An airport emergency resource dynamic scheduling system based on multi-modal data

Abstract

The application discloses an airport emergency resource dynamic scheduling system based on multi-modal data and belongs to the field of intelligent operation scheduling of civil aviation airports. The system comprises a multi-modal data acquisition and fusion preprocessing module, an emergency resource demand prediction module, an airport multi-dimensional operation safety risk assessment module, an emergency resource intelligent optimization scheduling module, a passenger emergency evacuation path planning module, an emergency scene multi-agent simulation deduction module, an instruction issuing and operation iteration module, multi-modal data global fusion, multi-algorithm collaborative linkage, covering resource prediction, risk assessment, people and vehicle scheduling, passenger evacuation, and preplan simulation full link, solving the problems of single data and isolated algorithm of traditional scheduling systems, and the like. Double models are combined to predict emergency resource demand, random forests are adapted to complex nonlinear support services, ARIMA is used to accurately capture flight delays and passenger flow time sequence fluctuations, and under extreme weather and emergencies, resource gaps can be recalculated in seconds to avoid resource shortages or idling.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation scheduling technology for civil aviation airports, and in particular to a dynamic scheduling system for airport emergency resources based on multimodal data. Background Technology

[0002] In both routine airport operations and emergency scenarios, ground support personnel, baggage carts, fire trucks, maintenance vehicles, security screening equipment, navigation equipment, and evacuation guidance personnel are all core emergency resources. The existing airport resource allocation model has significant shortcomings: Resource requirements rely on manual experience for estimation, lacking quantitative forecasting models. In the event of thunderstorms, widespread delays, or sudden safety incidents, it is impossible to quickly assess manpower and vehicle shortages, which can easily lead to resource shortages causing widespread flight delays or idle resources resulting in wasted operating costs. Traditional scheduling only allocates resources statically based on fixed flight plans, which cannot adapt to the dynamic changes in flights under emergency scenarios.

[0003] Ground special vehicles rely on manual command by controllers for their routes, without overall optimal route planning. Emergency rescue vehicles are prone to traffic congestion on the tarmac, delaying response time. At the same time, there is a lack of intelligent personnel and vehicle coordination and scheduling tools. Adjusting personnel and vehicles in emergency situations is time-consuming and cannot meet the needs of short-term surges in support.

[0004] The lack of real-time linkage between the airport operation safety risk, airspace obstacle intrusion, navigation equipment failure, and aircraft ground dangerous approach hazards and other hidden dangers makes it easy for secondary safety accidents to occur during emergency response. There is also a lack of multi-dimensional risk assessment methods to constrain resource allocation strategies.

[0005] Passenger emergency evacuation and diversion routes rely on fixed signs and are not planned in accordance with real-time passenger flow dynamics. This results in low evacuation efficiency in congested areas and exacerbates the pressure on emergency response.

[0006] Emergency response plans rely solely on manual simulations, making it impossible to quantitatively assess the effectiveness of different resource allocation schemes. The lack of simulation tools to verify the merits of scheduling schemes in advance results in numerous bottlenecks in the emergency response process.

[0007] The existing dispatch system only collects flight time-series data and does not integrate multimodal heterogeneous data such as aircraft spatial location, terrain and airspace clearance, equipment operating status, weather, and personnel location. The single data dimension leads to low accuracy of dispatch decisions, and various dispatch algorithms are independent of each other and cannot work together to complete the entire chain of emergency resource dispatch.

[0008] Therefore, an integrated emergency resource dynamic scheduling system is needed that integrates multimodal data, multiple prediction, evaluation, optimization, and simulation algorithms, and covers the entire process of resource demand prediction, safety risk verification, optimization of personnel and vehicle routes and scheduling, passenger evacuation, and contingency plan simulation. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention provides a dynamic scheduling system for airport emergency resources based on multimodal data, which solves the defects of existing airport emergency scheduling technologies, such as inaccurate resource prediction, delayed scheduling response, inability to provide early warning of safety risks, lack of simulation verification of solutions, inability to coordinate multiple types of algorithms, and inability to integrate and utilize multimodal data.

