Method for adjusting train operation based on passenger flow prediction

Abstract

The invention discloses a method for adjusting train operation based on predicted passenger flow, which comprises the following steps of: establishing a passenger flow dynamic model by combining the predicted passenger flow condition, and constructing a train adjustment optimization model based on the predicted passenger flow by combining the train dynamic model and considering the constraint condition and the target function of the model; in consideration of the real-time performance of train operation adjustment, a train adjustment strategy searching method based on MPC (model predictive control) is designed and used for solving a train adjustment optimization model; according to the method, passenger flow prediction and train operation adjustment are combined, and the train operation plan is optimized while the travel experience of passengers is ensured, so that the train operation adjustment scheme can well adapt to the dynamic change of the passenger flow, and the service level of urban rail transit is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of urban rail transit, and in particular to a method for adjusting train operation based on predicted passenger flow. Background Art

[0002] In recent years, with the rapid growth of urban scale and population in my country, as well as the substantial increase in the number of private cars, urban traffic congestion has become increasingly prominent. In this context, urban rail transit has gradually become a popular travel option. Compared with traditional public transportation, urban rail transit runs on dedicated lanes and is not interfered with by other means of transportation. Therefore, it has a higher operating speed and shorter stop time, allowing passengers to reach their destination faster, greatly shortening travel time. In addition, urban rail transit also provides convenient transfer services, further facilitating residents' daily travel. These characteristics make urban rail transit an important means to alleviate urban traffic pressure.

[0003] As cities expand, people's travel needs are also increasing. However, due to the diverse characteristics of passengers' travel needs, the passenger flow at a station will have different characteristics over time and geographical location. This unbalanced passenger flow distribution poses challenges to subway operation plans and passenger travel comfort. At present, most subway companies still rely on experience and professional judgment to determine operation plans, lacking relevant calculations and optimization, which leads to energy waste and reduced passenger travel comfort. Therefore, it is of great significance to accurately predict the passenger flow of the subway in the future and adjust the train operation strategy according to the temporal and spatial distribution of passenger flow.

[0004] The research directions of passenger flow forecasting are roughly divided into regular passenger flow forecasting and sudden large passenger flow forecasting. Regular passenger flow forecasting is crucial to daily subway operations and daily travel, while sudden large passenger flow forecasting can be predicted in advance through signaling data, information on activities around stations, and weather warnings. At present, many time-series-based forecasting methods are widely used, but these methods are highly dependent on historical data and cannot cope with situations such as data missing. Recent studies have begun to consider from a spatiotemporal perspective, using methods such as convolutional neural networks and graph neural networks for forecasting, and studying passenger flow from the perspective of the entire line. However, in a complex spatial environment, there is relatively little research on the impact of passenger flow on a single line.

[0005] In addition, most of the current passenger flow-based train operation adjustment methods use static passenger flow, which often leads to a large deviation between the assumed passenger flow and the actual passenger flow, reducing the accuracy of the train operation adjustment strategy. Combining dynamic passenger flow information with the train operation adjustment strategy, that is, using accurate passenger flow prediction results as auxiliary parameters for adjustment, can reduce passengers' travel time, improve passengers' travel experience, and at the same time improve train utilization and reduce energy consumption and operating costs.

[0006] In order to achieve real-time train operation adjustment based on predicted passenger flow, it is necessary to find a solution algorithm with strong real-time performance. However, existing train adjustment strategies often ignore the real-time performance of the algorithm. Although the Lagrangian relaxation method is commonly used, it takes a long time to solve. Heuristic algorithms and reinforcement learning algorithms (such as DQN and Q-Learning) have a higher solution speed, but the former relies on experience and the latter requires a large amount of sample data for learning.

[0007] Therefore, it is very important to design an effective short-term passenger flow prediction method for subways based on the characteristics of subway passenger flow and the use of urban network passenger flow data. This method can predict passenger flow for a period of time in the future and apply the prediction results to the real-time adjustment of trains while considering the real-time nature of the train adjustment algorithm. This will be an effective way to improve the service level of urban rail transit. Summary of the invention

[0008] The purpose of the present invention is to provide a method for adjusting train operation based on predicted passenger flow, so as to cope with the great challenge brought to the train operation adjustment process by the dynamic change of passenger flow.

