Chemical industrial park emergency processing method based on digital twinning

Abstract

The invention discloses a chemical industry park emergency processing method based on digital twinning. High-precision digital modeling is carried out on the physical environment of the chemical industry park, and dynamic changes in the park are sensed and reflected in real time; the Internet of Things equipment model is fused with the twin scene, the Internet of Things equipment model is matched with equipment information of an actual park enterprise, and collaborative linkage of the model and monitoring equipment is achieved; emergency scene simulation of various chemical industrial parks is provided, and different scenes are selected for drilling according to actual requirements; a chemical industrial park accident emergency resource scheduling model is constructed, a digital twinborn simulation technology and a hybrid intelligent optimization algorithm are fused, and resource allocation in a dynamic and complex environment is realized; and a collision-free safe path from the starting point to the ending point is found by comprehensively considering the necessary points and the bypassing areas. According to the invention, through integration of a high-precision digital twinning technology, a comprehensive emergency simulation drilling system and method are provided for the chemical industry park, so that the emergency response capability and the safety management level of the park are comprehensively optimized and improved.

Description

Technical Field

[0001] The present invention relates to the field of safety management, and in particular to an emergency response method for a chemical park based on digital twins. Background Art

[0002] In the field of chemical park safety management, traditional emergency simulation drills usually rely on physical simulation, paper simulation or simple computer simulation. Physical simulation sets up simulation scenes in chemical parks or specific sites and organizes personnel to conduct field exercises. It is costly and easily restricted by site, weather and other conditions. Paper simulation draws emergency flow charts, formulates emergency response plans, and conducts simulations on paper. Although it is low-cost, it lacks intuitiveness and interactivity, and it is difficult to fully evaluate the actual effect of emergency response. Computer simulation uses computer software to simulate emergency events in chemical parks. It has high flexibility and scalability, but is limited by the development level of computer technology, and the accuracy and fidelity of the simulation are often insufficient.

[0003] Traditional emergency simulation drills in chemical parks have deficiencies in accuracy, cost, real-time performance, and evaluation optimization. Summary of the invention

[0004] The core purpose of this invention is to provide a comprehensive emergency simulation drill system and method for chemical parks by integrating high-precision digital twin technology, so as to comprehensively optimize and enhance the emergency response capabilities and safety management level of the park.

[0005] The emergency response method for chemical parks based on digital twins includes the following steps:

[0006] Step S1: Build a high-precision digital twin model: Conduct high-precision digital modeling of the physical environment of the chemical park to accurately restore the equipment, facilities, and buildings in the park, and perceive and reflect the dynamic changes in the park in real time;

[0007] Step S2: The IoT device model is integrated with the twin scene to match the IoT device model with the device information of the actual park enterprise, thus realizing the coordinated linkage between the model and the monitoring equipment;

[0008] Step S3: Provide multiple chemical park emergency scenario simulations, including at least fire, explosion, and chemical leakage, and select different scenarios for drills according to actual needs;

[0009] Step S4: Construct a chemical park accident emergency resource scheduling model, integrate digital twin simulation technology and hybrid intelligent optimization algorithm, realize efficient resource (such as fire-fighting equipment, medical rescue materials, protective equipment, etc.) allocation in dynamic and complex environments, and significantly improve the scheduling response speed, resource utilization and accident response efficiency.

[0010] Step S5: Comprehensively consider the must-pass points and detour areas so that the target object can find a collision-free safe path from the starting point to the end point within the specified range of the park.

[0011] The present invention adopts the above technical solution and has the following advantages compared with the prior art:

[0012] Improve the realism and accuracy of emergency simulations: By building a digital twin model that is accurately mapped and highly compatible with the real chemical park, accurate simulation of various emergency events can be achieved, thereby improving the realism and accuracy of simulation exercises and making the exercise effects closer to actual accident scenarios.

[0013] Reduce the cost of emergency simulation drills: Relying on digital twin technology to conduct simulation drills, there is no need to invest a lot of space, equipment and manpower costs. The constructed simulation scenarios and accumulated data can be repeatedly called and deeply analyzed to avoid repeated waste of resources. While ensuring the quality and effect of the drills, the cost of the drills can be greatly reduced.

[0014] Enhance the real-time and effectiveness of emergency response: Through real-time data collection and transmission systems, real-time monitoring and response to emergency events can be achieved, and the initial signs, development trends and key change nodes of events can be quickly captured, thereby improving the speed and efficiency of emergency response.

[0015] Establish a scientific emergency assessment system: Use data mining and machine learning algorithms to conduct in-depth analysis and mining of simulation exercise data, accurately extract the key information and potential laws contained therein, and then establish scientific assessment standards and indicator systems to provide strong support for the optimization and improvement of emergency response plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The present invention DETAILED DESCRIPTION

[0017] The technical solution of the present invention is described in detail below with reference to the accompanying drawings:

[0018] like Figure 1 As shown in the figure, the steps of the chemical park emergency response method based on digital twin are as follows:

[0019] First, build a high-precision digital twin model: conduct high-precision digital modeling of the physical environment of the chemical park to achieve accurate restoration of equipment, facilities, buildings, etc. within the park, and perceive and reflect dynamic changes within the park in real time, such as equipment operating status, personnel flow, etc.

[0020] Second, the IoT device model is integrated with the twin scene to match the IoT device model with the equipment information of the actual park enterprises, thereby achieving coordinated linkage between the model and the monitoring equipment.

[0021] Third, it provides a variety of chemical park emergency scenario simulations, including fire, explosion, chemical leakage, etc., and different scenarios can be selected for drills according to actual needs.

[0022] Fourth, build a chemical park accident emergency resource dispatch model, integrate digital twin simulation technology and hybrid intelligent optimization algorithm, realize efficient resource (such as fire-fighting equipment, medical rescue materials, protective equipment, etc.) allocation in dynamic and complex environments, and significantly improve dispatch response speed, resource utilization and accident response efficiency.

[0023] Fifth, comprehensively consider the must-pass points and detour areas so that the target object can find a collision-free and safe path from the starting point to the end point within the specified scope of the park.

[0024] Sixth, after the drill is over, the drill process is evaluated and an evaluation report is generated, including drill time, task completion, personnel performance, etc. Users can summarize lessons learned and improve emergency plans based on the evaluation report.

[0025] During the functional execution process of the chemical park emergency simulation drill, executive-level personnel at various positions, commanders at all levels, expert teams, and guidance and control personnel participate together. The three-dimensional digital twin simulation drill process is completed according to the standard process control from the occurrence to the end of the disposal of an accident. At the same time, the entire drill process is tracked and recorded, thereby realizing the assessment and evaluation of the simulation drill process.

[0026] The technical solution of the present invention is described in detail below:

[0027] 1. Construction of high-precision digital twin chemical park scene

[0028] Digital twin visualization engine: Utilizes an efficient and stable 3D rendering engine to provide a foundation for the construction and visualization of twin scenes in chemical parks, including multi-dimensional spatial data visualization, rendering mechanism, environmental simulation, video scene fusion, etc. Based on the powerful rendering capabilities of the cloud rendering engine server and relying on network information flow, it provides refined scene display for each terminal while eliminating front-end hardware differentiation, achieving fast and efficient loading and browsing of scenes, and supporting user interactive operation scenarios.

[0029] 2. IoT device data fusion processing

[0030] The IoT device model is integrated with the twin scene to match the IoT device model with the equipment information of the actual park enterprises, realize the coordinated linkage between the model and the monitoring equipment, and collect the key location, video, electricity, sound, heat, light and other data of the monitored objects in real time. At the same time, it accesses the existing relevant monitoring, surveillance, and model data to jointly realize the dynamic collection of key data of the perceived objects.

