Dynamic cooperative scheduling method and system for urban regional tunnel network

Abstract

The invention discloses a dynamic cooperative scheduling method and system for an urban area tunnel network, and belongs to the technical field of tunnel network intelligent management, and the method comprises the steps of environment perception modeling, multi-dimensional prediction evaluation, dynamic cooperative scheduling and strategy execution optimization. According to the method, a space-time diagram multi-dimensional prediction method is adopted to perform multi-dimensional prediction evaluation, unified modeling and short-time prediction are performed on traffic operation, an air environment, an equipment state and an emergency channel condition in the tunnel network, potential congestion points, pollution hotspots and high-risk areas can be identified in advance, and the risk identification efficiency and accuracy are improved; a multi-objective optimization tunnel network intelligent scheduling method is adopted for dynamic cooperative scheduling, comprehensive balance of traffic smoothness, air quality and energy consumption is achieved, and on the premise that safety constraints are met, the performability of a scheduling strategy and the timeliness of emergency response are improved through layered optimization and an emergency priority strategy.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent management of tunnel networks, and particularly relates to a dynamic collaborative scheduling method and system for a city area tunnel network. BACKGROUND

[0002] The dynamic collaborative scheduling of a city area tunnel network is a kind of modeling analysis of multi-dimensional data related to the tunnel network by using artificial intelligence technology, aiming to comprehensively evaluate the operation state of the tunnel network, identify potential risk areas in advance, optimize traffic flow smoothness, air quality and equipment energy consumption, and ensure that the scheduling strategy can be quickly adjusted under different risk levels, so as to improve the operation efficiency, safety and emergency response capability of the tunnel network.

[0003] However, in the process of dynamic collaborative scheduling of a city area tunnel network, there are technical problems that traffic flow mutation is frequent, exhaust pollutants are easy to accumulate and diffuse in the tunnel, the operation state of the ventilation and smoke exhaust equipment is complex, and it is difficult to timely grasp the situation of the emergency passage when an accident occurs, thereby leading to the difficulty in timely identification of risks; there are technical problems that traffic efficiency, air quality, energy consumption and safety targets are difficult to balance, and the physical constraints of the equipment and the upper limit of the carrying capacity of the emergency passage are difficult to guarantee, and the scheduling strategy lacks differentiation under different risk levels. SUMMARY

[0004] In view of the above problems, in order to overcome the defects of the prior art, the application provides a dynamic collaborative scheduling method and system for a city area tunnel network, which creatively uses a spatiotemporal graph multi-dimensional prediction method for multi-dimensional prediction and evaluation, and uniformly models and short-time predicts the traffic operation, air environment, equipment state and emergency passage condition in the tunnel network, so as to identify potential congestion points, pollution hotspots and high-risk areas in advance, and improve the risk identification efficiency and accuracy; in view of the technical problems that traffic efficiency, air quality, energy consumption and safety targets are difficult to balance, and the physical constraints of the equipment and the upper limit of the carrying capacity of the emergency passage are difficult to guarantee, and the scheduling strategy lacks differentiation under different risk levels, the application creatively uses a multi-objective optimization tunnel network intelligent scheduling method for dynamic collaborative scheduling, realizes the comprehensive balance of traffic flow smoothness, air quality and energy consumption, and improves the executability of the scheduling strategy and the timeliness of the emergency response under the premise of meeting the safety constraints.

[0005] The technical scheme adopted by the application is as follows: the application provides a dynamic collaborative scheduling method for a city area tunnel network, which comprises the following steps:

[0006] Step S1: environment perception modeling;

[0007] Step S2: multi-dimensional prediction evaluation;

[0008] Step S3: dynamic collaborative scheduling;

[0009] Step S4: strategy execution optimization.

[0010] Further, in step S1, the environment perception modeling is used for perception modeling of the urban area tunnel network, specifically collecting traffic data, environment data, device state data and emergency passage data from the urban area tunnel network, and preprocessing, and then constructing a tunnel multi-relation spatio-temporal graph;

[0011] The tunnel multi-relation spatio-temporal graph is constructed, specifically by setting section nodes, device nodes and emergency passage nodes, and then setting traffic connection edges and device action edges to establish a tunnel multi-relation spatio-temporal graph;

[0012] The section node is specifically that each tunnel section is taken as a section node, and the section node attributes are set according to the traffic data and the environment data;

[0013] The device node is specifically that each key device is taken as a device node, the key device includes a fan, a gate valve, lighting facilities and a smoke exhaust device, and the device node attributes are set according to the device state data;

[0014] The emergency passage node is specifically that the emergency passage is taken as an emergency passage node, and the emergency passage node attributes are set according to the emergency passage data;

[0015] The traffic connection edge is specifically that if two tunnel sections are connected, a traffic connection edge is established between the corresponding two section nodes, and if a tunnel section and an emergency passage are adjacent, a traffic connection edge is established between the corresponding section node and the emergency passage node;

[0016] The device action edge is specifically that if a tunnel section is within the action range of a key device, a device action edge is established between the corresponding section node and the device node, and if an emergency passage is within the action range of a key device, a device action edge is established between the corresponding emergency passage node and the device node.