[0010] Technical Solution: To solve the above-mentioned technical problems, according to one aspect of the present invention, more specifically, an airport emergency resource dynamic scheduling system based on multimodal data, comprising seven functional modules: multimodal data acquisition and fusion preprocessing module, emergency resource demand prediction module, airport multi-dimensional operational safety risk assessment module, emergency resource intelligent optimization scheduling module, passenger emergency evacuation route planning module, emergency scenario multi-agent simulation and deduction module, and instruction issuance and operation and maintenance iteration module; The modules work together to take multimodal fusion data as input and output dynamic emergency resource dispatch instructions to the control tower, AOCC operation and control center, ground support, maintenance and security departments.

[0011] Furthermore, a multimodal data acquisition and fusion preprocessing module This module serves as the data foundation for this system, responsible for the acquisition of heterogeneous multimodal data across all dimensions, timeline synchronization, cleaning and normalization, and providing standardized input for all backend algorithm models; Data source classification (multimodal data): (1) Time-series business data: flight schedule, historical flight takeoff and landing / delay records, passenger flow / baggage throughput time-series data, personnel shift records, special vehicle operation logs; (2) Spatial geographic data: airport runway / apron / terminal layout, obstacle terrain elevation, aircraft GPS coordinates, vehicle real-time location, security checkpoint / boarding gate spatial coordinates; (3) Environmental meteorological data: real-time wind speed, rainfall, visibility, season, holidays, and schedules of major events; (4) Equipment status data: navigation facility signal parameters, special vehicle fault status, and security inspection equipment operating parameters; (5) Image and dynamic target data: apron monitoring screens, aircraft taxiing course, and real-time location of ground personnel; Data preprocessing workflow: Unify the timestamp alignment of all data sources, perform missing value imputation, outlier removal, and feature normalization; perform feature extraction on spatial image data, and perform difference and variance stabilization transformation on time series data; construct a unified feature dataset, distinguish between static features (airport geography, building, facility parameters) and dynamic real-time features (flight dynamics, vehicle location, weather, passenger flow), and push them synchronously to all backend algorithm modules.

[0012] Furthermore, an emergency resource demand forecasting module The system integrates the random forest prediction sub-model and the ARIMA flight situation assessment sub-model to dually calculate the demand for manpower, vehicles, and equipment resources in four emergency scenarios: routine operations, extreme weather, emergencies, and large-scale events. Random Forest Prediction Sub-model (Flight Support Resource Prediction) (1) Feature inputs: flight volume, weather, seasonal holidays, historical resource usage records; target variables are the demand for various resources such as ground support, security, baggage carts, and maintenance vehicles; (2) Model construction: Based on decision trees, samples are selected by self-sampling, feature subsets are randomly selected to complete node splitting, and multiple decision trees are integrated to construct a random forest; (3) Training and optimization: Historical operating data is used for training, with MSE (mean squared error) and R (reduction coefficient) as the benchmarks. 2 The coefficient of determination is used to assess accuracy; the model is optimized by adjusting the number of trees, the maximum tree depth, and the feature selection rules. (4) Emergency adaptation logic: When sudden events such as flight schedule changes, severe weather, or equipment failures trigger model recalculation, the predicted value of resource demand is quickly updated, and the shortage of personnel and vehicles is output.

[0013] ARIMA Flight Situation Analysis Submodel (1) Data input: flight takeoff and landing sequence, delay records, passenger flow, baggage handling volume, and meteorological time series data; (2) Feature processing: Time series difference eliminates nonstationarity, ACF / PACF function determines ARIMA(p,d,q) parameters, maximum likelihood method completes parameter estimation; residual white noise test completes model verification; (3) Predictive output: Predict flight delay duration, peak passenger flow, baggage handling load, supplement the short-term emergency dynamic demand of the random forest model, and jointly output the baseline of emergency resource demand; Module output: The predicted demand for manpower, special vehicles, and equipment in each time period is used as constraint parameters and sent to the emergency resource intelligent optimization and scheduling module.

