Urban rail train working diagram adjusting method, device, equipment, medium and product based on train bottom turnover relation prediction

Abstract

The invention discloses an urban rail train working diagram adjusting method and device based on train bottom turnover relation prediction, equipment, a medium and a product, and relates to the technical field of urban rail train operation control. The method comprises the following steps: constructing an urban rail train working diagram adjustment model according to acquired information data; performing linear conversion on constraint conditions and target functions in the urban rail train working diagram adjustment model by adopting a linearization method to obtain a converted urban rail train working diagram adjustment model; on the basis of a variable prediction model, according to the converted urban rail train working diagram adjustment model, predicting a vehicle bottom turnover variable to obtain a predicted value of the vehicle bottom turnover variable; performing variable assignment and optimization solution processing according to the predicted value of the vehicle bottom turnover variable to obtain a solution result; and determining the adjusted urban rail train working diagram according to the solving result. According to the invention, automatic adjustment of the urban rail train working diagram can be rapidly realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of urban rail train operation control, in particular to a kind of urban rail train diagram adjustment method, device, equipment, medium and product based on car bottom turnover relationship prediction. BACKGROUND

[0002] Urban rail transit has the characteristics of energy saving, punctuality, large capacity, no pollution and safety, and is deeply loved by citizens. In recent years, urban rail transit has developed rapidly, and the mileage of rail transit operation line has shown a growing trend year by year. However, due to the large scale and complexity of urban rail transit, its operation is inevitably affected by external interference, resulting in train delay. How to quickly restore normal operation and reduce the negative impact on passengers is a major challenge for the operation company.

[0003] Due to the small interval in urban rail transit, even slight delay will have a great impact on operation, so it is necessary to handle the emergency in a short time. Under normal circumstances, dispatchers rely on experience to adjust the train diagram to handle emergencies to avoid traffic conflicts. At the same time, adjustment strategies such as adding temporary passengers, midway turning back and skipping stops are taken to reduce the impact on passengers. However, dispatchers are busy and need to coordinate many departments, so it is difficult to make high-quality decisions in a short time. In order to solve the above problems, some researches abstract the train diagram adjustment process into a mixed integer programming model and solve it through commercial solvers or specially designed algorithms. However, with the increasing factors such as passenger flow, adjustment strategy and line condition, the complexity of the adjustment model will greatly increase, and it is currently impossible to automatically adjust the train diagram in a short time. SUMMARY

[0004] The purpose of the present application is to provide a kind of urban rail train diagram adjustment method, device, equipment, medium and product based on car bottom turnover relationship prediction, which can quickly realize the automatic adjustment of urban rail train diagram.

[0005] To achieve the above purpose, the present application provides the following scheme:

[0006] In a first aspect, the present application provides a kind of urban rail train diagram adjustment method based on car bottom turnover relationship prediction, comprising:

[0007] Obtain information data; the information data includes: urban rail basic line data, passenger flow data and delay data;

[0008] According to the information data, an urban rail train diagram adjustment model is constructed; the urban rail train diagram adjustment model includes: constraint condition and objective function;

[0009] The constraint condition and the objective function in the urban rail train diagram adjustment model are linearly converted by using a linearization method to obtain a converted urban rail train diagram adjustment model.

[0010] Based on a variable prediction model, the converted urban rail train diagram adjustment model is used to predict a car bottom turnover variable to obtain a predicted value of the car bottom turnover variable; the variable prediction model is a model that uses a neural network, is based on a known historical converted urban rail train diagram adjustment model corresponding to a train diagram adjustment scene and structure information of a car bottom turnover variable, performs feature conversion, graph convolution, and uses a binary cross-entropy loss function to train and perform value prediction processing; the car bottom turnover variable is a variable that determines the connection relationship between decision-making trains and trains.

[0011] According to the predicted value of the car bottom turnover variable, variable assignment and optimization solving processing are performed to obtain a solving result.

[0012] According to the solving result, an adjusted urban rail train diagram is determined.

[0013] In a second aspect, the present application provides an urban rail train diagram adjustment device based on car bottom turnover relationship prediction, comprising:

[0014] An information data acquisition module is configured to acquire information data; the information data includes urban rail basic line data, passenger flow data, and delay data.

[0015] A model construction module is configured to construct an urban rail train diagram adjustment model according to the information data; the urban rail train diagram adjustment model includes a constraint condition and an objective function.

[0016] A linearization conversion module is configured to linearly convert the constraint condition and the objective function in the urban rail train diagram adjustment model by using a linearization method to obtain a converted urban rail train diagram adjustment model.

[0017] A car bottom turnover variable prediction module is configured to predict a car bottom turnover variable based on a variable prediction model according to the converted urban rail train diagram adjustment model to obtain a predicted value of the car bottom turnover variable; the variable prediction model is a model that uses a neural network, is based on a known historical converted urban rail train diagram adjustment model corresponding to a train diagram adjustment scene and structure information of a car bottom turnover variable, performs feature conversion, graph convolution, and uses a binary cross-entropy loss function to train and perform value prediction processing; the car bottom turnover variable is a variable that determines the connection relationship between decision-making trains and trains.

[0018] A solving processing module is configured to perform variable assignment and optimization solving processing according to the predicted value of the car bottom turnover variable to obtain a solving result.

[0019] an adjusting module configured to determine an adjusted urban rail train diagram according to the solving result.

[0020] In a third aspect, the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the urban rail train diagram adjustment method based on the train stock turnover relationship prediction.

[0021] In a fourth aspect, the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the urban rail train diagram adjustment method based on the train stock turnover relationship prediction.

