Circular track RGV scheduling and avoiding method based on machine learning

Abstract

The invention discloses a circular track RGV scheduling and avoiding method based on machine learning, and the method comprises the following steps: S1, collecting the real-time data of RGVs, and carrying out the preprocessing; s2, generating a flow network to perform scheduling decision modeling, and defining a state space, an action space and a flow distribution model; s3, extracting an initial avoidance path; s4, adopting a fractional order differential evolutionary algorithm to optimize an avoidance path; s5, selecting a global optimal avoidance path; and S6, executing a scheduling decision. According to the method, intelligent optimization of scheduling and avoiding of the circular track RGV is realized, the scheduling efficiency is improved, the global optimality and the dynamic adaptability of avoiding path planning are remarkably improved, and the system can be ensured to adjust the optimal scheduling scheme in real time in a complex track environment.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and in particular to a method for dispatching and avoiding a circular rail RGV vehicle based on machine learning. Background Art

[0002] In modern intelligent manufacturing and automated logistics systems, the RGV trolley is widely used in warehousing logistics, intelligent production lines and automated distribution systems due to its efficient and accurate material transportation capabilities. However, with the expansion of production scale and the increase in logistics demand, the RGV trolley faces increasingly complex scheduling and avoidance problems during operation, including track resource conflicts, task priority scheduling, path optimization, and real-time avoidance decisions. Traditional scheduling methods, such as rule-based scheduling algorithms, genetic algorithms, ant colony algorithms, etc., can operate effectively under small-scale task scheduling, but in large-scale dynamic environments, it is difficult to achieve efficient and flexible scheduling and avoidance, resulting in reduced track resource utilization, limited task throughput, and even system congestion, which seriously affects production efficiency.

[0003] Most of the existing RGV scheduling and avoidance methods rely on fixed rules or heuristic optimization algorithms. For example, heuristic rule-based methods usually assign tasks and select paths for RGVs according to preset scheduling rules. Although they can achieve stable operation in a single environment, they lack the ability to adapt to complex dynamic environments. When the system task volume increases, track resources are tight, or task priorities change dynamically, rule-based methods are difficult to optimize paths in real time, resulting in frequent waiting or conflicts for vehicles. In addition, although optimization methods based on genetic algorithms, ant colony algorithms, etc. can improve path search capabilities to a certain extent, these algorithms usually rely on local search and heuristic evaluation, which are prone to fall into local optimality and cannot guarantee the optimal scheduling solution in a high-dimensional dynamic environment. In addition, the calculation convergence speed is slow, which makes it difficult to meet the real-time requirements of actual production.

[0004] In recent years, with the development of machine learning and artificial intelligence technology, data-driven methods such as reinforcement learning and deep learning have been gradually applied to RGV scheduling optimization problems. Among them, scheduling methods based on reinforcement learning, such as deep Q networks and policy gradient methods, can learn track scheduling strategies through interactive training and achieve adaptive optimization to a certain extent. However, the limitations of traditional reinforcement learning methods are: first, reinforcement learning often only focuses on the optimal solution of a single path, while in complex track avoidance problems, there may be multiple equivalent or better avoidance paths, and reinforcement learning models are difficult to effectively explore all possible solutions; second, reinforcement learning training relies on a large number of environmental interactions, and in a dynamic logistics environment, the system state changes frequently, resulting in low exploration efficiency and slow convergence during the training process, which affects the application effect of the model in actual scheduling.

[0005] In response to the above problems, generative flow networks have been proposed as a generative modeling method suitable for multimodal optimization problems. By maximizing the flow distribution from state to action, the generative flow network enables the system to generate multiple feasible scheduling schemes and select the optimal path through probabilistic sampling, avoiding the limitations of the single solution of traditional reinforcement learning methods. However, although the generative flow network can provide multiple candidate paths in scheduling optimization, its optimization process still requires additional global search capabilities to ensure that the final selected avoidance path reaches the global optimum under multi-objective optimization (such as track occupancy balance, energy consumption minimization, task throughput maximization, etc.). Therefore, there are still certain limitations in relying solely on generative flow networks for RGV scheduling, that is, it still relies on traditional optimization methods in the path optimization stage, and lacks global search capabilities for path optimization.