[0014] Furthermore, the airport's multi-dimensional operational safety risk assessment module It integrates four types of assessment algorithms to assess the safety risks of emergency scenarios in real time from four dimensions: apron taxiing, airspace clearance, navigation equipment, and overall support capabilities. It outputs the risk level and corrects the resource scheduling strategy in reverse to avoid secondary accidents caused by emergency response. Operational Assurance Capability Assessment Sub-model Based on CART Decision Tree Input the number of employees, weather, facility maintenance status, and real-time passenger flow characteristics, and construct a decision tree by recursively splitting using the Gini impurity criterion. Cross-validation and pruning are used to prevent overfitting. Real-time assessment of operational efficiency shortcomings in security checks, baggage handling, and apron support is conducted, and suggestions for strengthening support resources are output. OLS Obstacle Limitation Surface Clearance Assessment Submodel Based on civil aviation standards, conical and planar virtual restricted surfaces are constructed. The elevation coordinates of the surrounding terrain, buildings, trees, and antenna obstacles are imported. The algorithm automatically compares the position and height of obstacles with those of the restricted surfaces to identify the risk of airspace intrusion. In extreme weather emergency take-off and landing scenarios, the algorithm outputs airspace safety assessment results to constrain runway and flight scheduling plans. A sub-model for navigation facility failure assessment based on FTA fault tree and FMEA failure modes. Real-time acquisition of navigation equipment signal strength and error parameters; threshold detection to identify equipment anomalies; reverse tracing of fault root causes through fault tree analysis; failure mode analysis to assess the impact of faults on flight takeoff and landing; output of backup navigation equipment and maintenance personnel scheduling requirements, and inclusion in the scope of emergency resource scheduling. CNN+RNN multimodal fusion aircraft dangerous approach warning sub-model CNN extracts spatial features of apron monitoring and aircraft position, while RNN analyzes the temporal dynamic features of aircraft speed, heading, and vehicle position. Multimodal data is fused to predict the future positions of aircraft, ground vehicles, and personnel, identify dangerous approach events and issue real-time warnings. After a warning is triggered, vehicle driving paths and apron personnel scheduling are automatically adjusted to avoid collision risks.

[0015] Furthermore, the emergency resource intelligent optimization and scheduling module The system receives the resource requirements from the demand forecasting module and the safety constraints from the risk assessment module. It is divided into a special vehicle ant colony path planning subunit and a human-vehicle hybrid integer linear programming + random forest scheduling subunit to complete the dual optimization scheduling of vehicle routes and personnel shifts. Special vehicle ant colony path planning subunit (1) Applicable scenarios: daily ground support, aircraft emergency rescue, and route planning for fire and ambulance emergency vehicles; (2) Algorithm flow: Based on the airport's geographical road network, the path cost includes travel distance, congestion time, and safety control restrictions; ants select routes based on pheromone concentration and path cost probability; after completing the path solution, local search optimization is performed; pheromone evaporates over time, and the global optimal path strengthens the pheromone; after reaching the number of iterations or the solution quality threshold, the optimal path is output. (3) Emergency dynamic adaptation: In the event of a sudden accident, the road network congestion status is updated in real time, and the fastest passage route for rescue vehicles is generated again through iteration.

[0016] Intelligent scheduling sub-unit for people and vehicles (MILP + random forest fusion architecture) (1) Random forest pre-input: make demand forecasts for future flights, passenger flow and emergency events, and output the distribution of personnel and vehicle demand in each time period; (2) Mixed Integer Linear Programming (MILP) Modeling: Define binary decision variables for personnel, vehicles, and shifts; construct a dual objective function: minimize scheduling operation cost and maximize emergency service coverage; constraints include employee working hour regulations, personnel professional qualifications, vehicle maintenance restrictions, and minimum emergency resource availability; use the CPLEX / Gurobi solver to output the optimal shift plan; (3) Emergency dynamic adjustment: In the event of large-scale delays or emergencies, the MILP model is quickly recalculated to output the dispatching plan for overtime personnel and the activation of backup vehicles, taking into account both employee workload and service capacity.

[0017] Furthermore, a passenger emergency evacuation route planning module Based on Dijkstra's and A* heuristic path planning algorithms, dynamic planning of passenger routes is realized in two scenarios: regular boarding and emergency evacuation. Input parameters: terminal space topology, real-time queuing time for each channel, congestion points, temporary closure of channels, and flight evacuation requirements; Algorithm logic: The A* algorithm calculates the optimal path by combining physical distance and real-time congestion time cost; it monitors passenger flow distribution in real time, automatically diverts traffic to congested areas, and dynamically updates navigation routes; Dispatch linkage output: Push the evacuation route and personnel guidance needs to the emergency resource intelligent optimization and dispatch module, and simultaneously dispatch additional evacuation and security personnel.

