An adaptive method and system for multi-train scheduling parameters based on evolutionary optimization
By generating scheduling weight vectors through evolutionary optimization methods and combining them with the scheduling evaluation module to adjust train priorities and routes in real time, the flexibility and adaptability issues of existing railway scheduling systems in dynamic environments are solved, achieving efficient and safe train operation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- FUJIAN UNIV OF TECH
- Filing Date
- 2026-04-27
- Publication Date
- 2026-05-26
AI Technical Summary
The existing railway dispatching system lacks flexibility and adaptability when facing the dynamically changing train operation environment, resulting in increased manual intervention, high dispatching complexity, and difficulty in achieving safe and punctual operation.
An adaptive method for multi-train scheduling parameters based on evolutionary optimization is adopted. The scheduling weight vector is generated and updated through the algorithm optimization module, and real-time planning is carried out in combination with the scheduling evaluation module to form a self-learning closed loop, dynamically adjusting train priority and path to achieve adaptive scheduling.
It improves the adaptability and intelligence of the railway dispatching system, reduces manual intervention, enhances the safety and efficiency of train operation, and enables it to maintain good operating conditions in dynamic environments.
Smart Images
Figure CN122078469A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of train scheduling technology, and specifically relates to an adaptive method and system for multi-train scheduling parameters based on evolutionary optimization. Background Technology
[0002] With the continuous expansion of urban rail transit, intercity railways, and high-speed rail networks, train operation density is increasing year by year, significantly increasing the scheduling complexity of railway transportation systems. Modern railway scheduling systems are typically controlled centrally, combining information such as line status, train timetables, and section occupancy to comprehensively arrange train departure times and routes. At the operation control level, scheduling systems often employ methods based on fixed priorities or static rules, such as determining train passing order according to train type (passenger, freight), operating speed level, or planned departure time. Furthermore, some systems introduce optimization algorithms (such as shortest path search and spatiotemporal conflict detection) to calculate train routes to reduce delays and section conflicts.
[0003] However, these algorithms are often goal-oriented, focusing on fixed objectives such as shortest travel distance or minimum running time, lacking a comprehensive balance of factors such as line congestion, speed constraints, and fault recovery. In actual railway operations, the train operating environment is frequently affected by various uncertainties, such as temporary line maintenance, equipment failures, passenger flow fluctuations, and weather conditions. These dynamic changes place higher demands on the real-time performance and adaptability of the dispatching system. To ensure safe and punctual operation, dispatchers typically need to intervene manually to make temporary adjustments to train priorities or routes, thus increasing operational complexity and the dispatching burden. Summary of the Invention
[0004] This invention provides an adaptive method and system for multi-train scheduling parameters based on evolutionary optimization, aiming to solve at least one of the technical problems existing in the prior art.
[0005] The technical solution of this invention relates to an evolutionary optimization-based adaptive method for multi-train scheduling parameters. This method is applied to an evolutionary optimization-based adaptive system for multi-train scheduling parameters. The system includes an algorithm optimization module and a scheduling evaluation module connected sequentially. The evolutionary optimization-based adaptive method for multi-train scheduling parameters includes the following steps:
[0006] S100. The algorithm optimization module obtains basic parameters, generates and updates the parent matrix X and the scheduling weight vector w. The scheduling weight vector w aims at optimal scheduling and is used to characterize the weight ratio of multiple decision objectives in the train scheduling process.
[0007] S200. The algorithm optimization module evolves the scheduling weight vector w through iterative loops and sends the optimized scheduling weight vector w to the scheduling evaluation module.
[0008] S300, the scheduling evaluation module executes train operation planning according to the scheduling weight vector w, and determines the travel path, sequence and completion time of each train; the scheduling evaluation module calculates the comprehensive fitness value based on the train arrival rate, total operating cost and completion time operation results;
[0009] S400. After receiving the result packet returned by the scheduling evaluation module, the algorithm optimization module adjusts the scheduling weight vector w according to the comprehensive fitness value and performs result feedback and self-learning update.
[0010] Furthermore, step S100 includes:
[0011] S110, The algorithm optimization module obtains basic parameters;
[0012] S120. The algorithm optimization module randomly generates a parent matrix X(0), where each row of the parent matrix X(0) represents a candidate scheduling weight vector w.
[0013] S130. Based on simplex projection, ensure that all candidate scheduling weight vectors w are valid. The scheduling weight vector w includes three components. The three components represent the train distance priority weight values, respectively. Line congestion priority weight value and speed stability priority weight value The train distance priority weight value The aforementioned line congestion priority weight value and the speed stability priority weight value The sum is 1;
[0014] S140. Calculate the initial fitness value for each weight combination and find the best-performing individual. ;
[0015] S150. Establish an external memory bank to store the best individuals from each generation.
