A mine intelligent transportation system and method based on multi-machine cooperative scheduling

Abstract

The present application relates to the technical field of mine collaborative transportation, and discloses a mine intelligent transportation system and method based on multi-machine collaborative scheduling, which comprises the following steps: collecting and cleaning the working conditions and geometric data of locomotives, yards and track sections, calculating health scores and refreshing the dynamic upper limit of traction / braking, determining hydraulic availability and shield translation, dynamically tightening the speed limit / acceleration of a curve according to the load center of gravity deviation and wheel tread deviation, constructing an obstacle scene set and injecting a digital twin, establishing a robust / scene MILP model, and minimizing the weighted target composed of the construction period, energy consumption, risk occupancy, yard load and tail risk. Under the premise of ensuring the safety boundary and operation principle, the present application improves the plan stability and time window achievement rate, reduces conflicts, deadlocks and energy consumption / thermal peaks, and shortens the recovery time and tail risk in the obstacle scene. Through digital twinning and immutable audit chain, the present application realizes operation closed loop, and takes into account efficiency, risk and compliance.

Description

Technical Field

[0001] This invention relates to the field of collaborative transportation in mines, and more specifically, to an intelligent mine transportation system and method with multi-machine collaborative scheduling. Background Technology

[0002] In the scenario of rail / rubber-tired collaborative transportation in mines, there are multiple locomotives, multiple train sets, multiple yards and multiple track sections operating in parallel and converging. It is necessary to simultaneously meet multiple constraints such as track section occupancy and capacity mutual exclusion, dynamic head spacing, path continuity and passage time, track gauge compatibility, yard capacity and translation operation window, speed limit and acceleration on curves, traction-gradient-rolling resistance-acceleration coupling, braking safety distance, task time window and priority.

[0003] In particular, equipment health status, hydraulic availability, failure events, and congestion introduce significant uncertainties into the system. Traditional static timetables or single-scenario optimization / heuristic methods often struggle to balance safety mutual exclusion, global efficiency, and rapid recovery from anomalies. At the same time, the lack of strong consistency guarantees (such as token and transactional switch switching), the absence of risk measurement (tail risk), and rolling replanning mechanisms between scheduling and field control lead to congestion, increased energy consumption, and slow recovery under complex operating conditions. Summary of the Invention

[0004] This invention provides a multi-machine collaborative scheduling intelligent transportation system and method for mines, which solves the technical problem in related technologies of how to construct a robust, optimizable, and reprogrammable multi-machine collaborative scheduling and control system that can ensure the safety and consistency of track occupancy and turnout switching under uncertain conditions such as health, hydraulic pressure and obstacles.

[0005] This invention provides a multi-machine collaborative scheduling intelligent transportation method for mines, comprising the following steps:

[0006] S100, Health Perception and Digital Twin Initialization: Collect and clean the working condition and geometric data of locomotives, yards and track sections, calculate health scores and refresh the dynamic upper limit of traction / braking, determine hydraulic availability and shielding translation, dynamically tighten curve speed limit / acceleration based on load center of gravity offset and wheel track deviation, construct obstacle scene set and inject it into digital twin;

[0007] S200, Robust MILP / Scenario MILP Modeling and Objective Function Reconstruction: Establish a robust / scenario MILP model and minimize a weighted objective consisting of project duration, energy consumption, risk occupancy, parking lot load, and tail risk;

[0008] S300, Global Baseline Solving and Grouping / Path Generation: Solve robust / scenario MILP on the obstacle scenario set, generate task timetable, path selection, occupancy time sequence and depot translation sequence, and optimize grouping and segmented relay based on health score and working condition;

[0009] S400, Safety Mutual Exclusion and Turnout Atomic Switching Protocol Issued: Implement track segment token and lease management, turnout switching adopts two-stage submission, preparation stage verifies zero occupation and hydraulic availability, submission stage locks translation window, failure rollback and broadcast clearing and bypass replanning.

[0010] S500, edge control execution: executes speed-position-traction / braking closed-loop control at the edge controller, tracks target speed and target position and limits control saturation, has online application capabilities and safety limits, monitors total energy consumption and braking thermal load, and implements soft constraints and load reduction strategies on long slopes and congested sections.

[0011] S600, Event Triggered and Scrolling Time Domain Replanning: Set a scrolling window and lock historical decisions, resolve subproblems within the window according to event triggers, dynamically increase relevant weights and constraints, limit the maximum resolving frequency and follow the trigger priority, and inject rescue and bypass tasks when necessary to achieve rapid loss mitigation;

[0012] S700, MARL Local Collaboration and "Safety Fence" Shielding: Constructing agent state, the reward function considers multiple factors simultaneously, and adopts a safety fence shielding strategy to sample and execute only within the set of safe actions;

[0013] S800, Consistency Verification and Deployment: Injects device / operational condition obstacles into the digital twin to perform closed-loop consistency verification, assesses consistency and recovery time, deploys baseline plans and rolling policies on the edge controller, and enables immutable event logs and status snapshots to form auditable deployment and maintenance records.

[0014] Furthermore, in S200, the robust / scenario MILP model covers scenario factors such as track segment occupancy and capacity mutual exclusion, dynamic headway, path continuity and passage time, track gauge compatibility, yard capacity and translation shielding, curve speed limit / acceleration, traction-gradient-rolling resistance-acceleration coupling, braking safety distance, task time window and priority, rescue task injection for traction hook anomalies, and risk weighting.

[0015] Furthermore, the specific contextual factors include the following:

[0016] Track segment occupancy and capacity mutual exclusion constraints: Implement single-track capacity mutual exclusion and time step occupancy constraints to ensure that a track segment is uniquely occupied at any given time, and reserve a safe headway;

[0017] Dynamic headroom step calculation: The dynamic headroom step is calculated based on the current speed and braking health. In high-risk or heat-fading situations, the headroom is increased to ensure safe braking space.

[0018] Path continuity and passage time consistency: ensure that occupation only occurs on the selected path and meets the minimum number of steps determined by length / speed limit, avoiding "segment skipping" and "speed crossing";

[0019] Track gauge compatibility and wheel gauge constraints: Compatibility is determined by a decision function, and segments with mismatched wheel gauge or incompatible track gauge are prohibited from being included in the path selection;

[0020] Parking lot capacity and translation shielding: Implement parking lot capacity limit and translation operation shielding strategy. Translation is only allowed when hydraulic availability is available. During translation, the occupied space is cleared to avoid shearing conflict.

[0021] Curve speed limit / acceleration constraint: When a curve is occupied, dynamic constraints on the maximum permissible speed and maximum permissible acceleration are applied to prevent the risk of rollover and derailment caused by abnormal center of gravity / wheel track.

[0022] Traction-gradient-rolling resistance-acceleration coupling and capacity constraints: Calculate the traction force required for passage of the section and compare it with the dynamic traction limit. If the limit is exceeded, speed / load limit or detour is required to ensure dynamic feasibility.

