Method for adjusting and optimizing driving in recovery stage under uncertain interruption duration
By adopting a dual-target optimization model and a rolling time domain control algorithm in the rail transit interrupt recovery phase, the problems of rail transit interrupt recovery speed and impact of passenger travel are solved, and rapid recovery and precise adjustment are achieved.
Patent Information
- Application Number
- CN202411116671.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-14
- Publication Date
- 2024-11-22
AI Technical Summary
When handling rail transit interruptions, the prior art is difficult to quickly restore normal operation and ignore passenger travel benefits. It fails to effectively consider the recovery process after the interruption, resulting in the system impact not being completely eliminated.
A driving adjustment optimization method for recovery stage under uncertain interrupt duration was designed. By modeling the train timetable, the minimum interrupt recovery time and minimum passenger travel time were adopted as dual goals, combined with the rolling time domain control algorithm, the second phase was used. Distributed solution algorithm to reduce calculation time and improve accuracy and optimize driving adjustment solutions.
It realizes the fastest recovery of normal operation under uncertain interruptions, reduces the impact on passenger travel, improves the accuracy and speed of driving adjustments, and adapts to scheduling problems in dynamic environments.
Smart Images

Figure CN119005430A_ABST
Abstract
Description
Technical Field
[0001] The invention discloses a method for optimizing driving adjustment in a recovery phase under uncertain interruption duration, and belongs to the technical field of rail transportation. Background Art
[0002] As the proportion of urban rail transit in public transportation increases, urban public transportation is increasingly dependent on rail transit. When a sudden disturbance occurs in the urban rail transit system, causing delays or even interruptions, the impact is even greater, causing losses to social and economic development. The main deficiencies in the current domestic and foreign research on the optimization of train adjustment under subway operation interruptions are: most of the studies assume that the duration of the interruption is known, so these methods can only be used as a plan or an organizational plan for construction and maintenance in practice, and are difficult to apply directly in the case of real sudden interruptions; in actual operations, train dispatchers will choose to hold the train and wait before dispatching it according to the normal schedule, during which the travel interests of passengers are often ignored; in most studies, train adjustment is usually only for the duration of the operation interruption, without considering the recovery process after the interruption ends. From troubleshooting to the restoration of the operation diagram from the emergency timetable to the planned operation diagram, the impact of the interruption on the rail transit system has not been eliminated.
[0003] The rolling horizon control algorithm is a specific manifestation of the rolling optimization principle in model predictive control. The difference between predictive control and traditional optimal control is that its optimization target is not the entire production process, but consists of optimizations of multiple finite time periods. After the optimization of each time period is completed, the result is returned to the object, and then a new round of optimization begins with the new state of the object. Therefore, in a dynamic uncertain environment, the rolling horizon control can promptly detect real-time interference in the model and make timely adjustments to the interfered objects, reducing the errors caused by uncertain factors, and is targeted at scheduling problems in a dynamic environment. Summary of the invention
[0004] The purpose of the present invention is to overcome the problems existing in the prior art and provide a method for optimizing driving adjustment in the interruption recovery phase under uncertain interruption duration, which can solve the problem of urban rail transit resuming normal operation as quickly as possible under the condition of uncertain interruption duration while ensuring passenger travel.
[0005] In order to solve the above technical problems, the present invention models the train timetable, including the following steps:
[0006] S1. Traffic adjustment optimization based on minimum interruption recovery time to reduce the impact on passenger travel;
[0007] S2, measure the impact of traffic adjustment optimization on passengers based on passenger travel time;
[0008] S3. Based on the data of S1 and S2, a traffic adjustment optimization model for the interruption recovery phase is established: using the stop plan and the average stop time as decision variables, a dual-objective traffic adjustment optimization model is established with the fastest interruption recovery time as the primary objective and the minimum passenger travel time as the secondary objective. The secondary objective is used for selection only when the primary objective is equal;
[0009] S4. Design a two-stage distributed solution algorithm to reduce the time of a single calculation of the rolling horizon algorithm, improve the accuracy, and use the rolling horizon algorithm to solve the dual-objective driving adjustment optimization model under uncertain interruption time.
[0010] Furthermore, S1 specifically includes the following steps:
[0011] S101, interruption recovery phase end time Tr end The latest train number that "catches up" with the planned operation diagram is determined by the latest train number. The latest train number that "catches up" with the original daily planned operation diagram is the end time of the interruption recovery phase, as follows:
[0012] T end =Max{Tr n J}
[0013] In the formula, Tr n j It is the time when train number j catches up with the original daily planned operation diagram at node n.