[0016] Furthermore, in step S110,
[0017] The basic parameters include whether to use a fixed map instance enable flag, the fixed instance file path for loading the same line and train layout, the dynamic neighborhood parameter DNPS for controlling whether the weights participate in priority sorting, the single evaluation time limit parameter for constraining the scheduling runtime, the number of trains participating in the calculation in each round of scheduling, and the convergence judgment threshold for controlling the termination of optimization.
[0018] When the system is in dynamic priority sorting mode, that is, when the dynamic neighborhood parameter is 1, the train scheduling priority changes dynamically according to the scheduling weight vector w.
[0019] This allows different types of trains to obtain adaptive driving priority based on the actual conditions of the line.
[0020] Furthermore, in step S200, each iteration of the scheduling weight vector w includes the following steps:
[0021] S210. Select the optimal individual g(t-1) from the initialized parent matrix or the parent matrix of the previous iteration as the base point for the current search.
[0022] S220. Randomly rearrange the parent matrix twice to obtain P1(X) and P2(X);
[0023] S230. Calculate the difference term Δ used to reflect directional differences within the group.
[0024] ;
[0025] S240. Generate a binary mask matrix M(t-1) to control the dimension that needs to be updated, and adaptively generate a scaling matrix Rc(t-1) to control the update magnitude based on the historical step size.
[0026] S250. Based on the best individual from the previous iteration, generate new candidate vectors. :
[0027] ,
[0028] The results were then normalized to ;
[0029] S260. If the population stagnates, select an early excellent scheduling weight vector w from the external memory bank and replace the individual vectors in the current parent matrix X that are at the bottom of the preset proportion according to the comprehensive fitness value, or the individual vectors whose spatial distance between them is less than a preset distance threshold, in order to re-inject diversity.
[0030] S270. The new scheduling weight vector w and the running configuration are uniformly encapsulated into a parameter package and sent to the scheduling evaluation module for actual scheduling verification.
[0031] Furthermore, in step S270, the parameter package includes:
[0032] The system includes a scheduling weight vector w used for train sequencing and scheduling priority calculation, a dynamic neighborhood parameter used to control whether train priorities are dynamically adjusted, a single evaluation time limit parameter, the number of trains participating in the calculation in each round of scheduling, a convergence judgment threshold, a priority strategy number and neighborhood strategy used to control the way individuals are generated internally in the LNS algorithm, and a flag used to determine whether replanning of failed trains is allowed.
[0033] Furthermore, step S300 includes:
[0034] S310. If the environment variable specifies a fixed map instance, then the preset map file is read to load the railway network structure, train starting position, speed level and target position.
[0035] S320. If no fixed map instance is specified, the railway network generation mechanism will automatically construct the line topology and generate train operation task data based on the topology. At the same time, the corresponding fault occurrence probability and duration will be set for the train according to the fault parameters. The RailEnv environment object will be called to construct a complete scheduling environment and set the maximum number of running steps to realize the environment initialization operation.
[0036] S330: Receive and parse the encapsulated parameter packet. The scheduling evaluation module determines the scheduling priority of each train based on the allocation of the scheduling weight vector w. If the congestion weight is high, high-density sections are avoided. If the speed weight is high, express trains are given priority.
[0037] S340, the scheduling evaluation module performs scheduling based on the conflict search path planning module or the dynamic replanning module. In each round of scheduling, each train participating in the calculation performs path planning within the single evaluation time limit parameter. If a conflict or delay occurs, replanning is triggered.
[0038] S350, the scheduling evaluation module records the number of successfully arriving trains, the total number of running steps (soc) for all trains, the arrival time (mk) of the slowest train, and the actual running time after executing the scheduling.
[0039] The S360 scheduling evaluation module calculates the comprehensive fitness value and generates a result package, which is then sent to the algorithm optimization module.
[0040] Furthermore, in step S360, the overall fitness value is calculated as follows:
[0041] ,
[0042] It includes three components, each representing a train distance priority weight value. Line congestion priority weight value and speed stability priority weight value The train distance priority weight value The aforementioned line congestion priority weight value and the speed stability priority weight value The sum is 1;
[0043] The non-arrival penalty is defined as the product of the number of trains that failed to reach the target location and a preset penalty constant; the number of trains that failed to reach the target location is the total number of trains scheduled minus the number of trains that successfully arrived.
[0044] Furthermore, step S400 includes:
[0045] S410, Read the overall fitness value returned by the scheduling evaluation module;
[0046] S420. If the overall fitness value is better than the current best, then update the global best individual g(t); otherwise, maintain the old value or store the overall fitness value in the external memory.
[0047] S430: Update the binary mask matrix and scaling matrix to achieve adaptive adjustment;
[0048] S440: Generate the next generation of candidate population and resend the parameter packet to the scheduling evaluation module.
