CN105095984B - Real-time prediction method for subway train track - Google Patents

Abstract

The invention relates to a method for real-time prediction of subway train tracks, which comprises the following steps: first, according to the planned operation parameters of each train, a topological structure diagram of a rail transit network is generated; then, based on the topological structure diagram, the controllability and sensitivity of the train flow are analyzed Then, according to the planned operation parameters of each train, generate multi-train conflict-free running trajectories; then at each sampling time, based on the current running state of the train and the historical position observation sequence, predict the traveling position of the train at a certain time in the future. This method has high accuracy in predicting the trajectory of subway trains.

Description

A real-time prediction method of subway train trajectory

technical field

The invention relates to a real-time prediction method of a subway train trajectory, in particular to a real-time prediction method of a subway train trajectory based on a robust strategy.

Background technique

With the increasing scale of large and medium-sized cities in our country, the urban traffic system is facing more and more pressure, and vigorously developing the rail transit system has become an important means to solve urban traffic congestion. The national "Eleventh Five-Year Plan" outline pointed out that large cities and urban agglomerations where conditions permit should take rail transit as a priority area of development. my country is experiencing an unprecedented peak period of rail transit development. Some cities have shifted from line construction to network construction, and urban rail transit networks have gradually formed. In complex areas with dense rail transit network and train flow, the method of still adopting train operation plan combined with train interval allocation based on subjective experience gradually shows its backwardness. Due to the influence of various random factors, it is easy to cause congestion in traffic flow tactical management and reduce the safety of traffic system operation; (2) The train scheduling work focuses on maintaining the safe interval between individual trains, and has not yet risen to the strategic management of train flow. At the macro level; (3) The train deployment process mostly depends on the subjective experience of the front-line dispatchers, and the timing of the deployment is relatively random, lacking scientific theoretical support; (4) The deployment methods used by the dispatchers seldom take into account external interference factors The impact of the train allocation scheme is poor in robustness and availability.

The discussion objects of the existing literature are mostly for long-distance railway transportation, but there is still a lack of systematic design for the scientific regulation and control scheme of the urban subway transportation system under the conditions of large flow, high density and small interval operation. The train coordinated control scheme under the complex road network operating conditions needs to calculate and optimize the operating status of a single train on the regional traffic network at the strategic level, and implement collaborative planning for the traffic flow composed of multiple trains; while the multi-train Running conflict resolution is based on the prediction of subway train trajectories. The running state of trains often does not completely belong to a specific motion state. At present, there is no effective real-time prediction method for subway train trajectories.

Contents of the invention

The technical problem to be solved by the present invention is to provide a method for real-time prediction of subway train trajectory with better robustness and usability, and the method has higher prediction accuracy for subway train trajectory.

The technical scheme that realizes the object of the present invention is to provide a kind of real-time prediction method of subway train track, comprises the steps:

Step A, according to the planned operation parameters of each train, generate the topology structure diagram of the rail transit network;

Step B, based on the topology diagram of the rail transit network built in step A, analyze the controllability and sensitivity of the train flow;

Step C, according to the planned operation parameters of each train, on the basis of constructing the train dynamics model, establish a train operation conflict pre-allocation model according to the train operation conflict coupling points, and generate multiple train conflict-free running trajectories;

Step D. At each sampling time t, based on the current running state of the train and the historical position observation sequence, predict the traveling position of the train at a certain moment in the future; the specific process is as follows:

Step D1, train track data preprocessing, take the stop position of the train at the starting station as the origin of coordinates, and at each sampling moment, according to the obtained train original discrete two-dimensional position sequence x=[x ₁ , x ₂ , .. ., x _n ] and y=[y ₁ , y ₂ ,..., y _n ], use the first-order difference method to process it to obtain a new train discrete position sequence Δx=[Δx ₁ , Δx ₂ , .. ., Δx _n-1 ] and Δy=[Δy ₁ , Δy ₂ ,..., Δy _n-1 ], where Δx _i = _xi+1 -xi , Δy _i =y _{i +1} -y _i (i= 1,2,...-1);

Step D2, clustering the train track data, clustering the new discrete two-dimensional position sequence Δx and Δy of the train after processing by setting the number of clusters M′, and using the K-means clustering algorithm to cluster them respectively;