[0009] The objective of the present invention is achieved through the following technical solutions:

[0010] A method for adjusting train operation based on predicted passenger flow, comprising:

[0011] Combine the historical passenger flow data of each station to predict the passenger flow data of each station in the future time period;

[0012] The passenger flow dynamic model is constructed by using the predicted passenger flow data of each station in the future time period, and the train adjustment optimization model based on the predicted passenger flow is established by combining the train dynamic model and setting preset conditions;

[0013] The train adjustment strategy search method guided by MPC is adopted to solve the train adjustment optimization model based on predicted passenger flow on the basis of meeting the real-time requirements of train operation adjustment, and the operation plan of the train at each station is adjusted according to the solution results.

[0014] It can be seen from the technical solution provided by the present invention that, in combination with the predicted passenger flow, a passenger flow dynamic model is established, and in combination with the train dynamic model, the constraints and objective functions of the model are considered to construct a train adjustment optimization model based on the predicted passenger flow; the real-time nature of train operation adjustment is also taken into account, and a train adjustment strategy search method based on MPC (model predictive control) is designed to solve the train adjustment optimization model; the present invention combines passenger flow prediction with train operation adjustment, and optimizes the train operation plan while ensuring the travel experience of passengers, so that the train operation adjustment plan can adapt well to the dynamic changes of passenger flow and improve the service level of urban rail transit. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0016] Figure 1 A schematic diagram of a method for adjusting train operation based on predicted passenger flow provided by an embodiment of the present invention;

[0017] Figure 2 A schematic diagram of a basic graph processing process provided by an embodiment of the present invention;

[0018] Figure 3 A schematic diagram of a spatiotemporal graph neural network model provided by an embodiment of the present invention;

[0019] Figure 4 A schematic diagram of a spatiotemporal graph convolutional neural network structure provided by an embodiment of the present invention;

[0020] Figure 5 A schematic diagram of a train adjustment optimization model based on predicted passenger flow provided by an embodiment of the present invention;

[0021] Figure 6 A schematic diagram of a basic structure of a strategy network provided by an embodiment of the present invention;

[0022] Figure 7 A schematic diagram of the outbound passenger flow prediction result of a method for adjusting train operation based on predicted passenger flow provided by an embodiment of the present invention;

[0023] Figure 8 A schematic diagram of the prediction results of the incoming passenger flow of a method for adjusting train operation based on the predicted passenger flow provided by an embodiment of the present invention;

[0024] Fig. 9A schematic diagram of a train operation plan adjustment result of a method for adjusting train operation based on predicted passenger flow provided by an embodiment of the present invention;

[0025] Fig.10 A schematic diagram of the total objective function value result of a method for adjusting train operation based on predicted passenger flow provided by an embodiment of the present invention;

[0026] Fig.11 A schematic diagram of the results of various components of the objective function value of a method for adjusting train operation based on predicted passenger flow provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the protection scope of the present invention.

[0028] First, the terms that may be used in this article are explained as follows:

[0029] The terms "include", "comprises", "contains", "has" or other descriptions with similar semantics should be interpreted as non-exclusive inclusion. For example, including certain technical feature elements (such as raw materials, components, ingredients, carriers, dosage forms, materials, dimensions, parts, components, mechanisms, devices, steps, procedures, methods, reaction conditions, processing conditions, parameters, algorithms, signals, data, products or products, etc.) should be interpreted as including not only certain technical feature elements explicitly listed, but also other technical feature elements known in the art that are not explicitly listed.

[0030] The term "consisting of..." means excluding any technical feature elements not explicitly listed. If this term is used in a claim, it will make the claim closed, so that it does not contain technical feature elements other than the technical feature elements explicitly listed, except for the conventional impurities related to them. If this term only appears in a clause of a claim, it only limits the elements explicitly listed in the clause, and the elements recorded in other clauses are not excluded from the overall claim.

[0031] The following is a detailed description of a method for adjusting train operation based on predicted passenger flow provided by the present invention. The contents not described in detail in the embodiments of the present invention belong to the prior art known to professional and technical personnel in the field. If no specific conditions are specified in the embodiments of the present invention, the conventional conditions in the field or the conditions recommended by the manufacturer are followed. If the manufacturer of the reagents or instruments used in the embodiments of the present invention is not specified, they are all conventional products that can be purchased commercially.