[0049] Furthermore, this invention proposes an evolutionary optimization-based adaptive system for multi-train scheduling parameters to implement the aforementioned evolutionary optimization-based adaptive method for multi-train scheduling parameters. The evolutionary optimization-based adaptive system for multi-train scheduling parameters includes:
[0050] The algorithm optimization module is used to generate, update and distribute the scheduling weight vector w. The algorithm optimization module uses a heuristic search strategy to generate candidate scheduling weight vector w, and gradually improves train scheduling performance by combining random perturbation and local optimization. After each iteration, the algorithm optimization module records the scheduling weight vector w, the comprehensive fitness value and the corresponding scheduling scheme, and performs weight perturbation and local fine-tuning based on the historical data so as to gradually converge to a better scheduling scheme in multiple iterations.
[0051] The scheduling evaluation module is used to execute train operation planning and calculate the comprehensive fitness value. The algorithm optimization module and the scheduling evaluation module pass the scheduling weight vector w parameter through a unified interface or environment variables to realize fast communication and parameter isolation between different modules. The scheduling evaluation module can automatically select different scheduling frameworks for evaluation according to the system operation status, and use train operation efficiency and overall completion time as the main optimization criteria.
[0052] Furthermore, the present invention also proposes a computer-readable storage medium storing program instructions thereon, which, when executed by a processor, implement the aforementioned adaptive method for multi-train scheduling parameters based on evolutionary optimization. Attached Figure Description
[0053] Figure 1 This is a flowchart of an adaptive method for multi-train scheduling parameters based on evolutionary optimization.
[0054] Figure 2 This is a flowchart illustrating an adaptive method for multi-train scheduling parameters based on evolutionary optimization. Detailed Implementation
[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0056] The following will provide a clear and complete description of the concept, specific structure, and technical effects of the present invention in conjunction with the embodiments and accompanying drawings, so as to fully understand the purpose, solution, and effects of the present invention.
[0057] It should be noted that, unless otherwise specified, when a feature is referred to as "fixed" or "connected" to another feature, it can be directly fixed or connected to the other feature, or indirectly fixed or connected to the other feature. The singular forms "a," "described," and "the" used herein are also intended to include the plural forms, unless the context clearly indicates otherwise. Furthermore, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing particular embodiments only and not for limiting the invention. The term "and / or" as used herein includes any combination of one or more of the associated listed items.
[0058] It should be understood that although the terms first, second, third, etc., may be used in this disclosure to describe various elements, these elements should not be limited to these terms. These terms are only used to distinguish elements of the same type from each other. For example, a first element may also be referred to as a second element without departing from the scope of this disclosure, and similarly, a second element may also be referred to as a first element. Any and all instances or exemplary language (“e.g.,” “such as,” etc.) used herein are intended only to better illustrate embodiments of the invention and, unless otherwise required, do not impose a limitation on the scope of the invention. Furthermore, the industry term “pose” as used herein refers to the position and orientation of an element relative to a spatial coordinate system.
[0059] Reference Figure 1 and Figure 2 This invention provides an evolution-optimized method for adaptive multi-train scheduling parameters. This method is applied to an evolution-optimized multi-train scheduling parameter adaptive system, which includes an algorithm optimization module and a scheduling evaluation module connected sequentially. The evolution-optimized multi-train scheduling parameter adaptive method includes the following steps:
[0060] S100. The algorithm optimization module obtains basic parameters, generates and updates the parent matrix X and the scheduling weight vector w. The scheduling weight vector w aims at optimal scheduling and is used to characterize the weight ratio of multiple decision objectives in the train scheduling process.
[0061] S200. The algorithm optimization module evolves the scheduling weight vector w through iterative loops and sends the optimized scheduling weight vector w to the scheduling evaluation module.
[0062] S300, the scheduling evaluation module executes train operation planning according to the scheduling weight vector w, and determines the travel path, sequence and completion time of each train; the scheduling evaluation module calculates the comprehensive fitness value based on the train arrival rate, total operating cost and completion time operation results;
[0063] S400. After receiving the result packet returned by the scheduling evaluation module, the algorithm optimization module adjusts the scheduling weight vector w according to the comprehensive fitness value and performs result feedback and self-learning update.
[0064] Compared with existing technologies, the present invention has the following characteristics:
[0065] The aforementioned evolution-optimized adaptive method for multi-train scheduling parameters involves an algorithm optimization module generating candidate weights and sending parameter packets, a scheduling evaluation module performing scheduling based on the parameters, a scheduling evaluation module returning a comprehensive fitness value and operational indicators, and an algorithm optimization module updating weights and matrices using feedback. This process is repeated to form a self-learning optimization loop, ultimately allowing the system to automatically find the optimal three-dimensional weight ratio. This enables dynamic priority scheduling of trains on complex lines, achieving a more balanced and efficient operational effect.
