CN106672028A - Robust strategy based bi-level subway traffic flow optimal control method - Google Patents

Abstract

The invention relates to a robust strategy based bi-level subway traffic flow optimal control method. The method includes steps: generating a topological structure diagram of a rail transit network according to schedule running parameters of each train; analyzing controllability and sensitivity of a train flow on the basis of the topological structure diagram; generating a multi-train conflict-free running track according to the schedule running parameters of each train; at each sampling moment, predicting a moving position of each train at a certain future moment on the basis of a current running state and a historical position observation sequence of each strain, and establishing an observer from continuous dynamics of each train to discrete conflict logics to map the continuous dynamics to conflict states expressed by discrete observation values; when a system violates traffic control rules possibly, monitoring hybrid dynamics of a subway traffic hybrid system to provide warning information for a control center; finally, when the warning information appears, adopting a self-adaption control theoretical method for robust bi-level programming of train running tracks, and transmitting programming results to each train.

Description

The flow-optimized control method of double-deck subway transportation based on Robust Strategies

The application is Application No.：201510150696.0, invention and created name is《A kind of flow-optimized control of subway transportation Method》, the applying date is：The divisional application of the application for a patent for invention on March 31st, 2015.

Technical field

The present invention relates to a kind of flow-optimized control method of subway transportation, more particularly to a kind of double-deck ground based on Robust Strategies Iron traffic optimization control method.

Background technology

With the expanding day of China's big and medium-sized cities scale, Traffic Systems are faced with the increasing pressure, energetically Feasibility of developing track transportation system becomes the important means for solving urban traffic congestion.National Eleventh Five-Year Plan guiding principle is it is to be noted, that there is bar The big city and group of cities area of part is using track traffic as Priority setting.China is just experiencing a unprecedented rail Road transport development peak period, some cities have been turned to the construction of net by the construction of line, urban mass transit network progressively shape Into.In the complex region that Rail traffic network and train flow are intensive, still combined using train operation plan and be based on subjective experience Train interval dispensing mode gradually show its lag, be in particular in：(1) formulation of train operation plan timetable is simultaneously Not in view of the impact of various enchancement factors, easily cause traffic flow tactics and manage crowded, reduce the safety of traffic system operation Property；(2) train scheduling work lays particular emphasis on the personal distance kept between single train, and not yet rise to carries out strategic pipe to train flow The macroscopic aspect of reason；(3) subjective experience of a line dispatcher is depended on train allocation process, the selection for allocating opportunity is random more Property it is larger, lack scientific theory support；(4) the less shadow in view of external interference factor of allotment means that dispatcher is used Ring, the robustness and availability of train programs is poor.

It is directed to long-distance railway transportation more the discussion object of existing documents and materials, and is directed to big flow, high density and closely-spaced The Scientific Regulation scheme of the city underground traffic system under service condition still lacks system design.Under complicated road network service condition Train Coordinated Control Scheme need to carry out the running status of single vehicles in transportation network in region on strategic level to calculate and Optimization, and collaborative planning is implemented in the traffic flow to being made up of multiple trains；Pass through effective monitoring mechanism on pre- tactical level Adjust the subregional critical operational parameters in transportation network top to solve congestion problems, and ensure the fortune of all trains in the region Line efficiency；Then according to critical operational parameters adjusting the running status of related train on tactical level, single-row wheel paths are obtained Prioritization scheme, consideration train performance, scheduling rule and extraneous ring are changed into by the headway management of train from fixed manual type The factors such as border are in interior variable " microcosmic-macroscopic view-middle sight-microcosmic " Separation control mode.

The content of the invention

The technical problem to be solved in the present invention is to provide a kind of robustness and availability preferably based on the double of Robust Strategies The flow-optimized control method of layer subway transportation, the method can strengthen the subject of programs formulation and can effectively prevent subway train Operation conflict.

The technical scheme for realizing the object of the invention is to provide a kind of flow-optimized control of double-deck subway transportation based on Robust Strategies Method processed, comprises the steps：

Step A, according to the plan operational factor of each train, generate the topology diagram of Rail traffic network；

Step B, the topology diagram based on the Rail traffic network constructed by step A, analyze train flow controllability and The class feature of sensitiveness two；

Step C, according to the plan operational factor of each train, on the basis of Modeling Method for Train Dynamics is built, according to row Car operation conflict Coupling point sets up train running conflict and allocates model in advance, generates many train Lothrus apterus running orbits；

Step D, in each sampling instant t, based on the current running status of train and historical position observation sequence, to train The advanced positions at following certain moment are predicted；Its detailed process is as follows：

Step D1, train track data pretreatment, with train initiating station stop position as the origin of coordinates, adopt each The sample moment, according to the acquired original discrete two-dimensional position sequence x=[x of train₁,x₂,...,x_n] and y=[y₁,y₂,..., y_n], it is carried out using first-order difference method process new train discrete location sequence △ x=[the △ x of acquisition₁,△x₂,...,△ x_n-1] and △ y=[△ y₁,△y₂,...,△y_n-1], wherein △ x_i=x_i+1-x_i,△y_i=y_i+1-y_i(i=1,2 ..., n-1)；

