Multi-AGV path planning method with collision detection in matrix workshop

Abstract

The present invention relates to the technical field of path planning of AGV in intelligent manufacturing workshops, and in particular to a multi-AGV path planning method with collision detection in a matrix workshop. It includes: initializing algorithm parameters; initializing the population using a mixed initialization strategy; in the hiring bee stage, for each individual in the population, a vehicle merging strategy is adopted, and a local search operator is selected based on an adaptive selection strategy to perform individual evolution; in the observation bee stage, enhanced evolution is performed on individuals with better rankings, vehicle position information is updated, collision detection is performed on the vehicle's running process, and for possible collisions, the vehicle's driving path is replanned based on a collision avoidance strategy; in the reconnaissance bee stage, for individuals that have failed to evolve continuously, a ranking-based restart strategy is used to enhance the global exploration capability; it is determined whether the termination condition is met, and the optimal solution is finally output. The present invention is based on actual production application scenarios, improves production safety, and reduces AGV operating costs.

Description

Technical Field

[0001] The present invention relates to the technical field of AGV path planning in an intelligent manufacturing workshop, and in particular to a multi-AGV path planning method with collision detection in a matrix workshop. Background Art

[0002] With the continuous development of modern industry and the continuous expansion of enterprise production scale, the requirements for production efficiency and flexibility are gradually increasing. Traditional linear production workshops can no longer meet the needs of large-scale production. Against this background, the concept of distributed production came into being. As its typical layout form, the matrix workshop has become a very advantageous production mode with its modular design and flexible layout. By arranging production units in a matrix form, the matrix workshop significantly improves resource utilization efficiency, while enhancing the adaptability and dynamic scheduling capabilities of the production line.

[0003] In a matrix workshop, the efficiency of material transportation has a direct and far-reaching impact on the overall production process. Transportation time and waiting time are key factors that restrict the improvement of production efficiency and cause production bottlenecks. Optimizing these times has become the key to improving production efficiency. To meet this challenge, the introduction of intelligent scheduling technology provides a new solution. Among them, the application of multi-AGV (automatic guided vehicle) system is particularly important. As the core equipment for material transportation in a matrix workshop, the operating efficiency of AGV directly determines the execution effect of production tasks and the production efficiency of the entire workshop. By rationally planning the operation path and task allocation of AGV, not only can the transportation time be shortened, but also the additional consumption caused by waiting and improper scheduling can be effectively reduced, thereby greatly reducing the production cost of the workshop and improving production efficiency.

[0004] However, due to the complex layout of matrix workshops, the AGV operating environment has a high degree of uncertainty and faces many challenges. First, there are a large number of path intersections in the matrix workshop, and vehicles frequently intersect, which easily leads to collision risks. Secondly, due to the high driving frequency of AGVs, the scheduling system needs to handle dynamic task requests and path conflicts in real time. Once a vehicle collision or congestion occurs, it will not only cause stagnation at the intersection, but also cause delays in production tasks, affecting the production rhythm of the entire workshop. In addition, since the amount of goods unloaded at each task point is different, the vehicle load will change over time, resulting in fluctuations in real-time energy consumption. In actual operation, the AGV system also needs to comprehensively consider multiple factors such as energy consumption costs, intersection turning costs, and task priorities, which further increases the complexity of the scheduling problem.

[0005] In summary, in-depth research on the multi-AGV path planning problem with collision detection in a matrix workshop has important theoretical value and practical significance. This not only helps to solve the scheduling problem in a complex workshop environment, but also effectively reduces the operating cost of AGVs and provides strong support for improving the production efficiency of enterprises. Summary of the invention

[0006] The purpose of the present invention is to provide a multi-AGV path planning method with collision detection in a matrix workshop, which solves the problems of easy collisions between vehicles and high total vehicle operating costs in the current research on multi-AGVs in matrix workshops. It is based on actual production scenarios to effectively reduce operating costs, improve production efficiency and production safety.

[0007] The multi-AGV path planning method with collision detection in a matrix workshop provided by the present invention is characterized by comprising the following steps:

[0008] Step 1, parameter setting stage, initializing the adaptive artificial bee colony algorithm parameters;

[0009] Setting the population size Ps , observe the number of iterations of the bee stage r , the number of limited delayed evolutions in the scout bee stage , the number of times the local search operator can be repeatedly executed rep ;

[0010] Step 2, initialization phase, use the mixed initialization strategy to initialize the population, forming a population containing Ps The initial population of solutions is stored in the population POP middle;

[0011] Step 3, hiring bees stage, for solutions that exceed the expected number of vehicles, the vehicle merging strategy is adopted. For each solution in the population, the local search operator is selected for exploration using the adaptive selection strategy, and the solution that can reduce the total cost of vehicle operation is accepted, and the probability of the local search operator that generates the solution being selected is updated;

[0012] Step 4: Observation bee stage: sort all solutions in the population in ascending order according to the total vehicle operating cost, and select the top 40% of the total vehicle operating cost to store in the population with better performance. POP better ,for POP better The solution in is selected using an adaptive selection strategy. rThe first exploration is to select a local search operator for exploration, accept the solution with a reduced total vehicle operating cost, repeat the adaptive selection strategy, select a local search operator for exploration, initialize the vehicle information, update the vehicle information, check whether a collision occurs during the vehicle operation, and apply the collision avoidance strategy to plan the driving paths of all vehicles, and finally update the total vehicle operating cost;

[0013] Step 5, the scout bee stage, for the population POP Continuous For the solutions that failed in the second evolution, a ranking-based restart strategy is used to generate new solutions to replace the old ones;

[0014] Step 6: Update the solution with the minimum cost and determine whether the termination time has been reached. If so, terminate the evolution and output the solution with the minimum total vehicle operating cost and its corresponding vehicle driving path. Otherwise, execute step 3.