[0066] Specifically, the existing technology has the following drawbacks:
[0067] (1) Fixed priority and lack of flexible adjustment capability: Existing systems mostly adopt fixed train priority sorting strategies. When there is a running conflict or line congestion, the system cannot dynamically adjust the priority according to the real-time situation, which can easily cause trains to wait for each other and reduce traffic efficiency.
[0068] (2) Single scheduling objective, neglecting multi-indicator collaborative optimization: Traditional scheduling algorithms often only aim at the shortest travel distance or the minimum delay, without fully considering factors such as line occupancy, speed differences, and operational safety margins. This single objective orientation leads to some trains having the shortest path, but the overall network operating efficiency decreases, resulting in the problem of "local optima, global inefficiency".
[0069] (3) Delayed response to abnormal and sudden situations: When trains malfunction, lines are temporarily closed, or passenger flow surges, the existing dispatching system is unable to automatically generate new and reasonable operation plans in a short time. Usually, it is necessary to manually rearrange the train sequence and routes, which results in slow response speed and easy chain delays.
[0070] (4) High reliance on manual labor and insufficient level of intelligence: The scheduling process relies on the experience and judgment of dispatchers and manual intervention, resulting in a low degree of automation. As the scale of the railway network expands, manual intervention not only increases the workload but also limits the scalability and sustainable operation of the scheduling system.
[0071] (5) Lack of self-learning and optimization mechanisms: Most existing scheduling systems are static rule systems, which cannot automatically adjust parameters or optimize strategies based on historical operating results. The system cannot adaptively adjust scheduling weights according to different operating conditions (such as high-density sections and different train types), making it difficult to achieve continuous performance improvement under long-term operation.
[0072] Specifically, the present invention achieves the following technical effects:
[0073] (1) A two-way closed loop between the algorithm and the scheduling system is realized. In traditional systems, the algorithm is only responsible for calculating the result once, and the scheduling system is passively executed, with no feedback relationship between the two. This invention sets up a dual-module structure of "algorithm optimization end + scheduling evaluation end", which allows the algorithm to automatically adjust its own parameters according to the scheduling result. In this way, the system can continuously learn and correct the weights, forming a self-learning closed loop of "running - feedback - re-optimization", and realizing continuous improvement in long-term operation.
[0074] (2) It has the ability to automatically adjust weights without manual intervention. In the past, the system required manual resetting of parameters or priorities under different lines and different train densities. The algorithm optimization end of this invention will automatically generate and update the three-dimensional scheduling weight vector w (distance, congestion, speed), and automatically correct it according to the feedback results from the evaluation end. The dispatcher no longer needs to frequently adjust parameters manually, and the system can maintain good operating efficiency and safety under different scenarios.
[0075] (3) Supports dynamic priority strategy and can adapt to the operating status in real time. Traditional scheduling mostly adopts a fixed priority order, which cannot be adjusted with changes in line status. This invention uses the "Dynamic Neighborhood Parameters (DNPS)" mechanism to enable train running priorities to be updated in real time with changes in weight. When there is local congestion on the line or individual trains are delayed, the system can adjust the running order in an instant, effectively reducing mutual waiting and line congestion.
[0076] (4) Balancing global efficiency with local flexibility. This invention combines two frameworks: LNS (Large Neighborhood Search) and LNS+replan (Local Replanning), enabling rapid repair of local conflict areas while simultaneously optimizing the global scope. This design maintains overall operational efficiency and allows for rapid response in unexpected situations, significantly improving the robustness and stability of the system.
[0077] (5) It features a scalable and portable structural design. The optimization and evaluation ends adopt a modular architecture with clear interfaces, allowing for independent upgrades or replacements of algorithm modules without affecting other parts. This enables the system to easily integrate other optimization algorithms, scheduling strategies, or actual line data in the later stages, greatly improving the system's maintainability and scalability.
[0078] In summary, this invention establishes a dynamic closed loop between the algorithm optimization end and the scheduling evaluation end, thereby realizing the transformation of the train scheduling system from "fixed rules" to "adaptive learning," effectively improving the safety, efficiency, and scheduling intelligence level of train operation.
[0079] Specifically, referring to Table 1, the algorithm optimization module includes the following parameter descriptions:
[0080] Table 1:
[0081] Referring to Table 2, the scheduling evaluation module includes the following parameter descriptions:
[0082] Table 2:
[0083] Furthermore, refer to Figure 1 and Figure 2 Step S100 includes:
[0084] S110, The algorithm optimization module obtains basic parameters;
[0085] S120. The algorithm optimization module randomly generates a parent matrix X(0), where each row of the parent matrix X(0) represents a candidate scheduling weight vector w.