[0031] 3. Accurate simulation of emergency scenarios

[0032] Drill organizers are allowed to set up a variety of emergency drill scenarios based on the digital twin model according to different drill goals and needs. Different accident types (such as fire, explosion, leakage, etc.), accident locations (specific enterprises or areas within the park), accident scales (small-scale local accidents or large-scale park-level accidents), and initial environmental conditions (such as meteorological conditions, surrounding traffic conditions, etc.) can be selected. Advanced physical engines and simulation algorithms are used to simulate the development process of emergency events. The algorithm should be able to accurately simulate key processes such as fire spread, gas diffusion, and personnel evacuation, and consider the impact of environmental factors such as wind speed, temperature, and humidity on the simulation results.

[0033] (1) Construct a scenario of hazardous substance leakage and diffusion in a chemical park, and build a diffusion model of hazardous substances in the atmosphere, water, and soil based on the principles of fluid mechanics and mass transfer.

[0034] (2) Construct a fire spread scenario in a chemical park and use a model based on the principles of energy conservation and heat transfer to simulate the development of the fire.

[0035] (3) Construct explosion accident scenarios in chemical parks. For explosion accidents, construct a comprehensive model that includes the propagation of explosion shock waves, the scattering of explosion fragments, and the secondary disasters caused by the explosion (such as fire, spread of toxic substances, etc.).

[0036] 4. Emergency resource dispatch

[0037] Construct a chemical park accident emergency resource dispatch model, integrate digital twin simulation technology and hybrid intelligent optimization algorithm, realize efficient resource (such as fire-fighting equipment, medical rescue materials, protective equipment, etc.) allocation in dynamic and complex environments, and significantly improve dispatch response speed, resource utilization and accident response efficiency.

[0038] 4.1 Interaction Mechanism between Intelligent Agent and Digital Twin Scenario

[0039] (1) Resource Supply Agent

[0040] Through the digital twin scene, accurate resource reserve information of resource supply points can be obtained in real time, including detailed data such as the quantity, storage location, and validity period of each resource. At the same time, based on the prediction of resource consumption trends by the digital twin model, the resource allocation plan is decided in combination with the reinforcement learning algorithm.

[0041] Resource reserve information in the digital twin scenario: including the current actual quantity of various resources (such as firefighting equipment, protective equipment, medical supplies, etc.) at each supply point Store location coordinates, etc.

[0042] State Space:

[0043]

[0044] Action Space:

[0045]

[0046] Types of resources deployed at each supply point and quantity

[0047] Resource replenishment time information That is, the estimated time from issuing a resource replenishment request to the actual replenishment of resources.

[0048] Resource consumption rate predicted by digital twin The consumption rate of each resource per unit time is predicted based on factors such as accident type, scale and development trend.

[0049] Reward function R:

[0050] 1) Rescue time reward function:

[0051]

[0052] Where T windows The best rescue time window for accidents simulated by digital twins. (the completion time of the i-th rescue subtask);

[0053] 2) Resource Utilization Reward Function:

[0054]

[0055] in is the k-th resource demand predicted by the digital twin, and δ is the allowable error range.

[0056] Using the reinforcement learning algorithm, the agent is based on the current state S supply , through the policy network π θ (a|s) (network parameters) select action A supply , and learn and update according to the reward function R, the goal is to maximize the long-term cumulative reward (γ is the discount factor).

[0057] (2) Transportation Agent

[0058] Based on the dynamic information of the road network in the digital twin scene, such as real-time traffic flow, road conditions (whether it is congested, whether it is damaged by an accident, etc.), and the location and status information of the transportation tools in the digital twin space, the optimal transportation route is planned and the transportation strategy is adjusted. The interaction between the transportation agent and the digital twin scene also includes real-time simulation of the transportation process, predicting the problems that may be encountered in different transportation solutions and avoiding them in advance.

[0059] The road network information G=(V,E) in the digital twin scenario includes nodes V (resource supply points, accident sites, road intersections, etc.) and edges E (road connection relationships), as well as the basic attributes of each edge (such as length, width, etc.).

[0060] Real-time traffic conditions C ij :Provided by the fusion of IoT sensing devices and digital twin scenarios, including the congestion level of each road (using the congestion coefficient ρ ij Indicates that 0≤ρ ij≤1 , 0 means smooth traffic, 1 means complete congestion), traffic speed v ij And other information.

[0061] The location of the transport means L t and state S t :The real-time geographic location coordinates of the transport vehicle and the vehicle’s operating status (such as whether it is fully loaded, whether it is faulty, etc.) obtained through IoT sensing devices.

[0062] Real-time weather conditions in digital twin scenarios condition : Such as wind speed ω speed 、Wind direction direction , rainfall intensity r internsity Etc. These factors will affect transportation safety and efficiency.

[0063] Potential interference of other related accidents in the surrounding area on the transportation route A interferense : For example, information such as the scope and estimated duration of road closures or traffic control caused by nearby accidents.

[0064] State space: S transport = {G,C ij ,L t ,S t ,W condition ,A interferense}.

[0065] Action Space: A transport ={Route ij ,Vehicle m ,V adjust}.

[0066] Reward function:

[0067] Transportation cost reward function

[0068] Among them C transport =∑ i (C fuel-i +C depreciation-i +C labor-i ), C fuel-i =S ij ×c fuel ×(1+α ω ×ω speed +α r × intensity )(c fuel is the fuel consumption cost per unit distance, α ω and α r is the coefficient of the effect of wind speed and rainfall intensity on fuel consumption), ΔC transport C is the cost increase caused by dynamic changes in road conditions (such as temporary control or emergency repair), budget Budget for preset shipping costs.

[0069] The reinforcement learning algorithm is also used, according to S transport State selection action A transport , and learn and update based on the reward function to optimize the transportation strategy, reduce transportation costs and time, and ensure transportation safety.

[0070] (3) Rescue Scene Agent

[0071] The digital twin scene is used to fully perceive the details of the accident scene, such as the precise type of accident (not just fire, leakage, etc., but also the specific type of chemical leakage, fire combustion characteristics, etc.), the real-time development of the accident (the direction of fire spread, the scope and speed of leakage, etc.), and the use of existing resources on site. Through deep interaction with the digital twin scene, the rescue scene agent can more accurately assess resource needs and provide more targeted resource demand adjustment suggestions to other agents.

[0072] Accident information in the digital twin scenario: including accident type T accident (such as detailed classification of fire, explosion, leakage, etc.), real-time development information of the accident, such as the direction and speed of fire spread, the scope and speed of leakage, etc.

[0073] Usage of existing resources at the scene: Types of resources that have arrived at the accident scene and its quantity Resource usage progress (such as the consumption ratio of firefighting foam P foam-consume wait).

[0074] Accident development trend prediction based on digital twin scenario simulation accident-trend :For example, the estimated duration of the fire is T fire-predict 、The maximum range of the leak is predicted L max-spread-predict wait.

[0075] Resource demand adjustment suggestions: Propose the types of resources that need to be allocated to the resource supply agent and the transportation agent and quantity adjustment suggestions.

[0076] State Space:

[0077] The severity of the accident is S severity The rescue progress can be quantitatively evaluated based on indicators such as the scope of the accident, the amount of dangerous substance leakage, etc. rescue The completed rescue mission ratio or key indicators (such as the fire extinguishing ratio P fire-extinguish etc.)

[0078] Action Space:

[0079] Reward function:

[0080] Accident Loss Reward Function

[0081] Where L accident-predict Real-time assessment and prediction of accident losses for digital twin scenarios, including casualties prediction casaultu-predict , Property loss prediction property-predict and environmental damage prediction environment-predict etc., L. max is the maximum acceptable accident loss threshold.

[0082] Based on the reinforcement learning algorithm, according to the state S scene Generate action A scene , and learns and optimizes according to the reward function to improve rescue efficiency and reduce accident losses.

[0083] 4.2 Network flow optimization module enabled by digital twins

[0084] (1) Network construction and dynamic parameter update

[0085] Resource supply point information in the digital twin scenario: including the location coordinates of the supply point and the upper limit of resource reserves And the proportion of supply capacity reduction caused by surrounding accidents β i (For example, if a supply point is damaged due to a fire nearby and some materials cannot be deployed, β i >0).