[0015] Furthermore, in step 2, a complete solution is Indicates that the delivery request is initiated by each workstation and each delivery request is regarded as a task , is the task number, is the number of all workstations in the workshop, , is the number of tasks, , ,vehicle The execution order of tasks in Composition, of which Indicates the vehicle number. is the number of vehicles used, , a complete solution By vehicle Composed of 0s, that is .

[0016] Furthermore, in step 2, the rules of the hybrid initialization strategy are to generate 1 solution using the nearest neighbor individual heuristic rule, generate 1 solution using the nearest neighbor time heuristic rule, generate 1 solution using the maximum cargo demand heuristic rule, and randomly generate non-repeating remaining solutions using the improved forward insertion heuristic rule. Ps -3 solutions, forming a Ps The initial population of solutions is stored in the population POP middle.

[0017] Furthermore, in step 2, the nearest neighbor heuristic rule is that for each vehicle's work order, the next task is selected according to the minimum distance from the current vehicle's position, and the allocation is stopped until all tasks are assigned to the vehicle's work order, forming an initial solution;

[0018] The nearest neighbor time heuristic rule is that for each vehicle's work order, the next task is selected based on the minimum time difference between the time when the current task completes unloading and the time when the next task sends a request, until all tasks are assigned to the vehicle's work order and the allocation stops, forming an initial solution;

[0019] The heuristic rule for the maximum cargo demand is that for each vehicle's work order, the task with the largest amount of cargo among all task requests is selected as the next task until all tasks are assigned to the vehicle's work order, thus forming an initial solution.

[0020] The improved forward insertion heuristic rule is to randomly select a task and insert it into any position of the current vehicle. If it exceeds the weight and task time limit of the current vehicle, a new vehicle is added until all tasks are arranged and all solutions are stored in the population. POP middle.

[0021] Furthermore, in step 3, the total cost of vehicle operation is calculated as follows:

[0022]

[0023] in, In the vehicle scheduling problem, the goal is to minimize the total cost of vehicle operation. is the workstation number, k is the vehicle number, is the straight-line operating cost of the vehicle, is the turning cost of the vehicle, is the basic usage cost of each vehicle, The penalty cost for vehicles arriving overtime or early. is a binary value, such as workstation ,workstation If a vehicle passes through the channel between them, it is set to 1, otherwise it is set to 0. For workstations ,workstation The distance travelled between vehicles, For workstations ,workstation The number of turns between vehicles, is the number of vehicles departing from the garage, The latest delivery time of the vehicle. The actual delivery time of the vehicle.

[0024] Furthermore, in step 3, the rule of the vehicle merging strategy is that for all vehicles, the vehicles are sorted in descending order according to the number of tasks of each vehicle. The number of existing vehicles is , the expected number of vehicles is , the last mission attempt of the last car to the If the vehicle does not meet the load limit, power limit, and the latest time limit for the vehicle to arrive at the workstation, it will try to insert the The remaining tasks in the last vehicle are inserted until the last vehicle is empty, then the last vehicle is deleted and the number of existing vehicles meets the expected number of vehicles.

[0025] Furthermore, the formula for power limitation is:

[0026]

[0027] For workstations ,workstation The vehicle power consumption between , is the basic power consumption of the vehicle, is the real-time load of the vehicle, is the load power consumption of the vehicle, It is the total power of the vehicle in one mission cycle.

[0028] Furthermore, in step 3, the calculation formula for the expected number of vehicles is, , the number of tasks that the vehicle is expected to carry Determined by the vehicle models used in the workshop.

[0029] Furthermore, in step 3, the local search operator includes a local search operator for ordinary vehicles and a local search operator for key vehicles, where the key vehicle refers to the vehicle with the highest total vehicle operating cost, wherein the vehicle number , vehicle number , vehicle number ,

[0030] Local search operators for common vehicles include,

[0031] Insertion operator inside the vehicle, randomly select the vehicle Tasks in ,Task , the task Insert the tasks into the order of completion The previous position of the car changes the order of work that needs to be done by the car to form a new solution;

[0032] The exchange operator inside the vehicle, randomly select the vehicle Tasks in ,Task , the task ,Task The working order is exchanged to form a new solution;

[0033] Insertion operator between vehicles, randomly select vehicles ,vehicle , from the vehicle Randomly select tasks , from the vehicle Randomly select tasks , the task From vehicle Remove the task from the work order and Add to Vehicle Mission The previous position of the working sequence constitutes a new solution;

[0034] The exchange operator between vehicles, randomly selects vehicles ,vehicle , from the vehicle Randomly select tasks , from the vehicle Randomly select tasks , the task ,Task The working order is exchanged to form a new solution;

[0035] The inversion operator inside the vehicle, randomly select the vehicle , from the vehicle Randomly select tasks ,Task , the task ,Task The order of all the work in between is reversed to form a new solution;

[0036] The local search operators for key vehicles include,

[0037] Exchange operators within key vehicles, calculate and select key vehicles , from the vehicle Randomly select tasks ,Task , the task ,Task The working order is exchanged to form a new solution;

[0038] The exchange operator between the key vehicle and other vehicles, calculates and selects the key vehicle , and then randomly select a non-critical vehicle , from key vehicles Randomly select tasks , from the vehicle Randomly select tasks , the task ,Task The working order is exchanged to form a new solution;

[0039] The task removal operator in the key vehicle calculates and selects the key vehicles , and then randomly select a non-critical vehicle , from key vehicles Randomly select tasks , from the vehicle Randomly select tasks , the task Remove from the work order and move the task Insert into the selected task The previous position of , forming a new solution.