[0017] Further, in step S2, the multi-dimensional prediction evaluation is used for predicting node states and evaluating node risks, specifically according to the tunnel multi-relation spatio-temporal graph, a multi-dimensional prediction evaluation is performed by using a spatio-temporal graph multi-dimensional prediction method to obtain a multi-dimensional tunnel state prediction tensor, including the following steps:

[0018] Step S21: node feature encoding, specifically, first standardizing each node attribute to obtain a node feature vector, and then encoding the tunnel multi-relation spatio-temporal graph node and its neighbors through a constructed heterogeneous graph attention network to obtain a node high-dimensional feature vector;

[0019] The constructed heterogeneous graph attention network specifically receives the tunnel multi-relation spatio-temporal graph through an input layer, sets two layers of heterogeneous graph attention layers, and separately calculates attention coefficients for different types of neighbor nodes through a multi-head attention mechanism of each layer. Then, each node is aggregated through neighbor weighting to generate a node high-dimensional feature vector.

[0020] Step S22: short-time state prediction, specifically, according to the node high-dimensional feature vector and the tunnel multi-relation spatio-temporal graph, spatial convolution is performed through a standard spatio-temporal graph convolution network to capture the spatial dependency between nodes to obtain a node convolution feature matrix; a time series model is established for the node convolution feature matrix through a gated recurrent unit to obtain a node initial prediction state matrix, and a multi-modal coupling correction mechanism is introduced to weight and correct the node initial prediction state matrix to generate a node short-time prediction state matrix.

[0021] Step S23: comprehensive risk assessment, specifically, each node is scored according to the node short-time prediction state matrix, and the scores of different node types are fused by weighting to generate a node comprehensive risk value. The node comprehensive risk value is divided according to a risk threshold to obtain a node risk level label, and the node risk level label includes normal, alert, and emergency.

[0022] Step S24: predicted tensor integration, specifically, a multi-dimensional tunnel state prediction vector is constructed for each node according to the node short-time prediction state matrix, the node comprehensive risk value, and the node risk level label. Then, the multi-dimensional tunnel state prediction vectors of all nodes are stacked to generate a multi-dimensional tunnel state prediction tensor.

[0023] Further, in step S3, the dynamic collaborative scheduling, specifically, according to the tunnel multi-relation spatio-temporal graph and the multi-dimensional tunnel state prediction tensor, a multi-objective optimization tunnel network intelligent scheduling method is used for dynamic collaborative scheduling to obtain an optimal tunnel network scheduling strategy, including the following steps:

[0024] Step S31: multi-objective optimization modeling, specifically, according to the multi-dimensional tunnel state prediction tensor and the tunnel multi-relation spatio-temporal graph, a multi-objective optimization function including traffic smoothness, air quality, device energy consumption, and risk control is constructed in combination with device physical constraints and emergency passage carrying capacity;

[0025] Step S32: hierarchical strategy solving, specifically, hierarchical optimization solving is performed according to the node risk level label to obtain a hierarchical optimization scheduling action set.

[0026] The hierarchical optimization solving, in particular, adopts a rolling prediction control method for normal nodes and a distributed multi-agent reinforcement learning method for warning nodes.

[0027] The state space is a node short-time prediction state matrix, the action space is a vehicle speed guide, a fan speed regulation and a valve opening regulation, the reward function is the multi-objective optimization function, each agent independently decides in a local state, and collaborates in training through a central value network, so that a multi-objective balance of traffic flow smoothness, air quality and equipment energy consumption is achieved, and a scheduling action of the warning node is obtained.

[0028] Step S33: safety emergency adjustment, specifically, an emergency strategy library is constructed, and the emergency strategy library is called for adjustment when an emergency risk level is detected, so that an emergency scheduling action set is obtained, and the hierarchical optimization scheduling action set and the emergency scheduling action set are integrated, in the integration process, the emergency scheduling action set preferentially covers the hierarchical optimization scheduling action set, and the best tunnel network scheduling strategy is obtained.