[0022] In a fifth aspect, the present application provides a computer program product, comprising a computer program, and the computer program is executed by a processor to implement the urban rail train diagram adjustment method based on the train stock turnover relationship prediction.

[0023] According to the specific embodiments provided by the present application, the following technical effects are disclosed:

[0024] The present application provides an urban rail train diagram adjustment method, device, equipment, medium and product based on train stock turnover relationship prediction. The present application obtains information data, i.e. urban rail basic line data, passenger flow data and delay data, and constructs an urban rail train diagram adjustment model containing a target function and a constraint condition. Based on a variable prediction model, the train stock turnover variable is predicted according to the transformed urban rail train diagram adjustment model, and the value of the predicted train stock turnover variable is obtained. In other words, the correlation information between the train stock turnover variable and the constraint is extracted by the neural network, and the train stock turnover variable is predicted to reduce the model solution space. The variable is valued and optimized according to the value of the predicted train stock turnover variable, and the solving result is obtained, and then the adjusted urban rail train diagram is determined. The present application can realize the rapid and intelligent adjustment of the urban rail train diagram, reduce the disposal time of the urban rail sudden accident, and thus realize the automatic adjustment of the urban rail train diagram quickly. BRIEF DESCRIPTION OF DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0026] Figure 1 The flowchart of the urban rail train diagram adjustment method based on the train stock turnover relationship prediction;

[0027] Figure 2 An operation step schematic diagram of the urban rail train diagram adjustment method based on the car bottom turnover relationship prediction;

[0028] Figure 3 A line topology structure diagram;

[0029] Figure 4 A bipartite graph expression schematic diagram;

[0030] Figure 5 A car bottom turnover relationship variable prediction solving framework schematic diagram;

[0031] Figure 6 An assignment strategy schematic diagram of the car bottom turnover relationship variable;

[0032] Figure 7 An adjusted train diagram;

[0033] Figure 8 A structure diagram of the urban rail train diagram adjustment device based on the car bottom turnover relationship prediction;

[0034] Figure 9 A structure schematic diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0035] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0036] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0037] In an exemplary embodiment, as shown in Figure 1 A urban rail train diagram adjustment method based on a car bottom turnover relationship prediction is provided, comprising:

[0038] Step 100: obtaining information data. The information data includes: urban rail basic line data, passenger flow data and delay data. The line basic data includes: line station set, uplink station and downlink station information, depot information, planned train set, uplink planned train and downlink planned train information, added temporary passenger train set, uplink added temporary passenger train and downlink added temporary passenger train information, minimum stop time of train at uplink and downlink station, running time of train in section, turnaround station information, minimum turnaround time and maximum turnaround time of turnaround station, available train number information, planned train diagram, car capacity information. The passenger flow data includes: passenger arrival rate at each station, passenger alighting rate after arrival at each station. The delay information includes: station where train delay occurs, time when train delay occurs.

[0039] Step 200: constructing an urban rail train diagram adjustment model according to the information data. The urban rail train diagram adjustment model includes: constraint condition and objective function. The constraint condition is determined based on train running constraint and passenger boarding and alighting constraint. Specifically, the constraint condition includes: stop time constraint, delay information constraint, planned running constraint, section running time constraint, tracking interval constraint, turnaround constraint, car bottom turnover constraint, alighting passenger constraint, waiting passenger constraint, remaining car capacity constraint, boarding passenger constraint, train passenger capacity constraint and passenger retention constraint.

[0040] The objective function is determined by minimizing passenger retention number, timetable offset and canceled train number; the objective function specifically includes:

[0041] minω p ·f p +ω d ·f d +ω c ·f c .

[0042]

[0043] Wherein, ω p is the weight corresponding to the passenger retention number; f p is the total number of retained passengers of planned trains and added temporary passenger trains; ω d is the weight corresponding to the timetable offset; f d is the timetable offset; ω c is the weight corresponding to the canceled train; f c is the number of canceled trains; s is the station serial number; S is the line station set; zp k,s is the retained passengers of train k at station s; zp g,s is the retained passengers of temporary passenger train g at station s; k is the train serial number; K is the planned train set; g is the temporary passenger train serial number; G is the temporary passenger train set; a k,sis the arrival time of train k at station s; is the planned arrival time of train k at station s; k,s is the departure time of train k at station s; is the planned departure time of train k at station s; k is a variable.

[0044] ω p ,ω d ,ω c may be adjusted according to actual conditions and expectations.

[0045] Step 300: linearize the constraint conditions and objective functions in the urban rail train diagram adjustment model by using a linearization method to obtain the converted urban rail train diagram adjustment model.

[0046] The linearization method is to linearize the min function using less than or equal constraints, linearize the variable product using a large M, and linearize the absolute value function using auxiliary variables.

[0047] The linearization method includes:

[0048] Linearize the min function of the boarding passengers, and use the lower bound constraint to equivalently convert the original boarding passenger constraint:

[0049] sp k,s ≤cp k,s ,k∈K,s∈S

[0050] sp k,s ≤wp k,s ,k∈K,s∈S.

[0051] sp k,s ≤M·φ k ,k∈K

[0052] Linearize the variable product in the stranded passengers, and introduce M for constraint conversion. When φ k = 1, the number of stranded passengers is the number of waiting passengers minus the number of boarding passengers; when φ k = 0, the number of stranded passengers is 0:

[0053] zp k,s ≥wp k,s -sp k,s -M·(1-φ k ),k∈K,s∈S.