[0086] S130. Based on simplex projection, ensure that all candidate scheduling weight vectors w are valid. The scheduling weight vector w includes three components. The three components represent the train distance priority weight values, respectively. Line congestion priority weight value and speed stability priority weight value The train distance priority weight value The aforementioned line congestion priority weight value and the speed stability priority weight value The sum is 1;
[0087] S140. Calculate the initial fitness value for each weight combination and find the best-performing individual. ;
[0088] S150. Establish an external memory bank to store the best individuals from each generation.
[0089] In some embodiments, in step S130, all candidate scheduling weight vectors w are normalized so that the weight values of the normalized scheduling weight vector w are non-negative and their sum is 1.
[0090] In step S130, the train distance priority weight value Line congestion priority weight value and speed stability priority weight value It is used for both priority ranking and comprehensive fitness value evaluation coefficient.
[0091] Furthermore, refer to Figure 1 and Figure 2 In step S110,
[0092] The basic parameters include whether to use a fixed map instance enable flag, a fixed instance file path for loading the same line and train layout, a dynamic neighborhood parameter for controlling whether the weights participate in priority sorting, a single evaluation time limit parameter for constraining the scheduling runtime, the number of trains participating in the calculation in each round of scheduling, and a convergence judgment threshold for controlling the termination of optimization.
[0093] When the system is in dynamic priority sorting mode, that is, when the dynamic neighborhood parameter is 1, the train scheduling priority changes dynamically according to the scheduling weight vector w.
[0094] This allows different types of trains to obtain adaptive driving priority based on the actual conditions of the line.
[0095] Furthermore, refer to Figure 1 and Figure 2 In step S200, each iteration of the scheduling weight vector w includes the following steps:
[0096] S210. Select the optimal individual g(t-1) from the initialized parent matrix or the parent matrix of the previous iteration as the base point for the current search.
[0097] S220. Randomly rearrange the parent matrix twice to obtain P1(X) and P2(X);
[0098] S230. Calculate the difference term Δ used to reflect directional differences within the group.
[0099] ;
[0100] S240. Generate a binary mask matrix M(t-1) to control the dimension that needs to be updated, and adaptively generate a scaling matrix Rc(t-1) to control the update magnitude based on the historical step size.
[0101] S250. Based on the best individual from the previous iteration, generate new candidate vectors. :
[0102] ,
[0103] The results were then normalized to ;
[0104] S260. If the population stagnates, select an early excellent scheduling weight vector w from the external memory bank and replace the individual vectors in the current parent matrix X that are at the bottom of the preset proportion according to the comprehensive fitness value, or the individual vectors whose spatial distance between them is less than a preset distance threshold, in order to re-inject diversity.
[0105] S270. The new scheduling weight vector w and the running configuration are uniformly encapsulated into a parameter package and sent to the scheduling evaluation module for actual scheduling verification.
[0106] In step S260, if the population stagnates, an early excellent scheduling weight vector w is selected from the external memory bank to replace the individual vectors in the current parent matrix X that are ranked at the end of a preset proportion according to the comprehensive fitness value, or the individual vectors whose spatial distance between them is less than a preset distance threshold, so as to re-inject diversity.
[0107] Furthermore, refer to Figure 1 and Figure 2 In step S270, the parameter package includes:
[0108] The system includes a scheduling weight vector w used for train sequencing and scheduling priority calculation, a dynamic neighborhood parameter used to control whether train priorities are dynamically adjusted, a single evaluation time limit parameter, the number of trains participating in the calculation in each round of scheduling, a convergence judgment threshold, a priority strategy number and neighborhood strategy used to control the way individuals are generated internally in the LNS algorithm, and a flag used to determine whether replanning of failed trains is allowed.
[0109] Furthermore, refer to Figure 1 and Figure 2 Step S300 includes:
[0110] S310. If the environment variable specifies a fixed map instance, then the preset map file is read to load the railway network structure, train starting position, speed level and target position.
[0111] S320. If no fixed map instance is specified, the railway network generation mechanism will automatically construct the line topology and generate train operation task data based on the topology. At the same time, the corresponding fault occurrence probability and duration will be set for the train according to the fault parameters. The RailEnv environment object will be called to construct a complete scheduling environment and set the maximum number of running steps to realize the environment initialization operation.
[0112] S330: Receive and parse the encapsulated parameter packet. The scheduling evaluation module determines the scheduling priority of each train based on the allocation of the scheduling weight vector w. If the distance weight is higher, trains with shorter distances will be scheduled first. If the congestion weight is higher, high-density sections will be avoided. If the speed weight is higher, express trains will be given priority.
[0113] S340, the scheduling evaluation module performs scheduling based on the conflict search path planning module or the dynamic replanning module. In each round of scheduling, each train participating in the calculation performs path planning within the single evaluation time limit parameter. If a conflict or delay occurs, replanning is triggered.