[0006] In order to further improve the optimization ability of the avoidance path, the fractional differential evolution algorithm is introduced as a supplement to the generative flow network. The traditional differential evolution algorithm relies on fixed search steps and mutation strategies during the optimization process. It is easy to fall into the local optimum in a high-dimensional dynamic environment, which affects the optimization efficiency. The fractional differential evolution algorithm introduces fractional derivative calculations to enable the path optimization process to adaptively adjust the search step size and conduct dynamic search in combination with historical path information. It can not only maintain the global search ability in the early stage of optimization to avoid premature convergence, but also gradually converge to the local optimum in the later stage of optimization, thereby improving the optimization accuracy of the avoidance path. However, there are still few RGV scheduling optimization methods that combine the generative flow network with the fractional differential evolution algorithm in current research. The existing methods either rely on reinforcement learning for single path search or use traditional optimization algorithms for local path optimization. It is difficult to achieve the coordinated optimization of scheduling scheme generation and avoidance path optimization, resulting in the overall scheduling efficiency of the system still having room for improvement.

[0007] Based on the defects of the above-mentioned prior art, the present invention proposes a method for scheduling and avoiding RGV trolleys on a circular track based on a generative flow network and a fractional-order differential evolution algorithm. The method first uses a generative flow network to model task scheduling, models the RGV track state by constructing a state space, an action space and a flow distribution model, and generates multiple candidate scheduling schemes. Then, based on the scheduling scheme, a preliminary avoidance path is extracted and optimized using a fractional-order differential evolution algorithm. The global search capability of the avoidance path is improved through fractional-order derivative variation, adaptive step size adjustment and dynamic crossover strategy, so that the final selected avoidance path reaches the optimal state under multi-objective optimization. Finally, based on the optimized avoidance path, a path screening method is used to select the global optimal avoidance path, and it is transmitted to the central scheduling system to achieve real-time dynamic avoidance scheduling.

[0008] Compared with the traditional method, the method proposed in this invention can make full use of the multimodal optimization capability of the generated flow network, making the scheduling scheme more diverse, and at the same time combining the fractional differential evolution algorithm to improve the global search capability and convergence accuracy of the avoidance path. In addition, this method can adjust the avoidance path in real time according to the track occupancy, task time window changes and external interference, so that the RGV can achieve efficient avoidance in a complex dynamic environment, improve the utilization rate of track resources, and improve the overall throughput capacity of the logistics system.

[0009] Therefore, how to provide a circular track RGV trolley scheduling and avoidance method based on machine learning is a problem that technical personnel in this field urgently need to solve. Summary of the invention

[0010] One purpose of the present invention is to propose a method for scheduling and avoiding a circular rail RGV trolley based on machine learning. The present invention has the advantages of strong scheduling optimization capability, high avoidance path planning accuracy, and strong ability to adapt to complex dynamic environments.

[0011] A method for dispatching and avoiding a circular rail RGV vehicle based on machine learning according to an embodiment of the present invention includes the following steps: S1, collect real-time data of RGV car using sensors, perform preprocessing, and construct car state representation; S2. Use the generative flow network to model the scheduling decision of the vehicle state representation, including defining the state space, action space and flow distribution model, and based on the principle of maximizing the global task benefit, perform probability sampling on the state representation to generate candidate scheduling solutions; S3, extracting an initial avoidance path from the candidate scheduling schemes; S4, using an improved fractional differential evolution algorithm to optimize the initial avoidance path, including initializing candidate paths for the initial avoidance path, calculating fractional derivative changes of each candidate path, performing an adaptive crossover operation, and updating the candidate paths according to a predetermined selection criterion to obtain an optimized candidate avoidance path; S5. Based on the optimized candidate avoidance paths, a screening method is applied to select a global optimal avoidance path, and the global optimal avoidance path is transmitted to the central dispatching system; S6. Adjust the running status of the RGV vehicle according to the instructions of the central dispatching system to achieve real-time dynamic dispatching and avoidance.

[0012] Optionally, the S2 specifically includes: S21. Construct the state space of the trolley dispatching system and define the state space of the RGV trolley on the circular track ; S22. Construct the action space of the trolley dispatching system and define the action space of the RGV trolley ; S23, based on the generation flow network to build from the state space To the action space The scheduling traffic distribution model defines the scheduling revenue function Evaluate the pros and cons of different scheduling schemes, use probabilistic modeling methods to generate scheduling schemes, and schedule profit functions Designed to measure a specific state Next, execute the action The comprehensive benefits that can be obtained take into account the following factors: ; in, is the task completion rate benefit, For orbital conflict penalties, is the energy cost, For scheduling stability, , , , is the weight parameter, satisfying ; S24. Traffic modeling of generated flow networks based on the revenue function , the generation flow network samples the scheduling scheme through probability modeling, and generates candidate scheduling schemes based on the probability results. The probability modeling calculation is as follows: ; in, Indicates in status All the actions below, Indicates in status Next, execute the candidate action The dispatch benefit value obtained; S25. During the training of the generative flow network, each round of scheduling generates Group candidate scheduling scheme: ; in, Represents the number of sampling times, each Represents a complete car scheduling solution.