[0014] Furthermore, S2 specifically includes the following steps:
[0015] S201. Passenger travel time can be divided into platform waiting time and on-board time. Compared with on-board time, waiting on the platform is more easily perceived by passengers. Long-term platform detention will also cause passenger anxiety. Therefore, a penalty coefficient α is added to the platform waiting time in this model. w To reflect the impact of platform waiting time on passengers' travel time perception, as follows:
[0016] t p_total =α w t w_total +t c_total
[0017] Where, t p_total is the total passenger travel time, t w_total is the platform waiting time, t c_total The time in the car;
[0018] The total passenger platform waiting time and the total passenger on-board time can be obtained by integrating the number of people on the platform and on the train over time, as follows:
[0019]
[0020]
[0021] In the formula, p_w s The number of passengers waiting at the platform of station s, p_c j is the number of passengers on train j, Tr0 is the start time of the interruption recovery phase;
[0022] The passengers waiting on the platform and on the train are composed of passengers going to different destinations, as follows:
[0023]
[0024]
[0025] In the formula, p_w s,s′ is the number of passengers waiting for the train to station s' at platform s of station, pc j,s′ is the number of passengers on train number j who will arrive at s';
[0026] In this model, the OD of passengers in the passenger flow is known, and the passenger flow enters the platform at a fixed arrival rate, so the number of passengers waiting on the platform increases continuously, while the passengers waiting on the platform will only board the train when the train leaves the station, so the decrease in the number of passengers waiting on the platform is discrete. end ], the number of passengers waiting at station s and intending to go to station s' changes with time as follows:
[0027]
[0028] In the formula, is the number of passengers who board train j at station s and go to station s', is the departure time of train number j at the up / down track of station s;
[0029] The number of passengers in the carriage changes discretely. According to the assumption, when the train enters the station, the passengers get off, and the number of passengers on the train decreases. When the train leaves the station, the passengers get on, and the number of passengers on the train increases. For passengers on the train who need to go to a certain station, they will get off after the train arrives at the target station. Therefore, any passenger who satisfies T∈[Tr0,Tr end ] at time T, the change in the number of passengers on train number j who need to go to station s' satisfies the following formula: when the train arrives at station s', the number of passengers in the car with that station as their destination is 0.
[0030]
[0031] Furthermore, S3 specifically includes the following steps:
[0032] S301. Make assumptions based on the operational characteristics of urban rail transit systems and existing research;
[0033] S302, taking the stop plan and the average stop time as decision variables;
[0034] S303. In the interruption recovery phase, the main purpose of driving adjustment is to restore normal operation as soon as possible and reduce the impact of the interruption on passenger travel. This model selects the minimum recovery time and the minimum passenger travel time as dual optimization goals, hoping to find a driving adjustment plan with fast recovery and small impact on passengers. In general, the earlier the interruption is restored, the fewer passengers will be affected by the interruption. Therefore, the fastest interruption recovery time is set as the main goal, and the minimum passenger travel time is set as the secondary goal. The secondary goal is used for selection only when the main goal is not higher, as shown below:
[0035] MinZ=W*(Tr end -Tr0)+t p_total
[0036] Where Tr0 is the first time to "catch up" with the original train operation diagram, and W is the main target coefficient, which is a very large positive integer.
[0037] Furthermore, S4 specifically includes the following steps:
[0038] S401. The way of controlling the error of the rolling time domain algorithm comes from the continuous single calculation optimization. In order to improve the accuracy, it is necessary to reduce the time of single calculation. According to the model objectives and the characteristics of urban rail traffic adjustment problems, a two-stage step-by-step solution algorithm is independently designed. The model takes the interruption recovery time as the main goal, and the interruption recovery time is determined by the interruption recovery time of each train {Tr n j}The biggest moment in decision;
[0039] S402, in order to obtain {Tr n j} minTr , for the first stage, a heuristic search algorithm based on the best priority strategy and feasible solution range is designed;
[0040] S403, the {Tr n j} minTr As known quantities, the starting and ending positions of the trains are known in the second stage, which is a problem of finding the intermediate motion process based on the time and position spatial difference. A heuristic iterative algorithm is designed with the armistice plan and the average stop time as decision variables;
[0041] S404. Use the rolling horizon control algorithm to solve the driving adjustment in the recovery phase under the uncertain interruption duration, and execute steps S402 and S403 in each calculation window to obtain the optimal driving adjustment optimization solution in the interruption recovery phase.
[0042] Furthermore, S301 makes assumptions based on the operational characteristics of the urban rail transit system and in combination with existing research, which specifically includes the following steps:
[0043] (1) When the train adjustment begins, the train on the non-normal track must first complete this section of the path and return to the normal track. For example, if a down train on a single-track two-way route happens to be on the shared up track, it must first pass through this section of the track and return to the down track. It cannot be adjusted as an up train.
[0044] (2) A train is considered to have resumed operation only when its departure time “catches up” with the planned timetable. Operation adjustments during the recovery phase must not be completed until all trains resume the planned timetable.
[0045] (3) Only trains that start running after resumption are allowed to use the skipping scheme. Trains that have already started running before resumption are not allowed to use the skipping scheme to prevent passengers from being unable to reach their destinations.
[0046] (4) The change in interval running speed caused by skipping stops is not considered. In the model, skipping stops are only reflected as a stop time of 0.
[0047] (5) It is assumed that the time it takes for a train to pass through the station section is fixed and known, and the error caused by driving operations is not considered.