[0114] S350, the scheduling evaluation module records the number of successfully arriving trains, the total number of running steps (soc) for all trains, the arrival time (mk) of the slowest train, and the actual running time after executing the scheduling.
[0115] The S360 scheduling evaluation module calculates the comprehensive fitness value and generates a result package, which is then sent to the algorithm optimization module.
[0116] This process is repeated until the termination condition (convergence or timeout) is met. The optimal weights are then obtained.
[0117] This vector represents the most balanced and efficient combination of scheduling weights.
[0118] Furthermore, refer to Figure 1 and Figure 2 In step S360, the overall fitness value is calculated as follows:
[0119] ,
[0120] It includes three components, each representing a train distance priority weight value. Line congestion priority weight value and speed stability priority weight value The train distance priority weight value The aforementioned line congestion priority weight value and the speed stability priority weight value The sum is 1;
[0121] The non-arrival penalty is defined as the product of the number of trains that failed to reach the target location and a preset penalty constant; the number of trains that failed to reach the target location is the total number of trains scheduled minus the number of trains that successfully arrived.
[0122] The final evaluation module generates a result package:
[0123] {
[0124] "f": 1001607.4,
[0125] "sum_of_costs": 423,
[0126] "makespan": 59,
[0127] "arrived": 10,
[0128] "total_agents": 10,
[0129] "runtime_sec": 1.3
[0130] }
[0131] The result package is returned to the algorithm optimization module as a feedback signal for the next round of optimization.
[0132] Furthermore, refer to Figure 1 and Figure 2 Step S400 includes:
[0133] S410, Read the overall fitness value returned by the scheduling evaluation module;
[0134] S420. If the overall fitness value is better than the current best, then update the global best individual g(t); otherwise, maintain the old value or store the overall fitness value in the external memory.
[0135] S430: Update the binary mask matrix and scaling matrix to achieve adaptive adjustment;
[0136] S440: Generate the next generation of candidate population and resend the parameter packet to the scheduling evaluation module.
[0137] Furthermore, this invention proposes an evolutionary optimization-based adaptive system for multi-train scheduling parameters to implement the aforementioned evolutionary optimization-based adaptive method for multi-train scheduling parameters. The evolutionary optimization-based adaptive system for multi-train scheduling parameters includes:
[0138] The algorithm optimization module is used to generate, update and distribute the scheduling weight vector w. The algorithm optimization module uses a heuristic search strategy to generate candidate scheduling weight vector w, and gradually improves train scheduling performance by combining random perturbation and local optimization. After each iteration, the algorithm optimization module records the scheduling weight vector w, the comprehensive fitness value and the corresponding scheduling scheme, and performs weight perturbation and local fine-tuning based on the historical data so as to gradually converge to a better scheduling scheme in multiple iterations.
[0139] The scheduling evaluation module is used to execute train operation planning and calculate the comprehensive fitness value. The algorithm optimization module and the scheduling evaluation module pass the scheduling weight vector w parameter through a unified interface or environment variables to realize fast communication and parameter isolation between different modules. The scheduling evaluation module can automatically select different scheduling frameworks for evaluation according to the system operation status, and use train operation efficiency and overall completion time as the main optimization criteria.
[0140] In some embodiments, the system further includes a data interaction module and a storage module. The data interaction module is used to transmit the scheduling weight vector w and feedback information between the algorithm optimization module and the scheduling evaluation module. The storage module is used to record the historical weights, running data and final optimization results generated during the optimization process.
[0141] Furthermore, the present invention also proposes a computer-readable storage medium storing program instructions thereon, which, when executed by a processor, implement the aforementioned adaptive method for multi-train scheduling parameters based on evolutionary optimization.
[0142] It should be understood that the method steps in the embodiments of the present invention can be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium. The method can use standard programming techniques. Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, if necessary, the program can be implemented in assembly or machine language. In any case, the language can be a compiled or interpreted language. Furthermore, for this purpose, the program can run on a programmed application-specific integrated circuit (ASIC).
[0143] Furthermore, the procedures described herein may be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by the context. The procedures described herein (or variations and / or combinations thereof) may be executed under the control of one or more computer systems configured with executable instructions, and may be implemented by hardware or a combination thereof as code (e.g., executable instructions, one or more computer programs, or one or more applications) that commonly executes on one or more processors. The computer program comprises a plurality of instructions executable by one or more processors.
[0144] Furthermore, the method can be implemented operatively connected to any suitable type of computing platform, including but not limited to personal computers, minicomputers, mainframes, workstations, networked or distributed computing environments, standalone or integrated computer platforms, or communicating with charged particle tools or other imaging systems, etc. Aspects of the invention can be implemented as machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, optical read and / or write storage medium, RAM, ROM, etc., such that it is readable by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the processes described herein. Furthermore, the machine-readable code, or portions thereof, can be transmitted via wired or wireless networks. The invention described herein includes these and other different types of non-transitory computer-readable storage media when such media comprises instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor. When programmed according to the methods and techniques described in the invention, the invention may also include the computer itself.