[0013] Optionally, S3 specifically includes: based on the candidate scheduling schemes, using a multi-dimensional scheduling conflict analysis model to evaluate the track occupancy conflict risk of each scheduling scheme, establishing a conflict matrix, calculating the global avoidance weight vector based on the conflict matrix, constructing a multi-objective avoidance optimization function, defining the avoidance strategy space in the generated flow network through a reinforcement learning mechanism, and using a strategy sampling and dynamic adjustment mechanism to generate a preliminary avoidance path set for each car .

[0014] Optionally, the S4 specifically includes: S41, candidate avoidance path initialization, based on the preliminary avoidance trajectory set , for each path Conduct track occupancy assessment, calculate path fitness, combine the path weight matrix, and generate candidate path groups through adaptive initialization strategy ,in, Represents the candidate path group The first car feasible avoidance paths; S42. Performing mutation operation based on improved fractional-order differential, specifically including: S421. The core of fractional differential evolution algorithm lies in the order It directly affects the intensity of path disturbance and adopts a dynamic order adjustment strategy. The order changes adaptively with the optimization process: ; in, , denote the minimum and maximum order respectively, is the maximum number of optimization iterations, is the current iteration number, is the nonlinear control factor; S422. Based on the dynamic order adjustment, the avoidance path is mutated and calculated, and a fractional order gradient driving mechanism is introduced to enhance the directionality of path evolution. An adaptive step size adjustment strategy is adopted, and the path update method is as follows: ; in, , Represent the minimum and maximum step sizes, respectively. is the step attenuation coefficient, is the gamma function, For the Iterations, For path In the The coordinate value or trajectory status at the iteration; S43, using adaptive crossover operator to path group Perform cross fusion to obtain the optimized candidate avoidance path: ; in, and They are the avoidance paths of two different individual cars. is the dynamic adjustment coefficient; S44. Obtain an optimized candidate avoidance path set for the path group after mutation and crossover .

[0015] Optionally, the S5 specifically includes: S51, based on the optimized candidate avoidance path set , a comprehensive evaluation index is calculated for each path, including path avoidance cost, track occupancy balance, task completion efficiency and avoidance path stability. The path classification includes: High-priority avoidance path: If the path avoidance cost is lower than the minimum cost threshold, the track balance is higher than the maximum balance threshold, and the task completion efficiency of the avoidance path meets the optimal throughput rate, it is determined to be a high-priority avoidance path and is executed first; Dynamically adjust the avoidance path: If the path avoidance cost is between the minimum cost threshold and the maximum cost threshold, but there is a risk of local congestion in the track occupancy balance, and the task completion efficiency is affected by the dynamic changes in the environment, it is determined to dynamically adjust the avoidance path and set a dynamic optimization mechanism; Low-priority avoidance path: If the path avoidance cost is higher than the maximum cost threshold, the track resource occupancy rate is lower than the minimum balance threshold, and there is a bottleneck in the task completion efficiency, it is determined to be a low-priority avoidance path and is only used when there are no other feasible paths; S52. Filter the global optimal avoidance path. Consider the avoidance cost, track occupancy balance and task completion efficiency comprehensively. Use the following optimization criteria to filter the global optimal avoidance path. : Path avoidance cost optimization: select the vehicle avoidance path with the shortest avoidance time and the lowest energy consumption; Task completion efficiency constraint: If the path affects the task throughput, the avoidance trajectory is adjusted; Track resource occupancy balance: If the track occupancy balance of the global avoidance path is poor, the path will be re-optimized; S53, real-time monitoring data, evaluation Adaptability in a dynamic environment, categories include: Track status changes: If track congestion changes dynamically, adjust in real time Adapting avoidance paths to new track occupancy; Task time window changes: If the task priority changes and causes the avoidance path conflict, adjust To meet new task scheduling requirements; External interference factors: If the avoidance path is affected by external interference, the avoidance path is recalculated and the dynamic path correction mechanism is triggered; S54, executing the global optimal avoidance path and adjusting the dispatching parameters in real time, specifically including: Path execution optimization: When the car follows When avoiding an obstacle, the system monitors the trajectory of the car in real time and adjusts the path; Track conflict warning: If different vehicles execute avoidance paths at the same time, resulting in an increased risk of track resource conflict, the avoidance path re-optimization mechanism will be triggered; Speed ​​adjustment: If the speed of the vehicle on the avoidance path does not match the planning of the dispatching system, the vehicle's speed will be adjusted dynamically; Energy consumption optimization: When the car is executing the avoidance path, the energy consumption data is monitored in real time. If the energy consumption is found to be too high, the avoidance path or speed strategy is adjusted; S55, recording the avoidance path execution data and optimizing the path selection strategy, specifically including: Path execution record: compare the avoidance path actually executed by the car with the theoretical optimized path, analyze the path execution deviation, and record relevant data; Path adaptability evaluation: calculate the avoidance path adaptability under different environments and optimize the avoidance path selection strategy; Path optimization strategy update: Based on the path execution data, adjust the weight parameters in the avoidance path optimization model; S56. Upload the optimized avoidance path selection strategy to the central dispatching system for real-time update.