[0086] Accident site information: accident location coordinates, dynamic fluctuations in resource demand predicted by digital twin scenarios (As the incident develops, resource requirements change, such as the expansion of the danger zone or the increase in rescue difficulty).

[0087] Road network information: including nodes V' (resource supply points, accident sites, and possible transfer nodes) and edges E' (road connection relationships), as well as dynamic changes in road capacity monitored by the digital twin scenario (e.g. fire in surrounding buildings causing damage to some sections of the road, reducing road capacity) and real-time adjustment of transportation time (Calculated based on real-time traffic conditions, weather conditions and road condition changes, e.g. in is the dynamically adjusted transport speed).

[0088] The updated network flow model includes the adjusted network node set V', edge set E', edge capacity matrix C' (where ), the transport time matrix T' (where ) and the resource demand vector Q required (in

[0089] Model formula and logical relationship:

[0090] Construct network N = (V', E', C', T', D')

[0091] Where D' is the dynamic parameter matrix driven by the digital twin scenario, including and And other information.

[0092] For example, the resource supply constraint is adjusted to To consider the decline in supply capacity of the supply point; the capacity limit constraint is updated to Ensure that resource transportation flow is consistent with the actual road capacity.

[0093] (2) Dynamic adjustment of optimization objectives and constraints

[0094] The updated network flow model parameters (such as V′, E′, C′, T′, Q required wait).

[0095] Accident risk reduction degree P of digital twin scenario assessment risk-reduce :The proportion of accident diffusion range reduced by digital twin scenario simulation α spread-reduce 、Reduction ratio of dangerous substance concentration α concentration-reduce Quantitative evaluation of indicators such as R risk-reduce =ω1α spread-reduce +ω2αconcentration-reduce ,ω1 and ω2 are weight coefficients).

[0096] The final output is the optimized resource scheduling plan: determine the resource transportation flow on each transportation route The final allocation of resource supply points is allocated to meet the optimization objectives and constraints.

[0097] Model formula and logical relationship:

[0098] Objective function Z = λC total +(1-λ)T total +μR risk-reduce

[0099] in, The calculation is as described in the transportation agent section, T total =∑ (i,j)∈∈ T′ ij x ij .

[0100] Constraints:

[0101] 1) Resource supply constraints: (As mentioned above, considering the constraints after the reduction in supply capacity).

[0102] 2) Resource demand constraints:

[0103] 3) Flow conservation constraint: For the transit nodes k∈V′ except the resource supply point and the accident site, ∑ i:(i,k)∈E′ x ik ≤∑ j:(k,j)∈E′ x kj ,When considering new nodes or node failures that may appear in the digital twin scenario, the calculation scope and node relationships are dynamically adjusted.

[0104] 4) Capacity Limitation Constraints:

[0105] Adaptive network flow optimization algorithm is used to gradually approach the optimal solution by continuously searching for augmenting paths and adjusting network flows to meet the objective function and constraints.

[0106] 4.3 Deep Collaboration Mechanism between Digital Twins, Intelligent Agents, and Network Flows

[0107] (1) Real-time data sharing and synchronization

[0108] Real-time data in the digital twin scenario: including status data updated every second at the accident scene (such as the accident development trend and resource usage mentioned above), real-time traffic status data of the road network, and resource reserve and consumption data updated every second at resource supply points.

[0109] Decision information of multiple agents: resource allocation decision of resource supply agent and Transport agent's route selection ij , Choose Vehicle as your means of transportation m and transportation speed adjustment strategy V adjust , Suggestions for adjusting resource requirements of rescue scene agents and

[0110] The optimization results of the network flow optimization module: the optimized resource scheduling plan, that is, the resource transportation flow on each transportation route And other information.

[0111] The data in the digital twin scene are pushed to the multi-agent reinforcement learning module and the network flow optimization module respectively; the multi-agent decision information is passed to the digital twin scene for simulation evaluation, and the evaluation results are synchronized to the network flow optimization module; the optimization results of the network flow optimization module are fed back to the digital twin scene, and then passed to the multi-agent by the digital twin scene.

[0112] Establish data sharing interfaces and communication protocols to ensure real-time and accurate data transmission between digital twin scenarios, multi-agent reinforcement learning modules, and network flow optimization modules. For example, use technologies such as message queues or WebSocket to implement asynchronous data transmission and subscription publishing modes to ensure the timeliness and integrity of data.

[0113] (2) Collaborative decision-making and dynamic adjustment

[0114] Synchronized data and evaluation results of digital twin scenarios: such as simulation evaluation of the effects of multi-agent decisions (such as the possible resource arrival time caused by a certain resource allocation decision, the impact on the development of an accident, etc.), and simulation verification results of network flow optimization solutions in digital twin scenarios (such as whether there is a transportation bottleneck, whether resource allocation is reasonable, etc.).

[0115] The current status information of multiple agents: including the resource reserve change status S of the resource supply agent supply , the transportation process state S of the transportation agent transport , the latest state of the accident scene of the rescue scene agent S scene .

[0116] Optimization scheme information of network flow optimization module: such as resource transportation flow allocation Resource supply point allocation plan, etc.

[0117] Output dynamically adjusted decisions: Multiple agents adjust their decisions based on the feedback from the digital twin scenario and the network flow optimization results, such as the resource supply agent adjusting the amount and type of resource allocation and Transport Agent adjusts the transport route ij-new , transport speed V adjust-new , Rescue scene agent adjustment resource demand suggestions and The network flow optimization module further optimizes the resource scheduling scheme and generates new resource transportation flows based on the new decisions of multiple agents and the dynamic changes of digital twin scenarios. wait.

[0118] In the multi-agent reinforcement learning module, each agent recalculates the action selection strategy based on the new input state and the feedback information obtained from the digital twin scene and network flow optimization module. For example, the strategy update formula of the resource supply agent can be expressed as:

[0119]

[0120] where θ supply is the policy network parameter of the resource provisioning agent, α supply is the learning rate, is the policy function, For new resource allocation actions, For the new state, For new rewards (based on factors such as resource allocation effects evaluated by digital twin scenarios).

[0121] The network flow optimization module reconstructs the network flow model and performs optimization calculations based on the new decisions of multiple agents and the dynamic parameter changes in the digital twin scene (such as further deterioration or improvement of road conditions, sudden changes in resource requirements at the accident site, etc.). For example, when the transportation agent changes the transportation route because the digital twin scene indicates that the traffic conditions on a certain road section have deteriorated sharply, the network flow optimization module adjusts the parameters such as the edge capacity and transportation time related to the route accordingly, and re-solves the objective function:

[0122]

[0123] in The total cost, total time and degree of accident risk reduction calculated based on the new network flow model and accident development are respectively. The new resource transportation flow is obtained by solving To achieve dynamic optimization of resource scheduling solutions.

[0124] The entire collaborative decision-making and dynamic adjustment process is continuously cycled during the accident emergency response process until the accident is effectively controlled and handled. Each cycle is based on the latest data provided by the digital twin scene and the close collaboration between the modules, continuously optimizing the emergency resource scheduling strategy to adapt to the complexity, uncertainty and dynamic variability of chemical park accidents, maximizing resource scheduling efficiency, reducing accident losses and ensuring the safety and stability of the park.

[0125] 5. Emergency rescue and evacuation route planning

[0126] During the emergency response process of the chemical park, road conflicts or congestion are likely to occur due to the simple road network and weak traffic capacity of the park, resulting in evacuation failure or the inability of rescue vehicles to evacuate. Rescue / evacuation route planning comprehensively considers the necessary points and detour areas, so that the target object can find a collision-free safe path from the starting point to the end point within the specified scope of the park.

[0127] 5.1 Dynamic Grid Environment Modeling

[0128] The dynamic grid method is used for environmental modeling, and the environment of the chemical park is divided into small square cells, namely grids. Each grid represents an area in the environment and can be used to represent obstacles, free space and other information. By updating the status of the grid, an environmental map can be gradually constructed.