[0018] Furthermore, an emergency scenario multi-agent simulation and inference module Using the agent-based ABM multi-dimensional variable judgment algorithm model, we conduct pre-simulations of emergency scenarios such as fires, extreme weather, security incidents, and large-scale flight delays. We compare the execution effects of multiple resource scheduling schemes and select the optimal scheduling strategy for execution. Modeling steps: (1) Agent definition: Construct multiple intelligent agents such as passengers, ground staff, security personnel, special vehicles, and aircraft, and configure their respective behavior rules and attribute parameters; (2) Virtual environment construction: Replicating the entire space environment of the terminal, runway, apron, and evacuation exits; (3) Multi-dimensional variable interaction simulation: Input multi-dimensional variables such as weather, flight schedules, and resource allocation schemes to simulate the dynamic interaction between agents and between agents and the environment; (4) Simulation index output: total evacuation time, resource utilization rate, scale of flight delays, and frequency of safety risks; Scheduling decision support: Parallel simulation of multiple personnel and vehicle scheduling schemes, quantitative comparison of the advantages and disadvantages of each scheme, feedback of optimal resource allocation constraints to the scheduling module, and optimization of actual scheduling instructions.

[0019] Furthermore, the instruction issuance and operation and maintenance iteration module Dispatch instruction distribution: Simultaneously push vehicle routes, personnel scheduling, equipment allocation, passenger evacuation, and risk warning instructions to the AOCC Operation Control Center, control tower, ground dispatch terminal, maintenance workshop, and security management system; Model iteration and optimization: Continuously collect real operational data and accident handling records after emergency dispatch execution, and regularly retrain random forest, ARIMA, decision tree, and multimodal fusion early warning models; monitor the prediction accuracy of each algorithm, automatically adjust hyperparameters, and adapt to long-term changes in airport passenger flow, facilities, and routes; establish a model performance monitoring mechanism to prevent model degradation.

[0020] Furthermore, the complete system scheduling workflow Step 1: The multimodal data acquisition and fusion preprocessing module collects real-time time-series, spatial, equipment, and meteorological data from the airport across all dimensions, processes them in a unified and standardized manner, and then synchronizes them to all backend algorithm sub-modules. Step 2: The emergency resource demand forecasting module uses Random Forest and ARIMA to jointly calculate the total emergency resource demand for ground support, vehicles, and equipment for the current and future periods; Step 3: The airport multi-dimensional operational safety risk assessment module performs parallel detection of four types of risks: support capability, airspace clearance, navigation equipment, and dangerous aircraft approach, and outputs safety constraints. Step 4: The emergency resource intelligent optimization and scheduling module combines resource demand and safety constraints, generates the optimal route for special vehicles through the ant colony algorithm, and outputs an intelligent scheduling plan for personnel and vehicles through the MILP+random forest hybrid model. Step 5: The passenger emergency evacuation route planning module generates dynamic evacuation routes based on real-time passenger flow congestion status and simultaneously supplements the needs of evacuation personnel; Step 6: Input the scheduling plan to be executed into the emergency scenario multi-agent simulation and inference module, simulate the entire emergency response process, output the comparison results of multiple plans, and select the optimal scheduling strategy; Step 7: The instruction issuance and operation and maintenance iteration module distributes the optimized scheduling instructions to each business terminal, while continuously collecting real operation data and periodically iterating and optimizing all algorithm models.