[0016] The beneficial effects of the present invention are: (1) The present invention adopts generative flow network to model task scheduling. Compared with the traditional scheduling strategy based on rules or single reinforcement learning method, this method can generate multiple candidate scheduling schemes based on track resource status, task priority and real-time data of the vehicle, and optimize sampling through probability modeling. Generative flow network can avoid the limitation that traditional reinforcement learning method can only learn a single path solution, so that the scheduling system has stronger global optimization ability, ensuring that the RGV vehicle can flexibly select the optimal scheduling scheme in a complex task environment, and improve the system throughput and track resource utilization.

[0017] (2) The present invention uses a fractional differential evolution algorithm to optimize candidate avoidance paths, introduces dynamic order adjustment, adaptive step size adjustment and fractional gradient driving mechanism, enhances global search capability in the early stage of optimization, avoids path optimization from falling into local optimum, improves path convergence accuracy in the later stage of optimization, and makes the avoidance path more stable and reasonable. Compared with traditional genetic algorithms or fixed step size optimization methods, the fractional differential evolution algorithm has better search efficiency and can quickly re-optimize the avoidance path in the case of dynamic changes in track state, adjustment of task time window or external interference, thereby improving the real-time avoidance capability of the RGV vehicle.

[0018] (3) The scheduling avoidance method of the present invention can evaluate the path adaptability based on real-time data, and dynamically adjust the scheduling parameters in combination with multi-dimensional indicators such as track conflict warning, energy consumption monitoring, and task completion rate. By optimizing the path optimization model, the system can continuously learn and adjust the avoidance strategy during the actual operation process to ensure the operation efficiency and scheduling stability of the vehicle. In addition, combined with the real-time update mechanism of the central scheduling system, the method of the present invention can continuously optimize the avoidance path selection strategy, improve the intelligence level of the RGV vehicle, and ensure that the system maintains efficient operation for a long time in a complex dynamic environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is a flow chart of a method for scheduling and avoiding circular rail RGV vehicles based on machine learning proposed by the present invention. DETAILED DESCRIPTION

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0021] refer to Figure 1 , a circular track RGV trolley scheduling avoidance method based on machine learning, comprising the following steps: S1, collect real-time data of RGV car using sensors, perform preprocessing, and construct car state representation; S2. Use the generative flow network to model the scheduling decision of the vehicle state representation, including defining the state space, action space and flow distribution model, and based on the principle of maximizing the global task benefit, perform probability sampling on the state representation to generate candidate scheduling solutions; S3, extracting an initial avoidance path from the candidate scheduling schemes; S4, using an improved fractional differential evolution algorithm to optimize the initial avoidance path, including initializing candidate paths for the initial avoidance path, calculating fractional derivative changes of each candidate path, performing an adaptive crossover operation, and updating the candidate paths according to a predetermined selection criterion to obtain an optimized candidate avoidance path; S5. Based on the optimized candidate avoidance paths, a screening method is applied to select a global optimal avoidance path, and the global optimal avoidance path is transmitted to the central dispatching system; S6. Adjust the running status of the RGV vehicle according to the instructions of the central dispatching system to achieve real-time dynamic dispatching and avoidance.

[0022] In this implementation, S2 specifically includes: S21. Construct the state space of the trolley dispatching system and define the state space of the RGV trolley on the circular track ; S22. Construct the action space of the trolley dispatching system and define the action space of the RGV trolley ; S23, based on the generation flow network to build from the state space To the action space The scheduling traffic distribution model defines the scheduling revenue function Evaluate the pros and cons of different scheduling schemes, use probabilistic modeling methods to generate scheduling schemes, and schedule profit functions Designed to measure a specific state Next, execute the action The comprehensive benefits that can be obtained take into account the following factors: ; in, is the task completion rate benefit, For orbital conflict penalties, is the energy cost, For scheduling stability, , , , is the weight parameter, satisfying ; S24. Traffic modeling of generated flow networks based on the revenue function , the generation flow network samples the scheduling scheme through probability modeling, and generates candidate scheduling schemes based on the probability results. The probability modeling calculation is as follows: ; in, Indicates in status All the actions below, Indicates in status Next, execute the candidate action The dispatch benefit value obtained; S25. During the training of the generative flow network, each round of scheduling generates Group candidate scheduling scheme: ; in, Represents the number of sampling times, each Represents a complete car scheduling solution.