[0029] Further, in step S4, the strategy execution optimization, specifically, the action is issued to the traffic guidance system, the ventilation control system, the valve control system and the emergency passage management system according to the best tunnel network scheduling strategy, and the execution feedback is collected, and the scheduling action signal is obtained.

[0030] The application provides a kind of urban area tunnel network dynamic collaborative scheduling system, including: environment perception modeling module, multi-dimensional prediction evaluation module, dynamic collaborative scheduling module and strategy execution optimization module;

[0031] The environment perception modeling module is used for environment perception modeling, and the tunnel multi-relation space-time graph is obtained by environment perception modeling, and the tunnel multi-relation space-time graph is sent to the multi-dimensional prediction evaluation module and the dynamic collaborative scheduling module;

[0032] The multi-dimensional prediction evaluation module is used for multi-dimensional prediction evaluation, and the multi-dimensional tunnel state prediction tensor is obtained by multi-dimensional prediction evaluation, and the multi-dimensional tunnel state prediction tensor is sent to the dynamic collaborative scheduling module;

[0033] The dynamic collaborative scheduling module is used for dynamic collaborative scheduling, and the best tunnel network scheduling strategy is obtained by dynamic collaborative scheduling, and the best tunnel network scheduling strategy is sent to the strategy execution optimization module;

[0034] The strategy execution optimization module is used for strategy execution optimization, and the scheduling action signal is obtained.

[0035] The application has the following beneficial effects by using the above scheme:

[0036] (1) In the process of dynamic collaborative scheduling of urban regional tunnel network, there are frequent traffic flow mutations, tail gas pollutants are easy to accumulate and diffuse in tunnels, the operation state of ventilation and smoke exhaust equipment is complex, and the condition of emergency passage at the time of accident is difficult to grasp in time, thereby leading to the technical problem that the risk is difficult to identify in time, the space-time graph multi-dimensional prediction method is creatively used for multi-dimensional prediction and evaluation, the traffic operation, air environment, equipment state and emergency passage condition in the tunnel network are uniformly modeled and short-time predicted, potential congestion points, pollution hot spots and high-risk areas can be identified in advance, and the risk identification efficiency and accuracy are improved;

[0037] (2) In the process of dynamic collaborative scheduling of urban regional tunnel network, traffic efficiency, air quality, energy consumption and safety targets are difficult to balance, device physical constraints and emergency passage bearing upper limit are difficult to guarantee, and scheduling strategies under different risk levels lack differentiation, the multi-objective optimization tunnel network intelligent scheduling method is creatively used for dynamic collaborative scheduling, the comprehensive balance of traffic fluency, air quality and energy consumption is realized, and on the premise of meeting safety constraints, the executable of scheduling strategy and the timeliness of emergency response are improved through hierarchical optimization and emergency priority strategy. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 A flowchart of a dynamic collaborative scheduling method of an urban regional tunnel network is provided for the present application;

[0039] Figure 2 A module schematic diagram of a dynamic collaborative scheduling system of an urban regional tunnel network is provided for the present application;

[0040] Figure 3 A flowchart of step S2 is provided;

[0041] Figure 4 A flowchart of step S3 is provided.

[0042] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation on the present application. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application, obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments; based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0044] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0045] Embodiment one, refer to Figure 1 The present application provides a kind of urban area tunnel network dynamic coordination scheduling method, which comprises the following steps:

[0046] Step S1: environmental perception modeling;

[0047] Step S2: multi-dimensional prediction and evaluation;

[0048] Step S3: dynamic coordination scheduling;

[0049] Step S4: strategy execution optimization.

[0050] Embodiment two, refer to Figure 1 The embodiment is based on the above embodiment, in step S1, the environmental perception modeling is used for the perception modeling of urban area tunnel network, specifically from urban area tunnel network collection traffic data, environmental data, equipment state data and emergency passage data, and pre-processing is carried out, and then tunnel multi-relationship space-time graph is constructed;

[0051] The traffic data includes the speed, traffic flow, vehicle density and vehicle type of each section of the tunnel;

[0052] The environmental data includes CO concentration, NOx concentration, PM2.5 concentration, visibility, temperature and humidity in the tunnel;

[0053] The equipment state data includes fan speed, valve opening, lighting level and exhaust equipment state;

[0054] The emergency passage data includes emergency passage width, length, carrying capacity, temperature, smoke concentration and visibility;

[0055] The pre-processing includes data cleaning, uniform timestamp and alignment;

[0056] The tunnel multi-relationship space-time graph is constructed, specifically setting section node, equipment node and emergency passage node, and then setting traffic connection edge and equipment action edge to establish tunnel multi-relationship space-time graph;