Step D2, to train track data cluster, to train discrete two-dimensional position sequence △ x and △ y new after process, lead to Setting cluster number M' is crossed, using K-means clustering algorithms it is clustered respectively；

Step D3, parameter training is carried out using HMM to the train track data after cluster, by will place Train operation track data △ x and △ y after reason are considered as the aobvious observation of hidden Markov models, by setting hidden state number N' and parameter update period τ ', according to nearest T' position detection value and using the newest hidden Ma Erke of B-W algorithms rolling acquisition Husband's model parameter λ '；Specifically：Because the train track sets data length for being obtained is dynamic change, in order in real time with The state change of track train track, it is necessary to initial track HMM parameter lambda '=(π, A, B) on the basis of it is right It is readjusted, more accurately to speculate train in the position at following certain moment；Every period τ ', according to the T' of newest acquisition Individual observation (o₁,o₂,...,o_T') to track HMM parameter lambda '=(π, A, B) reevaluated；

Step D4, foundation HMM parameter, are obtained corresponding to current time observation using Viterbi algorithm Hidden state q；

Step D5, every the periodAccording to the HMM parameter lambda of newest acquisition '=(π, A, B) and nearest H Individual history observation (o₁,o₂,...,o_H), based on hidden state q at train current time, in moment t, by setting prediction time domain H', obtains position prediction value O of future time period train；

Step E, set up from train it is continuous dynamic to discrete conflict logic observer, by the continuous of subway transportation system Dynamic mapping is the conflict situation of discrete observation value expression；When system is possible to violate traffic control rule, to subway transportation The Hybrid dynamics behavior implementing monitoring of hybrid system, for control centre timely warning information is provided；

Step F, when warning information occurs, meeting train physical property, region hold stream constraint and track traffic scheduling On the premise of rule, by setting optimizing index function, Shandong is carried out to train operation track using Adaptive Control Theory method Rod dual layer resist, and program results is transferred to into each train, each train is received and performs train collision avoidance instruction until each train is equal Reach it and free terminal.

Further, the detailed process of step A is as follows：

Step A1, the database from subway transportation control centre extract the website letter stopped in each train travelling process Breath；

Step A2, the site information that each train is stopped is classified according to positive and negative two traffic directions, and will be same Same site on one traffic direction is merged；

Step A3, according to website amalgamation result, according to space layout form multiple websites before and after straight line connection of website.

Further, the detailed process of step B is as follows：

Step Bl, the Traffic flux detection model built in single subsegment；Its detailed process is as follows：

Step Bl.1, introducing state variable Ψ, input variable u and output variable Ω, wherein Ψ represents phase link between website The train quantity that certain moment is present in section, it includes single channel section and Multiple Sections two types, and u represents that track traffic dispatcher is directed to The Operation Measures that certain section is implemented, such as adjust train speed or change train in the station time, Ω represents certain section period On the train quantity left；

Step B1.2, by by time discretization, setting up shape such as Ψ (t+ △ t)=A₁Ψ(t)+B₁U (t) and Ω (t)=C₁ Ψ(t)+D₁Discrete time Traffic flux detection model in the single subsegment of u (t), wherein △ t represent sampling interval, Ψ (t) tables Show the state vector of t, A₁、B₁、C₁And D₁State-transition matrix, input matrix, the output measurement square of t are represented respectively Battle array and direct transmission matrix；

Step B2, the Traffic flux detection model built in many subsegments；Its detailed process is as follows：

Step B2.1, according to circuit space layout form and train flow historical statistical data, obtain each son of cross link Flow proportional parameter beta in section；

Step B2.2, according to the discrete time Traffic flux detection model in flow proportional parameter and single subsegment, build shape Such as Ψ (t+ △ t)=A₁Ψ(t)+B₁U (t) and Ω (t)=C₁Ψ(t)+D₁Discrete time traffic flow control in many subsegments of u (t) Simulation；

Step B3, according to the controllable factor matrix [B of Controlling model₁,A₁B₁,...,A₁ ^n-1B₁] order and numerical value n relation, Qualitative analysis its controllability, according to the sensitivity coefficient matrix [C of Controlling model₁(zI-A₁)^-1B₁+D₁], its input of quantitative analysis is defeated Go out sensitiveness, wherein n represents the dimension of state vector, and I represents unit matrix, and z is represented to original discrete time Traffic flux detection The element factor that model is changed.

Further, the detailed process of step C is as follows：

Step C1, train status transfer modeling, train is shown as between website along the process that track traffic road network runs Switching at runtime process, the website in train operation plan is arranged, and sets up single train switched and transferred between different websites Petri net model：(g, G, Pre, Post, are m) train section metastasis model to E=, and wherein g represents each sub- section, G tables between website Show the transfer point of train running speed state parameter, Pre and Post represent respectively between each sub- section and website before and after to connection Relation,The operation section residing for train is represented, wherein m represents model identification, Z⁺Represent Positive Integer Set；