[0023] Braking safety distance and thermal decay effect: assess braking distance and reserve a safety margin, reduce target speed and increase head-off distance during thermal decay;

[0024] Task time window and priority inheritance: All tasks are scheduled within the time window. Priority tasks are ensured not to default. If necessary, lower priority tasks are given priority or rescheduled.

[0025] Rescue / retrieval task injection for tow hook malfunction: When an abnormal tow hook status is detected, a rescue / retrieval task is immediately generated, relevant tokens are frozen, and rolling replanning and bypass paths are triggered.

[0026] Risk weight calculation and scenario risk modeling: Integrating slope, curvature, human factors and congestion signals to form risk weights that change with time and location for target weighting.

[0027] Furthermore, in S300, the optimization goal is to have high-health locomotives undertake critical paths and heavy loads, while low-health locomotives are subject to load and speed limits.

[0028] Furthermore, in the S600, the trigger priorities are, in order: safety, health, vehicle location, energy consumption, and cycle.

[0029] Furthermore, in the S700, the agent state includes neighborhood occupancy, parking lot status, task remaining, risk, health, and hydraulic pressure.

[0030] Furthermore, in S700, the reward function simultaneously considers time, safety, energy consumption, congestion, and yielding.

[0031] Furthermore, the safety barrier shielding strategy refers to directly shielding actions that are not in the set of safe actions, and only sampling and executing actions within the set of safe actions.

[0032] Furthermore, in the S800, the evaluation of consistency and recovery time includes timetables, tokens, switches, edge control, scroll optimization, and MARL barriers.

[0033] This invention also proposes a multi-machine collaborative scheduling intelligent transportation system for mines, comprising:

[0034] Equipment health perception and digital twin initialization module: aggregates sensor data to generate health scores, constrains traction / braking capabilities and constructs obstacle scenarios; completes parameter and scenario initialization in the digital twin;

[0035] Robust scheduling modeling and objective function module: Establish occupancy, capacity mutual exclusion, speed / acceleration, parking lot capacity and curve limit constraints; reconstruct the robust objective function with CVaR / scene weights;

[0036] Baseline solution and train formation path generation module: Solve robust / scenario MILP on the scenario set, and output task timetable, train formation path and depot translation sequence; formulate train formation strategy of "dual-machine segmented relay and depot staggered peak";

[0037] Safety Mutual Exclusion and Turnout Two-Phase Submission Module: Issues track segment tokens and lease revocation strategies, executes turnout 2PC preparation / submission and failure rollback; freezes adjacent track segments and broadcasts clearance when necessary;

[0038] Edge control execution and energy consumption constraint module: Locomotive-side closed-loop tracking of speed / position / traction braking to implement energy consumption and thermal load monitoring; speed limit and load reduction during health degradation to avoid slippage and thermal decay;

[0039] Event-driven scrolling replanning module: Fix historical decisions and update future timelines within a scrolling window; increase risk weights based on event priority and insert rescue / bypass tasks;

[0040] MARL Local Coordination and Safety Barrier Module: Learns yielding and cooperative behaviors based on state information; uses safety barriers to block illegal actions and reduce the risk of congestion and deadlock;

[0041] Consistency verification and deployment module: Inject equipment / operating condition obstacles into the twin environment to perform stress tests, verify the closed-loop consistency of "timetable, token, turnout, control, RHO, MARL guardrail", generate baseline plan and deploy, and enable immutable event logs and snapshots.

[0042] The beneficial effects of this invention are as follows:

[0043] This invention combines health-driven dynamic limits with robust / scenario-based MILP (including CVaR) through closed-loop integration, achieving strict adherence to capacity mutual exclusion and dynamic head-off constraints under single-line conditions;

[0044] Track segment tokens / leases and turnout 2PCs guarantee the atomicity and safe mutual exclusion of occupancy management and switching operations, significantly reducing the probability of conflicts, deadlocks and rollbacks;

[0045] The edge controller tracks and implements speed limits and braking thermal management on curves and long downhill sections to suppress energy consumption and heat peaks.

[0046] Event-triggered rolling replanning prioritizes "safety > health > parking lot > energy consumption > cycle" to minimize disturbances in unexecuted segments, significantly reducing recovery time and tail risk in obstacle scenarios;

[0047] MARL achieves micro-level yielding and congestion mitigation within safety barriers, improving collaborative efficiency without breaking hard constraints.

[0048] Digital twin stress testing and immutable event logs / state snapshots provide auditable and traceable release and operation records, which improves overall plan stability, window achievement rate and on-time rate of critical tasks, reduces both risk and energy consumption, and enhances security compliance and review capabilities.

[0049] In summary, while ensuring safety boundaries and operational principles, this approach significantly improves plan stability and window achievement rates, reduces conflicts, deadlocks, and energy / heat peaks, and shortens recovery time and tail risks in obstacle scenarios. Through digital twins and an immutable audit chain, it achieves a "verifiable, traceable, and reproducible" operational loop, balancing efficiency, risk, and compliance. Attached Figure Description

[0050] Figure 1 This is a flowchart of a multi-machine collaborative scheduling intelligent transportation method for mines according to the present invention;

[0051] Figure 2 This is a spatiotemporal occupancy Gantt chart in an example of the present invention;

[0052] Figure 3 This is a timing diagram of the mutual exclusion between turnout switching two-stage submission (2PC) and track segment token in S400 of the present invention;

[0053] Figure 4 This is the edge control closed-loop diagram in S500 of the present invention;

[0054] Figure 5 This is the event-triggered scroll replanning diagram in S600 of the present invention. Detailed Implementation

[0055] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0056] like Figure 1 - Figure 5 As shown, a multi-machine collaborative scheduling intelligent transportation method for mines includes the following steps:

[0057] S100, Health Perception and Digital Twin Initialization: Collect and clean the working condition and geometric data of locomotives, yards and track sections, calculate health scores and refresh the dynamic upper limit of traction / braking; determine hydraulic availability and shielding translation; dynamically tighten curve speed limit / acceleration based on load center of gravity offset and wheel track deviation; construct obstacle scene set and inject it into digital twin to form a reliable input for robust optimization and rolling replanning;

[0058] In one embodiment of the present invention, the following steps are specifically included:

[0059] S110, On-site data acquisition and cleaning: Collect operating condition flow and geometric load data of locomotives / yards / track sections, and perform time-series calibration and outlier filtering to form a clean dataset required for health assessment and limit refresh.

[0060] S120, Dynamic Health Score and Capability Limit: Calculate power health and braking health, and update the dynamic traction / braking capability limit accordingly, as a real-time parameter for subsequent constraints;

[0061] formula:

[0062] ;

[0063] ;

[0064] in, The power health score (0-1) of locomotive k at time t is given. The braking health score (0-1) of locomotive k at time t is given. This is the upper limit of the rated traction force (equipment parameter). This is the upper limit of the rated maximum braking deceleration (equipment parameter). The upper limit of the dynamic traction force of locomotive k (which varies with health status). The upper limit of dynamic braking deceleration for locomotive k (varies with health status);

[0065] S130, Parking lot hydraulic availability determination and translation operation shielding: Calculate the parking lot hydraulic availability binary indication based on the hydraulic pressure sensor value and threshold. When the hydraulic system is unavailable, shield the translation operation of the parking lot translation operation indication and freeze related resources.