[0023] In this embodiment, S3 specifically includes: based on the candidate scheduling schemes, using a multi-dimensional scheduling conflict analysis model to evaluate the track occupancy conflict risk of each scheduling scheme, establishing a conflict matrix, calculating the global avoidance weight vector based on the conflict matrix, constructing a multi-objective avoidance optimization function, defining the avoidance strategy space in the generated flow network through a reinforcement learning mechanism, and using a strategy sampling and dynamic adjustment mechanism to generate a preliminary avoidance path set for each vehicle. .

[0024] In this implementation manner, the S4 specifically includes: S41, candidate avoidance path initialization, based on the preliminary avoidance trajectory set , for each path Conduct track occupancy assessment, calculate path fitness, combine the path weight matrix, and generate candidate path groups through adaptive initialization strategy ,in, Represents the candidate path group The first car feasible avoidance paths; S42. Performing mutation operation based on improved fractional-order differential, specifically including: S421. The core of fractional differential evolution algorithm lies in the order It directly affects the intensity of path disturbance and adopts a dynamic order adjustment strategy. The order changes adaptively with the optimization process: ; in, , denote the minimum and maximum order respectively, is the maximum number of optimization iterations, is the current iteration number, is the nonlinear control factor; S422. Based on the dynamic order adjustment, the avoidance path is mutated and calculated, and a fractional order gradient driving mechanism is introduced to enhance the directionality of path evolution. An adaptive step size adjustment strategy is adopted, and the path update method is as follows: ; in, , Represent the minimum and maximum step sizes, respectively. is the step attenuation coefficient, is the gamma function, For the Iterations, For path In the The coordinate value or trajectory status at the iteration; S43, using adaptive crossover operator to path group Perform cross fusion to obtain the optimized candidate avoidance path: ; in, and They are the avoidance paths of two different individual cars. is the dynamic adjustment coefficient; S44. Obtain an optimized candidate avoidance path set for the path group after mutation and crossover .

[0025] In this implementation manner, S5 specifically includes: S51, based on the optimized candidate avoidance path set , a comprehensive evaluation index is calculated for each path, including path avoidance cost, track occupancy balance, task completion efficiency and avoidance path stability. The path classification includes: High-priority avoidance path: If the path avoidance cost is lower than the minimum cost threshold, the track balance is higher than the maximum balance threshold, and the task completion efficiency of the avoidance path meets the optimal throughput rate, it is determined to be a high-priority avoidance path and is executed first; Dynamically adjust the avoidance path: If the path avoidance cost is between the minimum cost threshold and the maximum cost threshold, but there is a risk of local congestion in the track occupancy balance, and the task completion efficiency is affected by the dynamic changes in the environment, it is determined to dynamically adjust the avoidance path and set a dynamic optimization mechanism; Low-priority avoidance path: If the path avoidance cost is higher than the maximum cost threshold, the track resource occupancy rate is lower than the minimum balance threshold, and there is a bottleneck in the task completion efficiency, it is determined to be a low-priority avoidance path and is only used when there are no other feasible paths; S52. Filter the global optimal avoidance path. Consider the avoidance cost, track occupancy balance and task completion efficiency comprehensively. Use the following optimization criteria to filter the global optimal avoidance path. : Path avoidance cost optimization: select the vehicle avoidance path with the shortest avoidance time and the lowest energy consumption; Task completion efficiency constraint: If the path affects the task throughput, the avoidance trajectory is adjusted; Track resource occupancy balance: If the track occupancy balance of the global avoidance path is poor, the path will be re-optimized; S53, real-time monitoring data, evaluation Adaptability in a dynamic environment, categories include: Track status changes: If track congestion changes dynamically, adjust in real time Adapting avoidance paths to new track occupancy; Task time window changes: If the task priority changes and causes the avoidance path conflict, adjust To meet new task scheduling requirements; External interference factors: If the avoidance path is affected by external interference, the avoidance path is recalculated and the dynamic path correction mechanism is triggered; S54, executing the global optimal avoidance path and adjusting the dispatching parameters in real time, specifically including: Path execution optimization: When the car follows When avoiding an obstacle, the system monitors the trajectory of the car in real time and adjusts the path; Track conflict warning: If different vehicles execute avoidance paths at the same time, resulting in an increased risk of track resource conflict, the avoidance path re-optimization mechanism will be triggered; Speed ​​adjustment: If the speed of the vehicle on the avoidance path does not match the planning of the dispatching system, the vehicle's speed will be adjusted dynamically; Energy consumption optimization: When the car is executing the avoidance path, the energy consumption data is monitored in real time. If the energy consumption is found to be too high, the avoidance path or speed strategy is adjusted; S55, recording the avoidance path execution data and optimizing the path selection strategy, specifically including: Path execution record: compare the avoidance path actually executed by the car with the theoretical optimized path, analyze the path execution deviation, and record relevant data; Path adaptability evaluation: calculate the avoidance path adaptability under different environments and optimize the avoidance path selection strategy; Path optimization strategy update: Based on the path execution data, adjust the weight parameters in the avoidance path optimization model; S56. Upload the optimized avoidance path selection strategy to the central dispatching system for real-time update.