[0032] like Figure 1 As shown, a method for adjusting train operation based on predicted passenger flow mainly includes the following steps:

[0033] Step 1: Passenger flow prediction.

[0034] In the embodiment of the present invention, the passenger flow data of each station in the future time period is predicted in combination with the historical passenger flow data of each station.

[0035] Preferably, graph data is constructed using stations and the connection relationships between stations. In the graph data, stations are nodes. If there is a connection relationship between stations, an edge is added between the corresponding nodes, and the characteristics of the nodes are passenger flow data. Through a graph convolutional neural network prediction model based on time and space, learning is carried out in combination with the historical passenger flow data of each station, and then the passenger flow data (i.e., passenger flow parameters) of each station in the future time period are predicted, which may specifically include the passenger exit rate and the passenger entry rate.

[0036] Preferably, the spatiotemporal graph convolutional neural network prediction model (denoted as ST-MGCN model) includes: a plurality of spatiotemporal convolutional blocks and a fully connected layer connected in sequence; wherein each spatiotemporal convolutional block includes a gated convolutional layer, a graph convolutional layer, a gated convolutional layer and a layer normalization layer connected in sequence; Figure 1 The ST-MGCN model structure in the left dashed box will be combined later Figure 3 Make an introduction.

[0037] In the embodiment of the present invention, the future time period mainly refers to the prediction time domain of the MPC in the subsequent step 3. The prediction time domain can be divided into multiple time steps, each time step represents an update point of the system state. Therefore, if the prediction time domain is divided into K stages, each stage contains M time steps, it means that the entire future time period contains K*M time steps.

[0038] Step 2: Establish a train adjustment optimization model based on predicted passenger flow.

[0039] In an embodiment of the present invention, a passenger flow dynamic model is constructed using the predicted passenger flow data of each station in the future time period, and a train adjustment optimization model based on the predicted passenger flow is established in combination with the train dynamic model and preset conditions. Preferably, the internal state variables of the train adjustment optimization model based on the predicted passenger flow are set: train arrival time, passenger capacity when the train arrives (i.e., the number of people in the car), and the number of people left by the previous train at the station; the passenger flow dynamic model of the station is established using the predicted passenger flow data of each station in the future time period, and in combination with the passenger capacity when the train arrives and the number of people left by the previous train at the station; the train dynamic model is established in combination with the train arrival time, the train stop time and the train running time between stations; the train adjustment optimization model based on the predicted passenger flow is established in combination with the passenger flow dynamic model, the train dynamic model and the preset conditions.

[0040] Preferably, the passenger flow dynamic model of the station is established by using the predicted passenger flow data of each station in the future time period, and combining the passenger capacity when the train arrives at the station and the number of people left by the previous train at the station, including: the passenger capacity when the train j arrives at the station n is represented as l j,n , using the predicted passenger flow data of station n in the future time period, that is, the passenger exit rate and passenger entry rate, determine the number of people getting on and off the train at station n, and the corresponding number is recorded as b j,n and d j,n , then the passenger capacity of train j leaving station n is expressed as: ld j,n = l j,n +b j,n -d j,n ; Among them, ld j,n is the passenger capacity of train j leaving station n.

[0041] Preferably, the train dynamic model is established by combining the train arrival time, the train stop time and the train running time between stations, including: the time when train j arrives at station n is represented as t j,n , according to the train stop time u j,n and station-to-station running time r j,n , the state transition is performed, then the time td when train j leaves station n j,n For: td j,n =t j,n +u j,n , the time t at which the train arrives at the next station, that is, station n+1 j,n+1 Expressed as: t j,n+1 =td j,n +r j,n .

[0042] Preferably, the preset conditions include: train stop time constraints, train station running time constraints, train front and rear vehicle safety distance constraints, and train passenger capacity constraints.

[0043] Step 3: Model solving and operation plan adjustment.

[0044] In an embodiment of the present invention, an MPC-guided train adjustment strategy search method is adopted to solve the train adjustment optimization model based on predicted passenger flow on the basis of meeting the real-time requirements of train operation adjustment, and the operation plan of the train at each station is adjusted according to the solution result. Specifically, solving the train adjustment optimization model based on predicted passenger flow includes solving the train stop time and the train operation time between stations of the train adjustment optimization model based on predicted passenger flow in the future time period.