[0048] (6) For the same train, except for skipped stations, it is assumed that the stop time at all stations is the same.
[0049] (7) Passengers enter the station at a fixed arrival rate. The OD of the arriving passengers at different stations is known, and passengers enter the platform directly without considering the passage time.
[0050] (8) Passenger boarding is subject to train capacity restrictions. Passengers who are unable to board will continue to wait for the next available train.
[0051] (9) Assume that the platform space is infinite and passengers entering the station are not constrained by the platform capacity.
[0052] (10) Assuming that passengers get on and off the train instantly, in order to make the calculation clear, it is stipulated that passengers get off when the train enters the station and get on when the train leaves the station.
[0053] (11) Automatic station-to-station blocking is used to control the train’s safe tracking distance.
[0054] Furthermore, S302 uses the stop time as a decision variable, and specifically includes the following steps:
[0055] 1) Train reachable node constraints: In this model, trains run along the main line, and there are no special crossing trains. Therefore, the node reached by a certain train is only related to the initial position of the train.
[0056]
[0057] 2) Tracking train safety constraints: According to the model assumptions, the automatic station-to-station blocking method is used to control the train after the interruption. In order to ensure that the tracking train safety constraints meet the automatic station-to-station blocking, the time when the train arrives at the exit node is required to meet the following requirements;
[0058]
[0059] 3) Adjacent node operation time constraint: Different from the interruption continuation stage, the interruption fault has been eliminated in this stage. Although the train operation status has not yet returned to normal, from the perspective of passengers, the interruption has ended. The operating company should provide normal quality service level, so the train should not be allowed to stop at the station for too long. Therefore, the stop time needs to meet the minimum stop time constraint and is also subject to the maximum stop time constraint;
[0060] The train's stop time at a station is affected by whether it skips a station. If the train skips a station, the stop time is 0. If the train stops at a station, the average stop time is used according to the model assumptions.
[0061]
[0062]
[0063] 4) Interruption recovery constraint: In normal operation management, the operation and timetable of the train will naturally have small errors. Short-term errors can be manually resolved by operators. Therefore, considering the complexity and necessity of the model calculation, in this model, the train departure time and the planned timetable train departure time are kept within a small error, and the train is considered to have completed the interruption recovery;
[0064]
[0065] 5) Carriage resource constraints: In this model, the correspondence constraints between car cradles and train numbers also need to be considered. In urban rail operations, the train number corresponds to the car cradle number. For example, in the form of train number 0125, 01 represents the car cradle number, and 25 represents the shift. As long as the train numbers operated by this car cradle are 01XX, the latter shift number will increase after each turnaround. Therefore, each car cradle can only correspond to a certain series of train numbers. During the interruption recovery process, when any train number j has been restored with the planned train number p_j as the target, other car cradle numbers j' can no longer be restored with the train number that uses the same car cradle as p_j in the planned train number as the target;
[0066]
[0067] 6) Passenger boarding constraints: According to the model assumption, passengers will only board trains that can reach their destinations. Therefore, when a train skips a station, passengers who are going to that station will not board the train, and passengers waiting at the skipped station cannot board the train. Therefore, only passengers whose train stops at that station and also at the target station can board the train.
[0068]
[0069] 7) Train capacity constraint: This model considers the impact of train capacity on passenger boarding. Therefore, at any time T, the number of passengers on the train should meet the train passenger capacity constraint and not be greater than the maximum number of passengers on the train.
[0070] pc j ≤C
[0071] When the sum of the number of passengers on board and the number of passengers in the carriage exceeds the maximum passenger capacity of the train, the passengers exceeding the capacity will not be able to board the train and will remain at the original platform waiting for the next train. As shown in formula (18), the number of passengers on board at train number j at s is subject to the train capacity constraint. The excess passengers cannot board the train, and the destinations of the remaining passengers are distributed in proportion to the passenger flow composition.
[0072]
[0073] It can be seen from the above technical solutions that the method for optimizing driving adjustment in the recovery phase under uncertain interruption duration provided by the present invention has achieved the following beneficial effects: the uncertain interruption duration has a great influence on the driving adjustment in the interruption recovery phase, and the unknown initial state brings great difficulty to the optimization of driving adjustment in interruption recovery. Therefore, a two-stage driving adjustment optimization model for driving adjustment in the interruption recovery phase based on the rolling time domain control algorithm is designed. A two-stage step-by-step solution algorithm is designed. In the first stage, the range of decision variables is searched with the main goal to narrow the range of solutions. In the second stage, it continues to iterate and search based on the results of the first stage with a secondary goal. Compared with directly iterating the decision variables, this algorithm is more directional and has a fast solution speed, and can meet the requirements of the rolling time domain control algorithm for calculation speed. BRIEF DESCRIPTION OF THE DRAWINGS
[0074] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The accompanying drawings are only provided for reference and explanation and are not intended to limit the present invention.