[0129] (1) Grid representation and initialization

[0130] Each grid has its corresponding state attribute, which can usually be defined as three basic states: idle (passable), obstacle (impassable), and unknown (initial undetected state).

[0131] Let S(G ijk ) represents the grid G ijk The state can be initialized as follows:

[0132]

[0133] At the beginning of the chemical park environment modeling, for those grids corresponding to the road areas that are obviously known, S(G ijk )=0; For the grids corresponding to the areas occupied by buildings, large chemical equipment, etc., S(G ijk )=1; and the grids corresponding to the areas that have not been detected by the sensor, S(G ijk )=-1.

[0134] The three-dimensional grid map M of the entire chemical park can be expressed as a collection of all grid states, namely:

[0135] M={S(G ijk)|i=1,2,...,I;j=1,2,...,J;k=1,2,...,K}

[0136] Among them, I, J, and K are the total number of raster maps in the three dimensions of rows, columns, and layers respectively.

[0137] (2) Grid status update based on sensor information

[0138] Assuming that the sensor detects the environment, there is a certain uncertainty in its detection results (for example, the sensor has a certain false alarm rate and missed alarm rate), and Bayesian filtering is used to update the probability estimate of the grid state.

[0139] Let P(S(G ijk )=s|Z 1:t ) indicates that after receiving a series of sensor observation values ​​Z from time 1 to time t 1:t After that, grid G ijk The probability of being in state s (s takes the value of 0, 1 or -1).

[0140] According to the Bayesian formula, the updating process is as follows:

[0141] 1) Prediction step (prior probability update):

[0142]

[0143] Here P(S(G ijk )=s|S(G ijk )=s',Z 1:t-1 ) represents the transition probability of the grid changing to state s at the current moment when the grid is in state s' and the previous observation value is known at the previous moment. It depends on factors such as the dynamic characteristics of the environment and the performance of the sensor;

[0144] P(S(G ijk )=s'|Z 1:t-1 ) is the probability that the previous grid is in state s', which comes from the update result of the previous round.

[0145] 2) Update step (update the posterior probability based on the current observation):

[0146]

[0147] Where P(Z t |S(G ijk )=s) means that in the grid G ijk When in state s, obtain the observation value Z at the current time t t The probability of this is determined by the observation model of the sensor.

[0148] For example, if the sensor detects an unknown (S(G ijk )=-1), there is an obstacle in the grid area (the observation result indicates an obstacle). Through the calculation process of the above Bayesian filter, the grid state can be updated to an obstacle (S(G ijk )=-1). When this probability exceeds a certain threshold (such as 0.8), it can be determined that its state is an obstacle.

[0149] (3) Grid map construction and visualization

[0150] In order to visualize the grid map in three-dimensional space, it is necessary to convert the row and column coordinates of the grid into actual three-dimensional space coordinates. Assume that the lengths of the sides of the grid in the x, y, and z directions are d x d y d z (Unit: meter or other length units), grid G ijk The coordinates of the lower left corner are (x ijk ,y ijk ,z ijk ), then:

[0151] x ijk =j×d x

[0152] y ijk =i×d y

[0153] z ijk =(k-1)×d z

[0154] In order to perform graphics drawing and other operations more accurately, it is necessary to obtain the center coordinates of the grid. Suppose the grid G ijk The center coordinates are (x ijk ,y ijk ,z ijk ), which is calculated as follows:

[0155]

[0156] 5.2 Mathematical modeling of bidirectional path planning

[0157] Emergency response is usually divided into two stages: internal emergency response within the park and coordinated emergency response within and outside the park. The goal is to plan the best two-way routes for emergency rescue and emergency evacuation at different stages to achieve high efficiency of rescue and evacuation while avoiding road conflicts and congestion.

[0158] The starting and ending points of emergency rescue and evacuation, the locations of nodes on the path, the travel time of each section, the rescue and evacuation flow, etc. are set as decision variables. By optimizing these variables, the optimal path and resource allocation plan are determined.

[0159] These include road capacity constraints, which is the maximum amount of traffic or pedestrian flow that each road section can accommodate; time constraints, such as completing rescue and evacuation within the specified emergency response time; and node flow balance constraints, which ensure that the flow entering and leaving each node is equal to ensure the continuity and rationality of the path.

[0160] (1) Basic elements of the model and variable definitions

[0161] 1) Spatial representation and grid definition

[0162] Assume that the three-dimensional space where the chemical park is located is discretized into a series of cubic grids, using G ijk to represent the grid located at the i-th row, j-th column, and k-th layer (i, j, and k are all positive integers).

[0163] 2) Decision variable setting

[0164] Introduce decision variable x ijk , used to represent the path planning situation. If a path passes through the grid G ijk , then x ijk =1; if the path does not pass through the grid, then x ijk =0.

[0165] 3) Definition of starting point and end point

[0166] Emergency rescue: Assume the coordinates of the emergency rescue starting point are (i r ,j r ,k r ), the corresponding rescue target end point coordinates are For example, the rescue starting point may be the grid location of the emergency rescue command center in the park, and the end point may be the grid corresponding to key rescue points such as the core area where the accident occurred.

[0167] Emergency evacuation: The coordinates of the emergency evacuation starting point are set as (i e ,j e ,k e ), the coordinates of the evacuation target endpoint are For example, the evacuation starting point is the grid position of the densely populated area in the park, and the end point is the grid corresponding to the safe assembly area outside the park.

[0168] 4) Definition of other related parameters

[0169] Distance metric: Represents the grid To Grid distance.

[0170] Travel time: Represents the grid To Grid The time required for passage is affected by many factors, such as the road conditions within the grid, the speed of people or vehicles, and whether there are dangerous situations.

[0171] Cost of passage: Used to measure from the grid To Grid The comprehensive travel cost is determined by taking into account many factors such as distance, concentration of hazardous substances, degree of road congestion, etc.

[0172] (2) Constraints

[0173] 1) Path connectivity constraints

[0174] In three-dimensional space, the grids on the path need to maintain connectivity, that is, for any two consecutive grids and They should be adjacent in space. For a three-dimensional grid, a grid usually has 26 neighboring grids (including the neighbors of the plane where it is located and the adjacent planes above and below).

[0175] Define the neighborhood function N(G i,j,k ) to represent the grid G i,j,k The mathematical formula of the path connectivity constraint is expressed as:

[0176]

[0177] This formula means that if a grid On the planned path (i.e. ), then at least one of its neighboring grids must also be on the path (that is, the corresponding The sum of is at least 1), thus ensuring that the planned path is continuous and traversable.

[0178] 2) Start and end point constraints

[0179] For emergency rescue paths and emergency evacuation paths, each has clear starting point and end point requirements, and the corresponding constraint formulas are as follows:

[0180] Emergency rescue path starting point constraints:

[0181]

[0182] The rescue path must start from the set rescue starting point, so the decision variable x corresponding to the starting point grid is assigned a value of 1.

[0183] Emergency rescue path endpoint constraints:

[0184]

[0185] To ensure that the rescue path can eventually reach the set rescue target destination, the decision variable x corresponding to the destination grid is also assigned a value of 1.

[0186] For emergency evacuation paths, there are similar constraints:

[0187] Emergency evacuation path starting point constraints:

[0188]

[0189] Emergency evacuation path endpoint constraints:

[0190]

[0191] 3) Flow conservation constraints (considering the situation where multiple agents are moving simultaneously)

[0192] When there are multiple rescue teams or multiple batches of evacuees planning routes at the same time, in order to ensure flow balance, the flow entering a grid should be equal to the flow leaving the grid.