[0045] Preferably, the train adjustment strategy search method guided by MPC, solving the train adjustment optimization model based on predicted passenger flow on the basis of meeting the real-time requirements of train operation adjustment, includes:

[0046] The train adjustment optimization model solution problem based on predicted passenger flow is decomposed into a K-stage optimization problem, and then the optimal strategy of each stage k is solved in turn, k = 1, 2, ..., K; that is, the future time period is divided into K stages, and each stage k includes M time steps;

[0047] Solving the optimal strategy for each stage k includes: based on the current stage system state, obtaining the sequence of train stop times in the internal state vector of the train adjustment optimization model based on predicted passenger flow [u k ,u k+1 ,...,u k+M-1 ], and the sequence of train running time between stations [r k ,r k+1 ,...,r k+M-1 ], the first control quantity u of the two sequences k and r k As the input of the train adjustment optimization model based on predicted passenger flow at stage k, the corresponding control action is decided according to the state, and the time required for decision-making meets the real-time requirements of train operation adjustment.

[0048] In an embodiment of the present invention, solving the train adjustment optimization model based on predicted passenger flow includes: solving the objective function of the train adjustment optimization model based on predicted passenger flow, the objective including predicting the total passenger travel time in M ​​steps at stage k, including the passenger time on the train and the passenger waiting time.

[0049] In order to more clearly demonstrate the technical solution and technical effects provided by the present invention, the method provided by the embodiment of the present invention is described in detail with reference to specific embodiments below.

[0050] 1. Prediction model based on spatiotemporal graph convolutional neural network.

[0051] In an embodiment of the present invention, the passenger flow and its spatiotemporal distribution of multiple stations and lines in the future are dynamically predicted based on the spatiotemporal graph convolutional neural network prediction model to obtain high-precision spatiotemporal change data of passenger flow.

[0052] like Figure 2 As shown in the figure, a basic graph is constructed based on the subway station structure, and the connected graph G in the basic graph d (V,E) is constructed using the physical connection relationship between stations, where A d is the adjacency matrix of the connected graph, assuming there are N stations. For the physical connection relationship of subway stations, 0 and 1 are used to indicate whether the stations are directly connected. If the stations are directly connected, it is set to 1, otherwise it is 0. The connection relationship between stations d m,n Indicates whether station m and station n are directly connected, d m,n It can be expressed as:

[0053]

[0054] Connected graph G d The adjacency matrix of (V,E) is constructed as follows:

[0055]

[0056] The adjacency matrix normalization is to normalize the adjacency matrix A d Perform normalization operation. d The specific operations for normalization are as follows:

[0057]

[0058] Where I is the identity matrix; D is the adjacency matrix A d The corresponding degree matrix, D, can be expressed as follows:

[0059]

[0060] Adjacency Matrix A d The symmetric normalized matrix can be expressed as

[0061] Figure 3 A schematic diagram of a graph convolutional neural network model based on time and space provided in an embodiment of the present invention; it mainly includes three parts: graph data, basic graph adjacency matrix and neural network structure. The graph data uses subway stations as nodes, and the node features are the passenger flow in and out of the station. In an embodiment of the present invention, the subway AFC card swiping data is processed into a time-series graph structure data F = {f1, f2, ..., f t ,...,f T}, T is the total time step and also the total length of F, f tis the passenger flow graph data of each station in t time step, and F is used as the feature matrix X. The neural network structure is based on the input graph data and the normalized adjacency matrix Learn the characteristics of passenger flow and predict passenger flow in the future.