[0021] The beneficial effects of the airport emergency resource dynamic scheduling system based on multimodal data of the present invention are as follows: Multimodal data is fully integrated across the entire domain, and multiple algorithms work together to cover the entire chain of resource prediction, risk assessment, personnel and vehicle scheduling, passenger evacuation, and contingency plan simulation, solving the pain points of traditional scheduling systems such as single data, isolated algorithms, and static resource allocation. The dual-model joint prediction of emergency resource demand, random forest adapts to complex nonlinear support operations, ARIMA accurately captures flight delays and passenger flow fluctuations, and can recalculate resource gaps in seconds under extreme weather and emergencies to avoid resource shortages or idleness and reduce airport operating costs. Multi-dimensional safety risk pre-assessment integrates four types of risk detection: airspace clearance, navigation, apron taxiing conflict, and overall support capabilities. It can avoid secondary safety accidents in advance during the resource scheduling stage and comprehensively improve the level of airport operation safety. Ant colony algorithm dynamically plans emergency rescue vehicle routes, significantly reducing traffic congestion on the tarmac and shortening emergency response time; MILP mixed integer programming combined with random forest prediction optimizes scheduling, balancing compliant working hours, labor costs, and emergency support capabilities, while improving employee job satisfaction. The ABM multi-agent simulation model can rehearse multiple emergency dispatch plans before an event occurs, quantitatively assess evacuation efficiency, resource utilization, and delay losses, optimize emergency plans, replace manual experience-based simulations, and improve the scientific nature of emergency response. A* / Dijkstra dynamic passenger evacuation route planning can divert congested passenger flow in real time, reduce passenger waiting and evacuation time, and improve passenger travel experience and satisfaction; The system is adapted to all airport emergency scenarios, including peak tourist season, large-scale events, extreme thunderstorms, aircraft malfunctions, navigation failures, apron conflicts, and fire evacuations. It takes into account both short-term emergency rapid dispatch and long-term human and equipment resource planning, and simultaneously supports long-term resource allocation decisions such as airport expansion and route adjustments. With a built-in model self-iterative operation and maintenance mechanism, it continuously optimizes the accuracy of each algorithm using real operational data, has excellent generalization ability, and can be adapted to airports with different terminal and runway layouts, thus improving flight punctuality and overall operational efficiency in the long term. Attached Figure Description

[0022] The present invention will now be described in further detail with reference to the accompanying drawings and specific implementation methods.

[0023] Figure 1 This is a schematic diagram of the system framework of the present invention. Detailed Implementation

[0024] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the present application can be combined with each other.

[0025] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] Reference Figure 1 An airport emergency resource dynamic scheduling system based on multimodal data is described below, using a typical emergency scenario of widespread flight delays caused by extreme thunderstorms as an example. Multimodal data acquisition and fusion: The system accesses real-time weather warnings (heavy rainfall, low visibility), flight dynamics (numerous flight delays / diversions), GPS positions of aircraft on the apron, real-time positioning of special vehicles, terminal passenger flow monitoring images, navigation equipment operating parameters, and airport airspace terrain data. After cleaning and preprocessing with a unified timestamp, the data is distributed to each module. Emergency resource demand forecasting: The random forest model combines thunderstorm weather, surge in flight volume, and holiday characteristics to predict the shortage of ground handling, shuttle buses, maintenance equipment, and comfort service personnel; the ARIMA model predicts the peak passenger flow and baggage handling in the next 4 hours, and jointly outputs the baseline of emergency resource demand. Multi-dimensional security risk assessment: (1) The CART decision tree assesses the current security check and baggage sorting capacity as insufficient and outputs suggestions for increasing manpower; (2) The OLS clearance algorithm verifies the elevation of obstacles around the airport under low visibility conditions, determines that the clearance risk of some runways has increased, and restricts the frequency of take-off and landing. (3) The FTA / FMEA detected that the navigation equipment signal was slightly affected by thunderstorms. The fault risk level was assessed, and backup navigation equipment and maintenance personnel were dispatched. (4) The CNN+RNN multimodal model identifies the significant increase in the density of aircraft and work vehicles on the apron after the delay, continuously outputs a dangerous approach warning, and constrains the vehicle speed and route. Intelligent optimization and scheduling of emergency resources: (1) The ant colony algorithm, combined with the real-time congestion status of the apron, replans the optimal passage route for shuttle buses, maintenance vehicles and emergency ambulances to avoid the congestion area; (2) MILP mixed integer programming takes the personnel and vehicle demand output by random forest as the target, and adds working hours and qualification constraints to quickly generate employee overtime schedules and backup vehicle activation schemes. Passenger emergency evacuation route planning: Identify passenger congestion in terminal check-in and waiting areas, dynamically switch diversion channels using the A* algorithm, and simultaneously push the deployment needs of diversion security personnel to the dispatch module; ABM Multi-Agent Simulation and Inference: Construct a thunderstorm delay simulation scenario, simulate two sets of scheduling and vehicle dispatching schemes in parallel, compare passenger delay time, resource utilization rate, and secondary taxiing conflict risk, and select the scheduling scheme with the lowest overall loss. Command issuance and model iteration: The system issues commands for vehicle routes, personnel overtime scheduling, maintenance equipment dispatch, passenger evacuation, and runway take-off and landing restrictions to AOCC, control tower, ground support, and security terminals. After the emergency response is completed, the system stores all the data from this thunderstorm emergency and automatically retrains the random forest, ARIMA, and multimodal early warning models in the background to optimize the prediction accuracy in thunderstorm scenarios.