The full operation profile hybrid system modeling of step C2, train, the operation by train between website is considered as continuous process, from The stress situation of train is set out, and according to energy model kinetics equation of the train in the different operation phase is derived, with reference to extraneous dry Factor is disturbed, is set up with regard to train in a certain operation phase speed v_GMapping function v_G=λ (T₁,T₂, H, R, α), wherein T₁、T₂、 H, R and α represent respectively tractive force of train, braking force of train, train resistance, train gravity and train status random fluctuation parameter；

Step C3, speculate by the way of emulation solution train track using mixing, by the way that by time subdivision, utilization state is continuous Distance of the characteristic Recursive Solution any time train of change in a certain operation phase away from initial rest position point,Wherein J₀For voyage of the initial time train away from initial rest position point, △ τ are the number of time window Value, J (τ) is distance of the τ moment train away from initial rest position point, thereby it is assumed that and obtains single-row wheel paths；

Step C4, train are modeled in station time probability distribution function, for specific run circuit, by transferring train each The dwell time data at station, obtain the dwell time probability distribution of different circuit difference website condition Trains；

Step C5, the Lothrus apterus robust track allotment of many trains coupling, according to each train in advance up to the time of conflict point, pass through Time segments division, in each sampling instant t, on the premise of random factor is incorporated, according to scheduling rule to being discontented near conflict point Implement robust secondary planning in the train track that sufficient personal distance is required.

Further, in step D, the value for clustering number M' is 4, and the value of hidden state number N' is 3, and parameter updates period τ ' For 30 seconds, T' was 10,For 30 seconds, H was 10, and prediction time domain h' is 300 seconds.

Further, the specific implementation process of step E is as follows：

The conflict hypersurface collection of functions of step E1, construction based on regulation rule：Set up hypersurface collection of functions is to reflect The contention situation of system, wherein, the continuous function h related to single train in the hypersurface that conflicts_IFor I type hypersurfaces, arrange with two The related continuous function h of car_IIFor Type-II hypersurface；

Step E2, set up by train continuous state to discrete conflict situation observer, build train in traffic network The safety regulation collection d that need to be met during operation_ij(t)≥d_min, wherein d_ijT () represents train i and train j between the reality of t Every d_minRepresent the minimum safety interval between train；

Step E3, based on person machine system is theoretical and complication system hierarchical control principle, according to train operation pattern, build Train monitor in real time mechanism of the people in loop, it is ensured that the operation of system designs the solution from conflicting to conflicting in safe reachable set The discrete watch-dog of section of slipping out of the hand, when the discrete observation vector of observer shows that safety regulation rally is breached, sends at once phase The warning information answered.

Further, the detailed process of step F is as follows：

Step F1, the analysis result based on step B and step E, it is determined that the traffic flow regulation measure specifically taken, including The speed of service and/or adjustment train of train are adjusted in the station class measure of time two, and using above regulation measure specifically Point and opportunity；

Step F2, termination reference point locations P, collision avoidance policy control time domain Θ, the trajectory predictions of the collision avoidance planning of setting train Time domain

Step F3, operation conflict Resolution process model building, the operation conflict Resolution in the above-listed workshop of Rail traffic network is considered as Based on the inside and outside dual planning problem of both macro and micro aspect, whereinRepresent outer layer plan model, i.e. rail Train flow flow-Density and distribution problem on road traffic network,Represent internal layer plan model, i.e. track traffic The state adjustment problem of single vehicles on section；F、x₁And u₁It is respectively object function, state vector and the decision-making of outer layer planning problem Vector, G (x₁,u₁)≤0 be outer layer planning constraints, f, x₂And u₂It is respectively object function, the state of internal layer planning problem Vector sum decision vector, g (x₂,u₂)≤0 is the constraints of internal layer planning, using the outer layer program results of macroscopic aspect as micro- The reference input of sight aspect internal layer planning；

Step F4, the variable bound modeling of operation conflict Resolution, build and include adjustable train quantity a, train speed ω and row Car is in variables such as station time γ in interior both macro and micro constraints：Wherein t need to implement the change of the section k of conflict Resolution Amount constraint can be described as：a_k(t)≤a_M、ω_k(t)≤ω_M、γ_k(t)≤γ_M, a_M、ω_M、γ_MRespectively maximum adjustable train number , in the station time, such variable of freeing can be subject to traffic flow distribution, train thing for amount, maximum train running speed and most long train The constraint of the aspect such as rationality energy and personal distance；

Step F5, Multi-objective Robust optimum road network flow allocation plan are solved：Based on cooperative collision avoidance trajectory planning thought, For different performance indications, by selecting different conflict Resolution object functions, in traffic flow operation macroscopic aspect base is solved In Euler's network model multiple target traffic flow optimum flow allocation plan and each control section is only real in Rolling Planning interval Apply its first Optimal Control Strategy；

Step F6, the train operation state adjustment of Multi-objective Robust optimum section：According to each section or zone flow configuration knot Really, evolutionary model is mixed based on train operation and Lagrangian plan model obtains optimum single vehicles controlled quentity controlled variable, generated optimum Single vehicles running orbit and each regulation and control train only implements its first Optimal Control Strategy in Rolling Planning interval；

Step F7, each train receive and perform train collision avoidance instruction；

Step F8, in next sampling instant, repeat step F5 to F7 is until each train reaches it and frees terminal.