[0062] Figure 4 A schematic diagram of a spatiotemporal graph convolutional neural network structure provided for an embodiment of the present invention provides an example of a neural network structure including two spatiotemporal convolution blocks and a fully connected output layer. Each spatiotemporal convolution block consists of a time-gated convolution layer, a spatial graph convolution layer, and another time-gated convolution layer. The time-gated convolution layer combines the gating mechanism with convolution, and can effectively capture long-term dependencies and nonlinear features in the data. As an intermediate layer of the gated convolution layer, the spatial graph convolution layer uses a bottleneck strategy to achieve scale and feature compression by downscaling and upscaling the channels of the previous and next layers, thereby improving the efficiency of the model. In order to cope with the covariance shift problem of the intermediate layer of the multi-layer neural network, a layer normalization operation is added after each spatiotemporal convolution block. The fully connected layer is responsible for mapping the data of the previous layer to output data. In the figure, C1 and C2 represent the number of input channels and output channels of the neural layer, respectively, and the specific settings are shown in the figure. The input layer contains the adjacency matrix A f , passenger flow data X, the output layer outputs the corresponding predicted passenger flow data. If the batch size of the neural network processing data is batch_size = B, the number of nodes is N, the node features include the inbound passenger flow and the outbound passenger flow, let the number of node features be W = 2, and the time slice size of a single sample data is L, then the input data format of the first gated convolution layer in the spatiotemporal convolution block 1 is: [B, N, L, W]. The number of input channels of the gated convolution layer is the number of features W.

[0063] 2. Train adjustment optimization model based on predicted passenger flow.

[0064] In an embodiment of the present invention, according to the passenger flow parameters provided by the spatiotemporal graph convolutional neural network prediction model, a passenger flow dynamic model of the station is established in combination with the number of train arrivals and the number of people left at the station by the previous train; a basic model for train operation adjustment is established according to the train arrival time, the train stop time and the train running time between stations; at the same time, with the goal of minimizing the total travel time of passengers, and from the perspective of the safety of train operation and the comfort of passenger travel, train stop time constraints, train running time constraints between stations, front and rear vehicle safety distance constraints and train passenger capacity constraints are added to establish a train adjustment optimization model based on predicted passenger flow.

[0065] Figure 5A schematic diagram of a train adjustment optimization model based on predicted passenger flow is provided for an embodiment of the present invention; its input is the passenger entry rate and passenger exit rate corresponding to the train arrival at the station, and the passenger entry rate and passenger exit rate are generated based on the spatiotemporal graph convolutional neural network model. The internal state variables of the train adjustment optimization model based on predicted passenger flow include the train arrival time, the passenger capacity when the train arrives at the station, and the number of people left by the previous train at the current station. The objective function is to predict the total passenger travel time at M steps in the current stage k, including the time the passengers are on the train and the waiting time of the passengers. The constraints include the train stop time constraint, the train station running time constraint, the train front and rear car safety distance constraint, and the train passenger capacity constraint. The solved control variables include the train stop time variable and the train station running time variable. Then, according to the control variables of stage k, combined with the predicted passenger flow parameters, the passenger flow dynamic model and the train dynamic model, the state variables of the train in the next stage can be obtained, and the above process is repeated continuously.

[0066] The passenger flow dynamic model and train dynamic model are described in detail as follows:

[0067] (1) Passenger flow dynamic model.

[0068] The passenger capacity of train j when it arrives at station n is represented by l j,n , using the predicted passenger flow data of station n in the future time period, that is, the passenger exit rate and passenger entry rate, determine the number of people getting on and off the train at station n, and the corresponding number is recorded as b j,n and d j ,n , then the passenger capacity of train j leaving station n is expressed as:

[0069] ld j,n = l j,n +b j,n -d j,n

[0070] Among them, ld j,n is the passenger capacity of train j leaving station n.

[0071] (2) Train dynamic model.

[0072] The time when train j arrives at station n is denoted as t j,n , according to the train stop time u j,n and station-to-station running time r j,n , the state transition is performed, then the time td when train j leaves station n j,n For: td j,n =t j,n +u j,n , the time t at which the train arrives at the next station, that is, station n+1 j,n+1 It is expressed as:

[0073]

[0074] 3. Optimization solution of train adjustment strategy based on MPC (model predictive control) and guidance strategy search method.

[0075] In the embodiment of the present invention, the predictive control advantages of MPC and the optimization ability of guiding strategy search are combined to solve the train operation adjustment model in real time to cope with complex and rapidly changing passenger flows and ensure the real-time and high efficiency of train operation.