[0066] formula:

[0067] ;

[0068] in, Let y be the hydraulic pressure sensor value of the parking lot at time t. The minimum hydraulic threshold (safety standard) that allows translational movement. The parking lot hydraulic availability is indicated by a binary value (1 for available, 0 for unavailable). Instructions are given for the horizontal movement of the parking lot (1 executes, 0 does not execute);

[0069] S140, speed limit / acceleration tightening on curves (considering center of gravity and track deviation): corrected using load center of gravity offset and track deviation. and Generates a cornering speed limit / acceleration limit that dynamically tightens with load and wheelbase;

[0070] formula:

[0071]

[0072] ;

[0073] in, The basic speed limit for track section e (on-site speed limit signs and standards). Speed ​​limit based on curvature of curve e (without considering load / track correction). The maximum acceleration of the base at curve e (without considering load / wheel track correction). , This is the speed limiting tightening factor (sensitivity to center of gravity shift and wheel track deviation). , This is the acceleration tightening coefficient (sensitivity to center of gravity shift and wheel track deviation). This is the offset of the load center of gravity of train set g (relative to the standard position). This is the wheel gauge deviation of locomotive k (relative to the standard track gauge).

[0074] S150, Obstacle Scenario Set Construction and Twin Injection: Establish a scenario library including power slump, hydraulic unavailability, brake heat fade, wheel track deviation, center of gravity shift, and tow hook detachment, and inject digital twins to drive robust optimization and rolling replanning (obstacle scenario set). );

[0075] S200, Robust MILP / Scenario MILP Modeling and Objective Function Reconstruction: Establish a robust MILP / scenario MILP model covering track segment occupancy and capacity exclusivity, dynamic headway, path continuity and through time, track gauge compatibility, yard capacity and translational shielding, curve speed limits / acceleration, traction-gradient-rolling resistance-acceleration coupling, braking safety distance, task time window and priority, rescue task injection for traction hook anomalies, and risk weights, using factors such as schedule, energy consumption, risk occupancy, yard load, and... Minimize the weighted objective of tail risk composition;

[0076] In one embodiment of the present invention, the following steps are specifically included:

[0077] S210, Establish a robust MILP / scenario-based MILP model:

[0078] Covering the following factors:

[0079] Track segment occupancy and capacity mutual exclusion constraints: Implement single-track capacity mutual exclusion and time step occupancy constraints to ensure that a track segment is uniquely occupied at any given time and reserve a safe headway;

[0080] formula:

[0081] ;

[0082] ;

[0083] in, Let the time t of train g on track section e be a binary variable (1 occupied, 0 idle). The capacity of track segment e (generally 1 for a single-track track). This is the dynamic head-off step count (calculated based on current speed and braking capacity).

[0084] Dynamic headroom step calculation (based on speed and braking capacity): Calculates dynamic headroom steps based on current speed and braking health, and increases headroom to ensure safe braking space during high-risk or heat-fading conditions.

[0085] formula:

[0086] ;

[0087] in, Let g be the speed of train set g at time t. The upper limit of dynamic braking deceleration for locomotive k (varies with health status). Let e ​​be the length of track segment e;

[0088] Path continuity and passage time consistency: ensure that occupation only occurs on the selected path and meets the minimum number of steps determined by length / speed limit, avoiding "segment skipping" and "speed crossing";

[0089] formula:

[0090] ;

[0091] ;

[0092] ;

[0093] in, Let g be a binary variable indicating whether train group g chooses to pass through track section e (1 means choose, 0 means do not choose). The minimum number of steps required to pass through track segment e (determined by the length and the basic speed limit). The basic speed limit for track segment e;

[0094] Track gauge compatibility and wheel gauge constraints: Compatibility is determined by a decision function, and segments with mismatched wheel gauge or incompatible track gauge are prohibited from being included in the path selection;

[0095] formula:

[0096] ;

[0097] in, This is a function to determine whether track segment e is compatible with train set g (including locomotive k wheel gauge deviation) (1 for compatible, 0 for incompatible). The track gauge property of train set g (determined by the formation and load geometry). This is the wheel gauge deviation of locomotive k (relative to the standard track gauge).

[0098] Parking lot capacity and translation shielding (hydraulic availability binding): Implement the parking lot capacity limit and translation operation shielding strategy, allowing translation only when hydraulic availability is available, and clearing the occupied space during translation to avoid shearing conflicts;

[0099] formula:

[0100] ;

[0101] ;

[0102] ;

[0103] in, Let g be the occupancy status of train g at time t in yard y (1 occupied, 0 idle). Let y be the maximum parking space capacity of parking lot y. Instructions are given for the translation of the parking lot y (1 executes, 0 does not execute). Hydraulic availability of parking lot y (1 available, 0 unavailable);

[0104] Curve speed limit / acceleration constraint (with center of gravity / wheel track influence): Apply dynamic constraints on the maximum permissible speed and maximum permissible acceleration when occupying a curve to prevent the risk of rollover and derailment caused by abnormal center of gravity / wheel track.

[0105] formula:

[0106] ;

[0107] ;

[0108] in, The maximum permissible speed of train g at time t on track segment e (varying with load / wheel gauge and curve characteristics). The maximum permissible acceleration of train set g at time t on track segment e (varying with load / wheel gauge and curve characteristics). Let g be the acceleration of train set g at time t;

[0109] Traction-gradient-rolling resistance-acceleration coupling and capacity constraints: Calculate the traction force required for passage of the section and compare it with the dynamic traction limit. If the limit is exceeded, speed / load limit or detour is required to ensure dynamic feasibility.

[0110] formula:

[0111] ;

[0112] ;

[0113] ;

[0114] ;

[0115] in, The required traction force (resultant force requirement) to pass through track segment e. Let g be the total mass of the train set (locomotive, flatcar, and load). It is the gravitational acceleration constant (approximately 9.81 m / s²). The slope angle (in radians or degrees) of track segment e. It is the rolling resistance coefficient (related to wheel-rail conditions). This is the wind resistance / additional drag coefficient (which can be approximated as linear at low speeds). This refers to the actual traction force command or measured value of train set g at time t. The upper limit of the dynamic traction force of locomotive k (which varies with health status). This refers to the maximum speed limit for train sets across the entire system (equipment and safety specifications).

[0116] Braking safety distance and thermal decay effect: assess braking distance and reserve a safety margin, reduce target speed and increase head-off distance during thermal decay;

[0117] formula:

[0118] ;

[0119] in, For locomotive k at speed The estimated braking distance is as follows. The upper limit of dynamic braking deceleration for locomotive k (considering thermal decay and health). Let e ​​be the length of track segment e. To allow for a safety margin (to prevent rear-end collisions / crossing the boundary);

[0120] Task time window and priority inheritance: All tasks are scheduled within the time window. Priority tasks are ensured not to default. If necessary, lower priority tasks are given priority or rescheduled.