[0040] Furthermore, in step 3, the steps of selecting the local search operator by the adaptive selection strategy are as follows: for each solution, record the 50 new solutions generated by each solution using the local search operator, and by judging whether the total vehicle operating cost of the new solution is less than the total vehicle operating cost of the original solution, count the number of times each local search operator successfully reduces the total vehicle operating cost, calculate the ratio of the number of successful times of each local search operator to the number of successful times of all local search operators as the probability of each local search operator being selected, and for a newly generated initial solution, use the roulette strategy to firstly select the 8 local search operators. The probability is normalized, and each of the 8 local search operators is assigned to an interval. The probability of interval 1 to interval 8 being selected is initialized to 1 / 8. The probability of each local search operator is accumulated to obtain an interval range of [0,1]. Intervals 1 to 8 correspond to the insertion operator inside the vehicle, the exchange operator inside the vehicle, the insertion operator between vehicles, the exchange operator between vehicles, the reversal operator inside the vehicle, the exchange operator within the key vehicle, the exchange operator between the key vehicle and other vehicles, and the task removal operator in the key vehicle. Next, a random number is generated in the range of [0,1]. rand , and based on the random number rand The corresponding local search operator is selected for the interval that it falls into, and each selected local search operator is repeatedly executed. rep times, based on whether the total vehicle operation cost of the generated solution is less than the total vehicle operation cost of the original solution, count the number of times each local search operator successfully reduces the total vehicle operation cost, calculate the ratio of the number of successes of each local search operator to the number of successes of all local search operators, update the probability of each local search operator being selected, and use the updated probability of the local search operator being selected as the basis for dividing the distribution interval of each local search operator in the next adaptive selection strategy.

[0041] Further, in step 4, the vehicle information is composed of {the X-axis coordinate of the vehicle's position, the Y-axis coordinate of the vehicle's position, the X-axis coordinate of the vehicle's target position, the Y-axis coordinate of the vehicle's target position, the vehicle's driving direction, whether the vehicle is at an intersection, whether the vehicle is in a waiting state, and whether the vehicle is running}. For the evolved solution, first initialize the vehicle's initial information, update the vehicle's next target, update the vehicle's moving direction, and let each vehicle start rolling at an interval of 1s. After each vehicle operation, update the vehicle's position information, and check whether the vehicle operation will collide through the vehicle's running direction and the distance between the vehicles. When two vehicles run in opposite directions in the same horizontal channel and the distance between the two vehicles is less than or equal to 5m, the two vehicles will collide; when two vehicles run in opposite directions in the same longitudinal channel and the distance between the two vehicles is less than or equal to 8m, the two vehicles will collide; if a collision occurs, use a collision avoidance strategy to modify the vehicle's driving path, avoid the collision in advance, and update the vehicle's latest total vehicle operating cost.

[0042] Furthermore, in step 4, the collision avoidance strategy includes the following event handling mechanisms:

[0043] The vehicles stop and wait. When a collision occurs during the operation of the detection vehicles, and the collision occurs at an intersection, one of the vehicles will stop in front of the intersection to give priority to the other vehicle;

[0044] Vehicle steering avoidance: when a collision is detected during the operation of the vehicle, and the collision occurs in the channel, and the detour of the vehicle at the intersection will not increase the vehicle's running distance, the vehicle will turn at this intersection;

[0045] The vehicle yields and waits. When a collision is detected during the operation of the vehicle, and the collision occurs in the channel, and the vehicle at the intersection will increase the vehicle's operating distance by detouring, the vehicle will turn right at this intersection, stop and wait for the oncoming vehicle to pass the intersection, and then resume operation.

[0046] Further, in step 5, for the number of evolutions exceeding the hysteresis For solutions without evolution, a ranking-based restart strategy is used. The top 20% solutions are retained according to the order of the total cost of vehicle operation of the solutions. For the top 20%-60% solutions, the key vehicles in the solutions are retained during the solution regeneration process, and the improved forward insertion heuristic rule is used to randomly generate the working order of other vehicles. For solutions ranked after 60%, the improved forward insertion heuristic rule is used to randomly generate the working order of all vehicles during the solution regeneration process.

[0047] Furthermore, in step 6, the termination time is fixed to 5 seconds.

[0048] The multi-AGV path planning method with collision detection in a matrix workshop provided by the present invention designs an effective hybrid initialization population method, and generates an initial population with high quality and diversity by initializing the population through a hybrid initialization strategy; 8 local search operators are designed to adapt to solutions in different stages for local search; secondly, an adaptive selection strategy is designed to select local search operators; in order to solve the problem of hysteresis evolution of solutions and falling into local optimality, a ranking-based restart strategy is proposed; in addition, in order to solve the collision problem in the matrix workshop, a collision avoidance strategy is proposed. In short, the application of the present invention can effectively solve the problem of multi-AGV path planning in a matrix production workshop, and by coordinating the running paths of all vehicles, it has the positive effects of reducing the total cost of vehicle operation, improving production efficiency, and ensuring the safety of the transportation process. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flow chart for realizing the present invention;

[0050] Figure 2 Schematic diagram of the vehicle parking waiting processing mechanism of the present invention;

[0051] Figure 3 Schematic diagram of the vehicle steering avoidance processing mechanism of the present invention;

[0052] Figure 4 Schematic diagram of the vehicle yield waiting processing mechanism of the present invention;

[0053] Figure 5 A layout diagram of a work shop in one embodiment of the present invention;

[0054] Figure 6 It is the average scatter plot of the present invention and five existing comparison algorithms;

[0055] Figure 7 It is a convergence curve diagram of the present invention and five existing comparison algorithms when the instance size is 30;

[0056] Figure 8 It is a three-dimensional diagram of the AGV scheduling path and time distribution of the present invention. DETAILED DESCRIPTION

[0057] like Figure 1 As shown, the multi-AGV path planning method with collision detection in a matrix workshop provided by the present invention is mainly implemented through the following steps.