Step D3, use the hidden Markov model to perform parameter training on the clustered train trajectory data, by treating the processed train trajectory data Δx and Δy as the obvious observations of the hidden Markov process, and by setting the hidden state The number N' and the parameter update period τ', based on the latest T' position observations and using the BW algorithm to scroll to obtain the latest hidden Markov model parameter λ'; specifically: since the obtained train trajectory sequence data length is dynamic In order to track the state change of the train trajectory in real time, it is necessary to readjust it on the basis of the initial trajectory hidden Markov model parameter λ′=(π, A, B), so as to more accurately predict the train’s future position at time; at intervals τ′, according to the latest T′ observed values (o ₁ , o ₂ ,..., o _T′ ) for trajectory hidden Markov model parameters λ′=(π, A, B) make a re-estimation;

Step D4, according to the Hidden Markov Model parameters, use the Viterbi algorithm to obtain the hidden state q corresponding to the observed value at the current moment;

Step D5, every time period , according to the latest hidden Markov model parameters λ′=(π, A, B) and the latest H historical observations (o ₁ , o ₂ ,..., o _H ), based on the hidden state of the train at the current moment q, at time t, by setting the prediction time domain h', the predicted position value O of the train in the future period is obtained, so that the trajectory of the subway train in the future period can be estimated rollingly at each sampling moment.

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

Step A1, from the database of the subway traffic control center, extract the site information that each train stops during operation;

Step A2, classify the station information where each train stops according to the positive and negative running directions, and merge the same stations in the same running direction;

Step A3, according to the site merging result, connect multiple sites before and after with straight lines according to the spatial layout of the sites.

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

Step B1, constructing a traffic flow control model on a single subsection; the specific process is as follows:

Step B1.1. Introduce state variable Ψ, input variable u and output variable Ω, where Ψ represents the number of trains that exist at a certain moment on the connected section between stations, which includes two types of single section and multi-section, and u indicates the rail transit dispatcher The dispatching measures implemented for a certain road section, such as adjusting the speed of the train or changing the time of the train at the station, etc., Ω represents the number of trains leaving on the road section in a certain period of time;

Step B1.2, by discretizing the time, establish the form such as Ψ(t+Δt)=A ₁ Ψ(t)+B ₁ u(t) and Ω(t)=C ₁ Ψ(t)+D ₁ u A discrete-time traffic flow control model on a single subsection of (t), where Δt represents the sampling interval, Ψ(t) represents the state vector at time t, and A ₁ , B ₁ , C ₁ and D ₁ represent the state at time t, respectively transfer matrix, input matrix, output measurement matrix and direct transfer matrix;

Step B2, constructing a traffic flow control model on multiple sub-sections; the specific process is as follows:

Step B2.1, according to the spatial layout form of the line and the historical statistical data of the train flow, obtain the traffic ratio parameter β on each sub-section of the crossing line;

Step B2.2. According to the traffic ratio parameter and the discrete-time traffic flow control model on a single subsection, construct a form such as Ψ(t+Δt)=A ₁ Ψ(t)+B ₁ u(t) and Ω(t) = Discrete-time traffic flow control model on multi-subsections of C ₁ Ψ(t)+D ₁ u(t);

Step B3, according to the relationship between the rank of the controllable coefficient matrix [B ₁ , A ₁ B ₁ ,..., A ₁ ^n-1 B ₁ ] of the control model and the value n, qualitatively analyze its controllability, according to the control model The sensitivity coefficient matrix [C ₁ (zI-A ₁ ) ^-1 B ₁ +D ₁ ], quantitative analysis of its input and output sensitivity, where n represents the dimension of the state vector, I represents the identity matrix, z represents the original discrete time The basic factor for the transformation of the traffic flow control model.

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

Step C1, train state transfer modeling, the process of the train running along the rail transit network is a dynamic switching process between stations, according to the station settings in the train operation plan, a Petri net model for a single train switching between different stations is established : E=(g, G, Pre, Post, m) is the train section transfer model, wherein g represents each sub-section between stations, G represents the transition point of the train speed state parameter, Pre and Post represent each sub-section and station respectively The forward and backward connection relationship between Indicates the running section of the train, where m represents the model identifier, and Z ⁺ represents a set of positive integers;

Step C2: Modeling the hybrid system of the train's full running profile, considering the running of the train between stations as a continuous process, starting from the stress situation of the train, deriving the dynamic equation of the train at different running stages according to the energy model, and combining external disturbance factors , establish a mapping function v _G = λ(T ₁ , T ₂ , H, R, α) about the speed v _G of the train in a certain running stage, where T ₁ , T ₂ , H, R and α represent the traction force of the train, Random fluctuation parameters of train braking force, train resistance, train gravity and train state;

Step C3, guessing and solving the train trajectory by means of hybrid simulation, and recursively solving the distance between the train at a certain operation stage and the initial stop position point at any time by subdividing the time and using the characteristics of continuous state change, Where J ₀ is the distance between the train and the initial stop at the initial moment, Δτ is the value of the time window, and J(τ) is the distance between the train and the initial stop at the time τ, from which the trajectory of a single train can be inferred;

Step C4, train station time probability distribution function modeling, for a specific operating line, by calling the train stop time data at each station, to obtain the train stop time probability distribution under the conditions of different lines and different stations;

Step C5, conflict-free robust trajectory allocation of multi-train coupling, according to the time when each train arrives at the conflict point, through the time period division, at each sampling time t, under the premise of incorporating random factors, according to the scheduling rules Robust quadratic programming for train trajectories that do not meet safety interval requirements.