[0075] Figure 1 A schematic diagram of the steps of a method for optimizing driving adjustment in a recovery phase under uncertain interruption duration according to the present invention;
[0076] Figure 2 It is a schematic diagram of a two-stage step-by-step solution model in the present invention;
[0077] Figure 3 A schematic diagram of a feasible solution that satisfies the train operation capacity in the first stage solution algorithm of the present invention;
[0078] Figure 4 It is a schematic diagram of the format of the queue to be searched sorted by time in the first stage solution algorithm of the present invention;
[0079] Figure 5 It is a schematic diagram of the reduction of the range of the better solution as the number of searches increases in the first stage solution algorithm of the present invention;
[0080] Figure 6 It is a flowchart of the second stage of the algorithm for solving the stop plan and average stop time iteration of the present invention;
[0081] Figure 7 It is a schematic diagram of optimization of interruption recovery driving adjustment under rolling time domain control in the present invention;
[0082] Figure 8 This is a flow chart for optimizing the interruption and resumption of driving adjustment in the present invention;
[0083] Fig. 9 This is a schematic diagram of various optimized windows for interruption recovery in the present invention, taking Nanjing Metro Line S1 as an example. DETAILED DESCRIPTION
[0084] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The accompanying drawings are only provided for reference and explanation and are not intended to limit the present invention.
[0085] The present invention considers a method for optimizing train operation adjustment in the recovery phase under uncertain interruption duration, and models the train operation adjustment, such as Figure 1 As shown, the following steps are included:
[0086] S1. Traffic adjustment optimization based on minimum interruption recovery time to reduce the impact on passenger travel;
[0087] S2, measure the impact of traffic adjustment optimization on passengers based on passenger travel time;
[0088] S3. Based on the data of S1 and S2, a traffic adjustment optimization model for the interruption recovery phase is established: using the stop plan and the average stop time as decision variables, a dual-objective traffic adjustment optimization model is established with the fastest interruption recovery time set as the primary objective and the minimum passenger travel time set as the secondary objective;
[0089] S4. Design a two-stage distributed solution algorithm to reduce the time of a single calculation of the rolling time domain algorithm and improve the accuracy. Use the rolling time domain algorithm to solve the dual-objective driving adjustment optimization model under uncertain interruption time and obtain the driving adjustment optimization plan.
[0090] like Figure 2-Figure 9 As shown,
[0091] S4 specifically includes the following steps:
[0092] S401. The way the rolling time domain algorithm controls errors comes from continuous single calculation optimization. In order to improve accuracy, it is necessary to reduce the time of single calculation. Therefore, this model does not directly use heuristic algorithms such as genetic algorithms in the solution process, but independently designs a two-stage step-by-step solution algorithm based on the model objectives and the characteristics of urban rail traffic adjustment problems. This model takes the interruption recovery time as the main goal, and the interruption recovery time is determined by {Tr n j}The biggest moment in decision making.
[0093] The first stage: According to the train operation constraints, search for the train interruption recovery time group ∑{Tr n j}, and find out the {Tr n j}, for the convenience of the following description, the interruption recovery time data set of each train found here is represented as {Tr n j} minTrIn this phase, a heuristic search algorithm based on the best-first strategy and the range of feasible solutions is designed and used.
[0094] The second stage: convert the first stage obtained {Tr n j} minTr As known quantities, they are added to the original model in the form of constraints, and the stop plan and average stop time are used as decision variables. The secondary goal of the model, the minimum passenger travel time, is used to iterate and solve the interruption recovery operation diagram. In this stage, a heuristic iterative algorithm with stop plan and average stop time as decision variables is designed and used. Figure 2 shown.
[0095] S402, the first stage solution steps are as follows, in order to obtain {Tr n j} minTr , the present invention designs a heuristic search algorithm based on the best priority strategy and feasible solution range:
[0096] (1) According to the train operation constraints, the shortest stop time operation diagram Min_P and the longest stop time operation diagram Max_P are generated based on the train positions in the line network. Figure 3 As shown in the figure, the red and blue segments are the running lines of a certain train in the running diagram Min_P and the running diagram Max_P respectively, and the dotted line segment is the running line of the planned running diagram. According to the interruption recovery constraint conditions, only the planned train departure node time between the red and blue running lines in the figure It may be the interruption and recovery moment of this train Tr n j ,like Figure 3 Shown in red circle.
[0097] (2) Search for the corresponding traceable nodes for each train. Sort the tracking nodes of each train in chronological order from early to late, and define them as standby in the queue format of {j: [(T, e)]}. Each data element can record the train number and bottom of the train in the planned operation diagram, the time of reaching the node, and the node location information, such as Figure 4 The {j: [(T, e)]} of each train is merged into ∑{j: [(T, e)]}. During the matching process, the queues of other trains are constrained by the matched trains.