[0054] Linearize the train diagram offset expression in the objective function, introduce two auxiliary variables u k,s ,v k,s , and equivalently convert the original expression to:

[0055]

[0056] Then, auxiliary variable w is introduced k,s The product is transformed as follows:

[0057] w k,s ≥u k,s +v k,s -M·(1-φ k )。

[0058]

[0059] where sp k,s is the number of passengers getting on; cp k,s is the remaining capacity; wp k,s is the total number of waiting passengers; and M is a very large positive number.

[0060] Step 400: Based on the variable prediction model, the transformed metro train diagram adjustment model is used to predict the car bottom turnover variable, and the value of the predicted car bottom turnover variable is obtained. The variable prediction model is a neural network based on the known historical transformed metro train diagram adjustment model corresponding to the train diagram adjustment scene and structure information of the car bottom turnover variable, which is a model for feature conversion, graph convolution, and value prediction processing by using binary cross entropy loss function; the car bottom turnover variable is a variable for deciding the connection relationship between train numbers.

[0061] The expression corresponding to the binary cross entropy loss function is:

[0062]

[0063] where L is the binary cross entropy loss function; n is the variable number; k is the train number; N and M are both constant quantities; f is the reverse direction train number; is the label value of the nth train connection variable; is the predicted value of the nth train connection variable.

[0064] N represents the number of train diagram adjustment scenes generated due to parameter changes. represents the label value of each variable, represents the predicted value of each variable. The loss L calculated is the sum of the differences between the predicted values and the label values of all variables in all instances.

[0065] Feature conversion is used to convert the variable node features and constraint node features in the transformed metro train diagram adjustment model to the same dimension, and to explore the interaction between the original features using high-dimensional features.

[0066] The variable node features are: (1) average coefficient of the variable in all constraints; (2) maximum value of the variable coefficient; (3) minimum value of the variable coefficient; (4) variance of the variable coefficient; (5) degree of the variable node; (6) train connection variable value in the planned running graph; (7) reciprocal of the distance between the predicted connection train number and the planned connection train number; (8) delay time. The variable features (1) to (5) are structural information of the variable in the mathematical model, and the variable features (6) to (8) are scenario information of the running graph adjustment. Specifically, the variable feature (6) indicates that most train connection relationships should be similar to the connection relationship in the planned running graph, and changes are made on this basis. The variable feature (7) indicates that other trains closer to the planned connection train are more likely to be connected. The variable feature (8) indicates that the delay duration will affect the train connection relationship, and the longer the delay, the later the connection train will be.

[0067] The constraint node features are: (1) average value of all variable coefficients in the constraint; (2) maximum value of all variable coefficients in the constraint; (3) minimum value of all variable coefficients in the constraint; (4) variance of all variables in the constraint; (5) right side value of the constraint; (6) degree of the constraint node; (7) constraint type, that is, greater than constraint, less than constraint or equal to constraint. The above features are used to describe the structural information of the constraint.

[0068] The processing process corresponding to the feature conversion specifically includes:

[0069]

[0070] Wherein, x v is the initial variable node feature in the model; x c is the initial constraint node feature; is the converted feature corresponding to x v ; is the converted feature corresponding to x c ; EMBEDDING is a fully connected neural network; v is the variable node sequence number; V I is the variable node set; c is the constraint node sequence number; C J is the constraint set.

[0071] The graph convolution is used to convolve the variable node features to the constraint node and convolve the constraint node information to the variable node. The processing process corresponding to the graph convolution specifically includes:

[0072]

[0073] Wherein, is the constraint node feature after the first convolution operation; σ r is a relu activation function; is the neural network weight matrix for the first variable node to constraint node direction convolution operation; CONCAT is a function for concatenating two feature vectors; is x c corresponds to the transformed feature; N(c) is the variable node connected to the constraint node; x e is the edge vector; is x v corresponds to the transformed feature; is the variable node feature after the first convolution operation; is the neural network weight matrix for the first constraint node to variable node direction convolution operation; N(v) is the constraint node connected to the variable node; is the constraint node feature after the second convolution operation; is the neural network weight matrix for the second variable node to constraint node direction convolution operation; is the variable node feature after the second convolution operation; is the neural network weight matrix for the second constraint node to variable node direction convolution operation.

[0074] The value prediction process is:

[0075] The single-layer perceptron is used to predict the train connection variable, and the process uses a sigmoid activation function:

[0076]

[0077] where σ s is the sigmoid activation function; W o is the perceptron weight matrix; is the variable feature after two convolution operations.

[0078] Since ξ k,f is a binary variable, its value can be 0 or 1, so the variable value prediction task can be regarded as a binary classification task. Therefore, a binary cross-entropy loss is used to train the model.

[0079] Step 500: According to the predicted value of the train bottom turnover variable, the variable assignment and optimization solving process are performed to obtain the solving result.

[0080] where the variable assignment and optimization solving process are performed according to the predicted value of the train bottom turnover variable to obtain the solving result, specifically including:

[0081] Based on the predicted values of the train bottom turnover variables, a part of the variables higher than the preset value and a part of the variables lower than the preset value are selected for fixed assignment processing; the variables within the preset value range are optimized and solved based on the solver to obtain a solution result. The assignment operation uses additional constraints to assign values to a part of the high and low prediction score variables.

[0082] The variable assignment operation is:

[0083] The k1 variables with the highest prediction scores and the k0 variables with the lowest prediction scores are fixed and assigned values, as shown in the following formula: Figure 4 The other variables are optimized and solved by a commercial solver in the sub-problems after the implementation of the fixing strategy. In addition, in order to avoid the infeasibility of the sub-problems or the generation of a large gap after fixing a part of the variables, the neighborhood parameters r0 and r1 are introduced, and the solver can search in the neighborhood space with a radius of r0 and r1. The above method can be realized by adding the following additional constraints:

[0084]

[0085]

[0086] wherein, ξ x is the train connection variable on which the assignment operation is taken; x is the variable number; U is the total number of train connection variables.