[0121] formula:

[0122] ;

[0123] ;

[0124] ;

[0125] in, , Let i be the upper and lower bounds of the time window for task i. , The start and finish times of task i are determined by the scheduler. The minimum time required to complete the task (determined by path length and speed limit). Task priority levels (high-priority tasks must be completed on time);

[0126] Rescue / retrieval task injection for tow hook malfunction: When an abnormal tow hook status is detected, a rescue / retrieval task is immediately generated, relevant tokens are frozen, and rolling replanning and bypass paths are triggered.

[0127] formula:

[0128] ;

[0129] in, The status of the traction coupler of train set g is indicated by two values ​​(1 for normal, 0 for abnormal). For rescue / recovery missions (automatically generated by the system and incorporated into the mission set);

[0130] Risk weight calculation and scenario risk modeling: Integrating slope, curvature, human factors and congestion signals to form risk weights that change with time and location for target weighting;

[0131] formula:

[0132] ;

[0133] in, Let e ​​be the comprehensive risk weight at time t. , , , The weighting coefficients for each risk source are: Let be the slope angle of track segment e; This refers to the curvature index (curvature strength). This is a risk signal for human factors (such as human intervention / construction overlap). This is a congestion / conflict risk signal (e.g., dense occupancy of adjacent segments).

[0134] S220, Robust Objective Function (including CVaR) Construction: Weighted minimization of multiple objectives including project duration, energy consumption, risk occupancy, parking lot load, and tail risk; calibration. and Enhance the ability to withstand extreme scenarios;

[0135] formula:

[0136] ;

[0137] ;

[0138] ;

[0139] ;

[0140] ;

[0141] ;

[0142] in, For the total construction period, For power consumption estimation, For discrete time steps, The target weighting coefficient (needs to be determined based on on-site considerations). Risk weighting For occupying variables; For parking lot capacity, In scene set The loss function approximates the above (late arrival / blocking / congestion / excessive energy consumption). Set the CVaR confidence level (e.g., 0.9 / 0.95). As an auxiliary variable for CVaR, For positive part operators ( ), CVaR refers to Conditional Value at Risk, a core quantitative tool used to measure the tail risk of a portfolio, representing the tail loss based on VaR at a specific confidence level.

[0143] S300, Global Baseline Solving and Grouping / Path Generation: In Scene Sets The robust / scenario MILP is solved to generate task timetables, path selection, occupancy time sequence and depot translation sequence; the grouping and segmented relay are optimized based on health score and working condition, with high-health locomotives undertaking critical paths and heavy loads, and low-health locomotives being subject to load and speed limits to ensure overall stability.

[0144] S310, Scene MILP Solving and Schedule Generation: In the scene set Solve robust / scene MILP above, and output task timetable, path selection, occupancy time sequence and depot translation sequence;

[0145] ;

[0146] ;

[0147] ;

[0148] ;

[0149] ;

[0150] ;

[0151] ;

[0152] in, (Source) ), (Reference Point) ), (Midpoint) For the train crew to assemble For the set of track segments, For a set of stations / nodes, Gather at the parking lot For discrete-time sets, For scene set, For task set, Does g occupy track segment e at time t? Whether to select track segment e for g; This indicates the occupancy of vehicle g in parking lot y (the parking lot is considered a special section). Capacity of track segment e (single track) ); Let e ​​be the dynamic head distance (number of steps) of track segment e at time t. In scene set The robust upper and lower bounds of the passage time are calculated by considering factors such as corner speed limit, gradient, and health ability. / Let v be the set of adjacent segments for both inbound and outbound traffic. / For the source / sink nodes of g; For the flow to maintain equilibrium constant, For track segment e, it indicates compatibility between the gauge / wheel gauge of the train; The track gauge parameter for train group g; Let K be the wheelbase deviation of locomotive. Let y be the capacity of the parking lot; Let y be the translational movement of the parking lot at time t; Hydraulic availability (1 for available). The earliest / latest start time for task i; Minimum duration of the task; The start / end time of the task. , The route segment selected for train group g; This is for timing purposes; Sequence for the translational movement of the parking lot.

[0153] S320, formation and segmented relay strategy: adopts the formation scheme of "A1 starting segment - E1 / E2 buffer - A2 relay", and implements load reduction and speed limit path for low health locomotives to ensure overall stability;

[0154] formula:

[0155] ;

[0156] ;

[0157] ;

[0158] ;

[0159] in, For locomotive assembly; A binary variable indicating whether locomotive k is assigned to train group g; Weighting of locomotive health (such as the importance of traction capacity). / Assess traction / braking health. For grouping healthy aggregates (weighted average or weakest value alternatives are also acceptable), use these aggregates. For the load of group g; The upper limit of the reference load; The load factor that tightens with health; Speed ​​limits are set for curves; For health-related speed limits, / Let e ​​be the gradient and curvature of track segment e; Path permission indication (masking high-risk segments for low health / heavy load). / This refers to the arrival / departure time at buffer point E1; This is to allow for a buffer period for relay waiting / handover. This indicates occupancy in E1;

[0160] S400, Safety Mutual Exclusion and Turnout Atomic Switching Protocol Issued (including Failure Rollback): Implement track segment token and lease management to ensure consistent occupancy and reclaimability after timeout; Turnout switching adopts a two-stage submission, the preparation stage verifies zero occupancy and hydraulic availability, the submission stage locks the translation window, failure rollback and broadcast clearing and bypass replanning to ensure the atomicity and safety mutual exclusion of switching.

[0161] S410, Track Segment Token and Lease Management: Issue occupancy tokens according to plan and set lease timeouts. When the timeout expires, the token is revoked, the adjacent segments are frozen, and the lease is broadcast to be cleared to avoid resource conflicts and deadlocks.

[0162] formula:

[0163] ;

[0164] in, The token lease status of track segment e at time t is 1 valid, 0 invalid / reclaimed. "Occupancy timeout" means that the planned occupancy steps and lease validity period have been exceeded (triggering reclamation and freezing).

[0165] S420, Two-stage submission and failure rollback of turnout: Verify zero occupancy and hydraulic availability in the preparation stage, and lock the translation window in the submission stage;

[0166] If it fails, a timeout rollback will be executed and a bypass replanning will be injected.

[0167] formula:

[0168] ;

[0169] in, For parking lot Y to be switched over, it must be "zero occupied". The submission phase can only proceed if hydraulics are available (translational movement is permitted).

[0170] S500, edge control execution (including health degradation and energy consumption constraints): executes speed-position-traction / braking closed-loop control at the edge controller, tracks target speed and target position and limits control saturation; online application capability and safety limits; monitors total energy consumption and braking thermal load, and implements soft constraints and load reduction strategies on long slopes and congested sections.