[0058] Step 1: Parameter setting stage. Initialize the adaptive artificial bee colony algorithm parameters and set the population size. Ps , observe the number of iterations of the bee stage r , the number of limited delayed evolutions in the scout bee stage , the number of times the local search operator can be repeatedly executed rep .in, Ps = 20, = 400, r = 60, rep = 10, the end time is 5s.

[0059] Step 2, initialization phase. Use the hybrid initialization strategy to initialize the population to form a population containing Ps The initial population of solutions is stored in the population POP Specifically, a complete solution, that is, individuals in the population, is used Indicates that the delivery request is initiated by each workstation and each delivery request is regarded as a task , is the task number, is the number of all workstations in the workshop, , is the number of tasks, , ,vehicle The execution order of tasks in Composition, of which Indicates the vehicle number. is the number of vehicles used, , a complete solution By vehicle Composed of 0s, that is .

[0060] The rules of the hybrid initialization strategy are to generate 1 solution using the nearest neighbor individual heuristic rule, 1 solution using the nearest neighbor time heuristic rule, 1 solution using the maximum cargo demand heuristic rule, and randomly generate non-repeating remaining solutions using the improved forward insertion heuristic rule. Ps -3 solutions, forming a Ps The initial population of solutions is stored in the population POP middle.

[0061] The nearest neighbor individual heuristic rule is that for each vehicle's work order, the next task is selected based on the minimum distance from the current vehicle's position, until all tasks are assigned to the vehicle's work order and the allocation is stopped, forming an initial solution; the nearest neighbor time heuristic rule is that for each vehicle's work order, the next task is selected based on the minimum time difference between the time when the current task completes unloading and the time when the next task sends a request, until all tasks are assigned to the vehicle's work order and the allocation is stopped, forming an initial solution; the maximum cargo demand heuristic rule is that for each vehicle's work order, the task with the largest amount of cargo in all task requests is selected as the next task, until all tasks are assigned to the vehicle's work order and the allocation is stopped, forming an initial solution; the improved forward insertion heuristic rule is that a task is randomly selected and inserted into any position of the current vehicle. If it exceeds the weight and task time limits of the current vehicle, a new vehicle is added until all tasks are arranged and all solutions are stored in the population. POP middle.

[0062] Step 3, hiring bees stage. For solutions that exceed the expected number of vehicles, the vehicle merging strategy is adopted. For each solution in the population, the local search operator is selected for exploration using the adaptive selection strategy, and the solution that can reduce the total cost of vehicle operation is accepted, and the probability of the local search operator that generates the solution being selected is updated.

[0063] Specifically, the total cost of vehicle operation is calculated as follows:

[0064]

[0065] in, In the vehicle scheduling problem, the goal is to minimize the total cost of vehicle operation. is the workstation number, k is the vehicle number, is the straight-line operating cost of the vehicle, is the turning cost of the vehicle, is the basic usage cost of each vehicle, The penalty cost for vehicles arriving overtime or early. is a binary value, such as workstation ,workstation If a vehicle passes through the channel between them, it is set to 1, otherwise it is set to 0. For workstations ,workstation The distance travelled between vehicles, For workstations ,workstation The number of turns between vehicles, is the number of vehicles departing from the garage, The latest delivery time of the vehicle. The actual delivery time of the vehicle.

[0066] The rule of the vehicle merging strategy is that for all vehicles, the vehicles are sorted in descending order according to the number of tasks of each vehicle. The number of existing vehicles is , the expected number of vehicles is , the last mission attempt of the last car to the If the vehicle does not meet the load limit, power limit, and the latest time limit for the vehicle to arrive at the workstation, it will try to insert the The remaining tasks in the last vehicle are inserted until the last vehicle is empty, then the last vehicle is deleted and the number of existing vehicles meets the expected number of vehicles.

[0067] The formula for power limit is:

[0068]

[0069] For workstations ,workstation The vehicle power consumption between , is the basic power consumption of the vehicle, is the real-time load of the vehicle, is the load power consumption of the vehicle, is the total power of the vehicle in a task cycle. , , .