[0087] Step 600: determining an adjusted urban rail train operation diagram according to the solution result.

[0088] The present application obtains urban rail basic line data, passenger flow data and delay data, constructs an objective function and train operation diagram adjustment constraints for minimizing the number of passengers stranded, the timetable offset and the number of canceled trains. The neural network extracts the correlation information between the train bottom turnover variables and the constraints, and predicts the train bottom turnover variables to narrow the model solution space. The additional constraints are used to assign values to a part of the high and low prediction score variables, and the sub-problems after the assignment are optimized and solved to output the adjusted train operation diagram, realize the rapid and intelligent adjustment of the urban rail train operation diagram, and reduce the disposal time of the urban rail sudden accident. In practical application, the method mentioned in the present application can be realized by the following steps.

[0089] A. First, obtain operation data according to the actual running line, which is line basic data, passenger flow data and delay data.

[0090] The specific steps of implementation A are:

[0091] A.1 Obtain basic route data based on actual operation, where the set of stations is represented as S = {1, 2, ..., 2N}, totaling 2N stations. Stations 1 to N represent stations in the upbound direction, and stations N+1 to 2N represent stations in the downbound direction. Figure 2 As shown. The first station and depot in the upward direction are connected. The train departs from the depot to perform its scheduled service, then returns to the depot for maintenance. There are no other depots. The set of planned train services is K = {1, 2, ..., 2M}, totaling 2M planned train services, of which train services 1 to M are in the upward direction, and train services M+1 to 2M are in the downward direction. The set of temporary passenger train services is G = {1, 2, ..., 2P}, totaling 2P additional train services, of which train services 1 to P are additional trains in the upward direction, and train services P+1 to 2P are additional trains in the downward direction. γ(k) represents the set of train services that may connect with train service k in the opposite direction. The minimum stopping time for each train at the station is... The travel time for each train on each section is T. s run The minimum turnaround time for each train at the origin and destination stations is... Maximum turnaround time is Train k is scheduled to arrive at station s at the following time: Train number k is scheduled to depart from station s at the following time:

[0092] A.2 Obtain passenger flow data for the route; the passenger alighting rate (λ) for train k at station s. k,s The passenger arrival rate at station s during the period from the departure of train k-1 to the departure of train k is τ. k,s .

[0093] A.3 Obtain train delay information. The delayed train number is k′, and the station where the delay occurs is s′. The delay time of the delayed train at the station is T. delay .

[0094] B. Based on the acquired operational data, construct a train timetable adjustment model that considers passenger services. This includes constructing constraints on train stops, planned operation, interval travel time, tracking intervals, turnaround, rolling stock turnover, boarding passengers, alighting passengers, remaining carriage capacity, waiting passengers, carriage passenger capacity, stranded passengers, and an objective function.

[0095] The specific steps to achieve B are as follows:

[0096] B.1 Establish train stopping constraints. To ensure passengers have sufficient time to board and alight at stations, the stopping time for both planned and temporary passenger trains must exceed a certain threshold:

[0097]

[0098] where d k,s , a k,s represent the departure time and arrival time of train k at station s respectively. φ k is a 0-1 variable, which equals to 1 if train k is not canceled, otherwise equals to 0. M is a large value, which is used to deal with the logical condition, i.e. when train k is canceled, the constraint is invalid.

[0099] B.2 Construct the delay information constraint, if train k' is delayed at station s' with delay time T delay , then the dwell time will also be increased:

[0100] d k′,s′ -a k′,s′ ≥ T delay .

[0101] B.3 Construct the planned operation constraint, before the delay occurs, each train should run strictly according to the planned operation diagram, and the arrival and departure time of each train should equal to the planned arrival and departure time:

[0102]

[0103]

[0104] B.4 Construct the interval running time constraint, the running time of each train in each interval is determined by the interval length, line conditions, etc., and is planned in the planned operation diagram by the operating company. The interval running time of each train can be guaranteed by the following constraints:

[0105] a k,s+1 -d k,s ≤ T s run + M(1 - φ i ), k e K U G, s e S \ {N, 2N}.

[0106] a k,s+1 -d k,s ≥ T s run - M(1 - φ i ), k e K U G, s e S \ {N, 2N}.

[0107] where a k,s+1 is the arrival time of train k at station s + 1; φ i is a 0-1 variable, which equals to 1 if train i is not canceled, otherwise equals to 0.

[0108] B.5 Constructing the tracking interval constraint, in order to avoid safety accidents, the minimum tracking interval needs to be met for the execution of two trains of adjacent train services, including the planned train service and the planned train service, and the temporary passenger train service and the temporary passenger train service. g,k represents whether the temporary passenger train service g is added after the planned train service k, and the minimum tracking interval constraint also needs to be met between the planned train service and the temporary passenger train service:

[0109] a k,s -d k-1,s ≥h min -M(1-x k ),k∈K\{1,M+1}∪G\{1,P+1},s∈S.

[0110] a g,s -d k,s ≥h min -M(1-x g,k ),k∈K,g∈G,s∈S.

[0111] a k+1,s -d g,s ≥h min -M(1-x g,k ),k∈K,g∈G,s∈S.

[0112] wherein d k-1,s is the departure time of train service k-1 at station s; h min is the minimum tracking interval; a g,s is the arrival time of temporary passenger train service g at station s; a k+1,s is the arrival time of train service k+1 at station s; d g,s is the departure time of temporary passenger train service g at station s; x g,k represents whether train service g is added after train service k.