[0027] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A dynamic scheduling system for airport emergency resources based on multimodal data, comprising a multimodal data acquisition and fusion preprocessing module, an emergency resource demand prediction module, an airport multi-dimensional operational safety risk assessment module, an intelligent optimization scheduling module for emergency resources, a passenger emergency evacuation route planning module, an emergency scenario multi-agent simulation and deduction module, and an instruction issuance and maintenance iteration module, characterized in that: The multimodal data acquisition and fusion preprocessing module is used to acquire time-series business data, spatial geographic data, environmental meteorological data, equipment status data, and image dynamic target multimodal heterogeneous data, and output a standardized feature dataset after unified timestamp synchronization, cleaning and normalization. The emergency resource demand prediction module integrates the random forest prediction sub-model and the ARIMA flight situation analysis sub-model. Based on the standardized feature dataset, it calculates the resource demand of ground support, security, special vehicles, and maintenance equipment in emergency scenarios and outputs the resource demand baseline. The airport multi-dimensional operation safety risk assessment module integrates the CART decision tree operation support assessment sub-model, the OLS airspace assessment sub-model, the FTA+FMEA navigation fault assessment sub-model, and the CNN+RNN multimodal aircraft dangerous approach warning sub-model, and outputs safety risk levels and scheduling constraints in parallel from four dimensions: support capability, airspace, navigation equipment, and apron taxiing conflict. The emergency resource intelligent optimization and scheduling module includes a special vehicle ant colony path planning subunit and a personnel and vehicle MILP+random forest scheduling subunit. Combining the resource demand baseline and safety scheduling constraints, it outputs the optimal driving route for emergency vehicles and the dynamic scheduling scheme for personnel and special vehicles, respectively. The passenger emergency evacuation route planning module is based on Dijkstra's and A* path algorithms. It dynamically generates passenger evacuation and boarding navigation routes by combining real-time terminal passenger flow congestion status, and outputs the evacuation personnel deployment requirements. The emergency scenario multi-agent simulation and inference module adopts the agent-based ABM multi-dimensional variable judgment algorithm to build an airport virtual environment and multiple types of intelligent agents, simulate the emergency response effects of multiple scheduling schemes, and select the optimal scheduling strategy. The instruction issuance and operation and maintenance iteration module is used to distribute scheduling instructions to various business terminals in the airport and collect emergency operation data to periodically iterate and optimize all algorithm models.

2. The airport emergency resource dynamic scheduling system based on multimodal data according to claim 1, characterized in that, The construction process of the random forest prediction sub-model includes: The input features are flight volume, weather, season, and historical resource usage records, and the target variables are the demand for various support resources. Self-service sampling selects samples, random selection of feature subsets completes decision tree node splitting, and multiple decision trees are integrated to build a model; Using MSE, R 2 The accuracy of the model is evaluated by indicators, and the model is optimized by adjusting the number of trees, the maximum depth of the trees, and the feature selection rules. Flight schedule changes, severe weather, and equipment failures trigger model recalculation, resulting in real-time updates of resource demand forecasts.

3. The airport emergency resource dynamic scheduling system based on multimodal data according to claim 1, characterized in that, The ARIMA flight situation assessment sub-model processing flow includes: The nonstationarity of flight and passenger flow time series data is eliminated by difference, and the model parameters p, d, and q are determined by ACF and PACF functions. The parameter estimation is completed by the maximum likelihood method. The residual white noise test verifies the effectiveness of the model, predicts flight delay duration, peak passenger flow, and baggage handling load, and outputs the baseline of emergency resource demand in conjunction with the random forest model.

4. The airport emergency resource dynamic scheduling system based on multimodal data according to claim 1, characterized in that, The implementation steps of the CNN+RNN multimodal aircraft dangerous approach warning sub-model are as follows: CNN extracts spatial features of apron images and aircraft coordinates, while RNN analyzes temporal features of aircraft speed, heading, and ground vehicle positions. It integrates multimodal features to predict the future locations of aircraft, vehicles, and personnel, identifies and warns of dangerous approach events, and automatically corrects vehicle routes and personnel scheduling strategies.