[0145] A computer program can be applied to input data to perform the functions described herein, thereby transforming the input data to generate output data stored in non-volatile memory. The output information can also be applied to one or more output devices, such as a display. In a preferred embodiment of the invention, the transformed data represents physical and tangible objects, including specific visual depictions of physical and tangible objects generated on the display.
[0146] The above description is merely a preferred embodiment of the present invention. The present invention is not limited to the above-described embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention, as long as they achieve the technical effects of the present invention by the same means, should be included within the scope of protection of the present invention. Within the scope of protection of the present invention, the technical solutions and / or implementation methods can have various modifications and variations.
Claims
1. An adaptive method for multi-train scheduling parameters based on evolutionary optimization, characterized in that, The evolution-optimized multi-train scheduling parameter adaptive method is applied to an evolution-optimized multi-train scheduling parameter adaptive system, which includes an algorithm optimization module and a scheduling evaluation module connected in sequence. The evolution-optimized multi-train scheduling parameter adaptive method includes the following steps: S100. The algorithm optimization module obtains basic parameters, generates and updates the parent matrix X and the scheduling weight vector w. The scheduling weight vector w aims at optimal scheduling and is used to characterize the weight ratio of multiple decision objectives in the train scheduling process. S200. The algorithm optimization module evolves the scheduling weight vector w through iterative loops and sends the optimized scheduling weight vector w to the scheduling evaluation module. S300, the scheduling evaluation module executes train operation planning according to the scheduling weight vector w, and determines the travel path, sequence and completion time of each train; the scheduling evaluation module calculates the comprehensive fitness value based on the train arrival rate, total operating cost and completion time operation results; S400. After receiving the result packet returned by the scheduling evaluation module, the algorithm optimization module adjusts the scheduling weight vector w according to the comprehensive fitness value and performs result feedback and self-learning update.
2. The adaptive method for multi-train scheduling parameters based on evolutionary optimization according to claim 1, characterized in that, Step S100 includes: S110, The algorithm optimization module obtains basic parameters; S120. The algorithm optimization module randomly generates a parent matrix, where each row of the parent matrix represents a candidate scheduling weight vector w. S130. Based on simplex projection, ensure that all candidate scheduling weight vectors w are valid. The scheduling weight vector w includes three components. The three components represent the train distance priority weight value, the line congestion priority weight value, and the speed stability priority weight value, respectively, and the sum of the train distance priority weight value, the line congestion priority weight value, and the speed stability priority weight value is 1. S140. Calculate the initial fitness value for each weight combination and find the best-performing individual. ; S150. Establish an external memory bank to store the best individuals from each generation.
3. The adaptive method for multi-train scheduling parameters based on evolutionary optimization according to claim 2, characterized in that, In step S110, The basic parameters include whether to use a fixed map instance enable flag, a fixed instance file path for loading the same line and train layout, a dynamic neighborhood parameter for controlling whether the weights participate in priority sorting, a single evaluation time limit parameter for constraining the scheduling runtime, the number of trains participating in the calculation in each round of scheduling, and a convergence judgment threshold for controlling the termination of optimization. When the system is in dynamic priority sorting mode, that is, when the dynamic neighborhood parameter is 1, the train scheduling priority changes dynamically according to the scheduling weight vector w.
4. The adaptive method for multi-train scheduling parameters based on evolutionary optimization according to claim 1, characterized in that, In step S200, each iteration of the scheduling weight vector w includes the following steps: S210. In the t-th iteration, select the best individual g(t-1) of the (t-1)-th generation as the base point for the current search; where t≥1, when t=1, select from the initialized parent matrix, and when t>1, select from the parent matrix updated in the previous iteration. S220. Randomly rearrange the parent matrix twice to obtain P1(X) and P2(X); S230. Calculate the difference term Δ used to reflect directional differences within the group. ； S240. Generate a binary mask matrix M(t-1) to control the dimension that needs to be updated, and adaptively generate a scaling matrix Rc(t-1) to control the update magnitude based on the historical step size. S250. Based on the best individual from the previous iteration, generate new candidate vectors. : ， The results were then normalized to ; S260. If the population stagnates, select an early excellent scheduling weight vector w from the external memory bank and replace the individual vectors in the current parent matrix X that are at the end of the preset proportion according to fitness value, or the individual vectors whose spatial distance between them is less than a preset distance threshold, in order to re-inject diversity. S270. The new scheduling weight vector w and the running configuration are uniformly encapsulated into a parameter package and sent to the scheduling evaluation module for actual scheduling verification.