[0057] The section node, specifically, each tunnel section is regarded as a section node, and the section node attribute is set according to traffic data and environmental data;

[0058] The device node, specifically taking each key device as a device node, the key device including a fan, a gate valve, a lighting facility and a smoke exhaust device, setting device node attributes according to device state data;

[0059] The emergency passage node, specifically taking an emergency passage as an emergency passage node, setting emergency passage node attributes according to emergency passage data;

[0060] The traffic connection edge, specifically if two tunnel sections are connected, a traffic connection edge is established between the corresponding two section nodes, and if a tunnel section and an emergency passage are adjacent, a traffic connection edge is established between the corresponding section node and the emergency passage node;

[0061] The device action edge, specifically if a tunnel section is within the action range of a key device, a device action edge is established between the corresponding section node and the device node, and if an emergency passage is within the action range of a key device, a device action edge is established between the corresponding emergency passage node and the device node.

[0062] Embodiment three, refer to Figure 1 and Figure 3 This embodiment is based on the above embodiment, in step S2, the multi-dimensional prediction evaluation is used to predict node state and evaluate node risk, specifically, according to the tunnel multi-relation space-time graph, a space-time graph multi-dimensional prediction method is used for multi-dimensional prediction evaluation, and a multi-dimensional tunnel state prediction tensor is obtained, including the following steps:

[0063] Step S21: node feature coding, specifically, first standardize each node attribute to obtain a node feature vector, and then encode the tunnel multi-relation space-time graph node and its neighbors through a constructed heterogeneous graph attention network to obtain a node high-dimensional feature vector;

[0064] The constructed heterogeneous graph attention network specifically receives the tunnel multi-relation space-time graph through an input layer, sets two layers of heterogeneous graph attention layers, the multi-head attention mechanism of each layer separately calculates the attention coefficients for different types of neighbor nodes, and then aggregates each node through neighbors to generate a node high-dimensional feature vector, the calculation formula is:

[0065] ;

[0066] In the formula, h v is the vth node high-dimensional feature vector, v is the node index, is a ReLU activation function, W0 is a node mapping weight matrix, x v is the vth node feature vector, a is the first index of the node type, A v is the neighbor type set of the vth node, u is the neighbor node index, Na (v) is the type a neighbor node set of the vth node, is the attention coefficient of the u th neighbor node in type a of the v th node, W [a] is the neighbor node encoding mapping weight matrix of type a, x u is the u th neighbor node feature vector;

[0067] Step S22: short-term state prediction, specifically, according to the node high-dimensional feature vector and the tunnel multi-relation space-time graph, the spatial convolution is performed through the standard space-time graph convolution network to capture the spatial dependence relationship between nodes, and the node convolution feature matrix is obtained; the node convolution feature matrix is modeled through the gate recurrent unit to obtain the node initial prediction state matrix, and then the multi-modal coupling correction mechanism is introduced to weight and correct the node initial prediction state matrix, and the node short-term prediction state matrix is generated.

[0068] The node short-term prediction state matrix includes the short-term prediction state of each node, wherein the section node prediction state includes vehicle speed, vehicle flow, vehicle density, CO concentration, NOx concentration, PM2.5 concentration, visibility, temperature and humidity, the equipment stage prediction state includes fan speed, valve opening, lighting level and exhaust equipment state, and the emergency passage node prediction state includes emergency passage temperature, smoke concentration and visibility.

[0069] The calculation formula of the multi-modal coupling correction mechanism is:

[0070] ;

[0071] In the formula, is the short-term prediction state of the v th node at the t+1 time step, t is the time step index, is the initial prediction state of the v th node at the t+1 time step, is the type a adjustment weight, is the neighbor node correction mapping weight matrix of type a, is the initial prediction state of the u th neighbor node at the t time step, is the initial prediction state of the v th node at the t time step;

[0072] Step S23: comprehensive risk assessment, specifically, according to the node short-term prediction state matrix, the risk score of each node is calculated, then the weighted fusion is performed for different node types to generate the node comprehensive risk value, and the node comprehensive risk value is divided according to the risk threshold to obtain the node risk level label, wherein the node risk level label includes normal, alert and emergency.