[0076] The optimization model solving method of the train adjustment strategy search method guided by MPC is used. The method decomposes the train operation adjustment optimization problem into K-stage optimization problems, and then solves the optimal strategy of each stage k (k=1, 2, ..., K) in turn; the train adjustment strategy search method guided by MPC is used to solve each stage k, which can be specifically described as: based on the current stage system state, the corresponding train stop time control sequence [u k ,u k+1 ,...,u k+M-1 ] and train station operation control sequence [r k ,r k+1 ,...,r k+M-1 ], and then the first control quantity u of the control sequence k and r k As the input of the model in the k stage, the MPC-guided train adjustment strategy search method is to decide the corresponding control action through the strategy network according to the state. The time required for decision-making is basically negligible compared with the minimum stop time. Therefore, it can meet the real-time requirements of train adjustment.

[0077] In the embodiment of the present invention, the decision-making process is described as follows: the policy network outputs the control variables (control actions) according to the state of the system at the current stage, that is, determines the stop time and the inter-station running time of the train. This control action is based on the optimal policy learned by the policy network, and the purpose is to minimize the objective function value, that is, the total travel time of passengers.

[0078] In the optimization model solving process of the train adjustment strategy search method guided by MPC, the role of the policy network is to provide decisions at each stage, that is, to output control variables (train stop time and inter-station running time), and MPC is responsible for solving the model objective function based on these control variables and updating the state. By learning historical data and simulation results, the policy network can predict the effect of taking specific control actions under given conditions and select control actions that can minimize the objective function. The training process of the train adjustment strategy search method guided by MPC is shown in Table 1.

[0079] Table 1: Training process of the MPC-guided train adjustment strategy search method

[0080]

[0081] In Table 1 above, state x k That is, the state variables corresponding to stage k, including the arrival time of the train, the number of passengers when the train arrives, and the number of people left by the previous train at the current station; the control variable o k is the control action obtained through the strategy network, that is, the train stop time variable and the train station running time variable, λ is the Lagrange multiplier, is the gradient of , η is the parameter used to control the update amplitude in gradient descent, τ represents the trajectory, τ=(x 1...T ,u 1...T ), contains the states of the previous T moments and the corresponding train stop time control sequence, θ is the parameter of the policy network, and Γ(.) is a function that transforms the objective function and constraints into an unconstrained objective function based on the Lagrange multiplier method.

[0082] The input data of the above strategy network is the state variables of the k-stage model, and the output control variables are the train stop time series and the train station operation control sequence. The strategy network π θ It consists of multiple fully connected layers, which can be divided into input layer, hidden layer and output layer. Figure 6 As shown. Exemplary: The number of input neurons in the input layer is the total number of state variables, and the number of output neurons is 128; the number of input neurons in the first hidden layer is 128, and the number of output neurons is 64; the number of input neurons in the second hidden layer is 64, and the number of output neurons is 32; the number of input neurons in the output layer is 32, and the number of output neurons is the total number of control variables. In order to prevent the policy network π θ Overfitting adds a dropout layer after each fully connected layer in the hidden layer, and the ratio of the dropout layer adopts the default ratio of 0.5.

[0083] After training is completed, the execution process is as follows:

[0084] (1) Obtaining the current state of the train: This step requires obtaining all relevant state information of the system, including state variables such as the arrival time of the train and the number of people on board.

[0085] (2) Solve the model objective function based on MPC and obtain the control variables: This step uses the model predictive control (MPC) method to solve the optimization model through the policy network according to the current state information to obtain the control variables, namely the train stop time and the train running time between stations. In this part, the policy network parameters are the results of the policy network training, which can quickly determine the optimal control action (i.e., control variables) according to the current state. By learning historical data and simulation results, the policy network parameters can predict the effect of taking a specific control action under a given state and select the control action that can minimize the objective function. Therefore, the "policy network parameters" play a role in step (2) of the execution process to assist the MPC method in solving the optimization problem and determining the optimal control variables.

[0086] (3) Combine the control variables to obtain the objective function value of the current stage and the state of the next stage: In this step, the train operation plan is actually adjusted according to the control variables obtained in the previous step, and the impact of this adjustment on the system state and the impact on the objective function value is observed, so as to obtain the state of the next stage.