[0171] In one embodiment of the present invention, the following steps are specifically included:

[0172] S510, Speed-Position-Traction / Brake Closed-Loop Control: Based on target speed / position tracking and limiting control saturation, traction and braking commands are executed in a closed loop within the upper limit of dynamic capability;

[0173] formula:

[0174] ;

[0175] in, The target velocity for track segment e (considering) (and safety constraints). For the target position / distance reference of track segment e; The actual position / distance of train set g. , For velocity and position error feedback gain; This is a saturation function (limiting control quantity);

[0176] S520, Capability and Safety Constraints Take Effect Online: Online Application , , Limits are set to prevent slippage and thermal decay while maintaining a safety boundary;

[0177] S530, Energy Consumption and Heat Load Monitoring: Calculation and Soft constraints and load reduction strategies are implemented on long slopes and congested sections to control peak energy consumption and heat load.

[0178] formula:

[0179] ;

[0180] ;

[0181] in, For total energy consumption estimation (traction power integral). For braking heat load estimation (approximate heat caused by speed reduction). This is the heat load weighting coefficient; For positive part operators ( );

[0182] S600, Event Triggered and Scrolling Time-Domain Replanning (RH, including obstacle priority): Sets a scrolling window and locks historical decisions, resolves sub-problems within the window based on event triggers, and dynamically increases relevant weights and constraints; limits the maximum resolving frequency and follows the trigger priority of "safety > health > parking lot > energy consumption > cycle", and injects rescue and bypass tasks when necessary to achieve rapid loss mitigation;

[0183] In one embodiment of the present invention, the following steps are specifically included:

[0184] S610, Scrolling window setting and history locking: Set a scrolling window and freeze historical decisions to ensure that re-decoding only affects future timelines and avoids backtracking impacts on already executed actions;

[0185] S620, Subproblem Resolving and Weight Adjustment: Resolve subproblems within the window and increase relevant weights and constraints; dynamically adjust target weights and constraint boundaries for events.

[0186] formula:

[0187] ;

[0188] in, The rolling window interval is [t, t+H]; H is the window length (seconds or intervals between steps). For local targets within the window (structure is consistent with the global target but localized). As the risk and safety-related weights / constraints increase with the event;

[0189] S630, Event Priority and Frequency Limitation: Triggered according to the priority of "Safety > Health > Parking Lot > Energy Consumption > Cycle", limiting the maximum re-resolution frequency and inserting rescue / bypass tasks to quickly stop losses;

[0190] S700, MARL Local Collaboration and "Safety Barrier" Shielding: Constructing an agent state that includes neighborhood occupancy, parking lot status, task remaining, risk, health, and hydraulic pressure; the reward function simultaneously considers schedule, safety, energy consumption, congestion, and yielding; and adopts a safety barrier shielding strategy to sample and execute only within the set of safe actions, ensuring that hard constraints cannot be broken.

[0191] In one embodiment of the present invention, the following steps are specifically included:

[0192] S710, agent state construction (including health and obstacle information): integrates neighborhood occupancy, parking lot status, task remaining window, risk and health scores to form a state vector for local collaborative decision-making;

[0193] formula:

[0194] ;

[0195] in, For the combined state vector, Let k be the set of neighborhood track segments of locomotive k; Observation of the time band occupied by the neighborhood (before and after) step), Let y be the state vector of the parking lot (capacity, translation window, frozen state). The positive portion of the remaining time window for the task; Risk weighting , For power and braking health rating, For hydraulic availability;

[0196] S720, reward function design (cycle-safety-energy consumption-congestion-yield): with changes in project duration, power consumption, risk occupancy, congestion / deadlock, and successful yielding as reward items, and the weights are optimized to achieve safety priority and collaborative gains;

[0197] formula:

[0198]

[0199] in, For the reward function, This represents the change in duration relative to the baseline. Estimate the power consumption of locomotive k; For time step, Variables indicating congestion or deadlock events; An indicator variable for successfully allowing / high-priority tasks to pass. The reward weighting coefficient (requires parameter tuning);

[0200] S730, Safety Barrier Shielding Strategy (Hard Constraints Cannot Be Broken): Actions not in the set of safe actions are directly shielded, and sampling and execution are only performed within the set of safe actions to ensure that existing constraints and mutual exclusion safety are always valid.

[0201] formula:

[0202]

[0203] in, Let k be the probability distribution of the original strategy for locomotive k. The set of security actions (determined by token / capacity / 2PC / health limit) To ensure that hard constraints cannot be broken, the strategy distribution is designed after the safety barrier is shielded.

[0204] S800, Consistency Verification and Deployment (Obstacle Injection Stress Test): Inject equipment / operating condition obstacles into the digital twin to perform closed-loop consistency verification, evaluate the consistency and recovery time of "timetable-token-turnout-edge control-rolling optimization-MARL guardrail"; deploy the baseline plan and rolling strategy to the edge controller and enable immutable event logs and state snapshots to form auditable deployment and operation records;

[0205] In one embodiment of the present invention, the following steps are specifically included:

[0206] S810, Scene Injection and Closed-Loop Consistency Verification: Inject equipment / operating condition obstacles in the twin environment to verify the consistency and recovery time of "timetable-token-turnout-edge control-rolling optimization-MARL guardrail".

[0207] Timetable consistency: ;

[0208] Token Consistency: ;

[0209] Turnout 2PC Consistency:

[0210] ;

[0211] Edge control limit consistency: ;

[0212] Rolling optimization consistency:

[0213] ;

[0214] Consistency score:

[0215] ;

[0216] ;

[0217] ;

[0218] Assuming the barrier is injected at t0, the recovery time is defined as:

[0219] ;

[0220] The tail risk of recovery time is measured using CVaR:

[0221] ;

[0222] in, This is a dynamic head-to-head distance; For track segment capacity; For occupying variables; Token holding status; For hydraulic availability, , This sets the speed limit and braking deceleration limit for curves. , For actual speed and braking, For the violation indication of dimension m; For dimension weights; For the verification time window; For consistency scoring, Inject moments into obstacles; For recovery time; CVaR confidence level; For CVaR auxiliary variables; For positive part operators; This is the expected value (which can be approximated by scenario weights).

[0223] S820, Baseline Plan Release and Operation Audit: Releases baseline plans and rolling policies to the edge controller, enables immutable event logs and status snapshots, and ensures that shift handover and auditing are traceable.

[0224] formula:

[0225] ;

[0226] ;

[0227] in, For log hashing, The log hash from the previous moment, For the initial log hash, For the starting event, For cryptographic hashing, Record events (including timestamps, entities, operations, parameters, and results);

[0228] index:

[0229] ;

[0230] ;

[0231] ;

[0232] in, For the first Item; For log hashing; For splicing operations; This refers to the number of log entries. The actual number of events. For log coverage; For chain integrity (1 if passed, 0 otherwise); To ensure consistency in deployment (configuration and deployment status must be consistent item by item, which is 1).

[0233] Based on the above process, the following example is given (to solve the coordination between the mine, locomotive and task).

[0234] like Figure 2 The spatiotemporal occupancy Gantt chart shown includes the topology and orbital segments:

[0235] Loading yard Grouping / Gathering Unloading site .