Further, in step F2, the next one for terminating reference point locations P for train stops website, and the value of parameter Θ is 300 seconds,Value be 300 seconds.

Further, the detailed process of step F5 is as follows：Order

WhereinRepresent distance between t train i present positions and next website square, P_i(t)=(x_it, y_it) represent t train i two-dimensional coordinate value,The next two-dimensional coordinate values for stopping website of train i are represented, The priority index of so t train i may be set to：

Wherein n_tRepresent there is the train number of conflict on t section, from the implication of priority index, train away from From next website more close to, its priority is higher；

Setting optimizing index

Wherein i ∈ I (t) represent train code and I (t)=1,2 ..., n_t, P_i(t+s △ t) represents train in moment (t + s △ t) position vector, Π represents control time, i.e., the time span of Future Trajectory planning, u from current time_iExpression is treated The optimal control sequence of the train i of optimization, Q_itFor positive definite diagonal matrix, its diagonal element refers to for train i in the priority of t Number λ_it, and

The present invention has positive effect：(1) the flow-optimized control of double-deck subway transportation based on Robust Strategies of the invention On the premise of track traffic control personal distance is met, based on the real-time position information of train, maintenance data digs method Pick means dynamic speculates train track；According to track traffic regulation rule, alarm is implemented in the conflict to being likely to occur, according to train Performance data and relevant constraint give each train planning conflict Resolution track；Train schedule is being configured When, it is contemplated that the probability distribution and the robustness of train schedule of all kinds of random factors of train are affected, strengthens configuration knot The availability of fruit.

(2) controllability and sensitivity analysis result of the present invention based on Rail traffic network topological structure, can hand over for subway Through-flow allotment time, the selection in allotment place and allotment means provide scientific basis, it is to avoid the randomness that regulation and control scheme is chosen.

(3) scene monitoring mechanism of the present invention based on constructed " people is in loop ", can be to train inside continuous variable Effecting reaction is made in time with the frequent interaction of external discrete event, overcomes the shortcoming of conventional open loop monitored off-line scheme.

(4) the dual layer resist scheme of train flow of the invention can not only reduce the solution dimension of Optimal Control Problem, also The practicality of regulation and control scheme can be strengthened, overcome model and algorithm in existing document only focus on train AT STATION to send out when Between, and lack the defect of control when running on concrete railroad section to train and prediction.

(5) present invention is based on constructed train operation track rolling forecast scheme, can in time incorporate train and transport in real time All kinds of disturbing factors in row, improve the accuracy of train trajectory predictions, overcome Conventional Off-line prediction scheme accuracy not high Shortcoming.

Description of the drawings

Fig. 1 is train flow analysis on Operating figure；

Fig. 2 is that Lothrus apterus 3D robusts track speculates figure；

Fig. 3 mixes monitoring figure for train operation state；

Fig. 4 frees figure for train running conflict optimum；

Fig. 5 is the schematic diagram of traffic flow bilayer allocation plan.

Specific embodiment

(embodiment 1)

A kind of flow-optimized control system of subway transportation, including it is wire topologies generation module, data transmission module, vehicle-mounted Terminal module, control terminal module and track monitoring module, track monitoring module is collected the status information of train and is supplied to Control terminal module.

The control terminal module includes following submodule：

Lothrus apterus Track Pick-up module before train operation：According to Train operation plan time of running table, train dynamicses are initially set up Model is learned, then train running conflict is set up according to train running conflict Coupling point and is allocated model in advance, ultimately produce Lothrus apterus row Car running orbit.

Train operation middle or short term Track Pick-up module：According to the train real time status information that track monitoring module is provided, profit With data mining model, thus it is speculated that the running orbit of train in future time period.

Train operation situation monitoring module：In each sampling instant t, based on the track estimation result of train, when between train When being possible to occur violating the situation of safety regulation, to its dynamic behaviour implementing monitoring and for control terminal warning information is provided.

Train collision avoidance track optimizing module：When train operation situation monitoring module sends warning information, train is being met On the premise of physical property, region hold stream constraint and track traffic scheduling rule, by setting optimizing index function, using adaptive Control theory method is answered to carry out robust dual layer resist to train operation track by control terminal module, and by data transmission module Program results is transferred to into car-mounted terminal module to perform.Train collision avoidance track optimizing module includes internal layer planning and outer layer planning two Class planning process.

Using the flow-optimized control of double-deck subway transportation based on Robust Strategies of the flow-optimized control system of above-mentioned subway transportation Method, comprises the following steps：

Step A, according to the plan operational factor of each train, generate the topology diagram of Rail traffic network；Its is concrete Process is as follows：

Step A1, the database from subway transportation control centre extract the website letter stopped in each train travelling process Breath；

Step A2, the site information that each train is stopped is classified according to positive and negative two traffic directions, and will be same Same site on one traffic direction is merged；

Step A3, according to website amalgamation result, according to space layout form multiple websites before and after straight line connection of website.