[0098] (3) Searching for the train interruption recovery time group ∑{Tr n j}, find the {Tr corresponding to the fastest interruption recovery time n j} minTrFirst, compare the first item in the queue of each train number {j: [(T, e)]}, and select the train number j with the latest time and its first time As the first matching [j, p_j], according to the model resource constraints, when a planned train has been selected as the tracking target, other trains can no longer track the planned trains that use the same car bottom as this train. Considering that there is no train overtaking in urban rail, it can be concluded that the car bottom order is fixed. Therefore, when the tracking target of one train is determined, the tracking targets of other trains can also be inferred from the relative order relationship between their car bottoms, which can greatly reduce the number of model resource constraint judgments and reduce the computational complexity. The reason for selecting the train j with the latest time node to start matching first is that the interruption recovery time is determined by the latest "caught" train, and the tracking range of each train is its corresponding {j: [(T, e)]}. Because the earliest time that this train can track is the latest, the tracking time of this train is matched first, and other trains with looser tracking ranges are selected in the remaining range after this train is matched. The individual that meets the most stringent conditions first is also a common priority strategy in planning problems.
[0099] So far, we can get a set of {Tr n j}1, the corresponding interrupt recovery time is {Tr n j The latest time in}1 is used as the temporary fastest interruption recovery time and plan, named Tr 临时 and {Tr n j} 临时 However, whether this group has the fastest interruption recovery time {Tr n j} minTr It is necessary to continue searching and judging and proceed to the following step (4).
[0100] (4) The first moment that has been selected Delete from {j: [(T, e)]}, then repeat step (3) to select the latest first train number j and its first time Generate {Tr n j}2, the corresponding interrupt recovery time is Tr2. Compare the search results with the existing results. If Tr2<Tr 临时 , then the current solution is better, and the current result is used to replace Tr 临时 and {Tr n j} 临时 , proceed to the following step (5).
[0101] (5) Determine whether there is a better solution based on the range of solutions. Determine the second priority time Is it greater than the existing optimal solution Tr? 临时 If yes, the search ends. If no, there may be a better solution, and step (4) is repeated. Step (5) is to sort {j: [(T, e)]} from small to large time in step (2) and start matching {Tr n j}, the interrupt recovery time Tr must be greater than or equal to the first time In the process of searching, there must be a better solution, which requires Tr<Tr 临时 , so when searching for , it is impossible for there to be a result that satisfies the conditions of a better solution. At this time, it can be considered that there is no need to search, even if there are still solutions that have not been searched.
[0102] like Figure 5 As shown, the blue circle represents the searchable nodes (T, e) in {j: [(T, e)]}, which are arranged in chronological order. The calculated first item (T, e) is excluded in subsequent searches, and the interruption and recovery time of the solution must belong to [(T, e)], which is the green dot in the figure. Its time must be greater than or equal to the first node. As shown in the figure, in the n+1 search, the possible interruption and recovery time range is already later than the current best result, so there is no point in continuing the search, so it can be stopped at the nth time.
[0103] This algorithm is a heuristic search algorithm that combines best-first search and feasible solution range. Compared with the ergodic search algorithm, this algorithm has fewer iterations and faster speed, which meets the requirements of the rolling time domain algorithm in the model of the present invention for calculation speed.
[0104] S403, such as Figure 6 As shown in the figure, in the second stage, the starting and ending positions of the train are known, which is a problem of finding the intermediate motion process based on the time and position space difference. Therefore, a heuristic iterative algorithm with the stop plan and the average stop time as decision variables is designed:
[0105] (1) Generate the initial average stop time and stop plan. The initial average stop time is the minimum stop time constrained by the model, and the initial stop plan is to stop at every stop.
[0106] (2) Input the current stop time and stop plan into the train operation model for calculation to determine whether the train recovery time meets the constraint conditions. If not, adjust and iterate according to the following situations: 1. If the calculated result reaches the target ahead of time, increase the stop time and repeat the current step for iteration; 2. If it is later than the target time, reduce the average stop time and repeat the current step. If the reduced average stop time does not meet the stop time constraint, add a skip station and enter step (3). If the constraint conditions are met, output the driving adjustment plan that meets the constraint conditions to the set of plans to be selected {P}, and end the iteration of this stop plan. If all stop plans (including those returned in step (3)) are calculated, enter step (4).
[0107] (3) In the schemes with the same number of skipped stations, there are also combinations of different skipped station positions, that is, stop plans. Therefore, this step will generate a number of different stop plans according to the number of stops / skips of the train, and recombining these stop plans and the minimum stop time into several individuals to iterate step (2) respectively. Select the optimal skip plan.
[0108] (4) Calculate the solutions in the set of solutions to be selected {P} and output the solution with the shortest passenger travel time as the optimal interruption recovery driving adjustment operation diagram.
[0109] S404. Rolling time domain control is usually used for long-period optimization problems and optimization problems with uncertain factors. The driving adjustment problem with uncertain interruption duration studied in the present invention is an optimization problem with time-varying factors. The interruption recovery driving adjustment belongs to an open-loop optimization problem in a finite time domain and is embedded in the full optimization cycle as an optimization window.