[0113] B.6 Constructing the turnaround constraint, when a train executes a train service, it needs to perform turnaround operation at the terminal station and execute the next train service in the opposite direction, the turnaround time is affected by the turnaround facility, and the turnaround operation needs to meet the minimum turnaround time and the maximum turnaround time. Use ξ k,f to represent whether train service k connects train service f in the opposite direction, and the turnaround constraint can be constructed as:

[0114]

[0115]

[0116] B.7 Construct the car bottom turnover constraint, any one train can only be executed by the train out of the section or the train that has executed the previous plan and passenger train. The train that has executed any one train can only return to the section or execute other planned and passenger trains. In addition, due to the limitation of vehicle resources, the train out of the section needs to be less than the total number of trains N1; in order to guarantee the quality of passenger service, the train returning to the section needs to be less than N2, and the above car bottom turnover relationship can be expressed as:

[0117]

[0118] Wherein, σ(k) represents the set of trains that may be connected with train k in the opposite direction; ξ k,g is a 0-1 variable, which is 1 if train k is connected with train g, otherwise 0; y k is a 0-1 variable, which is 1 if train k is executed by the train out of the section, otherwise 0; ξ f,k is a 0-1 variable, which is 1 if train f is connected with train k, otherwise 0; ξ g,k is a 0-1 variable, which is 1 if train g is connected with train k, otherwise 0; x k is a 0-1 variable, which is 1 if the train that executes train k will return to the vehicle depot, otherwise 0; y g is a 0-1 variable, which is 1 if train g is executed by the train out of the section, otherwise 0.

[0119] Urban rail operation is passenger-oriented service, aiming to reduce passenger waiting time and improve passenger riding experience. Therefore, in the process of train diagram adjustment, the number of waiting passengers should be reduced by using line and vehicle resources, considering the passenger experience. Based on urban passenger flow data, the dynamic process of passenger boarding and alighting is constructed.

[0120] B.8 Construct the alighting passenger constraint. The train is empty when it arrives at the starting station of the uplink and downlink directions, so there is no alighting passenger at the starting station. At subsequent stations, part of the passengers arrive at the destination and get off, and the passenger alighting rate is λ k,s , and the number of passengers alighting is:

[0121]

[0122] B.9 Construct the waiting passenger constraint. If the remaining capacity of the car of the previous train is limited, passengers cannot all board, so they will be stranded on the platform. Subsequent passengers arrive at a rate of τ k,s , and when the next train arrives, the total number of waiting passengers is the number of stranded passengers of the previous train plus the number of subsequent arriving passengers:

[0123] wp k,s = wp k-1,s -sp k-1,s +(d k,s -d k-1,s) x τ k,s , k e K, s e S.

[0124] B.10 Construct the remaining capacity of the car constraint, the remaining capacity of the train at the starting station of the up and down direction is the total capacity of the train. When the train arrives at the station, part of the passengers on the train get off, and part of the waiting passengers get on the train, and the passengers on the train change dynamically, and the remaining capacity is the remaining capacity of the car at the previous station minus the number of passengers getting on the train plus the number of passengers getting off the train:

[0125]

[0126] B.11 Construct the passenger boarding constraint: if the train runs normally, the number of boarding passengers is the minimum of the waiting passenger number and the remaining capacity of the car; if the train is canceled, the number of boarding passengers is 0. The number of boarding passengers can be expressed as:

[0127] sp k,s = min{cp k,s ,wp k,s , M·φ k}, k e K, s e S.

[0128] B.12 Construct the number of passengers on the train constraint, at the starting station of the up and down direction, the number of passengers on the train is equal to the number of boarding passengers. At subsequent stations, the number of passengers on the train is equal to the number of passengers on the train before the train arrives at the previous station plus the number of boarding passengers minus the number of passengers getting off the train. The above process can be specifically expressed as:

[0129]

[0130] B.13 Construct the stranded passenger constraint: if the waiting passengers on the platform cannot all get on the train due to the capacity limitation of the car, the stranded passengers will be generated, and the number of stranded passengers is equal to the number of waiting passengers minus the number of boarding passengers. If the train is canceled, the number of stranded passengers of the train is not calculated. The following constraint is used to represent the above process:

[0131] zp k,s = φ k ·(wp k,s -sp k,s ), k e K, s e S.

[0132] B.14 Construct the objective function: after the train delay occurs, it needs to be quickly handled and restored to normal, and the passenger experience needs to be considered, and more capacity is provided by adding temporary passengers and other strategies. Therefore, the model is adjusted from two aspects of company operation indicators and passenger service, three sub-objective functions of passenger stranded number, timetable offset and canceled train number are constructed, and weighted sum is performed.

[0133] The total number of stranded passengers is the sum of the stranded passengers of the planned trains and temporary passengers at all stations:

[0134]

[0135] Total schedule offset is the difference between actual arrival and departure time and planned arrival and departure time of all trains at stations:

[0136]

[0137] The number of train cancellations is:

[0138]

[0139] The final weighted overall objective function is:

[0140] minω p ·f p +ω d ·f d +ω c ·f c .

[0141] C. The model contains part of the nonlinear expression, which needs to be linearly converted to the constraint and objective function, so as to convert the model into a mixed integer programming model, which is convenient for subsequent solution optimization.

[0142] The specific implementation steps of C are as follows:

[0143] C.1 Linearization of min function of boarding passengers, minimization of stranded passengers is one of the optimization objectives in train operation diagram model, equivalent conversion of original boarding passenger constraint is realized by using lower bound constraint, the specific implementation is:

[0144] sp k,s ≤cp k,s ,k∈K,s∈S

[0145] sp k,s ≤wp k,s ,k∈K,s∈S.