[0193] Let f in (i,j,k) represents the flow into the grid, and (i,j,k) represents the flow out of the grid G ijk The mathematical expression of the flow conservation constraint is:

[0194]

[0195] In the specific calculation of f in (i,j,k) and f out (i,j,k), it needs to be determined based on the decision variable x entering and leaving the current grid through the neighborhood grid and the corresponding flow weight and other factors. The example is as follows:

[0196]

[0197] in and They represent the flow weights from different neighborhood grids to the current grid and from the current grid to different neighborhood grids respectively. These weights can be set according to factors such as road width, traffic capacity, and characteristics of different areas in the actual chemical park.

[0198] (3) Objective function

[0199] 1) Shortest path length goal

[0200] If the goal is to minimize the total length of the emergency rescue path and the emergency evacuation path, the following objective function can be constructed:

[0201]

[0202] This means that for all possible adjacent grid pairs, the distance between them is calculated and multiplied by the corresponding decision variable (this product will contribute to the total length only when both grids are on the path). Finally, all these products are summed up, and the result is the total length of the entire emergency rescue and emergency evacuation path. The goal is to minimize this total length.

[0203] 2) Shortest time goal

[0204] When it is desired to minimize the total time of emergency rescue and emergency evacuation, the objective function can be set as:

[0205]

[0206] here The calculation of (travel time) is usually determined based on the distance and the average travel speed of the corresponding road section. The specific calculation formula can be expressed as:

[0207]

[0208] in Represents the grid G i,j,k To Grid The average travel speed (units such as meters per second, etc.) of different areas (such as flat road areas, areas with obstacles, areas covered by hazardous substances, etc.) is different and needs to be set according to the specific conditions of each area in the actual chemical park.

[0209] 3) Comprehensively consider distance, danger level (for example, the danger level of different areas is measured based on factors such as the concentration of dangerous substances and high-temperature areas of fire), congestion and other factors to determine the travel cost When minimizing the total travel cost is used as the objective function:

[0210]

[0211] Cost of transportation The determination of can be achieved by constructing a cost function, for example:

[0212]

[0213] in Represents the grid To Grid Danger level indicators (which can be quantified based on actual relevant data such as concentration of dangerous substances and temperature), It represents the congestion index of the road section (which can be obtained through real-time monitoring data or estimation). α, β, and γ are corresponding weight coefficients, which are used to adjust the proportion of different factors in the total cost. Their values ​​can be reasonably set according to the environmental characteristics of the actual chemical park and the focus of emergency decision-making.

[0214] (4) Ant Colony Algorithm

[0215] The mathematical model for bidirectional path planning of emergency rescue and emergency evacuation in the three-dimensional environment constructed above usually belongs to integer programming problem (because the decision variable x i,j,k The value can only be 0 or 1).

[0216] For smaller problems, precise algorithms such as branch and bound and cutting plane methods can be used to solve them in order to obtain the optimal path planning solution. However, in actual chemical park scenarios, since the number of grids is often extremely large, the use of precise algorithms to solve them may face the problem of low computational efficiency and it is difficult to meet the real-time requirements of the emergency response stage.

[0217] Therefore, the actual application adopts the heuristic algorithm ant colony algorithm to find the optimal path planning scheme within an acceptable time range through continuous iterative search, so as to provide effective path guidance for the emergency rescue and emergency evacuation work in the emergency response stage of the chemical park, and ensure that the rescue and evacuation operations can be carried out efficiently and safely.

[0218] Through such comprehensive and detailed mathematical modeling of two-way path planning, various complex factors in emergency scenarios in chemical parks under three-dimensional environments are fully considered. Combined with corresponding objective functions, constraints and appropriate solution methods, it helps to plan scientific, reasonable and feasible emergency rescue and emergency evacuation routes.

[0219] 5.3 Intelligent obstacle avoidance model in dynamic environment

[0220] (1) Model basis and definition of related variables

[0221] 1) Space and grid representation

[0222] The three-dimensional space of the chemical park is divided into a series of cube-shaped grids G ijk (i, j, k represent row, column, and layer indices respectively), each grid has multiple attributes, such as whether it is an obstacle (with o ijk Indicates that ijk =1 means it is an obstacle, o ijk = 0 means no obstacle), concentration of dangerous substances c ijk , temperature t ijkEtc. These attributes will affect the passage and obstacle avoidance decisions of intelligent agents (such as rescue vehicles, evacuees, etc.).

[0223] 2) Agent-related variables

[0224] Assume the current grid position of the agent is (i α ,j α ,k α ), whose target position is (i t ,j t ,k t ), the moving speed vector of the agent is expressed as where v x 、v y 、v z are the velocity components of the agent in the x, y, and z directions respectively.

[0225] 3) Sensor sensing range and angle

[0226] Assume that the agent is equipped with sensors whose sensing range is R (unit: meter or other length units) in radius and [α min ,α max ], and in the vertical direction [β min ,β max ]. This means that the agent can only obtain environmental information within a certain angle and distance range to make obstacle avoidance decisions.

[0227] (2) Dynamic obstacle avoidance adjustment combined with path planning

[0228] 1) Path replanning trigger conditions

[0229] When the agent is moving along the planned path, if it encounters new obstacles, changes in dangerous areas, or other situations that make the original path impassable, it is necessary to trigger path replanning. For example, a threshold can be set. When the distance between the agent and the new obstacle is less than the threshold and the original path cannot be avoided, or when the concentration of dangerous substances exceeds the safety standard and the original path is unsafe, the path replanning mechanism is activated.

[0230] 2) Path re-planning and obstacle avoidance coordination

[0231] Once re-planning is triggered, the path needs to be re-planned in combination with the current obstacle avoidance situation and environmental information. The mathematical modeling method of bidirectional path planning mentioned above can be used, but at this time, the current position, speed and surrounding obstacle avoidance related information of the agent should be taken into account as constraints. For example, the obstacle avoidance related cost items are added to the objective function of path planning, so that the newly planned path not only meets the conventional optimization goals such as distance and time, but also avoids high-risk obstacle areas and dangerous areas as much as possible. The objective function can be modified to a form similar to the following (taking the comprehensive cost minimization goal as an example, the obstacle avoidance cost is added to the original basis):

[0232]

[0233] where c avoid (i,j,k,i2j2k2) represents the grid G ijk To Grid The obstacle avoidance cost can be determined based on factors such as the obstacle density and concentration of hazardous substances in the area. Through such a collaborative mechanism, the intelligent agent can continuously and effectively avoid obstacles and move towards the target in a dynamic environment.

[0234] The intelligent obstacle avoidance model composed of the above formulas and mechanisms can enable intelligent entities (rescue vehicles, evacuees, etc.) to effectively avoid obstacles and dangerous areas in the complex three-dimensional dynamic environment of the chemical park, ensuring the smooth progress of emergency rescue and emergency evacuation.

[0235] 6. Establishment of a scientific evaluation system

[0236] Collect and analyze drill data in real time, including the response time of emergency rescue forces, the rationality of resource allocation, the effect of accident control, etc. Formulate scientific evaluation standards and indicator systems, evaluate each link and overall effect of the drill by comparing with the preset evaluation standards and indicator systems, and provide feedback information to the drill organizers and participants in a timely manner, point out the problems and deficiencies in the drill, and put forward suggestions for improvement. For example, if it is found that the response time of the emergency rescue team is too long, it may be due to unreasonable traffic route planning or poor dispatching and command. The feedback module will prompt the drill organizer to adjust the rescue route or optimize the dispatching strategy. At the same time, the drill organizer can make real-time adjustments to the drill scene or emergency response strategy during the drill based on the feedback information, such as increasing rescue forces, changing rescue tactics, etc., to observe the impact of different strategies on the drill results and achieve dynamic optimization of the drill.

[0237] To sum up, the present invention creates technical solutions such as constructing a high-precision digital twin model, fusion processing of IoT equipment data, accurate simulation of emergency scenarios, efficient scheduling of emergency resources, rescue and evacuation route planning, real-time response and evaluation, etc., which comprehensively improves the emergency response capability and safety management level of the chemical park.