[0070] The formula for calculating the expected number of vehicles is: , the number of tasks that the vehicle is expected to carry , which is determined by the vehicle model used in the workshop. In this embodiment,

[0071] Local search operators include local search operators for common vehicles and local search operators for key vehicles. Key vehicles refer to vehicles with the highest total vehicle operating cost, where vehicle number , vehicle number , vehicle number ,

[0072] Local search operators for common vehicles include,

[0073] Insertion operator inside the vehicle, randomly select the vehicle Tasks in ,Task , the task Insert the tasks into the order of completion The previous position of the car changes the order of work that needs to be done by the car to form a new solution;

[0074] The exchange operator inside the vehicle, randomly select the vehicle Tasks in ,Task , the task ,Task The working order is exchanged to form a new solution;

[0075] Insertion operator between vehicles, randomly select vehicles ,vehicle , from the vehicle Randomly select tasks , from the vehicle Randomly select tasks , the task From vehicle Remove the task from the work order and Add to Vehicle Mission The previous position of the working sequence constitutes a new solution;

[0076] The exchange operator between vehicles, randomly selects vehicles ,vehicle , from the vehicle Randomly select tasks , from the vehicle Randomly select tasks , the task ,Task The working order is exchanged to form a new solution;

[0077] The inversion operator inside the vehicle, randomly select the vehicle , from the vehicle Randomly select tasks ,Task , the task ,Task The order of all the work in between is reversed to form a new solution;

[0078] The local search operators for key vehicles include,

[0079] Exchange operators within key vehicles, calculate and select key vehicles , from the vehicle Randomly select tasks ,Task , the task ,Task The working order is exchanged to form a new solution;

[0080] The exchange operator between the key vehicle and other vehicles, calculates and selects the key vehicle , and then randomly select a non-critical vehicle , from key vehicles Randomly select tasks , from the vehicle Randomly select tasks , the task ,Task The working order is exchanged to form a new solution;

[0081] The task removal operator in the key vehicle calculates and selects the key vehicles , and then randomly select a non-critical vehicle , from key vehicles Randomly select tasks , from the vehicle Randomly select tasks , the task Remove from the work order and move the task Insert into selected task The previous position of , forming a new solution.

[0082] The steps of selecting the local search operator by the adaptive selection strategy are as follows: for each solution, record the 50 new solutions generated by each solution using the local search operator, and by judging whether the total vehicle operating cost of the new solution is less than the total vehicle operating cost of the original solution, count the number of times each local search operator successfully reduces the total vehicle operating cost, calculate the ratio of the number of successful times of each local search operator to the number of successful times of all local search operators as the probability of each local search operator being selected, and for a newly generated initial solution, use the roulette strategy to first normalize the probabilities of the 8 local search operators , 8 local search operators are assigned to each interval, and the probability of interval 1 to interval 8 being selected is initialized to 1 / 8. The probability of each local search operator is accumulated to obtain an interval range of [0,1]. Interval 1 to interval 8 correspond to the insertion operator inside the vehicle, the exchange operator inside the vehicle, the insertion operator between vehicles, the exchange operator between vehicles, the reversal operator inside the vehicle, the exchange operator inside the key vehicle, the exchange operator between the key vehicle and other vehicles, and the task removal operator in the key vehicle. Next, a random number is generated in the range of [0,1]. rand , and based on the random number rand The corresponding local search operator is selected for the interval that it falls into, and each selected local search operator is repeatedly executed. reptimes, based on whether the total vehicle operation cost of the generated solution is less than the total vehicle operation cost of the original solution, count the number of times each local search operator successfully reduces the total vehicle operation cost, calculate the ratio of the number of successes of each local search operator to the number of successes of all local search operators, update the probability of each local search operator being selected, and use the updated probability of the local search operator being selected as the basis for dividing the distribution interval of each local search operator in the next adaptive selection strategy.

[0083] Step 4: Observation bee stage: sort all solutions in the population in ascending order according to the total vehicle operating cost, and select the top 40% of the total vehicle operating cost to store in the population with better performance. POP better ,for POP better The solution in is selected using an adaptive selection strategy. r It performs exploration again, selects the local search operator for exploration, accepts the solution with reduced total vehicle operating cost, repeatedly uses the adaptive selection strategy, selects the local search operator for exploration, then initializes the vehicle information, updates the vehicle information, checks whether a collision occurs during the vehicle operation, and applies the collision avoidance strategy to plan the driving paths of all vehicles, and finally updates the total vehicle operating cost.

[0084] Specifically, the vehicle information consists of {the vehicle's X-axis coordinate, the vehicle's Y-axis coordinate, the vehicle's target X-axis coordinate, the vehicle's target Y-axis coordinate, the vehicle's driving direction, whether the vehicle is at an intersection, whether the vehicle is in a waiting state, and whether the vehicle is running}. For the evolved solution, first initialize the vehicle's initial information, update the vehicle's next target, update the vehicle's moving direction, and let each vehicle start rolling at an interval of 1s. After each vehicle operation, update the vehicle's position information, and check whether the vehicle will collide based on the vehicle's running direction and the distance between the vehicles. When two vehicles are running in opposite directions in the same horizontal channel and the distance between the two vehicles is less than or equal to 5m, the two vehicles will collide; when two vehicles are running in opposite directions in the same longitudinal channel and the distance between the two vehicles is less than or equal to 8m, the two vehicles will collide; if a collision occurs, use a collision avoidance strategy to modify the vehicle's driving path, avoid the collision in advance, and update the vehicle's latest total vehicle operating cost.

[0085] The collision avoidance strategy includes the following event handling mechanisms:

[0086] The vehicle stops and waits, such as Figure 2 As shown, when a collision occurs during the operation of the detection vehicles, and the collision occurs at an intersection, one of the vehicles will stop before the intersection to give priority to the other vehicle to pass;

[0087] Vehicle steering avoidance, such as Figure 3As shown, when a collision occurs during the operation of the detection vehicle, and the collision occurs in the channel, and the vehicle at the intersection does not increase the vehicle's running distance by detouring, the vehicle will turn at this intersection;

[0088] Vehicles give way and wait, such as Figure 4 As shown, when a collision occurs during the operation of the detection vehicle, and the collision occurs in the channel, and the vehicle at the intersection will increase the vehicle's operating distance by detouring, the vehicle will turn right at this intersection, stop and wait for the oncoming vehicle to pass the intersection, and then resume operation.