[0087] The above is the main solution content provided by the embodiment of the present invention. In order to illustrate the performance of the present invention, Figure 7 to Figure 11 The result graph and performance comparison graph are shown. The dataset used is the AFC data of a certain city subway. There are two identical spatiotemporal convolution blocks in the spatiotemporal graph convolution neural network model, namely spatiotemporal convolution block 1 and spatiotemporal convolution block 2. Each convolution block contains a gated convolution layer, a spatial convolution layer and a gated convolution layer. The channels of these three layers are 64, 16, and 64 respectively, and the size of the temporal convolution kernel and the graph convolution kernel is 3. The batch size of the model processing data samples is B=50, the number of nodes is N=81, and the node features are divided into inbound passenger flow and outbound passenger flow. Let the number of node features be W=2, then the number of input channels of the gated convolution layer is the number of features W=2. The time slice size of a single sample data is L=24. Here, the past 24 time slices are used to predict the passenger flow of the next 3 time slices. The model is trained by minimizing the mean square error between the target value and the predicted value, and the learning rate of the model is 0.001. The simulation verification of train operation adjustment only considers the one-way line scenario of the subway, with a total of 14 stations, numbered from 67 to 80, and the specific time period is from 7:30 to 9:00 on weekdays. This time period is the morning rush hour period with obvious passenger flow characteristics. The initial values ​​of λ and η in the MPC-guided train adjustment strategy search method are set to 0.001, the goal of MPC is to minimize the objective function, and the prediction step size M=3, the strategy network π θ The goal is to minimize the difference between the policy network output and the MPC output. In a complete simulation, the initial stage of the simulation when k = 1 corresponds to 7:30 in the morning. When all trains have completed all stations, the simulation ends.

[0088] In the above figures, Figure 7 This is a schematic diagram of the outbound passenger flow prediction results; Figure 8 This is a schematic diagram of the prediction results of passenger flow entering the station; Fig. 9 This is a schematic diagram of the train operation plan adjustment results. The solid line is the planned operation timetable, and the dotted line is the actual operation timetable. Fig.10 : is a schematic diagram of the total objective function value result, wherein the left side of each sub-graph in the bar graph is the objective function value before adjustment, and thus is the objective function value after adjustment of the solution of the basic invention; Fig.11 Schematic diagram of the results of each component of the objective function value: part (a) is a schematic diagram of the time passengers spend on the bus before and after the adjustment, part (b) is the waiting time of the remaining passengers before and after the adjustment, and part (c) is the waiting time for passengers entering the station before and after the adjustment. In each sub-graph of the bar charts of parts (a) to (c), the left side is the time before the adjustment, and the right side is the time after the adjustment.

[0089] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above embodiments can be implemented by software, or by means of software plus necessary general hardware platforms. Based on such understanding, the technical solutions of the above embodiments can be embodied in the form of software products, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0090] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed in the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims. The information disclosed in the background technology section of this article is only intended to deepen the understanding of the overall background technology of the present invention, and should not be regarded as an admission or in any form that the information constitutes prior art known to those skilled in the art.

Claims

1. A method for adjusting train operation based on predicted passenger flow, characterized in that: include: Combine the historical passenger flow data of each station to predict the passenger flow data of each station in the future time period; The passenger flow dynamic model is constructed by using the predicted passenger flow data of each station in the future time period, and the train adjustment optimization model based on the predicted passenger flow is established by combining the train dynamic model and setting preset conditions; The train adjustment strategy search method guided by MPC is adopted to solve the train adjustment optimization model based on predicted passenger flow on the basis of meeting the real-time requirements of train operation adjustment, and the operation plan of the train at each station is adjusted according to the solution results.

2. A method for adjusting train operation based on predicted passenger flow according to claim 1, characterized in that: The method of combining the historical passenger flow data of each station to predict the passenger flow data of each station in the future time period includes: Graph data is constructed using stations and the connection relationship between stations. In the graph data, stations are nodes. If there is a connection relationship between stations, an edge is added between the corresponding nodes. The feature of the node is passenger flow data. Through the spatiotemporal-based graph convolutional neural network prediction model, combined with the historical passenger flow data of each station for learning, the passenger flow data of each station in the future time period is predicted.

3. The method for adjusting train operation based on predicted passenger flow according to claim 2, characterized in that: The spatiotemporal graph convolutional neural network prediction model includes: a plurality of spatiotemporal convolution blocks and a fully connected layer connected in sequence; wherein each spatiotemporal convolution block includes a gated convolution layer, a graph convolution layer, a gated convolution layer and a layer normalization layer connected in sequence.