Step B, the topology diagram based on the Rail traffic network constructed by step A, analyze train flow controllability and The class feature of sensitiveness two；Its detailed process is as follows：

Step Bl, see Fig. 1, build the Traffic flux detection model in single subsegment；Its detailed process is as follows：

Step Bl.1, introducing state variable Ψ, input variable u and output variable Ω, wherein Ψ represents phase link between website The train quantity that certain moment is present in section, it includes single channel section and Multiple Sections two types, and u represents that track traffic dispatcher is directed to The Operation Measures that certain section is implemented, such as adjust train speed or change train in the station time, Ω represents certain section period On the train quantity left；

Step B1.2, by by time discretization, setting up shape such as Ψ (t+ △ t)=A₁Ψ(t)+B₁U (t) and Ω (t)=C₁ Ψ(t)+D₁Discrete time Traffic flux detection model in the single subsegment of u (t), wherein △ t represent sampling interval, Ψ (t) tables Show the state vector of t, A₁、B₁、C₁And D₁State-transition matrix, input matrix, the output measurement square of t are represented respectively Battle array and direct transmission matrix；

Step B2, the Traffic flux detection model built in many subsegments；Its detailed process is as follows：

Step B2.1, according to circuit space layout form and train flow historical statistical data, obtain each son of cross link Flow proportional parameter beta in section；

Step B2.2, according to the discrete time Traffic flux detection model in flow proportional parameter and single subsegment, build shape Such as Ψ (t+ △ t)=A₁Ψ(t)+B₁U (t) and Ω (t)=C₁Ψ(t)+D₁Discrete time traffic flow control in many subsegments of u (t) Simulation；

Step B3, according to the controllable factor matrix [B of Controlling model₁,A₁B₁,...,A₁ ^n-1B₁] order and numerical value n relation, Qualitative analysis its controllability, according to the sensitivity coefficient matrix [C of Controlling model₁(zI-A₁)^-1B₁+D₁], its input of quantitative analysis is defeated Go out sensitiveness, wherein n represents the dimension of state vector, and I represents unit matrix, and z is represented to original discrete time Traffic flux detection The element factor that model is changed；

Step C, see Fig. 2, according to the plan operational factor of each train, on the basis of Modeling Method for Train Dynamics is built, Train running conflict is set up according to train running conflict Coupling point and allocate model in advance, generate many train Lothrus apterus running orbits；Its Detailed process is as follows：

Step C1, train status transfer modeling, train is shown as between website along the process that track traffic road network runs Switching at runtime process, the website in train operation plan is arranged, and sets up single train switched and transferred between different websites Petri net model：(g, G, Pre, Post, are m) train section metastasis model to E=, and wherein g represents each sub- section, G tables between website Show the transfer point of train running speed state parameter, Pre and Post represent respectively between each sub- section and website before and after to connection Relation,The operation section residing for train is represented, wherein m represents model identification, Z⁺Represent Positive Integer Set；

The full operation profile hybrid system modeling of step C2, train, the operation by train between website is considered as continuous process, from The stress situation of train is set out, and according to energy model kinetics equation of the train in the different operation phase is derived, with reference to extraneous dry Factor is disturbed, is set up with regard to train in a certain operation phase speed v_GMapping function v_G=λ (T₁,T₂, H, R, α), wherein T₁、T₂、 H, R and α represent respectively tractive force of train, braking force of train, train resistance, train gravity and train status random fluctuation parameter；

Step C3, speculate by the way of emulation solution train track using mixing, by the way that by time subdivision, utilization state is continuous Distance of the characteristic Recursive Solution any time train of change in a certain operation phase away from initial rest position point,Wherein J₀For voyage of the initial time train away from initial rest position point, △ τ are the number of time window Value, J (τ) is distance of the τ moment train away from initial rest position point, thereby it is assumed that and obtains single-row wheel paths；

Step C4, train are modeled in station time probability distribution function, for specific run circuit, by transferring train each The dwell time data at station, obtain the dwell time probability distribution of different circuit difference website condition Trains；

Step C5, the Lothrus apterus robust track allotment of many trains coupling, according to each train in advance up to the time of conflict point, pass through Time segments division, in each sampling instant t, on the premise of random factor is incorporated, according to scheduling rule to being discontented near conflict point Implement robust secondary planning in the train track that sufficient personal distance is required.

Step D, in each sampling instant t, based on the current running status of train and historical position observation sequence, to train The advanced positions at following certain moment are predicted；Its detailed process is as follows：

Step D1, train track data pretreatment, with train initiating station stop position as the origin of coordinates, adopt each The sample moment, according to the acquired original discrete two-dimensional position sequence x=[x of train₁,x₂,...,x_n] and y=[y₁,y₂,..., y_n], it is carried out using first-order difference method process new train discrete location sequence △ x=[the △ x of acquisition₁,△x₂,...,△ x_n-1] and △ y=[△ y₁,△y₂,...,△y_n-1], wherein △ x_i=x_i+1-x_i,△y_i=y_i+1-y_i(i=1,2 ..., n-1)；

Step D2, to train track data cluster, to train discrete two-dimensional position sequence △ x and △ y new after process, lead to Setting cluster number M' is crossed, using K-means clustering algorithms it is clustered respectively；