[0089] Step 5, the scout bee stage. POP Continuous For solutions that fail in the second evolution, a ranking-based restart strategy is used to generate new solutions to replace the old ones.

[0090] Specifically, for more than the number of delayed evolutions For solutions without evolution, a ranking-based restart strategy is used. The top 20% solutions are retained according to the order of the total cost of vehicle operation of the solutions. For the top 20%-60% solutions, the key vehicles in the solutions are retained during the solution regeneration process, and the improved forward insertion heuristic rule is used to randomly generate the working order of other vehicles. For solutions ranked after 60%, the improved forward insertion heuristic rule is used to randomly generate the working order of all vehicles during the solution regeneration process.

[0091] Step 6: Update the solution with the minimum cost. Determine whether the 5s termination time has been reached. If so, terminate the evolution and select the solution with the minimum total vehicle operation cost. Output this solution and its corresponding vehicle driving path. Otherwise, execute step 3.

[0092] According to the situation in which the calling unit in the matrix workshop sends material distribution requests to the control system within a certain period of time, the supply warehouse and workstation number, workstation X-axis coordinate, workstation Y-axis coordinate, call time, buffer inventory at the time of call, delivery time and other information are collected. This example is applicable to a matrix manufacturing workshop with a single warehouse and multiple AGVs of the same model. The goal is to minimize the total cost of vehicle operation.

[0093] like Figure 5As shown, by listing the distribution diagram of tasks in a workshop in a specific example T30I02 of the present invention, the layout of the workshop of the present invention is further described. The warehouse coordinates are (0,0), and the workstations in the workshop are connected by channels. The channels only allow vehicles to pass in one direction and turn at the intersection. The vehicle departure time, loading capacity, vehicle speed, unloading time, the maximum driving distance of a fully loaded vehicle and the maximum driving distance of an empty vehicle allowed by the power in a task cycle are 365s, 250kg, 1m / s, 15s, 142.85m and 500m respectively. The call units in the matrix workshop are arranged in a 10×10 grid, and the horizontal and vertical distances between each call unit are 5m and 8m respectively. In the following Table 1, the task data table of the example T30I02 is listed.

[0094] Table 1 Example data table

[0095]

[0096] In order to prove that the proposed method is superior to other methods in optimizing the operating cost of this problem, the adaptive artificial bee colony algorithm (AABC) of the present invention is compared with the artificial bee colony algorithm (ABC), discrete artificial bee colony algorithm (DABC), discrete weed algorithm (DIWO), improved genetic algorithm (IGA) and population-based iterative greedy algorithm (PIG). In order to reduce the error of the experiment and make the calculation results more general and effective, we experimented with 100 instances of different production scales, each instance was repeated 30 times, and the generated results were statistically generated. The present invention uses relative percentage deviation (relative percentage deviation, RPD ) as a performance evaluation indicator, RPD The calculation formula is expressed as , is the minimum total cost obtained by each algorithm when solving a certain instance, is the minimum total cost obtained by the six algorithms on the same instance. RPD The smaller the value of , the closer the solution to the minimum value is to the current algorithm in the same time, which also proves that the performance of the algorithm is better. Figure 6 As shown, the present invention has obvious advantages over the five existing comparison algorithms in solving this problem. In order to clearly illustrate the evolution process of the adaptive artificial bee colony algorithm of the present invention compared with the five existing comparison algorithms, Figure 7 The convergence of the example T30I02 is shown. The adaptive artificial bee colony algorithm and the five existing comparison algorithms are all set with a termination time of 5s in the same test environment. Figure 7As shown, the adaptive artificial bee colony algorithm of the present invention can provide the lowest total vehicle operation cost at any termination time, which reflects that the comprehensive performance of the method of the present invention is the best.

[0097] The state of the AGV running path obtained by the algorithm over time is mapped to three-dimensional space, such as Figure 8 shown. Figure 8 The figure shows the relationship between the driving path and time of each vehicle in the example T30I02 after being planned by the adaptive artificial bee colony algorithm, where the X-axis and Y-axis represent the position of the vehicle in the workshop, and the Z-axis represents the time. Figure 8 It can be seen that all vehicles did not collide at any time, and the path planning with the least number of turns was achieved while meeting the task constraints.

[0098] To sum up, through specific experimental verification, the application of the present invention solves the problem of multi-AGV path planning with collision detection in a matrix workshop, and solves the problems of easy collisions between vehicles, real-time energy consumption fluctuations, and high total vehicle operating costs that have not been solved in previous studies. It effectively reduces the total vehicle operating cost, improves production efficiency, ensures the safety of the transportation process, and has great practical application value.