Further, in step D, the value of the number of clusters M' is 4, the value of the number of hidden states N' is 3, the parameter update period τ' is 30 seconds, T' is 10, is 30 seconds, H is 10, and the prediction time domain h' is 300 seconds.

The present invention has positive effects: (1) the real-time prediction method of subway train trajectory of the present invention satisfies the rail traffic control safety interval under the premise, based on the real-time position information of the train rather than setting the specific location of the train before the prediction is implemented. Running status, using data mining means to dynamically infer the train trajectory. (2) The present invention is based on the constructed train trajectory rolling prediction scheme, which can be timely integrated into various interference factors in the real-time operation of the train, improves the accuracy of train trajectory prediction, and overcomes the shortcomings of conventional off-line prediction schemes that are not highly accurate. (3) Based on the controllability and sensitivity analysis results of the rail transit network topology, the present invention can provide a scientific basis for the prediction of the subway traffic flow trajectory, and avoid the arbitrariness of the selection of the prediction scheme.

Description of drawings

Figure 1 is an analysis diagram of train flow operation characteristics;

Figure 2 is a conflict-free 3D robust trajectory estimation diagram.

Detailed ways

(Example 1)

A subway traffic flow optimization control system includes a line topology generation module, a data transmission module, a vehicle terminal module, a control terminal module, and a trajectory monitoring module. The trajectory monitoring module collects train status information and provides it to the control terminal module.

The control terminal module includes the following submodules:

Conflict-free trajectory generation module before train operation: According to the planned operation timetable of the train, the train dynamics model is established first, and then the train operation conflict pre-allocation model is established according to the train operation conflict coupling point, and finally the conflict-free train operation trajectory is generated.

Short-term trajectory generation module for train operation: According to the real-time status information of the train provided by the trajectory monitoring module, use the data mining model to predict the trajectory of the train in the future period.

Train operation status monitoring module: at each sampling time t, based on the train trajectory estimation results, when there may be a violation of safety rules between trains, monitor its dynamic behavior and provide alarm information to the control terminal.

Train collision avoidance trajectory optimization module: when the train operation status monitoring module sends out an alarm message, on the premise of satisfying the physical performance of the train, the regional flow capacity constraints and the rail traffic dispatching rules, the adaptive control theory method is adopted by setting the optimization index function The control terminal module performs robust two-layer planning on the train trajectory, and transmits the planning results to the vehicle terminal module through the data transmission module for execution. The train collision avoidance trajectory optimization module includes two types of planning processes: inner planning and outer planning.

The method for real-time forecasting of the subway train trajectory using the above-mentioned subway traffic flow optimization control system comprises the following steps:

Step A, according to the planned operation parameters of each train, generate the topology structure diagram of the rail transit network; its specific process is as follows:

Step A1, from the database of the subway traffic control center, extract the site information that each train stops during operation;

Step A2, classify the station information where each train stops according to the positive and negative running directions, and merge the same stations in the same running direction;

Step A3, according to the site merging result, connect multiple sites before and after with straight lines according to the spatial layout of the sites.

Step B, based on the topological structure diagram of the rail transit network constructed in step A, analyze the controllability and sensitivity characteristics of the train flow; the specific process is as follows:

Step B1, see Fig. 1, construct the traffic flow control model on the single subsection; Its specific process is as follows:

Step B1.1. Introduce state variable Ψ, input variable u and output variable Ω, where Ψ represents the number of trains that exist at a certain moment on the connected section between stations, which includes two types of single section and multi-section, and u indicates the rail transit dispatcher The dispatching measures implemented for a certain road section, such as adjusting the speed of the train or changing the time of the train at the station, etc., Ω represents the number of trains leaving on the road section in a certain period of time;