[0238] The chemical park emergency response method based on digital twin created by the present invention can construct a multi-level, multi-dimensional and multi-grained chemical park emergency event handling model under different scenarios, allowing more alternative plans or decision theories that cannot be verified due to physical space limitations and reliance on real physical entities to become possible through continuous trial and error and optimization.

[0239] (1) Significantly improve the fidelity and accuracy of emergency simulation

[0240] Accurately simulate the development process of various emergency events in chemical parks, including fire spread, toxic gas diffusion, personnel evacuation, etc., to improve the realism and accuracy of simulation exercises.

[0241] (2) Significantly reduce the cost of emergency simulation drills

[0242] The present invention can realize comprehensive simulation and drill of emergency events without investing a lot of space, equipment and manpower costs. By reusing simulation scenes and data, the cost of drills is significantly reduced, while the efficiency and effect of drills are improved.

[0243] (3) Significantly enhance the real-time and effectiveness of emergency response

[0244] With the help of visualization technology and data analysis tools, the present invention can monitor and respond to emergency events occurring in the park in real time through a data acquisition and transmission system, intuitively display the development process and impact range of emergency events, and provide intuitive and clear information support for emergency decision-making.

[0245] The comprehensive emergency simulation and drill system and method for chemical parks based on high-precision digital twin technology takes high-precision digital twin technology as the core and combines advanced technologies such as the Internet of Things, big data, and cloud computing to build a virtual simulation environment that corresponds one-to-one with the physical environment of the chemical park. It can simulate various emergency accident scenarios and provide strong support for emergency management and safety training in chemical parks.

[0246] Build a high-precision digital twin model: Conduct high-precision digital modeling of the physical environment of the chemical park to accurately restore the equipment, facilities, buildings, etc. within the park, and perceive and reflect dynamic changes within the park in real time, such as equipment operating status, personnel flow, etc.

[0247] Scenario simulation: Provides a variety of chemical park emergency scenarios, including fire, explosion, chemical leakage, etc. Different scenarios can be selected for drills according to actual needs.

[0248] Resource scheduling: Build a chemical park accident emergency resource scheduling model, integrate digital twin simulation technology and hybrid intelligent optimization algorithm, realize efficient resource (such as fire-fighting equipment, medical rescue materials, protective equipment, etc.) allocation in dynamic and complex environments, and significantly improve scheduling response speed, resource utilization and accident response efficiency.

[0249] Rescue route planning: Comprehensively consider the must-pass points and detour areas, so that the target object can find a collision-free and safe path from the starting point to the end point within the specified scope of the park.

[0250] Evaluation and feedback: After the drill is completed, the drill process is evaluated and an evaluation report is generated, including drill time, task completion, personnel performance, etc. Users can summarize lessons learned and improve emergency plans based on the evaluation report.

Claims

1. The chemical park emergency response method based on digital twin is characterized by The steps include: Step S1: Build a high-precision digital twin model: Conduct high-precision digital modeling of the physical environment of the chemical park to accurately restore the equipment, facilities, and buildings in the park, and perceive and reflect the dynamic changes in the park in real time; Step S2: The IoT device model is integrated with the twin scene to match the IoT device model with the device information of the actual park enterprise, thus realizing the coordinated linkage between the model and the monitoring equipment; Step S3: Provide multiple chemical park emergency scenario simulations, including at least fire, explosion, and chemical leakage, and select different scenarios for drills according to actual needs; Step S4: Construct a chemical park accident emergency resource scheduling model, integrating digital twin simulation technology and hybrid intelligent optimization algorithm to achieve resource allocation in dynamic and complex environments; Step S5: Comprehensively consider the must-pass points and detour areas so that the target object can find a collision-free safe path from the starting point to the end point within the specified range of the park.

2. The chemical park emergency treatment method based on digital twin according to claim 1 is characterized in that The emergency resource scheduling model in step S4 includes constructing an interaction mechanism between the intelligent agent and the digital twin scene, and the specific process is as follows: Step S41-1: Interaction mechanism between resource supply agent and digital twin scene: Through the digital twin scenario, accurate resource reserve information of resource supply points is obtained in real time, including at least the quantity, storage location, and validity period of each resource. At the same time, resource allocation plans are decided based on the prediction of resource consumption trends by the digital twin model and the reinforcement learning algorithm. Resource reserve information in the digital twin scenario: including the current actual quantity of various resources at each supply point Store location coordinates; State Space: Where K is the total number of resource types; Action Space: Types of resources deployed at each supply point and quantity Resource replenishment time information That is, the estimated time from the issuance of a resource replenishment request to the actual replenishment of resources; Resource consumption rate predicted by digital twin The consumption rate of each resource per unit time is predicted based on the type, scale and development trend of the accident; Reward function R: 1) Rescue time reward function: Where T windows The best rescue time window for accidents simulated by digital twins. is the completion time of the i-th rescue subtask; 2) Resource Utilization Reward Function: in is the demand for the kth resource predicted by the digital twin, and δ is the allowable error range; Using the reinforcement learning algorithm, the agent is based on the current state S supply , through the policy network π θ (a|s) Select action A supply , and learn and update according to the reward function R, the goal is to maximize the long-term cumulative reward γ is the discount factor; Step S41-2: Interaction mechanism between the transport agent and the digital twin scene: Based on the dynamic information of the road network in the digital twin scene, the optimal transportation route is planned and the transportation strategy is adjusted. The interaction between the transportation agent and the digital twin scene also includes real-time simulation of the transportation process, predicting the problems that may be encountered in different transportation plans and avoiding them in advance. Road network information G = (V, E) in the digital twin scenario: including nodes V and edges E, as well as the basic attributes of each edge; Real-time traffic conditions C ij : Provided by the integration of IoT sensing devices and digital twin scenarios, including the congestion level of each road, traffic speed v ij The congestion coefficient ρ is used to describe the congestion degree. ij Indicates that 0≤ρ ij≤1 , 0 means unobstructed, 1 means completely blocked; The location of the transport means L t and state S t : The real-time geographic location coordinates of the transport vehicle and the operating status of the vehicle obtained through IoT sensing devices; Real-time weather conditions in digital twin scenarios condition : Including wind speed ω speed 、Wind direction direction , rainfall intensity r intensity ,These factors will affect transportation safety and efficiency; Potential interference of other related accidents in the surrounding area on the transportation route A interferense ; State space: S transport = {G,C ij ,L t ,S t ,W condition ,A interferense }; Action Space: A transport ={Route ij ,Vehicle m ,V adjust }; Reward function: Transportation cost reward function Among them C transport =∑ i (C fuel-i +C depreciation-i +C labor-i ), C fuel-i =S ij ×c fuel ×(1+α ω ×ω speed +α r × intensity );where c fuel is the fuel consumption cost per unit distance, α ω and α r is the coefficient of the effect of wind speed and rainfall intensity on fuel consumption, ΔC transport C is the cost increase caused by the dynamic change of road conditions. budget Budget for pre-set transportation costs; Using reinforcement learning algorithm, according to S transport State selection action A transport , and perform learning updates based on the reward function to optimize the transportation strategy; Step S41-3: Interaction mechanism between the rescue scene agent and the digital twin scene: By using the digital twin scene to fully perceive the details of the accident scene, the rescue scene agent can more accurately assess resource requirements through in-depth interaction with the digital twin scene and provide more targeted resource requirement adjustment suggestions to other agents; Accident information in the digital twin scenario: including accident type T accident , real-time development information of the accident; Usage of existing resources at the scene: Types of resources that have arrived at the accident scene and its quantity The progress of resource usage; Accident development trend prediction based on digital twin scenario simulation accident-trend ; Resource demand adjustment suggestions: Propose the types of resources that need to be allocated to the resource supply agent and the transportation agent and quantity Suggestions for adjustments; State Space: The severity of the accident is S severity According to the quantitative evaluation of the accident impact range and the amount of dangerous substance leakage, the rescue progress rescue Expressed as percentage of rescue missions completed or key indicators; Action Space: Reward function: Accident Loss Reward Function Where L accident-predict Real-time assessment and prediction of accident losses for digital twin scenarios, including casualties prediction casaultu-predict , Property loss prediction property-predict and environmental damage prediction environment-predict , L max is the maximum acceptable accident loss threshold; based on Reinforcement learning algorithm, based on the state S scene Generate action A scene , and learns and optimizes according to the reward function to improve rescue efficiency and reduce accident losses.