Claims

1. A multi-AGV path planning method with collision detection in a matrix workshop, characterized in that: The following steps are included: Step 1, parameter setting stage, initializing the adaptive artificial bee colony algorithm parameters; Setting the population size Ps , observe the number of iterations of the bee stage r , the number of limited delayed evolutions in the scout bee stage , the number of times the local search operator can be repeatedly executed rep ; Step 2, initialization phase, use the mixed initialization strategy to initialize the population, forming a population containing Ps The initial population of solutions is stored in the population POP middle; Step 3, hiring bee stage, for solutions that exceed the expected number of vehicles, the vehicle merging strategy is adopted. For each solution in the population, the local search operator is selected for exploration using the adaptive selection strategy, and the solution that can reduce the total cost of vehicle operation is accepted, and the probability of the local search operator that generates the solution being selected is updated; wherein, the local search operator includes the local search operator for the key vehicle, the key vehicle refers to the vehicle with the highest total cost of vehicle operation, and the local search operator for the key vehicle includes, Exchange operators within key vehicles, calculate and select key vehicles , from key vehicles Randomly select tasks ,Task , the task ,Task The working order is exchanged to form a new solution; The exchange operator between the key vehicle and other vehicles, calculates and selects the key vehicle , and then randomly select a non-critical vehicle , from key vehicles Randomly select tasks , from non-critical vehicles Randomly select tasks , the task ,Task The working order is exchanged to form a new solution; The task removal operator in the key vehicle calculates and selects the key vehicles , and then randomly select a non-critical vehicle , from key vehicles Randomly select tasks , from non-critical vehicles Randomly select tasks , the task Remove from the work order and move the task Insert into selected task The previous position of constitutes a new solution; Step 4: Observation bee stage: sort all solutions in the population in ascending order according to the total vehicle operating cost, and select the top 40% of the total vehicle operating cost to store in the population with better performance. POP better ,for POP better The solution in is selected using an adaptive selection strategy. r The first exploration is to select a local search operator for exploration, accept the solution with a reduced total vehicle operating cost, repeat the adaptive selection strategy, select a local search operator for exploration, initialize the vehicle information, update the vehicle information, check whether a collision occurs during the vehicle operation, and apply the collision avoidance strategy to plan the driving paths of all vehicles, and finally update the total vehicle operating cost; Step 5, the scout bee stage, for the population POP Continuous For the solutions that failed in the second evolution, a ranking-based restart strategy is used to generate new solutions to replace the old ones; Step 6: Update the solution with the minimum cost and determine whether the termination time has been reached. If so, terminate the evolution and output the solution with the minimum total vehicle operating cost and its corresponding vehicle driving path. Otherwise, execute step 3.

2. A multi-AGV path planning method with collision detection in a matrix workshop according to claim 1, characterized in that: In step 2, a complete solution is Indicates that the delivery request is initiated by each workstation, and each delivery request is regarded as a task , is the task number, is the number of all workstations in the workshop, , is the number of tasks, , ,vehicle The execution order of tasks in Composition, of which Indicates the vehicle number. is the number of vehicles used, , a complete solution By vehicle Composed of 0s, that is .

3. A multi-AGV path planning method with collision detection in a matrix workshop according to claim 2, characterized in that: The rules of the hybrid initialization strategy are to generate 1 solution using the nearest neighbor individual heuristic rule, 1 solution using the nearest neighbor time heuristic rule, 1 solution using the maximum cargo demand heuristic rule, and randomly generate non-repeating remaining solutions using the improved forward insertion heuristic rule. Ps -3 solutions, forming a Ps The initial population of solutions is stored in the population POP middle.

4. A multi-AGV path planning method with collision detection in a matrix workshop according to claim 3, characterized in that: The nearest neighbor heuristic rule is that for each vehicle's work order, the next task is selected based on the minimum distance from the current vehicle's position until all tasks are assigned to the vehicle's work order, forming an initial solution; The nearest neighbor time heuristic rule is that for each vehicle's work order, the next task is selected based on the minimum time difference between the time when the current task completes unloading and the time when the next task sends a request, until all tasks are assigned to the vehicle's work order and the allocation stops, forming an initial solution; The heuristic rule for the maximum cargo demand is that for each vehicle's work order, the task with the largest amount of cargo among all task requests is selected as the next task until all tasks are assigned to the vehicle's work order, thus forming an initial solution. The improved forward insertion heuristic rule is to randomly select a task and insert it into any position of the current vehicle. If it exceeds the weight and task time limit of the current vehicle, a new vehicle is added until all tasks are arranged and all solutions are stored in the population. POP middle.

5. A multi-AGV path planning method with collision detection in a matrix workshop according to claim 4, characterized in that: In step 3, the total cost of vehicle operation is calculated as, in, In the vehicle scheduling problem, the goal is to minimize the total cost of vehicle operation. is the workstation number, k is the vehicle number, is the straight-line operating cost of the vehicle, is the turning cost of the vehicle, is the basic usage cost of each vehicle, The penalty cost for vehicles arriving overtime or early. is a binary value, such as workstation ,workstation If a vehicle passes through the channel between them, it is set to 1, otherwise it is set to 0. For workstations ,workstation The distance travelled between vehicles, For workstations ,workstation The number of turns between vehicles, is the number of vehicles departing from the garage, The latest delivery time of the vehicle. The actual delivery time of the vehicle.

6. A multi-AGV path planning method with collision detection in a matrix workshop according to claim 5, characterized in that: The rule of the vehicle merging strategy is that for all vehicles, the vehicles are sorted in descending order according to the number of tasks of each vehicle. The number of existing vehicles is , the expected number of vehicles is , the last mission attempt of the last car to the If the vehicle does not meet the load limit, power limit, and the latest time limit for the vehicle to arrive at the workstation, it will try to insert the The last vehicle is inserted into the last vehicle until the remaining tasks in the last vehicle are empty, and then the last vehicle is deleted, and the number of existing vehicles meets the expected number of vehicles; the formula for power limit is, For workstations ,workstation The vehicle power consumption between , is the basic power consumption of the vehicle, is the real-time load of the vehicle, is the load power consumption of the vehicle, It is the total power of the vehicle in one mission cycle; The formula for calculating the expected number of vehicles is: , the number of tasks that the vehicle is expected to carry Determined by the vehicle models used in the workshop.