5. The adaptive method for multi-train scheduling parameters based on evolutionary optimization according to claim 4, characterized in that, In step S270, the parameter package includes: The system includes a scheduling weight vector w used for train sequencing and scheduling priority calculation, a dynamic neighborhood parameter used to control whether train priorities are dynamically adjusted, a single evaluation time limit parameter, the number of trains participating in the calculation in each round of scheduling, a convergence judgment threshold, a priority strategy number and neighborhood strategy used to control the way individuals are generated internally in the LNS algorithm, and a flag used to determine whether replanning of failed trains is allowed.
6. The adaptive method for multi-train scheduling parameters based on evolutionary optimization according to claim 1, characterized in that, Step S300 includes: S310. If the environment variable specifies a fixed map instance, then the preset map file is read to load the railway network structure, train starting position, speed level and target position. S320. If no fixed map instance is specified, the railway network generation mechanism will automatically construct the line topology and generate train operation task data based on the topology. At the same time, the corresponding fault occurrence probability and duration will be set for the train according to the fault parameters. The RailEnv environment object will be called to construct a complete scheduling environment and set the maximum number of running steps to realize the environment initialization operation. S330. Receive and parse the encapsulated parameter packet. The scheduling evaluation module performs a comprehensive weighted calculation based on the specific values of each weight component in the scheduling weight vector w, combined with the expected travel distance of each train, the congestion density of the planned route, and the train speed attributes, to obtain the scheduling priority score of each train and determine the priority passage order of the trains accordingly. S340, the scheduling evaluation module performs scheduling based on the conflict search path planning module or the dynamic replanning module. In each round of scheduling, each train participating in the calculation performs path planning within the single evaluation time limit. If a conflict or delay occurs, replanning is triggered. S350, the scheduling evaluation module records the number of successfully arriving trains, the total number of running steps (soc) for all trains, the arrival time (mk) of the slowest train, and the actual running time after executing the scheduling. The S360 scheduling evaluation module calculates the comprehensive fitness value and generates a result package, which is then sent to the algorithm optimization module.
7. The adaptive method for multi-train scheduling parameters based on evolutionary optimization according to claim 6, characterized in that, In step S360, the overall fitness value is calculated as follows: ， It includes three components, each representing a train distance priority weight value. Line congestion priority weight value and speed stability priority weight value The train distance priority weight value The aforementioned line congestion priority weight value and the speed stability priority weight value The sum is 1; The non-arrival penalty is defined as the product of the number of trains that failed to reach the target location and a preset penalty constant; the number of trains that failed to reach the target location is the total number of trains scheduled minus the number of trains that successfully arrived.
8. The adaptive method for multi-train scheduling parameters based on evolutionary optimization according to claim 1, characterized in that, Step S400 includes: S410, Read the overall fitness value returned by the scheduling evaluation module; S420. At the current iteration number t, compare the comprehensive fitness value of the current candidate scheduling weight vector w with the comprehensive fitness value of the previous generation's global best individual g(t-1); if the comprehensive fitness value is better than the comprehensive fitness value of the previous generation's global best individual, then update the candidate scheduling weight vector w to the current generation's global best individual g(t); otherwise, let the current generation's global best individual g(t) maintain the previous generation's value g(t-1), and store the candidate scheduling weight vector w and its comprehensive fitness value in the external memory bank; S430: Update the binary mask matrix and scaling matrix to achieve adaptive adjustment; S440: Generate the next generation of candidate population and resend the parameter packet to the scheduling evaluation module.
9. An evolution-optimized multi-train scheduling parameter adaptive system, used to implement the evolution-optimized multi-train scheduling parameter adaptive method as described in any one of claims 1 to 8, characterized in that, The aforementioned evolution-optimized multi-train scheduling parameter adaptive system includes: The algorithm optimization module generates, updates, and distributes the scheduling weight vector w. It employs a heuristic search strategy to generate candidate scheduling weight vectors w and gradually improves train scheduling performance through a combination of random perturbation and local optimization. After each iteration, the module records the scheduling weight vector w, the overall fitness value, and the corresponding scheduling scheme as historical data. Based on this historical data, it performs weight perturbation and local fine-tuning to gradually converge to a better scheduling scheme over multiple iterations. The scheduling evaluation module executes train operation planning and calculates the overall fitness value. The algorithm optimization module and the scheduling evaluation module communicate via a unified interface or environment variables to transfer the scheduling weight vector w parameter, enabling rapid communication and parameter isolation between different modules. The scheduling evaluation module can automatically select different scheduling frameworks for evaluation based on the system's operating status, using train operation efficiency and overall completion time as the primary optimization criteria.
10. A computer-readable storage medium, characterized in that, It stores program instructions, which, when executed by a processor, implement the evolution-optimized multi-train scheduling parameter adaptive method as described in any one of claims 1 to 8.