[0110] like Figure 7 As shown, the present invention makes adaptive improvements to the rolling time domain control algorithm flow chart, and in the optimistic estimation end time T o A fixed prediction horizon t ahead b As the beginning of the whole optimization cycle, in order to make the optimization results of each optimization window return to the object and take effect, it is necessary to make a short-term prediction of the train operation status at the sampling time, so the prediction time domain t is set. b , by predicting a period of time t after the sampling moment b The train position is taken as the starting state for each interruption recovery adjustment. For each optimization window t a Both require the completion of the restoration of the train operation diagram as the end state.
[0111] As the algorithm flow Figure 8 As shown, the first optimization window is at time T o -t b As the sampling time T now, to predict the time T now +t b The train status information is used as the initial information of the train to input the interruption recovery operation adjustment optimization model. The optimization window lasts until the train is restored to the planned operation diagram, and the optimized operation adjustment plan P is output. After each optimization window calculation is completed, it is determined whether the interruption fault is eliminated. If not, the window t is rolled forward. r , in T now +t r The next optimization window starts from the time as the new sampling time. If the fault elimination interruption ends, the latest calculated adjustment plan P is used to implement the interruption recovery driving adjustment.
[0112] Under this rolling optimization control, the maximum error between the actual interruption duration and the latest driving adjustment optimization solution is the rolling time domain t r , assuming that the probability is evenly distributed, the theoretical average error is The expected value variance of the error is It can be seen that the rolling time domain t r The smaller it is, the less impact the uncertainty interruption will have on the optimization result, and the higher the accuracy. However, due to the limitation of program computing power, the rolling time domain t r Should be at least greater than the program calculation time t cp , in order to keep up with the changes in the online scene.
[0113] Considering the accuracy of model prediction, the prediction time domain t b The smaller the better. The longer the prediction time, the greater the error between the actual scene and the train position state predicted by the model. In order to make the calculated driving adjustment plan effective on the controlled object, it is necessary to ensure that the start time of the optimization plan is later than the time when the calculation is completed, that is, T now +t b ≥T now +t cp , it is easy to get the prediction time domain t b It should also be greater than or equal to the program calculation time t cp .
[0114] In summary, the window time settings in the rolling horizon control model under the research problem of the present invention satisfy the following formula:
[0115] t cp ≤t r
[0116] t cp ≤t b
[0117] Where, t cp Calculate the time taken by the program, t r is the rolling window, t b is the prediction window.
[0118] The model is verified using Nanjing Metro Line S1:
[0119] Nanjing Metro Line S1 uses 6-car B-type trains, with a single car load of 240 people and a single train load limit of 1,440 people. The passenger arrival rate used in the calculation example is based on the daily passenger flow of Line S1. According to the 20-year data, the average daily passenger flow is 90,000 people, the operating time is 16 hours, and the average passenger flow rate of the entire line is 1.56 people / second. Considering that the interruption time is the off-peak time period, the passenger flow arrival rate of the entire line is α = 1.5 people / second. Assume that the interruption time is 12:08 (43680) noon on a normal working day, the time in brackets is in seconds, and the interruption area occurs on the up track of Zhengfangzhong Road Station. The dispatcher estimates the end time of the interruption at 14:00, with an optimistic estimate of 13:40 (49200) and a pessimistic estimate of 14:20 (51600). During the interruption recovery phase, the driving adjustment takes the optimistic estimate time in advance of the single calculation time as the start time of the first optimization window. The optimization cycle is from 13:40 (49200) to the completion of the schedule recovery. After the first optimization window calculation is completed, a new optimization cycle starts with the current time until the fault is repaired. In the example, the possible fault exclusion range is set to [49200,51600], and the optimization during the interruption recovery phase will cover this range.
[0120] The default rolling window is equivalent to the program calculation time. In the trial calculation, it was found that the calculation amount of some windows was small, and the actual calculation time was less than 1 second. However, the difference of 1 second is too small for the scene change, and there is basically no change. In order to ensure a certain difference, for the window with program calculation time less than 5 seconds, the minimum rolling window is set to 5 seconds. The minimum stop time is set to 20 seconds, and the maximum stop time in the interruption recovery stage is 60 seconds. The platform waiting time penalty coefficient is set to 1 in the calculation of the present invention.
[0121] The calculation of the interruption recovery phase starts with the train status at 13:40 (49200) as the initial state to start the first optimization window. After the calculation of the first optimization window is completed, the window will be rolled back once. When the calculation time is less than the minimum rolling time, the next optimization window calculation is performed according to the minimum rolling time of 5 seconds. The calculation and rolling are repeated until the fault is eliminated and the interruption recovery begins. In the example, the present invention completely calculates the entire estimation period [49200, 51600], a total of 384 windows, and the rolling time between each window varies between 5 and 20 seconds. The calculation time of most solutions is within 5 seconds, and very few solutions take a long time due to the large complexity of the stop plan. It can be seen that the rolling time domain algorithm model can generate an interruption recovery operation diagram with an average error of no more than 10 seconds within the possible interval. Therefore, in the actual interruption scenario, when the uncertain interruption duration is determined, the model can provide a recovery plan that adapts to the current scenario under real-time calculation, achieving the research purpose.