Step D3, parameter training is carried out using HMM to the train track data after cluster, by will place Train operation track data △ x and △ y after reason are considered as the aobvious observation of hidden Markov models, by setting hidden state number N' and parameter update period τ ', according to nearest T' position detection value and using the newest hidden Ma Erke of B-W algorithms rolling acquisition Husband's model parameter λ '；Specifically：Because the train track sets data length for being obtained is dynamic change, in order in real time with The state change of track train track, it is necessary to initial track HMM parameter lambda '=(π, A, B) on the basis of it is right It is readjusted, more accurately to speculate train in the position at following certain moment；Every period τ ', according to the T' of newest acquisition Individual observation (o₁,o₂,...,o_T') to track HMM parameter lambda '=(π, A, B) is reevaluated；

Step D4, foundation HMM parameter, are obtained corresponding to current time observation using Viterbi algorithm Hidden state q；

Step D5, every the periodAccording to the HMM parameter lambda of newest acquisition '=(π, A, B) and nearest H History observation (o₁,o₂,...,o_H), based on hidden state q at train current time, in moment t, by setting prediction time domain h', Obtain position prediction value O of future time period train；

The value of above-mentioned cluster number M' is 4, and the value of hidden state number N' is 3, and parameter renewal period τ ' is 30 seconds, and T' is 10,For 30 seconds, H was 10, and prediction time domain h' is 300 seconds.

Step E, see Fig. 3, set up from the continuous dynamic of train to the observer of discrete conflict logic, by subway transportation system Continuous dynamic mapping be discrete observation value expression conflict situation；When system is possible to violate traffic control rule, over the ground The Hybrid dynamics behavior implementing monitoring of iron traffic hybrid system, for control centre timely warning information is provided；

The specific implementation process of step E is as follows：

The conflict hypersurface collection of functions of step E1, construction based on regulation rule：Set up hypersurface collection of functions is to reflect The contention situation of system, wherein, the continuous function h related to single train in the hypersurface that conflicts_IFor I type hypersurfaces, arrange with two The related continuous function h of car_IIFor Type-II hypersurface；

Step E2, set up by train continuous state to discrete conflict situation observer, build train in traffic network The safety regulation collection d that need to be met during operation_ij(t)≥d_min, wherein d_ijT () represents train i and train j between the reality of t Every d_minRepresent the minimum safety interval between train；

Step E3, based on person machine system is theoretical and complication system hierarchical control principle, according to train operation pattern, build Train monitor in real time mechanism of the people in loop, it is ensured that the operation of system designs the solution from conflicting to conflicting in safe reachable set The discrete watch-dog of section of slipping out of the hand, when the discrete observation vector of observer shows that safety regulation rally is breached, sends at once phase The warning information answered.

Step F, see Fig. 4, when warning information occurs, train physical property, region hold stream constraint and track is handed over meeting On the premise of logical scheduling rule, by setting optimizing index function, using Adaptive Control Theory method to train operation track Robust dual layer resist is carried out, and program results is transferred to into each train, each train is received and performs train collision avoidance instruction until each Train reaches it and frees terminal；Its detailed process is as follows：

Step F1, the analysis result based on step B3 and step E3, it is determined that the traffic flow regulation measure specifically taken, bag The speed of service and/or adjustment train of adjustment train are included in the class measure of station time two, and using the concrete of above regulation measure Place and opportunity；

Step F2, termination reference point locations P, collision avoidance policy control time domain Θ, the trajectory predictions of the collision avoidance planning of setting train Time domain

The next one for terminating reference point locations P for train stops website, and the value of parameter Θ is 300 seconds,Value be 300 Second；

Step F3, operation conflict Resolution process model building, the operation conflict Resolution in the above-listed workshop of Rail traffic network is considered as Based on the inside and outside dual planning problem of both macro and micro aspect, Fig. 5 is seen, whereinRepresent outer layer planning mould Train flow flow-Density and distribution problem in type, i.e. track traffic road network,Internal layer plan model is represented, i.e., The state adjustment problem of single vehicles on track traffic section；F、x₁And u₁Be respectively the object function of outer layer planning problem, state to Amount and decision vector, G (x₁,u₁)≤0 be outer layer planning constraints, f, x₂And u₂It is respectively the target of internal layer planning problem Function, state vector and decision vector, g (x₂,u₂)≤0 is the constraints of internal layer planning, by the outer layer planning knot of macroscopic aspect Reference input of the fruit as the planning of microcosmic point internal layer；

Step F4, the variable bound modeling of operation conflict Resolution, build and include adjustable train quantity a, train speed ω and row Car is in variables such as station time γ in interior both macro and micro constraints：Wherein t need to implement the change of the section k of conflict Resolution Amount constraint can be described as：a_k(t)≤a_M、ω_k(t)≤ω_M、γ_k(t)≤γ_M, a_M、ω_M、γ_MRespectively maximum adjustable train number , in the station time, such variable of freeing can be subject to traffic flow distribution, train thing for amount, maximum train running speed and most long train The constraint of the aspect such as rationality energy and personal distance；

Step F5, Multi-objective Robust optimum road network flow allocation plan are solved：Based on cooperative collision avoidance trajectory planning thought, For different performance indications, by selecting different conflict Resolution object functions, in traffic flow operation macroscopic aspect base is solved In Euler's network model multiple target traffic flow optimum flow allocation plan and each control section is only real in Rolling Planning interval Apply its first Optimal Control Strategy；Its detailed process is as follows：Order