[0146] sp k,s ≤M·φ k ,k∈K

[0147] C.2 Linearization of variable product in stranded passengers, constraint conversion is realized by introducing large M. When φ k = 1, the number of stranded passengers is the number of waiting passengers minus the number of boarding passengers; when φ k = 0, the number of stranded passengers is 0:

[0148] zp k,s ≥wp k,s -sp k,s -M·(1-φ k), k e K, s e S.

[0149] C.3 Linearize the running graph offset expression in the objective function, introduce two auxiliary variables u k,s , k,s , and equivalently convert the original expression to:

[0150]

[0151] Next, introduce an auxiliary variable w k,s , and convert the product expression:

[0152] w k,s ≥ u k,s + v k,s - M·(1- φ k ).

[0153]

[0154] The above train diagram adjustment model can handle adjustment scenarios of different delay durations, and optimize the train bottom turnover relationship and arrival and departure times of each train. However, each time the solution is solved for different scenarios, it needs to be re-optimized and cannot utilize historical information. A large number of 0-1 variables in the model will greatly increase the model solving time, so it is considered to solve after predicting part of the variables to speed up the solving speed.

[0155] D. The train bottom turnover variable is a key variable in the train diagram adjustment model and is closely related to the delay duration. The structural information of the train bottom turnover variable is extracted using a graph convolutional neural network, and the train connection relationship is predicted, thereby reducing the solution space. Figure 3 The line topology graph is as follows:

[0156] The specific implementation steps of D are as follows:

[0157] D.1 The mixed integer programming model is graphically expressed, and the mathematical structure is converted into a graph structure to extract variable features. A total of n variables and m constraints in the model are defined. The variables and constraints in the train diagram adjustment model are expressed as a bipartite graph G=(V,E,C) with nodes and edges. The bipartite graph contains two different types of nodes, namely variable nodes and constraint nodes. The edges only exist between the variable nodes and the constraint nodes, and there is no connection between the variable nodes and the variable nodes, and there is also no connection between the constraint nodes and the constraint nodes. As shown in Figure 4 , where V represents a set containing n variable nodes, C represents a set containing m constraint nodes, and E represents a set of edges connecting variable nodes and constraint nodes.

[0158] D.2For the train diagram adjustment problem, variable node features and constraint node features are constructed. There are 8 dimensions for variable features, which are: (1) the average coefficient of the variable in all constraints; (2) the maximum value of the variable coefficient; (3) the minimum value of the variable coefficient; (4) the variance of the variable; (5) the degree of the variable node; (6) the value of the train connection variable in the planned train diagram; (7) the inverse of the distance between the predicted train connection number and the planned train connection number; (8) the delay time. The features (1) to (5) are the structural information of the variable in the mathematical model, and the features (6) to (8) are the scenario information of the train diagram adjustment. Specifically, feature (6) indicates that most train connection relationships should be similar to those in the planned train diagram and change on this basis. Feature (7) indicates that other trains closer to the planned connection train are more likely to be connected. Feature (8) indicates that the duration of the delay will affect the train connection relationship, and the longer the delay, the later the connection train will be.

[0159] There are 7 dimensions for constraint node features, which are: (1) the average value of all variable coefficients in the constraint; (2) the maximum value of all variable coefficients in the constraint; (3) the minimum value of all variable coefficients in the constraint; (4) the variance of all variables in the constraint; (5) the right side value of the constraint; (6) the degree of the constraint node; (7) the constraint type, i.e. greater than, less than or equal to constraint. The above features are used to describe the structural information of the constraint.

[0160] In addition, if a train connection variable appears in the constraint, an edge exists between the constraint node and the variable node to connect them, and the edge weight is the coefficient of the variable in the constraint.

[0161] D.3Collect the dataset, set different delay times, delayed trains and delay locations according to the actual operation situation, and use commercial solvers for optimization solution. Collect the solution values of the train connection variables in the optimal solution, and use the variable solution values as the labels for the subsequent prediction task. At the same time, collect the graph structure information of the model under different delay conditions, record the variable node features, constraint node features and edge features. In order to ensure the effectiveness of the machine learning model training, the above collected data is divided into training set, validation set and test set in the ratio of 7:2:1, which is used for subsequent model training.

[0162] D.4Use neural networks to perform feature convolution operations on nodes and predict train connection variables, as shown in the framework Figure 5 The framework includes feature conversion, graph convolution and value prediction.

[0163] First, use a fully connected neural network to convert the variable node and constraint node features to the same dimension, and use high-dimensional features to explore the interaction between the original features. Assume that the initial variable node features are x v, the initial constraint node feature is x c , The converted features are respectively denoted as x

[0164]

[0165] Then, the converted features are subjected to convolution operation, and two times of convolution are respectively performed on the two types of nodes. The specific sequence is: the variable node feature information is convolved to the constraint node, and the constraint node information is convolved to the variable node. The above process is repeated twice, and the relationship between the two types of nodes is fully extracted. The above process is:

[0166]

[0167] x e is an edge vector, which is used to determine the weight relationship between the variable node and the constraint node. N(c) represents the neighbor node of the constraint node, i.e., the variable node connected to the constraint node. N(v) represents the neighbor node of the variable node, i.e., the constraint node connected to the variable node. CONCAT is used to concatenate two feature vectors, and W is the weight value between neurons in the neural network. σ r is a relu activation function.