4. The method for adjusting train operation based on predicted passenger flow according to claim 1, characterized in that: The method of constructing a passenger flow dynamic model by using the predicted passenger flow data of each station in the future time period, and establishing a train adjustment optimization model based on the predicted passenger flow by combining the train dynamic model and setting preset conditions includes: Set the internal state variables of the train adjustment optimization model based on predicted passenger flow: train arrival time, train passenger capacity when arriving, and the number of passengers left by the previous train at this station; The passenger flow dynamic model of the station is established by using the predicted passenger flow data of each station in the future time period, and combining the passenger volume when the train arrives at the station and the number of passengers left by the previous train at the station; the train dynamic model is established by combining the train arrival time, train stop time and train running time between stations; By combining the passenger flow dynamic model, the train dynamic model and setting preset conditions, a train adjustment optimization model based on predicted passenger flow is established.

5. The method for adjusting train operation based on predicted passenger flow according to claim 4, characterized in that: The preset conditions include: train stop time constraints, train station running time constraints, train front and rear vehicle safety distance constraints, and train passenger capacity constraints.

6. The method for adjusting train operation based on predicted passenger flow according to claim 4, characterized in that: The passenger flow dynamic model of the station is established by using the predicted passenger flow data of each station in the future time period, and combining the passenger volume when the train arrives at the station and the number of people left by the previous train at the station, including: The passenger capacity of train j when it arrives at station n is represented by l j,n , using the predicted passenger flow data of station n in the future time period, that is, the passenger exit rate and passenger entry rate, determine the number of people getting on and off the train at station n, and the corresponding numbers are recorded as b j,n and d j,n , then the passenger capacity of train j leaving station n is expressed as: ld j,n =l j,n +b j,n -d j,n Among them, ld j,n is the passenger capacity of train j leaving station n.

7. The method for adjusting train operation based on predicted passenger flow according to claim 4, characterized in that: The train dynamic model is established by combining the train arrival time, the train stop time and the train running time between stations, including: The time when train j arrives at station n is denoted as t j,n , according to the train stop time u j,n and station-to-station running time r j,n , the state transition is performed, then the time td when train j leaves station n j,n For: td j,n =t j,n +u j,n , the time t at which the train arrives at the next station, that is, station n+1 j,n+1 Expressed as: t j,n+1 =td j,n +r j,n .

8. A method for adjusting train operation based on predicted passenger flow according to claim 4, 5, 6 or 7, characterized in that: Solving the train adjustment optimization model based on predicted passenger flow includes: Solve the train adjustment optimization model based on predicted passenger flow to obtain the train stop time and train running time between stations in the future time period.

9. The method for adjusting train operation based on predicted passenger flow according to claim 1, characterized in that: The train adjustment strategy search method guided by MPC, solving the train adjustment optimization model based on predicted passenger flow on the basis of meeting the real-time requirements of train operation adjustment, includes: The train adjustment optimization model solution problem based on predicted passenger flow is decomposed into a K-stage optimization problem, and then the optimal strategy of each stage k is solved in turn, k = 1, 2, ..., K; that is, the future time period is divided into K stages, and each stage k includes M time steps; Solving the optimal strategy for each stage k includes: based on the current stage system state, obtaining the sequence of train stop times in the internal state vector of the train adjustment optimization model based on predicted passenger flow [u k ,u k+1 ,...,u k+M-1 ], and the sequence of train running time between stations [r k ,r k+1 ,...,r k+M-1 ], the first control quantity u of the two sequences k and r k As the input of the train adjustment optimization model based on predicted passenger flow at stage k, the corresponding control action is decided according to the state, and the time required for decision-making meets the real-time requirements of train operation adjustment.

10. A method for adjusting train operation based on predicted passenger flow according to claim 1 or 9, characterized in that: The solving of the train adjustment optimization model based on predicted passenger flow includes: solving the objective function of the train adjustment optimization model based on predicted passenger flow, wherein the objective includes predicting the total passenger travel time in M ​​steps at stage k, including the passenger time on the train and the passenger waiting time.

Images