[0073] The calculation formula of the node comprehensive risk value is:

[0074] ;

[0075] wherein R v is the node comprehensive risk value, b is the node type second index, s is the section node type, d is the device node type, e is the emergency passage node type, is the node self weight, is an indicator function, which takes the value of 1 when the condition is satisfied, otherwise takes the value of 0, and is used to ensure that the ontology risk only takes effect on the self node type, is the type of the vth node, f b (·) is the risk score function for the type b, is the short-time prediction state of the vth node in the type b, is the neighbor node weight of the type b, is the set of neighbor nodes of the vth node in the type b, is the weight of the uth neighbor node in the type b to the vth node, is the short-time prediction state of the uth neighbor node in the type b;

[0076] Preferably, the risk score function is set for different node types, specifically, the risk score of the section node type is obtained by a lightweight LSTM model, the traffic congestion risk and the environmental abnormal risk are obtained, the risk score of the device node type is obtained by an isolation forest model, the device abnormal risk is obtained, and the risk score of the emergency passage node type is obtained by a graph neural network model, and the emergency passage traffic safety risk is obtained.

[0077] Step S24: prediction tensor integration, specifically, a multi-dimensional tunnel state prediction vector is constructed for each node according to the node short-time prediction state matrix, the node comprehensive risk value and the node risk level label, and then the multi-dimensional tunnel state prediction vectors of all nodes are stacked to generate a multi-dimensional tunnel state prediction tensor;

[0078] The calculation formula of the multi-dimensional tunnel state prediction vector is:

[0079] ;

[0080] wherein RiskLevel v is the risk level label of the vth node;

[0081] The calculation formula of the multi-dimensional tunnel state prediction tensor is:

[0082] ;

[0083] wherein, is the multi-dimensional tunnel state prediction tensor, specifically, the multi-dimensional tunnel state prediction tensor of all nodes in a multi-dimensional tunnel state prediction vector of a time step, is a prediction time span, V is a tunnel multi-relation spatio-temporal graph node set, is a two-dimensional tensor with a size of , |V| is the number of tunnel multi-relation spatio-temporal graph nodes, and F is the dimension of the multi-dimensional tunnel state prediction vector.

[0084] By performing the above operations, for the technical problems that in the process of dynamic collaborative scheduling of urban area tunnel network, traffic flow mutation frequently occurs, exhaust pollutants are easy to accumulate and diffuse in tunnels, the operation state of ventilation and smoke exhaust equipment is complex, and the situation of emergency passages at the time of accidents is difficult to grasp in time, thereby leading to difficulty in timely identification of risks, the present scheme creatively uses a spatio-temporal graph multi-dimensional prediction method for multi-dimensional prediction evaluation, unified modeling and short-time prediction of traffic operation, air environment, equipment state and emergency passage conditions in the tunnel network, which can identify potential congestion points, pollution hotspots and high-risk areas in advance, and improve the risk identification efficiency and accuracy.

[0085] In the embodiment four, referring to Figure 1 and Figure 4 , the embodiment is based on the above-mentioned embodiments, in step S3, the dynamic collaborative scheduling is specifically performed by using a multi-objective optimization tunnel network intelligent scheduling method according to the tunnel multi-relation spatio-temporal graph and the multi-dimensional tunnel state prediction tensor, to obtain an optimal tunnel network scheduling strategy, including the following steps:

[0086] Step S31: multi-objective optimization modeling, specifically, a multi-objective optimization function including traffic fluency, air quality, equipment energy consumption and risk control is constructed according to the multi-dimensional tunnel state prediction tensor and the tunnel multi-relation spatio-temporal graph, in combination with device physical constraints and emergency passage bearing capacity; the multi-objective optimization function is used as a reward function of reinforcement learning in step S32 for solving a hierarchical strategy;

[0087] The calculation formula of the multi-objective optimization function is:

[0088] ;

[0089] In the formula, r v (t) is a multi-objective optimization function of the tth time step, w1 is a traffic fluency weight, is a traffic fluency loss term of the tth time step, w2 is an air quality weight, is an air quality loss term of the tth time step, w3 is an equipment energy consumption weight, is an equipment energy consumption loss term of the tth time step, w4 is a risk control weight, and R v (t) is a node comprehensive risk value of the tth time step, is a device physical constraint weight, is the device physical constraint violation penalty at time step t, is the emergency channel carrying weight, is the emergency channel carrying capacity penalty at time step t;

[0090] Preferably, the calculation formula of the traffic flow loss term is:

[0091] ;

[0092] Where, is the smooth speed of the segment node, v v (t) is the actual average velocity of the segment node at the tth time step, is the density weight, is the density of segment nodes at time step t, is the critical density of segment nodes;

[0093] Preferably, the calculation formula of the air quality loss term is:

[0094] ;

[0095] Where p is the air quality impact factor index, is the weight of the air quality influencing factor, is the measured concentration of the air quality impact factor at time step t, is the threshold concentration of air quality impact factors;

[0096] Preferably, the calculation formula for the equipment energy loss term is:

[0097] ;

[0098] Where, is the fan energy consumption coefficient, is the normalized value of the fan speed at the tth time step, Valve energy consumption coefficient, is the normalized value of the valve action at the tth time step;

[0099] Preferably, the calculation formula for the device physical constraint violation penalty is:

[0100] ;

[0101] Where, is the component of the action space parameter at the vth node in the tth time step, is the upper bound of the component physical constraint, is the lower bound of the component physical constraint;

[0102] Preferably, the calculation formula of the emergency passage carrying capacity penalty is:

[0103]

[0104] In the formula, is the total number of emergency passages associated with the vth node, c is the emergency passage index, Load c is the emergency passage load of the cth emergency passage, Cap c is the upper limit of the carrying capacity of the cth emergency passage;

[0105] Step S32: hierarchical strategy solving, specifically, hierarchical optimization solving is performed according to the node risk level label to obtain a hierarchical optimization scheduling action set;

[0106] The hierarchical optimization solving specifically refers to using a rolling prediction control method for normal nodes and using a distributed multi-agent reinforcement learning method for alert nodes;

[0107] Preferably, the rolling prediction control method specifically extracts the predicted traffic flow, vehicle speed and air quality index of the normal node according to the multi-dimensional tunnel state prediction tensor, combines with the device physical constraint violation penalty, takes the traffic fluency and energy consumption balance as the target, uses the rolling time domain optimization method to generate a candidate control action sequence, executes the control action at the first time, reacquires the prediction result at the next time and updates the solving to obtain the scheduling action of the normal node;

[0108] Preferably, the distributed multi-agent reinforcement learning method specifically uses the MADDPG algorithm realized under the PyTorch1.12 environment, models the section node, device node and emergency passage node as agents respectively;

[0109] Wherein, the state space is the node short-time prediction state matrix, the action space is the vehicle speed guidance, fan speed regulation and valve opening regulation, the reward function is the multi-objective optimization function, each agent makes independent decision under the local state and cooperatively trains through the central value network to realize the multi-objective balance of traffic fluency, air quality and device energy consumption, and obtains the scheduling action of the alert node;

[0110] Step S33: safety emergency adjustment, specifically, an emergency strategy library is constructed, and the emergency strategy library is called for adjustment when an emergency risk level is detected to obtain an emergency scheduling action set, the hierarchical optimization scheduling action set and the emergency scheduling action set are integrated, and in the integration process, the emergency scheduling action set preferentially covers the hierarchical optimization scheduling action set to obtain the best tunnel network scheduling strategy;

[0111] ​Preferably, the emergency scheduling action set, specifically for constructing an emergency strategy library based on an action priority sequence, the priority is from high to low in turn:

[0112] The first emergency strategy guarantees personnel safety, specifically including turning on emergency passage lighting, broadcasting evacuation instructions, and blocking the entrance of dangerous sections; the second emergency strategy controls fire and smoke diffusion, specifically including running the fan at full speed, opening the smoke exhaust valve, and closing unnecessary air inlet valves; the third emergency strategy maintains traffic order, specifically including traffic induction diversion and adjacent section speed limit operation; and the fourth emergency strategy reduces equipment risk, specifically including preventing equipment overload and dynamically switching standby fans or valves.

[0113] By performing the above operations, for the technical problems that traffic efficiency, air quality, energy consumption, and safety goals are difficult to balance in the dynamic collaborative scheduling process of the urban area tunnel network, device physical constraints and upper limits of emergency passage carrying capacity are difficult to guarantee, and scheduling strategies lack differentiation under different risk levels, the present scheme creatively adopts a multi-objective optimization tunnel network intelligent scheduling method for dynamic collaborative scheduling, realizes the comprehensive balance of traffic fluency, air quality, and energy consumption, and on the premise of meeting safety constraints, improves the executability of the scheduling strategy and the timeliness of the emergency response through hierarchical optimization and emergency priority strategies.

[0114] Embodiment five, refer to Figure 1 This embodiment is based on the above-mentioned embodiments, and in step S4, the strategy execution optimization is specifically for issuing actions to the traffic induction system, the ventilation control system, the valve control system, and the emergency passage management system according to the optimal tunnel network scheduling strategy, and collecting execution feedback to obtain scheduling action signals.