5. The airport emergency resource dynamic scheduling system based on multimodal data according to claim 1, characterized in that, The algorithm flow of the special vehicle ant colony path planning subunit is as follows: The path cost is determined by the distance to the airport road network, the time spent in congestion, and security control; ants choose their route based on pheromone concentration and the probability of path cost. After finding the route solution, perform local search optimization. Path pheromones evaporate over time, while the globally optimal path enhances pheromones. After reaching the preset number of iterations or the solution quality threshold, the optimal passage route for emergency vehicles is output. In the event of a sudden event, the road network status is refreshed in real time and the calculation is re-iterated.

6. The airport emergency resource dynamic scheduling system based on multimodal data according to claim 1, characterized in that, The workflow of the human-vehicle MILP+random forest scheduling sub-unit is as follows: Random forest models predict the distribution of emergency resource demand across different time periods; Mixed Integer Linear Programming (MILP) defines binary decision variables for personnel, vehicles, and shifts, and constructs a dual objective function that minimizes operating costs and maximizes emergency service coverage. Set work hour regulations, personnel qualifications, vehicle maintenance, and minimum resource availability constraints, and output the scheduling plan through the solver; Large-scale flight delays and emergencies trigger model re-solution, requiring rapid deployment of overtime staff and backup vehicles.

7. The airport emergency resource dynamic scheduling system based on multimodal data according to claim 1, characterized in that, The execution steps of the OLS net clearance assessment sub-model are as follows: Construct virtual surfaces to restrict cone-shaped and planar obstacles in accordance with civil aviation standards; Collect elevation coordinates of the terrain, buildings, trees, and antenna obstacles around the airport; The algorithm compares the coordinates and height of obstacles with the parameters of the limiting surface to identify the risk of airspace intrusion. Output airspace safety assessment results in extreme weather emergency take-off and landing scenarios to constrain runway and flight scheduling.

8. The airport emergency resource dynamic scheduling system based on multimodal data according to claim 1, characterized in that, The process of the FTA+FMEA navigation facility fault assessment sub-model is as follows: Real-time acquisition of navigation device signal strength and error parameters; threshold detection to identify device malfunctions. Fault Tree Analysis (FTA) is used to trace the root cause of the fault in reverse, and Failure Mode and Effects Analysis (FMEA) is used to analyze the impact of the fault on flight takeoff and landing. Output the needs for backup navigation equipment and maintenance personnel, and incorporate them into the constraints of emergency resource allocation.

9. The airport emergency resource dynamic scheduling system based on multimodal data according to claim 1, characterized in that, The ABM modeling steps of the emergency scenario multi-agent simulation and inference module include: Define multiple types of intelligent agents, including passengers, ground staff, security personnel, special vehicles, and aircraft, and configure their behavioral rules; Build a virtual airport environment that includes terminals, runways, aprons, and evacuation exits; input multi-dimensional variables such as weather, flights, and resource scheduling to simulate dynamic interactions between intelligent agents and between intelligent agents and the environment; Statistical analysis of evacuation time, resource utilization rate, and flight delay scale indicators; comparison of multiple scheduling schemes; and output of the optimal strategy.

10. The airport emergency resource dynamic scheduling system based on multimodal data according to claim 1, characterized in that, The scheduling workflow is as follows: Step 1: The multimodal data acquisition and fusion preprocessing module collects and standardizes all-dimensional airport operation data, and distributes it to all algorithm sub-modules; Step 2: The emergency resource demand forecasting module calculates the total emergency resource demand using a combination of random forest and ARIMA. Step 3: The airport's multi-dimensional operational safety risk assessment module performs four types of risk detection in parallel and outputs scheduling safety constraints; Step 4: The emergency resource intelligent optimization and scheduling module generates vehicle routes through the ant colony algorithm and outputs personnel and vehicle scheduling schemes through the MILP+random forest model. Step 5: The passenger emergency evacuation route planning module generates dynamic evacuation routes based on real-time passenger flow and supplements the needs of evacuation personnel; Step 6: The multi-agent simulation and deduction module for emergency scenarios simulates multiple scheduling schemes and selects the optimal scheduling strategy; Step 7: The instruction issuance and operation and maintenance iteration module distributes scheduling instructions to each business terminal, collects real operation data, and periodically iterates and optimizes all algorithm models.

Images