[0168] The convolved variable node fully receives the structure information from the constraint node, and then uses a single-layer perceptron to predict the train connection variable. This process uses a sigmoid activation function:

[0169]

[0170] Since ξ k,f is a binary variable, its value can be 0 or 1, so the variable value prediction task can be regarded as a binary classification task. Therefore, a binary cross-entropy loss is used to train the model. The loss function is:

[0171]

[0172] In the above equation, N represents the number of running graph adjustment scenarios due to parameter changes. represents the label value of each variable, represents the predicted value of each variable. The calculated loss L is the sum of the differences between the predicted values and the label values of all variables in all instances. After inputting the basic line data, passenger flow data and delay data, the trained neural network can predict the train connection variable in the model.

[0173] E.Because of the complexity of the parameters and structure of the neural network, there are many potential hooking possibilities between trains, and the difficulty of the prediction task is greatly increased. Therefore, after the prediction task is completed, a value assignment strategy is adopted, and only variables with high and low prediction scores are assigned values.

[0174] The specific operation steps of E are as follows:

[0175] The k1 variables with the highest prediction scores and the k0 variables with the lowest prediction scores are fixed and assigned values, as shown in Figure 6 . Other variables are optimized and solved by a commercial solver in the sub-problem after the implementation of the fixing strategy. In addition, in order to avoid the infeasibility of the sub-problem or the generation of a large gap after fixing part of the variables, the neighborhood parameters r0 and r1 are introduced, and the solver can search in the neighborhood space with a radius of r0 and r1. The above method can be realized by adding the following additional constraints:

[0176]

[0177] Experimental data on the test set show that the method based on the prediction of the train bottom turnover relationship has a significant acceleration effect in all instances, and the calculation efficiency is improved by more than 80% compared with the original CPLEX. Although the prediction method does not converge to the global optimal value in 8% of the instances, the gap with the optimal value is kept within 5%. Figure 7 The train working diagram adjustment result after a 500-second delay of the 12th train at the Rongjingdong Street Station in a certain subway line is shown. The red line represents the uplink train, the green line represents the downlink train, and the blue line represents the added train.

[0178] In an exemplary embodiment, as shown in Figure 8 , a city rail train working diagram adjustment device based on train bottom turnover relationship prediction is provided, comprising:

[0179] An information data acquisition module is configured to acquire information data, wherein the information data includes city rail basic line data, passenger flow data, and delay data.

[0180] A model construction module is configured to construct a city rail train working diagram adjustment model based on the information data, wherein the city rail train working diagram adjustment model includes constraint conditions and an objective function.

[0181] A linearization conversion module is configured to linearize the constraint conditions and the objective function in the city rail train working diagram adjustment model using a linearization method to obtain a converted city rail train working diagram adjustment model.

[0182] The vehicle bottom turnover variable prediction module is configured to predict a vehicle bottom turnover variable based on a variable prediction model and according to the converted metro train diagram adjustment model, and obtain a predicted value of the vehicle bottom turnover variable. The variable prediction model is a model that uses a neural network, is based on a known historical converted metro train diagram adjustment model corresponding to a train diagram adjustment scene and structure information of a vehicle bottom turnover variable, and is configured to perform feature conversion, graph convolution, training using a binary cross-entropy loss function, and value prediction processing. The vehicle bottom turnover variable is a variable that determines the connection relationship between decision-making train numbers and train numbers.

[0183] The solving processing module is configured to perform variable assignment and optimization solving processing according to the predicted value of the vehicle bottom turnover variable, and obtain a solving result.

[0184] The adjustment module is configured to determine an adjusted metro train diagram according to the solving result.

[0185] In an exemplary embodiment, a computer device is provided, which can be a server or a terminal. An internal structure diagram of the computer device can be as shown in Figure 9 The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the computer device is configured to store metro train diagram adjustment data predicted based on a vehicle bottom turnover relationship. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with external terminals through a network connection. The computer program is executed by the processor to implement a metro train diagram adjustment method based on a vehicle bottom turnover relationship prediction.

[0186] Those skilled in the art can understand that Figure 9 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0187] In an exemplary embodiment, a computer device is provided, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in each of the above method embodiments.

[0188] In an exemplary embodiment, a computer readable storage medium storing a computer program is provided, the computer program, when executed by a processor, implements the steps in the above method embodiments.

[0189] In an exemplary embodiment, a computer program product is provided, comprising a computer program, the computer program, when executed by a processor, implements the steps in the above method embodiments.

[0190] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of the relevant data need to comply with relevant regulations.

[0191] The technical features of the above embodiments can be combined in any manner. To make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present disclosure.

[0192] The principles and implementation modes of the present application are described by applying specific examples herein, and the above embodiment descriptions are only used to help understand the method of the present application and its core idea; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation modes and application ranges will have changes. In conclusion, the content of the present description should not be understood as a limitation of the present application.

Claims

1. A method for adjusting a city rail train diagram based on a car bottom turnover relationship prediction, characterized in that, The method comprises the following steps: obtaining information data; the information data comprises urban rail basic line data, passenger flow data and delay data; constructing an urban rail train diagram adjustment model according to the information data; the urban rail train diagram adjustment model comprises constraint conditions and a target function; linearly converting the constraint conditions and the target function in the urban rail train diagram adjustment model by using a linearization method to obtain a converted urban rail train diagram adjustment model; predicting a car bottom turnover variable value based on a variable prediction model and the converted urban rail train diagram adjustment model, wherein the variable prediction model is a model that uses a neural network, performs feature conversion, graph convolution and value prediction processing based on a known historical converted urban rail train diagram adjustment model corresponding to a train diagram adjustment scene and structure information of the car bottom turnover variable, and is trained by using a binary cross-entropy loss function; the car bottom turnover variable is a variable that determines the connection relationship between decision-making trains and trains; performing variable assignment and optimization solving processing according to the predicted car bottom turnover variable value to obtain a solving result; determining an adjusted urban rail train diagram according to the solving result.