[0236] Track section:

[0237] e1: (2.2km, +1.8% slope);

[0238] e2: (2.8km, +0.6% slope);

[0239] e3: (2.6km, sharp bends, curvature approximately 7.5°);

[0240] e4: (4.2km, -2.3% long downhill);

[0241] e5: Bypass (3.0km, -0.8% slope);

[0242] Turnouts: J1, J2 (hydraulic). Single-track operation, track section capacity 1; dynamic headway. Adjust according to braking capacity and risk.

[0243] Locomotive health score (calculated by S100):

[0244] L1: ; (Good traction / braking capability, suitable for critical sections and heavy loads).

[0245] L2: ; (Load and speed limits apply);

[0246] L3: ; (Mobile / rescue backup);

[0247] Hydraulic availability and shielding (S100): Between 10:15 and 10:45, resulting in (Entry, exit, and pan / grouping operations are prohibited in this window).

[0248] Transportation tasks and priorities:

[0249] T1: 2200t The time window [09:00, 11:00] has a high priority.

[0250] T2: 1500t, Time window [09:10, 10:50], priority is medium.

[0251] S100, Data Acquisition and Limitation Application;

[0252] Data acquisition and cleaning: Integrating locomotive operating conditions (temperature / current / brake heat), track geometry (gradient / curvature / gauge), yard status (occupancy / hydraulics), and on-site events.

[0253] Dynamic capabilities and safety limits (example):

[0254] L1: Dynamic traction limit Dynamic braking deceleration limit .

[0255] L2: Dynamic Traction Limit Dynamic braking deceleration limit .

[0256] Speed ​​limit on curves: E3 due to curvature and load eccentricity. Dynamically tightened to approximately 36 km / h (more stringent under heavy load).

[0257] Blocking: 10:15–10:45 The translation / grouping operation is masked (affecting the "commit phase" of 2PC).

[0258] Obstacle Scene Set This includes measures to address issues such as decreased friction in rainy weather, hydraulic fluctuations, and traction hook anomalies, and is used for robust optimization and rolling replanning.

[0259] S200, Robust / Scene MILP Modeling (Constraints and Objectives);

[0260] Key constraints:

[0261] Track section capacity mutual exclusion and occupancy consistency; dynamic head spacing ;

[0262] Path continuity and transit time constraints; parking lot capacity and shielding ;

[0263] Track gauge compatible Locomotives with large wheel track deviations should avoid the sharp curve section E3;

[0264] Traction-gradient-rolling resistance-acceleration coupling; braking safety distance and thermal load limit;

[0265] Task time window and priority; rescue task injection when towing hook malfunctions;

[0266] Risk weight E3 (sharp bend) and E4 (long downhill slope) have higher weighting;

[0267] Weighted objectives (example weights): Construction period 0.45, Energy consumption 0.25, Risk occupancy 0.10, Parking lot load 0.10. ( Tail risk 0.10.

[0268] S300, in scene set The above solution is consistent with the initial plan;

[0269] Route selection:

[0270] L1 (Executing T1, under heavy load and with relatively high health): (Responsible for key sections).

[0271] L2 (Executing T2, with load and speed limits): (Bypassing e3 to reduce risk).

[0272] Example of time indication (including acceleration / deceleration and safety margin):

[0273] L1: Departure 09:08; e1 09:08–09:13; e2 09:15–09:22; e3 09:25–09:33; e4 09:36–09:48; Arrival 09:50; Uninstallation 20 minutes; T1 completed 10:10.

[0274] L2: Departure 09:12; e1 09:12–09:18; e2 09:20–09:29; e5 09:32–09:46; e4 09:49–10:03; Arrival 10:05; Uninstallation 15 minutes; T2 completed 10:20.

[0275] Occupancy and headroom: maintained in e4 Minute headway; e3 is used only by L1; L2 uses e5 bypass.

[0276] Translational shielding: Plan to avoid entering / leaving between 10:15 and 10:45. The operation ensures There is no need for grouping or translation at this time.

[0277] like Figure 3The timing diagram shown is for the turnout switching two-phase commit (2PC) and track segment token mutual exclusion. Execution S400, token / lease and turnout 2PC (atomicity and mutual exclusion).

[0278] Track segment token Trains must have a token before entering E3 / E4 and must return it upon leaving the depot to avoid ghost occupation and conflicts.

[0279] Two-stage submission for turnout switching (2PC):

[0280] Preparation phase: Verify zero occupancy and hydraulic availability of the target track section ( ).

[0281] Commit phase: Lock the short panning window and write to the immutable event log; if the commit fails, roll back and broadcast a bypass replanning request.

[0282] In this example: because of advance avoidance The window was blocked, and both 2PC commits were successful without any rollback.

[0283] like Figure 4 The edge control closed-loop diagram shown illustrates the execution of S500, which is the edge control closed-loop (speed / position / traction / braking).

[0284] track , Limit the saturation of the control quantity; implement braking thermal management and segmented load reduction on e4 (long downhill) to smooth the speed profile.

[0285] Online application limits: , , .

[0286] Energy consumption and heat load (estimated):

[0287] L1: ;

[0288] L2: ;

[0289] The two cars had brake heat at E4. Peak values ​​are under control.

[0290] like Figure 5 The event-triggered scrolling replanning graph shown executes S600, scrolling the window and triggering the replanning.

[0291] Scrolling window Lock the executed historical decisions; limit the maximum re-decomposition frequency.

[0292] Trigger priority: "Safety > Health > Parking Lot > Energy Consumption > Cycle".

[0293] Event example:

[0294] 10:25 Light rain intensifies: Improve E3 / E4 performance and The target speed is reduced; only the unexecuted segments are slightly re-decoded, which does not affect the completion time of T1 / T2.

[0295] If the L2 braking temperature is detected to be close to the threshold, the speed curve of e4 will be adjusted downward and the head distance will be increased to ensure safety redundancy.

[0296] S700, MARL guardrails and yielding;

[0297] Agent state: Neighborhood occupancy Parking lot status Remaining tasks ,risk ,healthy / Hydraulics .

[0298] Incentives: Considering overall timeline, safety, energy consumption, congestion, and yielding; only applicable to... Internal sampling and execution.

[0299] Purpose: To implement micro-yield and speed fine-tuning, in line with the MILP plan, without violating hard constraints.

[0300] S800, Digital Twin Verification and Audit;

[0301] Injection obstacles (hydraulic unavailability, decreased friction in rainy weather, temporary speed limit), closed-loop verification of the consistency and recovery time of "timetable-token-turnout-edge control-rolling optimization-MARL guardrail".

[0302] Deploy baseline plans and rolling policies to the edge controller; enable immutable event logs and state snapshots to create auditable records for easy post-event traceability and compliance.

[0303] It should be noted that the following parameters also need explanation:

[0304] Scene set A set of obstacles and operating conditions used for robust / scenario optimization and digital twin verification. It can include scenarios such as weather (rainfall intensity / temperature), equipment (hydraulic status / traction hook malfunction), and operations (temporary speed limit / blockage); each scenario is assigned a weight. Based on an approximate probability distribution.

[0305] Dynamic head distance On-orbit segment At different times The minimum time / space interval required to ensure safety and braking redundancy. This interval is dynamically adjusted based on factors such as braking capacity, gradient, curvature, and risk weight; single-line capacity. In this case, window constraints are often used to achieve the following: .

