a scheduling scheme.

[0186] The scheduling decision generation module is used to construct a decision function based on the scheduling plan and priority score, and use the genetic algorithm to solve it to obtain the scheduling decision.

[0187] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0188] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for intelligent scheduling and optimization of emergency resources based on large model analysis, characterized in that: include: Collect emergency event data and environmental data and pre-process them to obtain pre-processed data; Extracting emergency event features and environmental features from the preprocessed data, constructing a prediction model based on a long short-term memory network, inputting the emergency event features and the environmental features, and outputting a list of required emergency resources; Constructing an emergency resource evaluation model based on the analytic hierarchy process to obtain the priority scores of the required emergency resources; Collecting emergency resource data for each resource in the required emergency resource list, building a multi-objective optimization model based on the emergency resource data, the emergency event data, and the environmental data, and solving the multi-objective optimization model using a particle swarm algorithm to generate a scheduling plan; According to the scheduling scheme, a decision function is constructed in combination with the priority score, and a genetic algorithm is used to solve it to obtain a scheduling decision.

2. The method for intelligent scheduling and optimization of emergency resources based on large model analysis according to claim 1 is characterized in that: The emergency event data includes event type, event time, and event scope; the emergency resource data includes resource type, resource quantity, and resource distribution; and the environmental data includes geographic information and meteorological information.

3. The method for intelligent scheduling and optimization of emergency resources based on large model analysis according to claim 1 is characterized in that: The process of extracting emergency event features and environmental features from the pre-processed data includes: The continuous data in the preprocessed data is converted into discrete category data using an equal-width discretization method. The conversion process includes: Dividing the continuous data in the preprocessed data into k intervals, calculating the interval width and the boundary of each interval, and obtaining discrete category data; Using the feature selection method based on information gain, the information gain of each discrete category data for emergency event and environment classification is calculated; Category data with information gain greater than the preset threshold are selected as the final emergency event features and environmental features.

4. The method for intelligent scheduling and optimization of emergency resources based on large model analysis according to claim 1 is characterized in that: The process of building a prediction model based on LSTM networks includes: Collect historical emergency event data and historical environmental data; Preprocessing the historical emergency event data and the historical environment data, extracting historical emergency event features and historical environment features, and marking a historical emergency resource list; A long short-term memory network structure is set to obtain a basic model, wherein the basic model includes an input layer, a hidden layer, and an output layer; the input layer is used to receive historical emergency event characteristics and historical environmental characteristics; the hidden layer includes multiple long short-term memory network units, which are used to capture long-term dependencies between the historical emergency event characteristics and historical environmental characteristics; and the output layer is used to output a predicted emergency resource list; The basic model is trained by taking historical emergency event characteristics and historical environmental characteristics as input and the corresponding historical emergency resource list as output; The stochastic gradient descent algorithm is used to update the model parameters, and the mean square error is used as the loss function. The model parameters that meet the preset accuracy are retained to obtain the prediction model.

5. The method for intelligent scheduling and optimization of emergency resources based on large model analysis according to claim 1 is characterized in that: The process of obtaining the priority score of the required emergency resources includes: The emergency resource assessment problem is decomposed into a target layer, a criterion layer, and a solution layer. The target layer is used to determine the priority of the required emergency resources; the criterion layer includes three criteria, corresponding to the importance, availability, and timeliness of the required emergency resources; and the solution layer is used to represent each emergency resource in the required emergency resource list. For each criterion in the criterion layer, the relative importance of various resources in the solution layer is compared to construct a judgment matrix; Use the eigenroot method to calculate the maximum eigenroot of the judgment matrix and its corresponding eigenvector, and normalize the eigenvector to obtain the weight vector; Multiply the weight of each criterion by the weight of the resource in the corresponding solution layer, and sum them up to obtain the comprehensive priority score of each resource.

6. The method for intelligent scheduling and optimization of emergency resources based on large model analysis according to claim 1 is characterized in that: The process of building a multi-objective optimization model includes: Setting an objective function, wherein the objective function includes minimizing resource transportation costs, minimizing resource response time, and maximizing resource supply and demand matching; Determine the constraints and obtain a multi-objective optimization model. The constraints include supply constraints, demand constraints and non-negative constraints. The supply constraints indicate that the transportation volume of each resource from each supply point shall not exceed its supply volume; the demand constraints indicate that the volume of each resource obtained by each demand point must meet its basic demand; and the non-negative constraints indicate that the transportation quantity cannot be a negative number.

7. The method for intelligent scheduling and optimization of emergency resources based on large model analysis according to claim 1 is characterized in that: The process of using the particle swarm algorithm to solve the multi-objective optimization model includes: Randomly initialize a group of particles, each particle represents a possible scheduling solution; For each particle, its fitness value is calculated according to the objective function of the multi-objective optimization model; Update the particle position and velocity, terminate the algorithm when the preset maximum number of iterations is reached, and output the scheduling plan corresponding to the global optimal position as the approximate optimal solution of the multi-objective optimization model.

8. The method for intelligent scheduling and optimization of emergency resources based on large model analysis according to claim 1 is characterized in that: The process of constructing a decision function in combination with the priority score includes: Define the priority weight coefficient, introduce the priority weight to adjust the objective function according to the multi-objective optimization model, and construct the decision function, resource supply cost decision function, resource scheduling time decision function and resource demand satisfaction degree decision function; Genetic algorithm is used to solve the decision function and obtain the optimized scheduling decision.

9. The method for intelligent scheduling and optimization of emergency resources based on large model analysis according to claim 8 is characterized in that: The process of solving the decision function using genetic algorithms includes: Using real number coding, the scheduling quantity of each resource in the scheduling scheme is used as the gene of the chromosome. A certain number of individuals are randomly generated in the feasible solution space to form the initial population. Each individual represents a possible solution of the decision function. The weighted summation method is used to transform multiple decision functions into a single-objective fitness function; Randomly select k individuals from the current population to form a tournament group, compare the fitness values ​​of the k individuals, and select the individual whose fitness reaches the preset threshold as the parent; Randomly pair two of the selected parent individuals. For each pair of parent individuals, randomly select a crossover point and exchange the genes of the two parent individuals after the crossover point to generate two offspring individuals. Set the crossover probability, and perform a crossover operation with the crossover probability for each pair of parent individuals; For each individual in the offspring population, a preset mutation probability is used to decide whether to mutate. If the individual needs to mutate, one or more genes on the chromosome are randomly selected for mutation. For the selected genes, a random mutation amount is generated according to the Gaussian distribution, and the gene value is updated. For each individual in the mutated offspring population, recalculate the fitness value, merge the offspring population with the parent population, and select individuals in descending order of fitness value to form the next generation population, keeping the population size unchanged; When the maximum number of iterations is reached, the iteration is stopped and the individual with the highest fitness in the population is output, which is the optimal scheduling solution.

10. An emergency resource intelligent scheduling optimization system based on large model analysis, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, it implements the emergency resource intelligent scheduling optimization method based on large model analysis as described in any one of claims 1 to 9.

Images