3. The chemical park emergency treatment method based on digital twin according to claim 2 is characterized in that The above step S4 integrates digital twin simulation technology and hybrid intelligent optimization algorithm, including establishing a network flow optimization module enabled by digital twins; the specific process is as follows: Step S42-1: Network construction and dynamic parameter update Resource supply point information in the digital twin scenario: including the location coordinates of the supply point and the upper limit of resource reserves And the proportion of supply capacity reduction caused by surrounding accidents β i , β i >0; Accident site information: accident location coordinates, dynamic fluctuations in resource demand predicted by digital twin scenarios Road network information: including the network node set V' and edge set E', as well as the dynamic changes in road capacity monitored by the digital twin scenario Real-time adjustment of transportation time Updated network flow model: including the adjusted network node set V′, edge set E′, edge capacity matrix C′, transportation time matrix T′, and resource demand vector Q required ; Model formula and logical relationship: Construct a network N = (V′, E′, C′, T′, D′); where D′ is the dynamic parameter matrix driven by the digital twin scenario, including and information; Step S42-2: Dynamic adjustment of optimization objectives and constraints Accident risk reduction degree R of digital twin scenario assessment risk-reduce :The reduction ratio of accident diffusion range simulated by digital twin scenario α spread-reduce 、Reduction ratio of dangerous substance concentration α concentration-reduce Quantitative evaluation of indicators; The final output is the optimized resource scheduling plan: determine the resource transportation flow on each transportation route The final allocation of resource supply points to meet the optimization objectives and constraints; Model formula and logical relationship: Objective function Z = λC total +(1-λ)T total +μR risk-reduce ;in, The calculation is as described in the transportation agent section, T total =∑ (i,j)∈∈ T′ ij x ij ; Constraints: 1) Resource supply constraints: 2) Resource demand constraints: 3) Flow conservation constraint: For the transit nodes k∈V′ except the resource supply point and the accident site, ∑ i:(i,k)∈E′ x ik ≤∑ j:(k,j)∈E′ x kj , dynamically adjust the calculation scope and node relationships when considering new nodes or node failures that may appear in the digital twin scenario; 4) Capacity Limit Constraints: Adaptive network flow optimization algorithm is used to gradually approach the optimal solution by continuously searching for augmenting paths and adjusting network flows to meet the objective function and constraints.

4. The chemical park emergency treatment method based on digital twin according to claim 3 is characterized in that The above step S4 integrates digital twin simulation technology and hybrid intelligent optimization algorithm, and also includes establishing a deep collaboration mechanism between digital twin, intelligent agent and network flow; the specific process is: Step S43-1: Real-time data sharing and synchronization Real-time data in digital twin scenarios: including status data updated every second at the accident scene, real-time traffic status data of the road network, and resource reserve and consumption data updated every second at resource supply points; Decision information of multiple agents: resource allocation decision of resource supply agent and Transport agent's route selection ij , Choose Vehicle as your means of transportation m and transportation speed adjustment strategy V adjust , Suggestions for adjusting resource requirements of rescue scene agents and The optimization results of the network flow optimization module: the optimized resource scheduling plan, that is, the resource transportation flow on each transportation route Push the data in the digital twin scene to the multi-agent reinforcement learning module and the network flow optimization module respectively; The multi-agent decision information is passed to the digital twin scene for simulation evaluation, and the evaluation results are synchronized to the network flow optimization module; Feedback the optimization results of the network flow optimization module to the digital twin scene, which is then passed to multiple agents; Step S43-2: Collaborative decision making and dynamic adjustment Synchronous data and evaluation results of digital twin scenarios: simulation evaluation of multi-agent decision-making effects, simulation verification results of network flow optimization solutions in digital twin scenarios; The current status information of multiple agents: including the resource reserve change status S of the resource supply agent supply , the transportation process state S of the transportation agent transport , the latest state of the accident scene of the rescue scene agent S scene ; Optimization scheme information of network flow optimization module: resource transportation flow allocation Resource supply point deployment plan; Output dynamically adjusted decisions: Multiple agents adjust their decisions based on the feedback from the digital twin scenario and the network flow optimization results, including the resource supply agent adjusting the amount and type of resource allocation and Transport Agent adjusts the transport route ij-new , transport speed V adjust-new , Rescue scene agent adjustment resource demand suggestions and The network flow optimization module further optimizes the resource scheduling scheme and generates new resource transportation flows based on the new decisions of multiple agents and the dynamic changes of digital twin scenarios. In the multi-agent reinforcement learning module, each agent recalculates the action selection strategy based on the new input state and the feedback information obtained from the digital twin scene and network flow optimization module; the strategy update formula of the resource supply agent is expressed as: where θ supply is the policy network parameter of the resource provisioning agent, α supply is the learning rate, is the policy function, For new resource allocation actions, For the new state, For new rewards; The network flow optimization module reconstructs the network flow model and performs optimization calculations based on the new decisions of the multi-agents and the dynamic parameter changes in the digital twin scenario; the network flow optimization module adjusts the parameters of the edge capacity and transportation time related to the route accordingly, and re-solves the objective function: in The total cost, total time and degree of accident risk reduction calculated based on the new network flow model and accident development are respectively. The new resource transportation flow is obtained by solving To achieve dynamic optimization of resource scheduling solutions.

5. The chemical park emergency treatment method based on digital twin according to claim 1 is characterized in that The above step S5 finds a collision-free safe path, including emergency rescue and evacuation path planning, and the specific process is as follows: During the emergency response process of the chemical park, road conflicts or congestion are likely to occur due to the simple road network and weak traffic capacity of the park, resulting in evacuation failure or the inability of rescue vehicles to evacuate. Rescue / evacuation route planning comprehensively considers the necessary points and detour areas, so that the target object can find a collision-free safe path from the starting point to the end point within the specified scope of the park. Step S51-1 Dynamic grid method environment modeling The dynamic grid method is used for environmental modeling, and the environment of the chemical park is divided into small square cells, namely grids. Each grid represents an area in the environment and is used to represent information about obstacles and free space. By updating the status of the grid, an environmental map is gradually constructed. Step S51-2: Grid representation and initialization Each grid has its corresponding state attribute, which can usually be defined as three basic states: idle, obstacle and unknown; Let S(G ijk ) represents the grid G ijk The state is initialized as follows: At the beginning of the chemical park environment modeling, for those grids corresponding to the road areas that are obviously known, S(G ijk )=0; for the grids corresponding to the areas occupied by buildings and large chemical equipment, S(G ijk )=1; and the grids corresponding to the areas that have not been detected by the sensor, S(G ijk )=-1; The three-dimensional grid map M of the entire chemical park is represented as a collection of all grid states, namely: M={S(G ijk )|i=1,2,...,I;j=1,2,...,J;k=1,2,...,K} Among them, I, J, and K are the total number of raster maps in the three dimensions of rows, columns, and layers; Step S51-3: Grid status update based on sensor information Assuming that the sensor detects the environment, there is a certain degree of uncertainty in its detection results, and Bayesian filtering is used to update the probability estimate of the grid state; Let P(S(G ijk )=s|Z 1:t ) indicates that after receiving a series of sensor observation values ​​Z from time 1 to time t 1:t After that, grid G ijk The probability of being in state s; s takes the value of 0, 1 or -1; According to the Bayesian formula, the updating process is as follows: 1) Prediction step, i.e. prior probability update: Here P(S(G ijk )=s|S(G ijk )=s',Z 1:t-1 ) represents the transition probability of the grid changing to state s at the current moment when the grid is in state s' and the previous observation value is known at the previous moment, which depends on factors such as the dynamic characteristics of the environment and the performance of the sensor; P(S(G ijk )=s'|Z 1:t-1 ) is the probability that the previous grid is in state s', which comes from the update result of the previous round; 2) Update step, that is, update the posterior probability based on the current observation: Where P(Z t |S(G ijk )=s) means that in the grid G ijk When in state s, obtain the observation value Z at the current time t t The probability of , which is determined by the sensor's observation model; Step S51-4: Grid map construction and visualization In order to visualize the grid map in three-dimensional space, it is necessary to convert the row and column coordinates of the grid into actual three-dimensional space coordinates; let the lengths of the sides of the grid in the x, y, and z directions be d x d y d z , grid G ijk The coordinates of the lower left corner are (x ijk ,y ijk ,z ijk ), then: x ijk =j×d x and ijk =i×d y With ijk =(k-1)×d z In order to perform graphics drawing and other operations more accurately, it is necessary to obtain the center coordinates of the grid. Suppose the grid G ijk The center coordinates are (x ijk ,y ijk ,z ijk ), which is calculated as follows:

6. The chemical park emergency treatment method based on digital twin according to claim 5 is characterized in that Finding a collision-free safe path in the above step S5 also includes mathematical modeling of bidirectional path planning, dividing the emergency response into two stages: internal emergency response in the park and coordinated emergency response inside and outside the park; the goal is to plan the optimal bidirectional path for emergency rescue and emergency evacuation respectively at different stages to achieve high efficiency of rescue and evacuation, while avoiding road conflicts and congestion; the specific process is: Step S52-1: Model basic elements and variable definitions 1) Spatial representation and grid definition Assume that the three-dimensional space where the chemical park is located is discretized into a series of cubic grids, using G ijk to represent the grid at the i-th row, j-th column, and k-th layer, where i, j, and k are all positive integers; 2) Decision variable setting Introduce decision variable x ijk , used to represent the path planning situation; If a path passes through the grid G ijk , then x ijk =1; If the path does not pass through the grid, then x ijk =0; 3) Definition of starting point and end point Emergency rescue: Assume the coordinates of the emergency rescue starting point are (i r ,j r ,k r ), the corresponding rescue target end point coordinates are Emergency evacuation: The coordinates of the emergency evacuation starting point are set as (i e ,j e ,k e ), the coordinates of the evacuation target endpoint are 4) Definition of other related parameters Distance metric: Represents the grid To Grid distance; Travel time: Represents the grid To Grid The time required for the journey; Cost of passage: Used to measure from the grid To Grid The comprehensive travel cost is determined by comprehensively considering the distance, concentration of hazardous substances, and degree of road congestion to determine the specific value; Step S52-2: Constraints 1) Path connectivity constraints In three-dimensional space, the grids on the path need to maintain connectivity, that is, for any two consecutive grids and They should be adjacent in space. For a three-dimensional grid, a grid usually has 26 neighboring grids, that is, the neighborhood of the plane where the grid is located and the adjacent planes above and below. Define the neighborhood function N(G i,j,k ) to represent the grid G i,j,k The mathematical formula of the path connectivity constraint is expressed as: This formula means that if a grid On the planned path (i.e. ), then at least one of its neighboring grids must also be on the path, that is, the grid corresponding to its neighboring grid The sum of is at least 1, which ensures that the planned path is continuous and traversable; 2) Start and end point constraints For emergency rescue paths and emergency evacuation paths, each has clear starting point and end point requirements, and the corresponding constraint formulas are as follows: Emergency rescue path starting point constraints: The rescue path must start from the set rescue starting point, so the decision variable x corresponding to the starting point grid is assigned a value of 1. Emergency rescue path endpoint constraints: To ensure that the rescue path can eventually reach the set rescue target destination, the decision variable x corresponding to the destination grid is also assigned a value of 1; For the emergency evacuation path, the starting point constraints of the emergency evacuation path are: Emergency evacuation path endpoint constraints: 3) Flow conservation constraint, that is, considering the simultaneous movement of multiple agents When there are multiple rescue teams or multiple groups of evacuees planning routes at the same time, in order to ensure flow balance, the flow entering a grid should be equal to the flow leaving the grid; Let f in (i,j,k) represents the flow into the grid, and (i,j,k) represents the flow out of the grid G ijk The mathematical expression of the flow conservation constraint is: In the specific calculation of f in (i,j,k) and f out (i,j,k), it is necessary to determine based on the decision variable x entering and leaving the current grid through the neighborhood grid and the corresponding flow weight and other factors: in and Respectively represent the flow weights from different neighborhood grids to the current grid and from the current grid to different neighborhood grids; Step S52-3: Target function 1) Shortest path length goal If the goal is to minimize the total length of the emergency rescue path and the emergency evacuation path, the following objective function is constructed: This means that for all possible adjacent grid pairs, the distance between them is calculated and multiplied by the corresponding decision variables. Finally, all these products are summed up. The result is the total length of the entire emergency rescue and emergency evacuation path. The goal is to minimize this total length. 2) Shortest time goal When it is desired to minimize the total time of emergency rescue and emergency evacuation, the objective function is set as: Passing time The calculation of is usually determined based on the distance and the average speed of the corresponding road section. The specific calculation formula is expressed as: in Represents the grid G i,j,k To Grid The average speed of travel; 3) Comprehensively consider the distance, danger level, and congestion to determine the travel cost When minimizing the total travel cost is used as the objective function: Cost of transportation The determination of can be achieved by constructing a cost function, for example: in Represents the grid To Grid The risk index, It represents the congestion index of the road section. α, β, and γ are the corresponding weight coefficients, which are used to adjust the proportion of different factors in the total cost.

7. The chemical park emergency treatment method based on digital twin according to claim 6 is characterized in that Finding a collision-free safe path in the above step S5 also includes establishing a dynamic environment intelligent obstacle avoidance model. The specific process is as follows: Step S53-1: Model basis and related variable definition 1) Space and grid representation The three-dimensional space of the chemical park is divided into a series of cube-shaped grids G ijk , i, j, and k represent row, column, and layer indexes, respectively, and each grid has multiple attributes; 2) Agent-related variables Assume the current grid position of the agent is (i α ,j α ,k α ), whose target position is (i t ,j t ,k t ), the moving speed vector of the agent is expressed as where v x 、v y 、v z are the velocity components of the agent in the x, y, and z directions respectively; 3) Sensor sensing range and angle Assume that the agent is equipped with a sensor whose sensing range is R in radius and [α min ,α max ], and in the vertical direction [β min ,β max ]; Step S53-2: Dynamic obstacle avoidance adjustment combined with path planning 1) Path replanning trigger conditions When the agent is moving along the planned path, if it encounters new obstacles, changes in dangerous areas, or other situations that make the original path impassable, it is necessary to trigger path replanning. Set a threshold. When the distance between the agent and the new obstacle is less than the threshold and the original path cannot be avoided, or when the concentration of dangerous substances exceeds the safety standard and makes the original path unsafe, the path replanning mechanism is activated. 2) Path re-planning and obstacle avoidance coordination Once re-planning is triggered, the path needs to be re-planned in combination with the current obstacle avoidance situation and environmental information, and a two-way path planning mathematical modeling method is used. However, the current position, speed, and surrounding obstacle avoidance related information of the agent should be taken into consideration as constraints at this time; the objective function is modified to be similar to the following form, taking the comprehensive cost minimization objective as an example, and adding the obstacle avoidance cost on the original basis: where c avoid (i,j,k,i2j2k2) represents the grid G ijk To Grid The obstacle avoidance cost is determined according to the obstacle density and concentration of hazardous substances in the area. Through such a collaborative mechanism, the intelligent agent can continuously and effectively avoid obstacles and move towards the target in a dynamic environment.

Images