Step B1.2, by discretizing the time, establish a shape such as Ψ(t+Δt)=A ₁ Ψ(t)+B ₁ u(t) and Ω(t)=C ₁ Ψ(t)+D ₁ u A discrete-time traffic flow control model on a single subsection of (t), where Δt represents the sampling interval, Ψ(t) represents the state vector at time t, and A ₁ , B ₁ , C ₁ and D ₁ represent the state at time t, respectively transfer matrix, input matrix, output measurement matrix and direct transfer matrix;

Step B2, constructing a traffic flow control model on multiple sub-sections; the specific process is as follows:

Step B2.1, according to the spatial layout form of the line and the historical statistical data of the train flow, obtain the traffic ratio parameter β on each sub-section of the crossing line;

Step B2.2. According to the traffic ratio parameter and the discrete-time traffic flow control model on a single subsection, construct a form such as Ψ(t+Δt)=A ₁ Ψ(t)+B ₁ u(t) and Ω(t) = Discrete-time traffic flow control model on multi-subsections of C ₁ Ψ(t)+D ₁ u(t);

Step B3, according to the relationship between the rank of the controllable coefficient matrix [B ₁ , A ₁ B ₁ ,..., A ₁ ^n-1 B ₁ ] of the control model and the value n, qualitatively analyze its controllability, according to the control model The sensitivity coefficient matrix [C ₁ (zI-A ₁ ) ^-1 B ₁ +D ₁ ], quantitative analysis of its input and output sensitivity, where n represents the dimension of the state vector, I represents the identity matrix, z represents the original discrete time The basic factors for the transformation of the traffic flow control model;

Step C, see Figure 2, according to the planned operation parameters of each train, on the basis of constructing the train dynamics model, establish a train operation conflict pre-allocation model according to the train operation conflict coupling points, and generate multiple train conflict-free running trajectories; the specific process as follows:

Step C1, train state transfer modeling, the process of the train running along the rail transit network is a dynamic switching process between stations, according to the station settings in the train operation plan, a Petri net model for a single train switching between different stations is established : E=(g, G, Pre, Post, m) is the train section transfer model, wherein g represents each sub-section between stations, G represents the transition point of the train speed state parameter, Pre and Post represent each sub-section and station respectively The forward and backward connection relationship between Indicates the running section of the train, where m represents the model identifier, and Z ⁺ represents a set of positive integers;

Step C2: Modeling the hybrid system of the train's full running profile, considering the running of the train between stations as a continuous process, starting from the stress situation of the train, deriving the dynamic equation of the train at different running stages according to the energy model, and combining external disturbance factors , establish a mapping function v _G = λ(T ₁ , T ₂ , H, R, α) about the speed v _G of the train in a certain running stage, where T ₁ , T ₂ , H, R and α represent the traction force of the train, Random fluctuation parameters of train braking force, train resistance, train gravity and train state;

Step C3, guessing and solving the train trajectory by means of hybrid simulation, and recursively solving the distance between the train at a certain operation stage and the initial stop position point at any time by subdividing the time and using the characteristics of continuous state change, Where J ₀ is the distance between the train and the initial stop at the initial moment, Δτ is the value of the time window, and J(τ) is the distance between the train and the initial stop at the time τ, from which the trajectory of a single train can be inferred;

Step C4, train station time probability distribution function modeling, for a specific operating line, by calling the train stop time data at each station, to obtain the train stop time probability distribution under the conditions of different lines and different stations;

Step C5, conflict-free robust trajectory allocation of multi-train coupling, according to the time when each train arrives at the conflict point, through the time period division, at each sampling time t, under the premise of incorporating random factors, according to the scheduling rules Robust quadratic programming for train trajectories that do not meet safety interval requirements.

Step D. At each sampling time t, based on the current running state of the train and the historical position observation sequence, predict the traveling position of the train at a certain moment in the future; the specific process is as follows:

Step D1, train track data preprocessing, take the stop position of the train at the starting station as the origin of coordinates, and at each sampling moment, according to the obtained train original discrete two-dimensional position sequence x=[x ₁ , x ₂ , .. ., x _n ] and y=[y ₁ , y ₂ ,..., y _n ], use the first-order difference method to process it to obtain a new train discrete position sequence Δx=[Δx ₁ , Δx ₂ , .. ., Δx _n-1 ] and Δy=[Δy ₁ , Δy ₂ ,..., Δy _n-1 ], where Δx _i = _xi+1 -xi , Δy _i =y _{i +1} -y _i ( i=1,2,...,n-1);

Step D2, clustering the train track data, clustering the new discrete two-dimensional position sequence Δx and Δy of the train after processing by setting the number of clusters M′, and using the K-means clustering algorithm to cluster them respectively;