[0122] See the diagram for optimization window and rolling window during calculation. Fig. 9 ,Each horizontal column in the histogram represents the optimization window of this calculation.,As the number of scrolling times increases, the starting point of the optimization window also moves forward with,time.
[0123] The fastest interruption recovery time searched in the optimization window with an interruption duration of 49200 is 14:01 (50430); in the example, the start time of the optimization window closest to the interruption duration of 49550 is 49553, and the fastest interruption recovery time searched by this solution is 50530; the optimization window is 49990, and the fastest interruption recovery time searched by this solution is 51680; the optimization window is 50150. The biggest feature of this window is that compared with earlier windows such as the 49990 window shown above, the fastest interruption recovery time searched is earlier, only 51580. Although this window starts recovery later, the fastest recovery time is earlier than the earlier solution. Due to the change in the order of the bottom of the car, the shortest interruption recovery time searched also changes. In some cases, early recovery is not as good as late recovery. It also shows that in this case, even if the fault has been eliminated, it is an optional strategy to let the train continue to run for a period of emergency operation and start interruption recovery at the appropriate position. The same situation also occurs in the optimization window 50920 and window 50990 during the rolling process. The interruption recovery time of the former is 52100, while that of the latter is 51950. The optimization window 51600 is the latest time in the estimated interruption duration and is also the last optimization window in this example. The fastest interruption recovery time found is 52940.
[0124] The rolling horizon control algorithm is effective for optimizing the operation adjustment in the interruption recovery phase with uncertain interruption duration, and six interruption durations are selected for display analysis in the results. The analysis results show that the position of the train at the beginning of recovery will affect the subsequent recovery efficiency.
Claims
1. A method for optimizing driving adjustment during the recovery phase under uncertain interruption duration, characterized in that Modeling train operation adjustments includes the following steps: S1. Traffic adjustment optimization based on minimum interruption recovery time to reduce the impact on passenger travel; S2, measure the impact of traffic adjustment optimization on passengers based on passenger travel time; S3. Based on the data of S1 and S2, a traffic adjustment optimization model for the interruption recovery phase is established: using the stop plan and the average stop time as decision variables, a dual-objective traffic adjustment optimization model is established with the fastest interruption recovery time as the primary objective and the minimum passenger travel time as the secondary objective. The secondary objective is used for selection only when the primary objective is equal; S4. Design a two-stage distributed solution algorithm to reduce the time of a single calculation of the rolling horizon algorithm, improve the accuracy, and use the rolling horizon algorithm to solve the dual-objective driving adjustment optimization model under uncertain interruption time.
2. The method for optimizing driving adjustment in the recovery phase under uncertain interruption duration according to claim 1 is characterized in that: S1 specifically includes the following steps: S101, interruption recovery phase end time Tr end The latest train number that "catches up" with the original daily schedule is determined by the train number that "catches up" with the original daily schedule, which is the end time of the interruption recovery phase, as follows: Tr end =Max{Tr n j } In the formula, Tr n j It is the time when train number j catches up with the original daily planned operation diagram at node n.
3. The method for optimizing driving adjustment in the recovery phase under uncertain interruption duration according to claim 1, characterized in that: S2 specifically includes the following steps: S201. Passenger travel time can be divided into platform waiting time and on-board time. Compared with on-board time, waiting on the platform is more easily perceived by passengers. Long-term platform detention will also cause passenger anxiety. Therefore, in this model, a penalty coefficient α is added to the platform waiting time. w To reflect the impact of platform waiting time on passengers' travel time perception, as follows: t p_total =α w t w_total +t c_total In the formula, t p_total is the total passenger travel time, t w_total is the platform waiting time, t c_total For the time in the car.
4. The method for optimizing driving adjustment in the recovery phase under uncertain interruption duration according to claim 1, characterized in that: S3 specifically includes the following steps: S301. Make assumptions based on the operational characteristics of urban rail transit systems and existing research; S302, taking the stop plan and the average stop time as decision variables; S303. In the interruption recovery phase, the main purpose of driving adjustment is to restore normal operation as soon as possible and reduce the impact of the interruption on passenger travel. This model selects the minimum recovery time and the minimum passenger travel time as dual optimization goals, hoping to find a driving adjustment plan with fast recovery and small impact on passengers. In general, the earlier the interruption is restored, the fewer passengers will be affected by the interruption. Therefore, the fastest interruption recovery time is set as the main goal, and the minimum passenger travel time is set as the secondary goal. The secondary goal is used for selection only when the main goal is not higher, as follows: MinZ=W*(Tr end -Tr0)+t p_total Where Tr0 is the first time to "catch up" with the original train operation diagram, and W is the main target coefficient, which is a very large positive integer.