WhereinRepresent distance between t train i present positions and next website square, P_i(t)=(x_it, y_it) represent t train i two-dimensional coordinate value,The next two-dimensional coordinate values for stopping website of train i are represented, The priority index of that so t train i may be set to：

Wherein n_tRepresent there is the train number of conflict on t section, from the implication of priority index, train away from From next website more close to, its priority is higher；

Setting optimizing index

Wherein i ∈ I (t) represent train code and I (t)=1,2 ..., n_t, P_i(t+s △ t) represents train in moment (t + s △ t) position vector, Π represents control time, i.e., the time span of Future Trajectory planning, u from current time_iExpression is treated The optimal control sequence of the train i of optimization, Q_itFor positive definite diagonal matrix, its diagonal element refers to for train i in the priority of t Number λ_it, and

Step F6, the train operation state adjustment of Multi-objective Robust optimum section：According to each section or zone flow configuration knot Really, evolutionary model is mixed based on train operation and Lagrangian plan model obtains optimum single vehicles controlled quentity controlled variable, generated optimum Single vehicles running orbit and each regulation and control train only implements its first Optimal Control Strategy in Rolling Planning interval；

Step F7, each train receive and perform train collision avoidance instruction；

Step F8, in next sampling instant, repeat step F5 to F7 is until each train reaches it and frees terminal.

Obviously, above-described embodiment is only intended to clearly illustrate example of the present invention, and is not to the present invention The restriction of embodiment.For those of ordinary skill in the field, it can also be made on the basis of the above description The change or variation of its multi-form.There is no need to be exhaustive to all of embodiment.And these belong to this Obvious change that bright spirit is extended out or among changing still in protection scope of the present invention.

Claims (1)

1. a kind of flow-optimized control method of double-deck subway transportation based on Robust Strategies, it is characterised in that comprise the steps：

Step A, according to the plan operational factor of each train, generate the topology diagram of Rail traffic network；

Step B, the topology diagram based on the Rail traffic network constructed by step A, analyze the controllability and sensitivity of train flow Two class features of property；

Step C, according to the plan operational factor of each train, on the basis of Modeling Method for Train Dynamics is built, according to train fortune Row conflict Coupling point sets up train running conflict and allocates model in advance, generates many train Lothrus apterus running orbits；

Step D, in each sampling instant t, based on the current running status of train and historical position observation sequence, to train future The advanced positions at certain moment are predicted；Its detailed process is as follows：

Step D1, train track data pretreatment, with train initiating station stop position as the origin of coordinates, in each sampling Carve, according to the acquired original discrete two-dimensional position sequence x=[x of train₁,x₂,...,x_n] and y=[y₁,y₂,...,y_n], adopt It is carried out with first-order difference method processing and obtain new train discrete location sequence △ x=[△ x₁,△x₂,...,△x_n-1] and △ y=[△ y₁,△y₂,...,△y_n-1], wherein △ x_i=x_i+1-x_i,△y_i=y_i+1-y_i(i=1,2 ..., n-1)；

Step D2, to train track data cluster, to train discrete two-dimensional position sequence △ x and △ y new after process, by setting Surely number M' is clustered, using K-means clustering algorithms it is clustered respectively；

Step D3, parameter training is carried out using HMM to the train track data after cluster, after it will process Train operation track data △ x and △ y be considered as the aobvious observation of hidden Markov models, by set hidden state number N' and Parameter updates period τ ', according to nearest T' position detection value and using the newest Hidden Markov mould of B-W algorithms rolling acquisition Shape parameter λ '；Specifically：Because the train track sets data length for being obtained is dynamic change, for real-time tracking row The state change of wheel paths, it is necessary to initial track HMM parameter lambda '=(π, A, B) on the basis of it is heavy to its New adjustment, more accurately to speculate train in the position at following certain moment；Every period τ ', see according to T' of newest acquisition Measured value (o₁,o₂,...,o_T') to track HMM parameter lambda '=(π, A, B) is reevaluated；

Step D4, foundation HMM parameter, obtain hidden corresponding to current time observation using Viterbi algorithm State q；

Step D5, every the periodAccording to the HMM parameter lambda of newest acquisition '=(π, A, B) and nearest H history Observation (o₁,o₂,...,o_H), based on hidden state q at train current time, in moment t, by setting prediction time domain h', obtain Position prediction value O of future time period train；

Step E, set up from train it is continuous dynamic to discrete conflict logic observer, by the continuous dynamic of subway transportation system It is mapped as the conflict situation of discrete observation value expression；When system is possible to violate traffic control rule, subway transportation is mixed The Hybrid dynamics behavior implementing monitoring of system, for control centre timely warning information is provided；

Step F, when warning information occurs, meeting train physical property, region hold stream constraint and track traffic scheduling rule On the premise of, by setting optimizing index function, robust is carried out to train operation track using Adaptive Control Theory method double Layer planning, and program results is transferred to into each train, each train is received and performs train collision avoidance instruction until each train is reached It frees terminal；