[0306] Hydraulic availability / shielding When the hydraulic system of the depot / turnout is at a certain time Unavailable When ), mask variables This indicates that operations such as entering, exiting, panning, and grouping are prohibited during this window; planning and 2PC submissions must both avoid this window.

[0307] Track capacity The maximum number of trains that can pass simultaneously on a single-track section in the same direction at any given time; this example uses... (Mutually exclusive).

[0308] Track gauge compatible Determine the wheel track deviation of the train set / locomotive ( ) and track gauge tolerance ( This involves determining whether the binary constraints meet safety requirements on a specific track segment (especially curves). If incompatible, selecting that segment is prohibited. ).

[0309] Risk weight Weighting the risk level of track segments and time points (e.g., sharp curves e3 and long downhill slopes e4 have higher weights) is used to tighten the objective function and limits; events such as rain or congestion can improve the response. .

[0310] Health score Braking health The overall health and braking health of locomotive k typically range from 0 to 1. High health leads to higher traction / braking limits and more lenient speed plans; low health results in load and speed restrictions and avoidance of high-risk sections.

[0311] Speed ​​limit on curves Based on curvature The speed limit is dynamically tightened by load eccentricity and wheel-rail conditions; it is even stricter under heavy load on sharp curves (e3).

[0312] Traction-gradient-rolling resistance-acceleration coupling: a combination of kinematic / dynamic equilibrium constraints, typically in the form of... ,in Including slope resistance Rolling resistance Additional resistance to curves, etc.; when braking in parallel, the upper limit of deceleration and thermal load must be met.

[0313] Track segment token / lease A token must be held before entering a segment and must be returned upon leaving; the token carries a TTL (Trust Time Limit) to prevent "ghost occupancy". This works in conjunction with capacity exclusion constraints to prevent conflicts.

[0314] Two-stage turnout submission (2PC): Preparation stage checks target track section for zero occupancy and hydraulic availability. During the commit phase, the translation window is briefly locked and written to an immutable log; if it fails, it rolls back and triggers a bypass replanning.

[0315] Edge control closed loop: Performing closed-loop tracking of speed-position-traction / braking on the field controller ( , Furthermore, saturation prevention measures should be implemented for the control quantity; the sampling period and delay must be matched with safety requirements.

[0316] Soft constraints and load reduction: To manage the risk of braking heat or congestion on long slopes, punitive constraints are imposed on the target speed / power, and risks and energy consumption peaks are reduced through temporary load reduction or segmented relay.

[0317] Construction period (make span) Total time from the start of the first task to the completion of the last task; used to measure overall efficiency.

[0318] Total energy consumption The power-time integral is often expressed as... Approximate representation; used for energy consumption optimization and operation and maintenance assessment.

[0319] Braking thermal load The heat power or temperature surcharge consumed by braking can be normalized to [0,1] to monitor peak values ​​and safety redundancy; it is related to the speed curve and load reduction strategy on the downhill section.

[0320] and Tail risk measurement. Quantile threshold, The expected tail loss exceeding this threshold; The value is usually taken as 0.90–0.95.

[0321] Scrolling window In time Nearby planning / re-resolution horizon; "Lock history" indicates that implemented decisions will not be changed; Limit the maximum re-resolution frequency to avoid oscillations.

[0322] MARL (Multi-Agent Reinforcement Learning): Introduces agent policies in multi-train cooperative scenarios for micro-level yielding and speed fine-tuning; rewards take into account schedule, safety, energy consumption, congestion, and yielding.

[0323] Safety Actions Collection In state The set of actions that satisfy hard constraints such as tokens, capacity, 2PC, and health limits; the strategy uses a shielding method to ensure that the action does not go out of bounds.

[0324] Micro-level yielding: Under the premise of meeting hard constraints, fine-grained speed / timing adjustments are made to achieve coordinated passage and conflict resolution between adjacent segments; consistent with the MILP baseline plan.

[0325] Immutable event logs and hash chains: Record event hashes in a format that ensures integrity and audit traceability.

[0326] State snapshot: Periodically records the key states of the controller and scheduler (tokens, occupancy, timetables, 2PC events, etc.) for backtracking and compliance auditing.

[0327] "Transfer / Grouping / Passing" in the parking lot:

[0328] Translation: Track switching or short-distance movement within the parking lot;

[0329] Grouping: Reorganizing vehicles / locomotives to match mission loads;

[0330] It will allow trains to pass each other / give way in a single-track environment.

[0331] Bypass (e5): An alternative route that bypasses the main curve / congested section, used for risk avoidance and improved robustness; may increase mileage but reduce the risk of accidents / heat load.

[0332] This invention also proposes a multi-machine collaborative scheduling intelligent transportation system for mines, comprising:

[0333] Equipment health perception and digital twin initialization module: aggregates sensor data to generate health scores, constrains traction / braking capabilities and constructs obstacle scenarios; completes parameter and scenario initialization in the digital twin;

[0334] Robust scheduling modeling and objective function module: Establish constraints such as occupancy, capacity mutual exclusion, speed / acceleration, parking lot capacity and curve limits; reconstruct the robust objective function with CVaR / scene weights;

[0335] Baseline solution and train formation path generation module: Solve robust / scenario MILP on the scenario set, and output task timetable, train formation path and depot translation sequence; formulate train formation strategy of "dual-machine segmented relay and depot staggered peak";

[0336] Safety Mutual Exclusion and Turnout Two-Phase Submission Module: Issues track segment tokens and lease revocation strategies, executes turnout 2PC preparation / submission and failure rollback; freezes adjacent track segments and broadcasts clearance when necessary;

[0337] Edge control execution and energy consumption constraint module: Locomotive-side closed-loop tracking of speed / position / traction braking to implement energy consumption and thermal load monitoring; speed limit and load reduction during health degradation to avoid slippage and thermal decay;

[0338] Event-Driven Rolling Replanning (RHO) module: Fixes historical decisions and updates future timelines within a rolling window; increases risk weights based on event priority and inserts rescue / bypass tasks;

[0339] MARL Local Coordination and Safety Barrier Module: Learns yielding and cooperative behaviors based on state information; uses safety barriers to block illegal actions and reduce the risk of congestion and deadlock;

[0340] Consistency verification and deployment module: Inject equipment / operating condition obstacles into the twin environment to perform stress tests, verify the closed-loop consistency of "timetable, token, turnout, control, RHO, MARL guardrail", generate baseline plan and deploy, and enable immutable event logs and snapshots.

[0341] The embodiments of the present invention have been described above, but the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention, all of which are within the protection scope of the present invention.