[0115] Embodiment six, refer to Figure 2 This embodiment is based on the above-mentioned embodiments, and the present application provides a kind of urban area tunnel network dynamic collaborative scheduling system, comprising: environmental perception modeling module, multi-dimensional prediction and evaluation module, dynamic collaborative scheduling module and strategy execution optimization module;

[0116] The environmental perception modeling module is used for environmental perception modeling, and through environmental perception modeling, a tunnel multi-relation space-time graph is obtained, and the tunnel multi-relation space-time graph is sent to the multi-dimensional prediction and evaluation module and the dynamic collaborative scheduling module;

[0117] The multi-dimensional prediction and evaluation module is used for multi-dimensional prediction and evaluation, and through multi-dimensional prediction and evaluation, a multi-dimensional tunnel state prediction tensor is obtained, and the multi-dimensional tunnel state prediction tensor is sent to the dynamic collaborative scheduling module;

[0118] The dynamic coordination scheduling module is configured to perform dynamic coordination scheduling, obtain an optimal tunnel network scheduling strategy through the dynamic coordination scheduling, and send the optimal tunnel network scheduling strategy to the policy execution optimization module.

[0119] The policy execution optimization module is configured to perform policy execution optimization and obtain a scheduling action signal.

[0120] It should be noted that the relational terms herein such as first and second and the like are used only to differentiate one entity or action from another, and do not necessarily require or imply any actual such relationship or order between or among the entities or actions. Also, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0121] Although the embodiments of the present application have been shown and described, it should be understood by those ordinary skilled in the art that various changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the present application.

[0122] The above describes the present application and its embodiments, which are not restrictive, and the embodiments shown in the drawings are only one of the embodiments of the present application, and the actual structure is not limited thereto. In general, if a person skilled in the art is inspired by it, without departing from the purpose of the present application, without creative design, similar structure and embodiments of the technical solution can be designed, which should belong to the protection scope of the present application.

Claims

1. A method for dynamic collaborative scheduling of urban area tunnel networks, characterized by: The method comprises the following steps: Step S1: environment perception modeling, for perception modeling of the urban area tunnel network, to obtain a tunnel multi-relation spatio-temporal graph; Step S2: multi-dimensional prediction and evaluation, for predicting node states and evaluating node risks, specifically, according to the tunnel multi-relation spatio-temporal graph, a multi-dimensional prediction and evaluation is performed by using a spatio-temporal graph multi-dimensional prediction method to obtain a multi-dimensional tunnel state prediction tensor, comprising the following steps: step S21: node feature coding; step S22: short-time state prediction; step S23: comprehensive risk evaluation; step S24: prediction tensor integration; Step S3: dynamic collaborative scheduling, specifically, according to the tunnel multi-relation spatio-temporal graph and the multi-dimensional tunnel state prediction tensor, a multi-objective optimization tunnel network intelligent scheduling method is used for dynamic collaborative scheduling to obtain an optimal tunnel network scheduling strategy, comprising the following steps: step S31: multi-objective optimization modeling; step S32: hierarchical strategy solving; step S33: safety emergency adjustment; Step S4: strategy execution optimization, to obtain a scheduling action signal.

2. The dynamic coordinated scheduling method of a city area tunnel network according to claim 1, characterized in that: In step S21, the node feature coding, specifically, first standardizing each node attribute to obtain a node feature vector, and then encoding the tunnel multi-relation spatio-temporal graph node and its neighbors by constructing a heterogeneous graph attention network to obtain a node high-dimensional feature vector; The construction of the heterogeneous graph attention network specifically receives the tunnel multi-relation spatio-temporal graph through an input layer, sets two layers of heterogeneous graph attention layers, the multi-head attention mechanism of each layer separately calculates attention coefficients for different types of neighbor nodes, and then aggregates each node through neighbor weighting to generate a node high-dimensional feature vector; In step S22, the short-time state prediction, specifically, according to the node high-dimensional feature vector and the tunnel multi-relation spatio-temporal graph, a standard spatio-temporal graph convolution network is used for spatial convolution to capture the spatial dependency between nodes to obtain a node convolution feature matrix; A node initial prediction state matrix is obtained by time series modeling of the node convolution feature matrix through a gated recurrent unit, and a node short-time prediction state matrix is generated by introducing a multi-modal coupling correction mechanism to weight and correct the node initial prediction state matrix; In step S23, the comprehensive risk evaluation, specifically, according to the node short-time prediction state matrix, a risk score is given to each node, then different node types are weighted and fused to generate a node comprehensive risk value, and the node comprehensive risk value is divided according to a risk threshold to obtain a node risk level label, the node risk level label includes normal, alert and emergency; In step S24, the prediction tensor integration, specifically, according to the node short-time prediction state matrix, the node comprehensive risk value and the node risk level label, a multi-dimensional tunnel state prediction vector is constructed for each node, and then all node multi-dimensional tunnel state prediction vectors are stacked to generate a multi-dimensional tunnel state prediction tensor.