[0026] Example: In order to verify the feasibility of the present invention in implementation, the present invention is applied to the automated logistics and transportation system of a large-scale intelligent manufacturing factory. The factory uses circular rail RGV trolleys for production material transportation, covering multiple production workshops. The total length of the track between the workshops is about 10,000 meters, and the number of RGV trolleys reaches 100. Due to the heavy production tasks, track resources often conflict, resulting in queuing of trolleys, affecting the overall logistics efficiency. The traditional scheduling system adopts a rule-based fixed scheduling strategy. When faced with sudden tasks and track congestion, it is unable to make dynamic adjustments, resulting in production delays. Therefore, the factory hopes to improve the scheduling efficiency of the RGV trolley and the optimality of the avoidance path through intelligent scheduling optimization, reduce system congestion, and increase production logistics throughput.

[0027] The method of the present invention is deployed in the central dispatching system of the factory, and the specific application process is as follows: First, sensors are used to collect data on the real-time position, task status, and track occupancy of the RGV trolley, and preprocessing is performed to construct a trolley state representation. Then, scheduling modeling is performed based on the generative flow network to generate multiple candidate scheduling schemes, and a probabilistic sampling method is used to select a scheme that meets the task requirements. Next, a preliminary avoidance path is extracted from the scheduling scheme, and optimized using a fractional-order differential evolution algorithm. The avoidance path is optimized using fractional-order derivative calculations and an adaptive step-size adjustment strategy to reduce track conflicts. Finally, based on the optimized candidate avoidance path, the global optimal avoidance path is screened out and transmitted to the central dispatching system to achieve real-time dynamic scheduling of the RGV trolley.

[0028] In order to compare the performance of the present invention with that of the traditional method, we conducted a 30-day experimental test in the production environment of the factory and collected key indicator data, including task completion rate, track congestion rate, vehicle waiting time, avoidance path optimization rate, etc. The experiment was divided into two groups: The first group (traditional method): adopts a rule-based fixed scheduling strategy, where all vehicles run along a preset path and execute fixed avoidance rules when encountering conflicts; The second group (the method of the present invention): the generation flow network proposed in the present invention is combined with the fractional-order differential evolution algorithm combinatorial optimization method for intelligent scheduling and avoidance.

[0029] The experimental results are as follows: Table 1: RGV trolley scheduling optimization effect data table

[0030] Experimental results show that the intelligent scheduling and avoidance method of the present invention is significantly superior to traditional methods in multiple key performance indicators. First, in terms of task completion rate, the method of the present invention has increased by 18.4%, from the original 78.2% to 92.5%, which shows that the generative flow network can generate better scheduling solutions in complex task scenarios to ensure the smooth execution of tasks. At the same time, the optimization capability of the fractional-order differential evolution algorithm further enhances the rationality of the avoidance path, enabling the RGV car to complete tasks more efficiently in complex track environments and avoid task delays or failures caused by scheduling conflicts.

[0031] In terms of track congestion rate, the method of the present invention reduces the congestion rate from 35.6% to 12.8%, a decrease of 64.0%, and significantly reduces track resource conflicts. Traditional scheduling methods rely on fixed avoidance rules and cannot dynamically adjust the path, resulting in multiple vehicles concentrating in the same area, forming traffic congestion. The method of the present invention generates multiple candidate scheduling schemes by generating flow networks, and combines the fractional differential evolution algorithm to optimize the avoidance path, which can reasonably allocate track resources and avoid excessive concentration of vehicles in certain areas, thereby effectively alleviating congestion.

[0032] The average waiting time of the trolley was also significantly reduced, from 112.5 seconds in the traditional method to 43.7 seconds, a reduction of 61.1%. This means that the method of the present invention can allow the RGV trolley to complete the task more smoothly without being stuck for a long time due to track blockage. The reduction in waiting time also indirectly improves the task throughput capacity. The data shows that the task throughput has increased from 320 pieces / hour to 415 pieces / hour, an increase of 29.7%, indicating that the method of the present invention can complete more tasks in the same time, greatly improving the operating efficiency of the logistics system.