Claims

1. A mine intelligent transportation method of multi-machine cooperative scheduling, characterized in that, The method comprises the following steps: S100, collecting the working condition and geometric data of the cleaning locomotive, yard, and track section, calculating the health score and refreshing the dynamic upper limit of traction / braking, determining the hydraulic availability and shielding translation, dynamically tightening the curve speed limit / acceleration according to the load center of gravity deviation and wheel tread deviation, constructing an obstacle scene set and injecting a digital twin; S200, a robust / scenario MILP model is established to minimize the weighted target composed of construction period, energy consumption, risk occupancy, yard load, and tail risk; S300, solving the robust / scenario MILP on the obstacle scene set, generating the task schedule, path selection, occupancy time sequence, and yard translation sequence, and optimizing the marshalling and segmentation relay according to the health score and working condition; S400, implementing track segment token and lease management, adopting two-phase commit for switch switching, verifying zero occupancy and hydraulic availability in the preparation phase, locking the translation window in the commit phase, failing to roll back and broadcasting the evacuation and bypass re-planning; S500, performing speed-position-traction / brake closed-loop control on the edge controller, tracking the target speed and target position and limiting the control quantity saturation, applying the online capability and safety limit, monitoring the total energy consumption and brake heat load, and implementing soft constraint and load reduction strategy on long slope and congestion section; S600, set a rolling window and lock the historical decision, re-solve the sub-problem in the window according to the event trigger, dynamically improve the related weight and constraint, limit the maximum re-solution frequency and follow the trigger priority, inject rescue and bypass tasks to achieve fast stop loss; S700, construct the agent state, the reward function considers multiple factors, and the safety barrier shielding strategy only samples and executes in the safe action set; S800, inject equipment / working condition obstacles into the digital twin for closed-loop consistency checking, evaluate consistency and recovery time, online baseline plan and rolling strategy on edge controller, and enable immutable event log and state snapshot to form auditable release and operation records.

2. The mine intelligent transportation method of multi-machine cooperative scheduling according to claim 1, characterized in that, In S200, the robust / scenario MILP model covers track occupancy and capacity mutual exclusion, dynamic headway, path continuity and passing time, track gauge compatibility, yard capacity and translation shielding, curve speed limit / acceleration, traction-slope-rolling resistance-acceleration coupling, brake safety distance, task time window and priority, rescue task injection of traction hook anomaly, and scene factors of risk weight.

3. The mine intelligent transportation method of multi-machine cooperative scheduling according to claim 1, characterized in that, The scene factors specifically include the following: Track occupancy and capacity mutual exclusion constraint: implement single track capacity mutual exclusion and time step occupancy constraint to ensure unique track occupancy at the same time and reserve safety headway; Dynamic headway step calculation: calculate the dynamic headway steps according to the current speed and brake health, and increase the headway to ensure safe braking space in high-risk or thermal decay; Path continuity and passing time consistency: ensure that the occupancy only occurs on the selected path and meet the minimum passing steps determined by the length / speed limit, avoiding "segment skipping" and "overspeed crossing"; Track gauge compatibility and wheel tread constraint: determine compatibility through a determination function, and prohibit segments with incompatible wheel tread or track gauge from being included in path selection; Parking capacity and translation shield: Perform parking lot capacity upper limit and translation action shielding strategy, only allow translation when hydraulic availability is available, clear the occupation during translation to avoid shear conflict; Bend speed limit / acceleration constraint: When the bend is occupied, dynamic constraints of dynamic maximum allowed speed and maximum allowed acceleration are applied to prevent the risk of tilting and derailment caused by abnormal center of gravity / wheelbase; Traction-slope-rolling resistance-acceleration coupling and capacity constraint: Calculate the required traction force for segment passage and compare it with the dynamic traction upper limit. If it exceeds, limit speed / load or change route to ensure dynamic feasibility; Braking safety distance and thermal decay effect: Evaluate the braking distance and reserve a safety margin. Reduce the target speed and increase the headway when the thermal decay occurs; Task time window and priority inheritance: All tasks are scheduled within the time window, and priority tasks ensure non-default; Rescue / recovery task injection for abnormal traction hook: Generate a rescue / recovery task immediately when an abnormal traction hook state is detected, freeze the related token, and trigger rolling re-planning and bypass path; Risk weight calculation and scenario risk modeling: Combine slope, curvature, human factors, and congestion signals to form a risk weight that changes with time and location for target weighting.

4. The mine intelligent transportation method of multi-machine cooperative scheduling according to claim 1, characterized in that, In S300, the optimization target is to have high-health locomotives undertake the critical path and heavy load, and low-health locomotives limit load and speed.

5. The mine intelligent transportation method of multi-machine cooperative scheduling according to claim 1, characterized in that, In S600, the priority triggering is safety, health, parking, energy consumption, and cycle in turn.

6. The mine intelligent transportation method of multi-machine cooperative scheduling according to claim 1, characterized in that, In S700, the agent state includes neighborhood occupation, parking state, task remaining, risk, health, and hydraulic.

7. The mine intelligent transportation method of multi-machine cooperative scheduling according to claim 1, characterized in that, In S700, the reward function considers schedule, safety, energy consumption, congestion, and yielding.

8. The mine intelligent transportation method of multi-machine cooperative scheduling according to claim 1, characterized in that, The safety guardrail shielding strategy means directly shielding actions that are not in the safe action set, and only sampling and executing within the safe action set.

9. The mine intelligent transportation method of multi-machine cooperative scheduling according to claim 1, characterized in that, In S800, the consistency and recovery time length evaluation includes schedule, token, turnout, edge control, rolling optimization, and MARL guardrail.

10. A mine intelligent transportation system of multi-machine cooperative scheduling, characterized in that, The steps for performing a mine intelligent transportation method as claimed in any one of claims 1-9, comprising: Device health perception and digital twin initialization module: aggregate sensor streams to generate a health score, constrain traction / braking capacity, and construct obstacle scenarios; complete parameter and scenario initialization in the twin; Robust scheduling modeling and objective function module: establish occupancy, capacity mutual exclusion, speed / acceleration, parking capacity, and bend limit constraints; reconstruct a robust objective function containing CVaR / scenario weight; Baseline solution and marshalling path generation module: solve robust / scenario MILP on the scenario set, output task schedule, train set path, and parking translation sequence; develop a "two-machine segmented relay, parking peak-shifting" marshalling strategy; Safety mutual exclusion and turnout two-phase commit module: issue track segment tokens and lease recovery strategy, perform turnout 2PC preparation / commit and failure rollback; When the token lease expires, recycle the token, freeze the adjacent segment, and broadcast the eviction; when the turnout execution fails, roll back and inject a bypass re-planning; Edge control execution and energy consumption constraint module: locomotive end closed-loop tracking speed / position / traction braking, implement energy consumption and thermal load monitoring; limit speed and load when health degrades to avoid slipping and thermal decay; Event-driven rolling re-planning module: fix historical decisions, update future timing within rolling window; boost risk weight by event priority, insert rescue / bypass tasks; MARL local coordination and safety fence module: learn yielding and coordination behavior based on state information; use safety fence to shield illegal actions, reduce congestion and deadlock risk; Consistency check and online release module: inject device / working condition obstacles into twin environment for stress testing, check the closed-loop consistency of "timetable, token, turnout, control, RHO, MARL fence"; generate baseline plan and release, enable immutable event log and snapshot.

Images