3. The method of claim 2, wherein: In step S31, the multi-objective optimization modeling, specifically, according to the multi-dimensional tunnel state prediction tensor and the tunnel multi-relation spatio-temporal graph, a multi-objective optimization function is constructed by combining device physical constraints and emergency passage carrying capacity, which includes traffic fluency, air quality, device energy consumption and risk control; In step S32, the layered strategy is solved, specifically, a layered optimization solution is obtained by performing layered optimization solving according to the node risk level label; The layered optimization solving specifically refers to adopting a rolling prediction control method for normal nodes and a distributed multi-agent reinforcement learning method for alert nodes; The state space is the node short-time prediction state matrix, the action space is the vehicle speed guidance, fan speed regulation and valve opening regulation, the reward function is the multi-objective optimization function, each agent independently decides in the local state, and is collaboratively trained through a central value network to achieve multi-objective balance of traffic flow smoothness, air quality and equipment energy consumption, and the scheduling action of the alert node is obtained; In step S33, the safety emergency adjustment is performed, specifically, an emergency strategy library is constructed, and the emergency strategy library is called for adjustment when an emergency risk level is detected, to obtain an emergency scheduling action set, and the layered optimization scheduling action set and the emergency scheduling action set are integrated, in which the emergency scheduling action set has priority over the layered optimization scheduling action set, to obtain the best tunnel network scheduling strategy.

4. The dynamic coordinated scheduling method of a city area tunnel network according to claim 3, characterized in that: In step S4, the strategy execution optimization is performed, specifically, the action is sent to the traffic guidance system, the ventilation control system, the valve control system and the emergency passage management system according to the best tunnel network scheduling strategy, and the execution feedback is collected, to obtain the scheduling action signal.

5. The dynamic coordinated scheduling method of a city area tunnel network according to claim 4, characterized in that: In step S1, the environment perception modeling is performed, specifically, traffic data, environmental data, device state data and emergency passage data are collected from the urban area tunnel network, preprocessed, and then a tunnel multi-relation spatio-temporal graph is constructed.

6. The dynamic coordinated scheduling method of a city area tunnel network according to claim 5, characterized in that: In step S1, the tunnel multi-relation spatio-temporal graph is constructed, specifically, section nodes, device nodes and emergency passage nodes are set, traffic connection edges and device action edges are set, and the tunnel multi-relation spatio-temporal graph is established; The section node specifically refers to regarding each tunnel section as a section node, and setting the section node attributes according to the traffic data and environmental data; The device node specifically refers to regarding each key device as a device node, the key devices including fans, valves, lighting facilities and smoke exhaust equipment, and setting the device node attributes according to the device state data; The emergency passage node specifically refers to regarding the emergency passage as an emergency passage node, and setting the emergency passage node attributes according to the emergency passage data; The traffic connection edge specifically refers to establishing a traffic connection edge between the corresponding two section nodes if two tunnel sections are connected, and establishing a traffic connection edge between the corresponding section node and emergency passage node if the tunnel section and the emergency passage are adjacent; The device action edge specifically refers to establishing a device action edge between the corresponding section node and device node if the tunnel section is within the action range of the key device, and establishing a device action edge between the corresponding emergency passage node and device node if the emergency passage is within the action range of the key device.

7. A dynamic collaborative scheduling system for urban area tunnel network, which is used to implement the dynamic collaborative scheduling method for urban area tunnel network according to any one of claims 1-6. The system comprises an environment perception modeling module, a multi-dimensional prediction evaluation module, a dynamic collaborative scheduling module and a strategy execution optimization module.

8. The dynamic coordinated scheduling system of urban area tunnel network according to claim 7, characterized in that: The environment perception modeling module is configured to perform environment perception modeling, obtain a tunnel multi-relation space-time graph through the environment perception modeling, and send the tunnel multi-relation space-time graph to the multi-dimensional prediction evaluation module and the dynamic cooperative scheduling module. The multi-dimensional prediction evaluation module is configured to perform multi-dimensional prediction evaluation, obtain a multi-dimensional tunnel state prediction tensor through the multi-dimensional prediction evaluation, and send the multi-dimensional tunnel state prediction tensor to the dynamic cooperative scheduling module. The dynamic cooperative scheduling module is configured to perform dynamic cooperative scheduling, obtain an optimal tunnel network scheduling strategy through the dynamic cooperative scheduling, and send the optimal tunnel network scheduling strategy to the policy execution optimization module. The policy execution optimization module is configured to perform policy execution optimization, and obtain a scheduling action signal.

Images