The detailed process of step F is as follows：

Step F1, the analysis result based on step B and step E, it is determined that the traffic flow regulation measure specifically taken, including adjustment The speed of service of train and/or adjustment train in the station class measure of time two, and the specified place using above regulation measure and Opportunity；

Step F2, termination reference point locations P, collision avoidance policy control time domain Θ, the trajectory predictions time domain of the collision avoidance planning of setting train Υ；The next one for terminating reference point locations P for train stops website, and the value of parameter Θ is 300 seconds, and the value of γ is 300 seconds；

Step F3, operation conflict Resolution process model building, the operation conflict Resolution in the above-listed workshop of Rail traffic network are considered as and are based on The inside and outside dual planning problem of both macro and micro aspect, whereinRepresent that outer layer plan model, i.e. track are handed over The online train flow flow-Density and distribution problem of path,Represent internal layer plan model, i.e. track traffic section The state adjustment problem of upper single vehicles；F、x₁And u₁Be respectively outer layer planning problem object function, state vector and decision-making to Amount, G (x₁,u₁)≤0 be outer layer planning constraints, f, x₂And u₂Be respectively the object function of internal layer planning problem, state to Amount and decision vector, g (x₂,u₂)≤0 is the constraints of internal layer planning, using the outer layer program results of macroscopic aspect as microcosmic The reference input of aspect internal layer planning；

Step F4, the variable bound modeling of operation conflict Resolution, structure exists comprising adjustable train quantity a, train speed ω and train The variables such as time γ stand in interior both macro and micro constraints：Wherein t need to implement the variable of the section k of conflict Resolution about Beam can be described as：a_k(t)≤a_M、ω_k(t)≤ω_M、γ_k(t)≤γ_M, a_M、ω_M、γ_MRespectively maximum adjustable train quantity, most , in the station time, such variable of freeing can be subject to traffic flow distribution, train physical for big train running speed and most long train The constraint of the aspect such as energy and personal distance；

Step F5, Multi-objective Robust optimum road network flow allocation plan are solved：Based on cooperative collision avoidance trajectory planning thought, for Different performance indications, by selecting different conflict Resolution object functions, solve in traffic flow operation macroscopic aspect and are based on Europe It is only implemented in the multiple target traffic flow optimum flow allocation plan of pull-up network model and respectively control section in Rolling Planning interval First Optimal Control Strategy；The detailed process of step F5 is as follows：Order

$d_{it}^2=\left|\right|P_i\left(t\right)-P_i^f|{|}_2^2={(x_{it}-x_i^f)}^2+{(y_{it}-y_i^f)}^2,$

WhereinRepresent distance between t train i present positions and next website square, P_i(t)=(x_it,y_it) table Show the two-dimensional coordinate value of t train i,Represent the next two-dimensional coordinate values for stopping website of train i, that so t The moment priority index of train i may be set to：

${\mathrm{\λ}}_{it}=100\frac{d_{it}^{-2}}{\underset{i=1}{\overset{n_t}{\Σ}}{d}_{it}^{-2}},$

Wherein n_tRepresent there is the train number of conflict on t section, from the implication of priority index, under train distance One website is nearer, and its priority is higher；

Setting optimizing index

$\begin{array}{c}{J}^{*}(u_1^{\left(t\right)},u_1^{(t+\mathrm{\Δ}t)},\mathrm{...},u_1^{(t+p\mathrm{\Δ}t)},\mathrm{...},u_{n_t}^{\left(t\right)},u_{n_t}^{(t+\mathrm{\Δ}t)},\mathrm{...},u_{n_t}^{(t+\Π\mathrm{\Δ}t)})\\ =\underset{s=1}{\overset{\Π}{\Σ}}\underset{i=1}{\overset{n_t}{\Σ}}{\mathrm{\λ}}_{it}\left|\right|P_i(t+s\mathrm{\Δ}t)-P_i^f|{|}_2^2\\ =\underset{s=1}{\overset{\Π}{\Σ}}\underset{i=1}{\overset{n_t}{\Σ}}{\left(P_i\right(t+s\mathrm{\Δ}t)-P_i^f)}^T{Q}_{it}\left(P_i\right(t+s\mathrm{\Δ}t)-P_i^f)\end{array},$

Wherein i ∈ I (t) represent train code and I (t)=1,2 ..., n_t, P_i(t+s △ t) represents train in moment (t+s △ T) position vector, Π represents control time, i.e., the time span of Future Trajectory planning, u from current time_iRepresent to be optimized Train i optimal control sequence, Q_itFor positive definite diagonal matrix, its diagonal element is priority index of the train i in t λ_it, and

Step F6, the train operation state adjustment of Multi-objective Robust optimum section：According to each section or zone flow configuration result, base Mix evolutionary model and Lagrangian plan model in train operation and obtain optimum single vehicles controlled quentity controlled variable, generate the single-row of optimum Car running orbit and respectively regulation and control train only implement its first Optimal Control Strategy in Rolling Planning interval；

Step F7, each train receive and perform train collision avoidance instruction；

Step F8, in next sampling instant, repeat step F5 to F7 is until each train reaches it and frees terminal.