5. The method for optimizing driving adjustment in the recovery phase under uncertain interruption duration according to claim 1, characterized in that: S4 specifically includes the following steps: S401. The way of controlling the error of the rolling time domain algorithm comes from the continuous single calculation optimization. In order to improve the accuracy, it is necessary to reduce the time of single calculation. According to the model objectives and the characteristics of urban rail traffic adjustment problems, a two-stage step-by-step solution algorithm is independently designed. The model takes the interruption recovery time as the main goal, and the interruption recovery time is determined by the interruption recovery time of each train {Tr n j }The biggest moment in decision; S402, in order to obtain {Tr n j } minTr , for the first stage, a heuristic search algorithm based on the best priority strategy and feasible solution range is designed; S403, the {Tr n j } minTr As known quantities, the starting and ending positions of the trains are known in the second stage, which is a problem of finding the intermediate motion process based on the time and position space difference. A heuristic iterative algorithm is designed with the armistice plan and the average stop time as decision variables; S404. Use the rolling horizon control algorithm to solve the driving adjustment in the recovery phase under the uncertain interruption duration, and execute steps S402 and S403 in each calculation window to obtain the optimal driving adjustment optimization solution in the interruption recovery phase.
6. The method for optimizing driving adjustment in the recovery phase under uncertain interruption duration according to claim 4 is characterized in that: S301 specifically includes the following steps: (1) When the train adjustment begins, the train on the non-normal track must first complete this section of the path and return to the normal track. For example, if a down train on a single-track two-way route happens to be on the shared up track, it must first pass through this section of the track and return to the down track. It cannot be adjusted as an up train. (2) A train is considered to have resumed operation only when its departure time "catches up" with the planned timetable. Operation adjustments during the recovery phase must not be completed until all trains resume the planned timetable. (3) Only trains that start running after resumption are allowed to use the skipping scheme. Trains that have already started running before resumption are not allowed to use the skipping scheme to prevent passengers from being unable to reach their destinations. (4) The change in interval running speed caused by skipping is not considered. In the model, skipping is only manifested as a stop time of 0. (5) It is assumed that the time it takes for a train to pass through the station section is fixed and known, and the error caused by driving operations is not considered. (6) For the same train, except for skipped stations, it is assumed that the stop time at all stations is the same. (7) Passengers enter the station at a fixed arrival rate. The OD of the arriving passengers at different stations is known, and passengers enter the platform directly without considering the passage time. (8) Passenger boarding is subject to train capacity restrictions. Passengers who are unable to board will continue to wait for the next available train. (9) Assume that the platform space is infinite and passengers entering the station are not constrained by the platform capacity. (10) Assuming that passengers get on and off the train instantly, in order to make the calculation clear, it is stipulated that passengers get off when the train enters the station and get on when the train leaves the station. (11) Automatic station-to-station blocking is used to control the train’s safe tracking distance.
7. The method for optimizing driving adjustment in the recovery phase under uncertain interruption duration according to claim 5, characterized in that: S404 specifically includes the following steps: Rolling time domain control is usually used for long-period optimization problems and optimization problems with uncertain factors. The driving adjustment problem with uncertain interruption duration studied in this invention is an optimization problem with time-varying factors. Interruption recovery driving adjustment is an open-loop optimization problem in a finite time domain and is embedded in the full optimization cycle as an optimization window. The present invention makes adaptive improvements to the rolling time domain control algorithm flow chart, and in the optimistic estimation end time T o A fixed prediction horizon t ahead b As the beginning of the whole optimization cycle, in order to make the optimization results of each optimization window return to the object and take effect, it is necessary to make a short-term prediction of the train operation status at the sampling time, so the prediction time domain t is set. b , by predicting a period of time t after the sampling moment b The train position is taken as the starting state for each interruption recovery adjustment. For each optimization window t a Both require the completion of the restoration of the train operation diagram as the end state. The first optimization window starts at time T o -t b As the sampling time T now , to predict the time T now +t b The train status information is used as the initial information of the train to input the interruption recovery operation adjustment optimization model. The optimization window lasts until the train is restored to the planned operation diagram, and the optimized operation adjustment plan P is output. After each optimization window calculation is completed, it is determined whether the interruption fault is eliminated. If not, the window t is rolled forward. r , in T now +t r The next optimization window starts from the time as the new sampling time. If the fault elimination interruption ends, the latest calculated adjustment plan P is used to implement the interruption recovery driving adjustment. Under this rolling optimization control, the maximum error between the actual interruption duration and the latest driving adjustment optimization solution is the rolling time domain t r , assuming that the probability is evenly distributed, the theoretical average error is The expected value variance of the error is It can be seen that the rolling time domain t r The smaller it is, the less impact the uncertainty interruption will have on the optimization result, and the higher the accuracy. However, due to the limitation of program computing power, the rolling time domain t r At least should be greater than the program calculation time t cp , in order to keep up with the changes in the online scene. Considering the accuracy of model prediction, the prediction time domain t b The smaller the better. The longer the prediction time, the greater the error between the actual scene and the train position state predicted by the model. In order to make the calculated driving adjustment plan effective on the controlled object, it is necessary to ensure that the start time of the optimization plan is later than the time when the calculation is completed, that is, T now +t b ≥T now +t cp , it is easy to get the prediction time domain t b It should also be greater than or equal to the program calculation time t cp .