Step D3, use the hidden Markov model to perform parameter training on the clustered train trajectory data, by treating the processed train trajectory data Δx and Δy as the obvious observations of the hidden Markov process, and by setting the hidden state The number N' and the parameter update period τ', based on the latest T' position observations and using the BW algorithm to scroll to obtain the latest hidden Markov model parameter λ'; specifically: since the obtained train trajectory sequence data length is dynamic In order to track the state change of the train trajectory in real time, it is necessary to readjust it on the basis of the initial trajectory hidden Markov model parameter λ′=(π, A, B), so as to more accurately predict the train’s future position at time; at intervals τ′, according to the latest T′ observed values (o ₁ , o ₂ ,..., o _T′ ) for trajectory hidden Markov model parameters λ′=(π, A, B) make a re-estimation;

Step D4, according to the Hidden Markov Model parameters, use the Viterbi algorithm to obtain the hidden state q corresponding to the observed value at the current moment;

Step D5, every time period , according to the latest hidden Markov model parameters λ′=(π, A, B) and the latest H historical observations (o ₁ , o ₂ ,..., o _H ), based on the hidden state of the train at the current moment q, at time t, by setting the prediction time domain h', the predicted position value O of the train in the future period is obtained, so that the trajectory of the subway train in the future period can be estimated rollingly at each sampling moment;

The value of the above clustering number M' is 4, the value of the number of hidden states N' is 3, the parameter update period τ' is 30 seconds, T' is 10, is 30 seconds, H is 10, and the prediction time domain h' is 300 seconds.

Apparently, the above-mentioned embodiments are only examples for clearly illustrating the present invention, rather than limiting the implementation of the present invention. For those of ordinary skill in the art, on the basis of the above description, other changes or changes in different forms can also be made. It is not necessary and impossible to exhaustively list all the implementation manners here. And these obvious changes or modifications derived from the spirit of the present invention are still within the protection scope of the present invention.

Claims (1)

1. a real-time prediction method of subway train track, is characterized in that comprising the steps: Step A, according to the planned operation parameters of each train, generate the topology structure diagram of the rail transit network; Step B, based on the topology diagram of the rail transit network built in step A, analyze the controllability and sensitivity of the train flow; Step C, according to the planned operation parameters of each train, on the basis of constructing the train dynamics model, establish a train operation conflict pre-allocation model according to the train operation conflict coupling points, and generate multiple train conflict-free running trajectories; Step D. At each sampling time t, based on the current running state of the train and the historical position observation sequence, predict the traveling position of the train at a certain moment in the future; the specific process is as follows: Step D1, train track data preprocessing, take the stop position of the train at the starting station as the origin of coordinates, and at each sampling time, according to the acquired original discrete two-dimensional position sequence of the train x=[x ₁ ,x ₂ ,.. .,x _n ] and y=[y ₁ ,y ₂ ,...,y _n ], use the first-order difference method to process them to obtain a new train discrete position sequence △x=[△x ₁ ,△x ₂ ,...,△x _n-1 ] and △y=[△y ₁ ,△y ₂ ,...,△y _n-1 ], where △x _i = _{xi +1} -xi ,△y _i =y _i+1 -y _i (i=1,2,...,n-1); Step D2, clustering the train track data, clustering the new discrete two-dimensional position sequence △x and △y of the train after processing by setting the number of clusters M', using the K-means clustering algorithm to cluster them respectively ; Step D3. Use the Hidden Markov Model to perform parameter training on the clustered train trajectory data. By treating the processed train trajectory data △x and △y as the obvious observations of the hidden Markov process, by setting The number of hidden states N' and the parameter update period τ', according to the latest T' position observations and the BW algorithm to obtain the latest hidden Markov model parameters λ'; is dynamic. In order to track the state changes of the train trajectory in real time, it is necessary to readjust it on the basis of the initial trajectory hidden Markov model parameter λ'=(π,A,B), so as to more accurately speculate that the train is at The position at a certain moment in the future; every time period τ', according to the latest T' observations (o ₁ ,o ₂ ,...,o _T' ) for trajectory hidden Markov model parameters λ'=(π, A, B) re-estimate; Step D4, according to the Hidden Markov Model parameters, use the Viterbi algorithm to obtain the hidden state q corresponding to the observed value at the current moment; Step D5, every time period According to the latest hidden Markov model parameters λ'=(π,A,B) and the latest H historical observations (o ₁ ,o ₂ ,...,o _H ), based on the hidden state q of the train at the current moment , at time t, by setting the prediction time domain h', the predicted position value O of the train in the future period is obtained, so that the trajectory of the subway train in the future period can be estimated rollingly at each sampling moment.