[0033] In terms of avoidance path optimization, the method of the present invention achieved an avoidance path optimization rate of 85.2%. Compared with the traditional method that completely relies on fixed rules, the present method can adaptively adjust the avoidance path in a dynamic track environment to ensure that the car chooses the optimal driving trajectory and avoids unnecessary detours or waiting. This optimization not only reduces system congestion, but also improves the consistency of overall task scheduling.

[0034] In addition, the average energy consumption was reduced by 24.2%, from 9.5kWh to 7.2kWh. This change is mainly due to the intelligent path planning that reduces unnecessary long-distance driving and frequent starts and stops of the car, allowing the car to operate more energy-efficiently. The traditional method often leads to an increase in the driving distance of the car due to the fixed avoidance mechanism. The present invention uses intelligent scheduling and avoidance optimization to enable the car to choose a shorter and more reasonable path, reducing unnecessary energy consumption and improving the overall energy-saving efficiency of the factory.

[0035] Based on the above data analysis, the method of the present invention not only improves the scheduling efficiency of the RGV car, but also significantly reduces track congestion, improves task throughput, and achieves significant optimization in energy consumption control. By combining the generative flow network and the fractional differential evolution algorithm, the present invention realizes intelligent and dynamic scheduling optimization, ensures that the car can operate efficiently in a complex track environment, and can be widely used in multiple industries such as intelligent manufacturing, automated warehousing, and unmanned distribution, improving the overall operation efficiency of the logistics system.

[0036] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for dispatching and avoiding RGV vehicles based on machine learning, characterized in that: The steps include: S1, collect real-time data of RGV car using sensors, perform preprocessing, and construct car state representation; S2. Use the generative flow network to model the scheduling decision of the vehicle state representation, including defining the state space, action space and flow distribution model, and based on the principle of maximizing the global task benefit, perform probability sampling on the state representation to generate candidate scheduling solutions; S3, extracting an initial avoidance path from the candidate scheduling schemes; S4, using an improved fractional differential evolution algorithm to optimize the initial avoidance path, including initializing candidate paths for the initial avoidance path, calculating fractional derivative changes of each candidate path, performing an adaptive crossover operation, and updating the candidate paths according to a predetermined selection criterion to obtain an optimized candidate avoidance path; S5. Based on the optimized candidate avoidance paths, a screening method is applied to select a global optimal avoidance path, and the global optimal avoidance path is transmitted to the central dispatching system; S6. Adjust the running status of the RGV vehicle according to the instructions of the central dispatching system to achieve real-time dynamic dispatching and avoidance.

2. The method for dispatching and avoiding circular rail RGV vehicles based on machine learning according to claim 1 is characterized in that: The S2 specifically includes: S21. Construct the state space of the trolley dispatching system and define the state space of the RGV trolley on the circular track ; S22. Construct the action space of the trolley dispatching system and define the action space of the RGV trolley on the circular track ; S23, based on the generation flow network to build from the state space To the action space The scheduling traffic distribution model defines the scheduling revenue function Evaluate the pros and cons of different scheduling schemes, use probabilistic modeling methods to generate scheduling schemes, and schedule profit functions Designed to measure a specific state Next, execute the action The comprehensive benefits that can be obtained take into account the following factors: ； in, is the task completion rate benefit, For orbital conflict penalties, is the energy cost, For scheduling stability, , , , is the weight parameter, satisfying ; S24. Flow modeling of generated flow networks based on the revenue function , the generation flow network samples the scheduling scheme through probability modeling, and generates candidate scheduling schemes based on the probability results. The probability modeling calculation is as follows: ； in, Indicates in status All the actions below, Indicates in status Next, execute the candidate action The dispatch benefit value obtained; S25. During the training of the generative flow network, each round of scheduling generates Group candidate scheduling scheme: ； in, Represents the number of sampling times, each Represents a complete car scheduling solution.

3. The method for dispatching and avoiding circular rail RGV vehicles based on machine learning according to claim 1 is characterized in that: S3 specifically includes: based on the candidate scheduling schemes, using a multi-dimensional scheduling conflict analysis model to evaluate the track occupancy conflict risk of each scheduling scheme, establishing a conflict matrix, calculating the global avoidance weight vector based on the conflict matrix, constructing a multi-objective avoidance optimization function, defining the avoidance strategy space in the generated flow network through a reinforcement learning mechanism, and using a strategy sampling and dynamic adjustment mechanism to generate a preliminary avoidance path set for each car. .