2. The urban rail train working diagram adjustment method based on car bottom turnover relationship prediction according to claim 1, characterized in that, The target function is determined by minimizing the number of passengers stranded, the time table offset and the number of canceled trains; the target function specifically comprises: min ω p · f p + ω d · f d + ω c · f c ; ω p is the weight corresponding to the number of passengers stranded; f p is the total number of stranded passengers for the planned train and the added temporary passenger train; ω d is the weight corresponding to the timetable offset; f d is the timetable offset; ω c is the weight corresponding to the canceled train; f c is the number of canceled trains; s is the station number; S is the set of stations on the line; zp k,s is the stranded passengers of train k at station s; zp g,s is the stranded passengers of temporary passenger train g at station s; k is the train number; K is the set of planned trains; g is the temporary passenger train number; G is the set of temporary passenger trains; a k,s is the arrival time of train k at station s; is the planned arrival time of train k at station s; d k,s is the departure time of train k at station s; is the planned departure time of train k at station s; φ k is the variable.

3. The urban rail train working diagram adjustment method based on car bottom turnover relationship prediction according to claim 1, characterized in that, The expression corresponding to the binary cross-entropy loss function is: Wherein, L is a binary cross-entropy loss function; n is a variable serial number; k is a train serial number; N, M are constant quantities; f is a reverse train; is the label value of the nth train connection variable; is the predicted value of the nth train connection variable.

4. The urban rail train working diagram adjustment method based on car bottom turnover relationship prediction according to claim 1, characterized in that, The feature conversion is used to convert the variable node features and the constraint node features in the converted urban rail train diagram adjustment model to the same dimension, and explore the interaction relationship between the original features by using high-dimensional features; The processing process corresponding to the feature conversion specifically comprises: where x v is the initial variable node feature in the model; x c is the initial constraint node feature; is the converted feature for x v ; is the converted feature for x c ; EMBEDDING is a fully connected neural network; v is the variable node sequence number; V I is the variable node set; c is the constraint node sequence number; C J is the constraint set.

5. The urban rail train working diagram adjustment method based on car bottom turnover relationship prediction according to claim 4, characterized in that, The graph convolution is used to convolve the variable node features in the train diagram adjustment model to the constraint nodes, and convolve the constraint node information to the variable nodes; the processing process corresponding to the graph convolution specifically comprises: wherein, is the constraint node feature after the first convolution operation; σ r is the relu activation function; is the neural network weight matrix for the first variable node to constraint node direction convolution operation; CONCAT is a function to concatenate two feature vectors; is the variable node feature after the first convolution operation; x c is the transformed feature; N(c) is the variable node connected to the constraint node; x e is the edge vector; is the transformed feature; x v is the transformed feature; is the constraint node feature after the first convolution operation; x is the neural network weight matrix for the first constraint node to variable node direction convolution operation; N(v) is the constraint node connected to the variable node; is the constraint node feature after the second convolution operation; x is the neural network weight matrix for the second variable node to constraint node direction convolution operation; is the variable node feature after the second convolution operation; x is the neural network weight matrix for the second constraint node to variable node direction convolution operation.

6. The urban rail train working diagram adjustment method based on car bottom turnover relationship prediction according to claim 1, characterized in that, The variable assignment and optimization solving processing according to the predicted car bottom turnover variable value to obtain a solving result, specifically comprises: Based on the predicted car bottom turnover variable value, a part of variables higher than a preset value and a part of variables lower than the preset value are selected for fixed assignment processing; The variables within the preset value range are optimized and solved based on a solver to obtain a solving result.

7. A device for adjusting an urban rail train diagram based on a car bottom turnover relationship prediction, characterized in that, The method comprises the following steps: an information data obtaining module is configured to obtain information data; the information data comprises urban rail basic line data, passenger flow data and delay data; a model constructing module is configured to construct an urban rail train diagram adjustment model according to the information data; the urban rail train diagram adjustment model comprises constraint conditions and a target function; a linearization converting module is configured to linearly convert the constraint conditions and the target function in the urban rail train diagram adjustment model by using a linearization method to obtain a converted urban rail train diagram adjustment model; The vehicle bottom turnover variable prediction module is configured to perform prediction of a vehicle bottom turnover variable based on a variable prediction model according to the converted urban rail train diagram adjustment model, and obtain a value of the predicted vehicle bottom turnover variable; the variable prediction model is a model that is trained by using a neural network, based on historical converted urban rail train diagram adjustment model corresponding to a train diagram adjustment scene and structure information of a known vehicle bottom turnover variable, and performs feature conversion, graph convolution, and value prediction processing by using a binary cross-entropy loss function; the vehicle bottom turnover variable is a variable that determines a hook connection relationship between decision train numbers and train numbers. The solving processing module is configured to perform variable assignment and optimization solving processing according to the value of the predicted vehicle bottom turnover variable, and obtain a solving result. The adjustment module is configured to determine an adjusted urban rail train diagram according to the solving result.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the urban rail train diagram adjustment method based on vehicle bottom turnover relationship prediction according to any one of claims 1-6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the urban rail train diagram adjustment method based on vehicle bottom turnover relationship prediction according to any one of claims 1-6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the urban rail train diagram adjustment method based on vehicle bottom turnover relationship prediction according to any one of claims 1-6.

Images