7. A multi-AGV path planning method with collision detection in a matrix workshop according to claim 6, characterized in that: The local search operator includes a local search operator for ordinary vehicles, defining the vehicle number , vehicle number , vehicle number , Local search operators for common vehicles include, Insertion operator inside the vehicle, randomly select the vehicle Tasks in ,Task , the task Insert the tasks into the order of completion The previous position of the car changes the order of work that needs to be done by the car to form a new solution; The exchange operator inside the vehicle, randomly select the vehicle Tasks in ,Task , the task ,Task The working order is exchanged to form a new solution; Insertion operator between vehicles, randomly select vehicles ,vehicle , from the vehicle Randomly select tasks , from the vehicle Randomly select tasks , the task From vehicle Remove the task from the work order and Add to Vehicle Mission The previous position of the working sequence constitutes a new solution; The exchange operator between vehicles, randomly selects vehicles ,vehicle , from the vehicle Randomly select tasks , from the vehicle Randomly select tasks , the task ,Task The working order is exchanged to form a new solution; The inversion operator inside the vehicle, randomly select the vehicle , from the vehicle Randomly select tasks ,Task , the task ,Task The order of all the work in between is reversed to form a new solution.

8. A multi-AGV path planning method with collision detection in a matrix workshop according to claim 7, characterized in that: The steps of selecting the local search operator by the adaptive selection strategy are as follows: for each solution, record the 50 new solutions generated by each solution using the local search operator, and by judging whether the total vehicle operating cost of the new solution is less than the total vehicle operating cost of the original solution, count the number of times each local search operator successfully reduces the total vehicle operating cost, calculate the ratio of the number of successful times of each local search operator to the number of successful times of all local search operators as the probability of each local search operator being selected, and for a newly generated initial solution, use the roulette strategy to first normalize the probabilities of the 8 local search operators , 8 local search operators are assigned to each interval, and the probability of interval 1 to interval 8 being selected is initialized to 1 / 8. The probability of each local search operator is accumulated to obtain an interval range of [0,1]. Interval 1 to interval 8 correspond to the insertion operator inside the vehicle, the exchange operator inside the vehicle, the insertion operator between vehicles, the exchange operator between vehicles, the reversal operator inside the vehicle, the exchange operator inside the key vehicle, the exchange operator between the key vehicle and other vehicles, and the task removal operator in the key vehicle. Next, a random number is generated in the range of [0,1]. rand , and based on the random number rand The corresponding local search operator is selected for the interval that it falls into, and each selected local search operator is repeatedly executed. rep times, based on whether the total vehicle operation cost of the generated solution is less than the total vehicle operation cost of the original solution, count the number of times each local search operator successfully reduces the total vehicle operation cost, calculate the ratio of the number of successes of each local search operator to the number of successes of all local search operators, update the probability of each local search operator being selected, and use the updated probability of the local search operator being selected as the basis for dividing the distribution interval of each local search operator in the next adaptive selection strategy.

9. A multi-AGV path planning method with collision detection in a matrix workshop according to claim 8, characterized in that: In step 4, the vehicle information is composed of {the X-axis coordinate of the vehicle's position, the Y-axis coordinate of the vehicle's position, the X-axis coordinate of the vehicle's target position, the Y-axis coordinate of the vehicle's target position, the vehicle's driving direction, whether the vehicle is at an intersection, whether the vehicle is in a waiting state, and whether the vehicle is running}. For the evolved solution, first initialize the vehicle's initial information, update the vehicle's next target, update the vehicle's moving direction, and let each vehicle start rolling with an interval of 1s. After each vehicle operation, update the vehicle's position information, and check whether the vehicle operation will collide through the vehicle's running direction and the distance between the vehicles. When two vehicles are running in opposite directions in the same horizontal channel and the distance between the two vehicles is less than or equal to 5m, the two vehicles will collide; when two vehicles are running in opposite directions in the same longitudinal channel and the distance between the two vehicles is less than or equal to 8m, the two vehicles will collide; if a collision occurs, use the collision avoidance strategy to modify the vehicle's driving path, avoid the collision in advance, and update the vehicle's latest vehicle operating total cost. Among them, the collision avoidance strategy includes the following event processing mechanisms, The vehicles stop and wait. When a collision occurs during the operation of the detection vehicles, and the collision occurs at an intersection, one of the vehicles will stop in front of the intersection to give priority to the other vehicle; Vehicle steering avoidance: when a collision is detected during the operation of the vehicle, and the collision occurs in the channel, and the detour of the vehicle at the intersection will not increase the vehicle's running distance, the vehicle will turn at this intersection; The vehicle yields and waits. When a collision is detected during the operation of the vehicle, and the collision occurs in the channel, and the vehicle at the intersection will increase the vehicle's operating distance by detouring, the vehicle will turn right at this intersection, stop and wait for the oncoming vehicle to pass the intersection, and then resume operation.

10. A multi-AGV path planning method with collision detection in a matrix workshop according to claim 9, characterized in that: In step 5, for the number of evolutions exceeding the hysteresis For solutions without evolution, a ranking-based restart strategy is used. The top 20% solutions are retained according to the order of the total cost of vehicle operation of the solutions. For the top 20%-60% solutions, the key vehicles in the solutions are retained during the solution regeneration process, and the improved forward insertion heuristic rule is used to randomly generate the working order of other vehicles. For solutions ranked after 60%, the improved forward insertion heuristic rule is used to randomly generate the working order of all vehicles during the solution regeneration process.

Images