4. The method for dispatching and avoiding circular rail RGV vehicles based on machine learning according to claim 1 is characterized in that: The S4 specifically includes: S41, candidate avoidance path initialization, based on the preliminary avoidance trajectory set , for each path Conduct track occupancy assessment, calculate path fitness, combine the path weight matrix, and generate candidate path groups through adaptive initialization strategy ,in, Represents the candidate path group The first car feasible avoidance paths; S42. Performing mutation operation based on improved fractional-order differential, specifically including: S421. The core of fractional differential evolution algorithm lies in the order It directly affects the intensity of path disturbance and adopts a dynamic order adjustment strategy. The order changes adaptively with the optimization process: ； in, , denote the minimum and maximum order respectively, is the maximum number of optimization iterations, is the current iteration number, is the nonlinear control factor; S422. Based on the dynamic order adjustment, the avoidance path is mutated and calculated, and a fractional order gradient driving mechanism is introduced to enhance the directionality of path evolution. An adaptive step size adjustment strategy is adopted. The path update method is as follows: ； in, , Represent the minimum and maximum step sizes, respectively. is the step attenuation coefficient, is the gamma function, For the Iterations, For path In the The coordinate value or trajectory status at the iteration; S43, using adaptive crossover operator to path group Perform cross fusion to obtain the optimized candidate avoidance path: ； in, and They are the avoidance paths of two different individual cars. is the dynamic adjustment coefficient; S44. Obtain an optimized candidate avoidance path set for the path group after mutation and crossover .

5. The method for dispatching and avoiding circular rail RGV vehicles based on machine learning according to claim 1 is characterized in that: The S5 specifically includes: S51, based on the optimized candidate avoidance path set , a comprehensive evaluation index is calculated for each path, including path avoidance cost, track occupancy balance, task completion efficiency and avoidance path stability. The path classification includes: High-priority avoidance path: If the path avoidance cost is lower than the minimum cost threshold, the track balance is higher than the maximum balance threshold, and the task completion efficiency of the avoidance path meets the optimal throughput rate, it is determined to be a high-priority avoidance path and is executed first; Dynamically adjust the avoidance path: If the path avoidance cost is between the minimum cost threshold and the maximum cost threshold, but there is a risk of local congestion in the track occupancy balance, and the task completion efficiency is affected by the dynamic changes of the environment, it is determined to dynamically adjust the avoidance path and set a dynamic optimization mechanism; Low-priority avoidance path: If the path avoidance cost is higher than the maximum cost threshold, the track resource occupancy rate is lower than the minimum balance threshold, and there is a bottleneck in the task completion efficiency, it is determined to be a low-priority avoidance path and is only used when there are no other feasible paths; S52. Filter the global optimal avoidance path. Consider the avoidance cost, track occupancy balance and task completion efficiency comprehensively. Use the following optimization criteria to filter the global optimal avoidance path. : Path avoidance cost optimization: select the vehicle avoidance path with the shortest avoidance time and the lowest energy consumption; Task completion efficiency constraint: If the path affects the task throughput, the avoidance trajectory is adjusted; Track resource occupancy balance: If the track occupancy balance of the global avoidance path is poor, the path will be re-optimized; S53, real-time monitoring data, evaluation Adaptability in a dynamic environment, categories include: Track status changes: If track congestion changes dynamically, adjust in real time Adapting avoidance paths to new track occupancy; Task time window changes: If the task priority changes and causes the avoidance path conflict, adjust To meet new task scheduling requirements; External interference factors: If the avoidance path is affected by external interference, the avoidance path is recalculated and the dynamic path correction mechanism is triggered; S54, executing the global optimal avoidance path and adjusting the dispatching parameters in real time, specifically including: Path execution optimization: When the car follows When avoiding an obstacle, the system monitors the trajectory of the car in real time and adjusts the path; Track conflict warning: If different vehicles execute avoidance paths at the same time, resulting in an increased risk of track resource conflict, the avoidance path re-optimization mechanism will be triggered; Speed ​​adjustment: If the speed of the vehicle on the avoidance path does not match the planning of the dispatching system, the vehicle's speed will be adjusted dynamically; Energy consumption optimization: When the car is executing the avoidance path, the energy consumption data is monitored in real time. If the energy consumption is found to be too high, the avoidance path or speed strategy is adjusted; S55, recording the avoidance path execution data and optimizing the path selection strategy, specifically including: Path execution record: compare the avoidance path actually executed by the car with the theoretical optimized path, analyze the path execution deviation, and record relevant data; Path adaptability evaluation: calculate the avoidance path adaptability under different environments and optimize the avoidance path selection strategy; Path optimization strategy update: Based on the path execution data, adjust the weight parameters in the avoidance path optimization model; S56. Upload the optimized avoidance path selection strategy to the central dispatching system for real-time update.

Images