An evolutionary computation-based dynamic path multi-agv charging pile site selection optimization method and system

Abstract

The application discloses an evolutionary computing-based dynamic path multi-AGV charging pile site selection optimization method and system. The method comprises the following steps: determining a plurality of workstation positions as candidate charging pile arrangement points, and representing whether each position is arranged with a charging pile as a binary decision variable; based on a multi-objective optimization model, taking the minimization of the number of charging piles, the maximization of charging pile coverage and utilization as the target, under the constraint conditions of meeting at least one charging pile arrangement, energy accessibility and full coverage, an evolutionary algorithm is used to iteratively optimize the initial population, and a charging pile layout scheme suitable for multiple sets of AGV operation routes is generated; through multi-scenario energy constraint verification and a self-adaptive repair mechanism, it is ensured that the layout scheme still meets the AGV charging demand under the route change; finally, the scheme quality is further improved through local optimization and simulated annealing strategy. The application can realize the comprehensive optimization of the economy, coverage balance and system robustness of the charging pile layout.

Description

Technical Field

[0001] This invention relates to the field of automated logistics technology, specifically to a dynamic path multi-AGV charging pile location optimization method and system based on evolutionary computation, and more particularly to a charging pile layout planning method based on multi-objective optimization and constraint satisfaction under the condition that the Automated Guided Vehicle (AGV) running path has multiple scenario changes. Background Technology

[0002] In automated terminals, Automated Guided Vehicles (AGVs) are key equipment for horizontal container transport. Due to battery capacity limitations, AGVs need to be recharged in a timely manner during operation to avoid vehicle delays and congestion caused by insufficient power or path conflicts, thus ensuring transportation efficiency and system safety.

[0003] Currently, charging methods for AGVs mainly include battery swapping, opportunistic charging, and a hybrid "battery swapping + opportunistic charging" mode. Battery swapping requires the construction of dedicated swapping stations, which is costly, and AGVs need to be taken offline for battery swapping, affecting operational continuity. Opportunistic charging, on the other hand, allows AGVs to charge at workstations during task breaks, facilitating uninterrupted operation and thus becoming a focus of research and application. Under opportunistic charging, the location and layout of charging stations directly affect the charging accessibility of AGVs, system operating efficiency, and infrastructure costs.

[0004] Most existing research on charging pile site selection optimization is based on the assumption of fixed paths and static process sequences, using methods such as binary coding to optimize energy balance in a single operating scenario. However, in actual flexible production environments, AGV operating routes are dynamically adjusted according to tasks, and the dwell time at each workstation fluctuates. Changes in path length and task load further increase energy consumption uncertainty. This makes it difficult for charging pile deployment schemes based on fixed path optimization to adapt to multi-scenario and dynamically changing actual working conditions when switching to other operating routes, as problems such as AGV battery power falling below the safety threshold, insufficient charging pile coverage, or uneven utilization may occur.

[0005] Furthermore, this problem is essentially a multi-objective combinatorial optimization problem with multiple constraints, and the solution space grows exponentially with the number of candidate points. Existing methods often struggle to meet the energy constraints and full coverage requirements of AGVs under all possible operating routes while ensuring economy (minimum number of charging piles), accessibility (sufficient coverage), and resource balance (high utilization). In particular, they lack systematic modeling and robust optimization mechanisms for path variability.

[0006] Therefore, how to design a charging pile site selection method that can simultaneously optimize economic costs, coverage performance, and resource utilization, and ensure energy safety in all scenarios, in a flexible environment where AGV operating paths are uncertain and dwell times fluctuate, has become a key technical problem that urgently needs to be solved in the scheduling and planning of automated terminals. Summary of the Invention

[0007] To address the above technical problems, this invention proposes a dynamic path multi-AGV charging pile location optimization method based on evolutionary computation, comprising the following steps: S1. Input a set of q different automated guided vehicle (AGV) operation route scenarios. Each route contains an ordered workstation access sequence and the corresponding AGV dwell time. Input the AGV's equipment parameters, including battery capacity, energy consumption rate per unit distance, safe power threshold MinCharge%, and charging power of the charging pile. Set the core parameters of the algorithm. S2. Encode the positions of the m candidate workstations into binary decision vectors. ,in This indicates that a charging station is deployed at the i-th workstation. It indicates that no charging piles will be deployed; a multi-objective optimization function is constructed, the objective function includes minimizing the number of charging piles f1(X), minimizing the farthest unreachable distance to maximize coverage f2(X), and maximizing the average utilization rate of charging piles f3(X); the constraints include: at least one charging pile must be deployed, energy constraint g2: the automated guided vehicle must be able to travel back and forth between any two charging piles without falling below the minimum energy threshold, and full coverage constraint g3: all transportation paths are covered by charging piles; S3. Randomly generate N binary vectors of length m that satisfy at least one charging pile constraint as the initial population; S4. Global optimization based on genetic algorithm: the population is iterated through multiple generations. In each generation, fitness score evaluation, selection, crossover and mutation operations are performed, and the resulting offspring individuals are checked for energy constraints in multiple scenarios. If the check fails, an adaptive repair mechanism is triggered. The specific operation of constraint verification and repair is as follows: For the offspring individuals generated through crossover and mutation, they are sequentially substituted into each set of route scenarios to perform energy trajectory simulation and constraint checks; if the energy reachability constraint or full coverage constraint is violated under any route, an adaptive repair mechanism is triggered: ① Locating the breach point: The algorithm locates the route. The specific location where the energy is depleted (i.e., the energy level first falls below the threshold); ② Perform repair: Based on the principle of "minimum movement cost", find the nearest feasible location among the candidate workstations near the default point that is not currently equipped with a charging pile, and change the value of its decision variable x from 0 to 1 (i.e., add a charging pile here). ③ Iteration and Verification: After completing the above repairs, resubmit the modified scheme X' to all route scenarios for a complete constraint verification. If a violation is still found under any route, repeat steps ① and ② until the fixed layout scheme X can simultaneously satisfy the energy constraint g2 and the full coverage constraint g3 under all preset route scenarios.

[0008] S5. Based on local search, fine optimization is performed on the solution set obtained by global optimization. The 2-opt or Or-opt local search operator is used to perturb the solution set. The simulated annealing criterion is combined to accept inferior solutions in order to escape local optima. At the same time, multi-scenario constraint verification and repair are performed on the new solutions again. The 2-opt local search operation is as follows: randomly select two workstation indices i and j, and exchange their charging pile layout status x. i With x j If the new solution satisfies the constraints and the fitness score is improved or accepted according to the simulated annealing probability, then update the current solution; The Or-opt local search operation involves selecting three or more consecutive workstation subsequences, rotating their charging pile deployment status, generating new solutions, and evaluating their feasibility and fitness scores.

[0009] S6. From the feasible solutions that satisfy all hard constraints, prioritize the solution with the fewest number of charging piles. If the number of piles is the same, select the solution with the best overall coverage and utilization rate as the final charging pile layout solution.

[0010] Furthermore, the core algorithm parameters in step S1 include population size N, maximum number of generations T1, and crossover probability P. c Probability of mutation P m .

[0011] Furthermore, the energy constraint g2(X) in step S2 is expressed as: the path distance between any two adjacent charging piles i and j ,satisfy: ,like ,but In the formula, d ij This represents the path distance between workstations i and j when charging piles are placed at workstations i and j; r c C represents the energy consumption per unit distance of the automated guided vehicle (AGV); C represents the battery capacity of the AGV; MinCharge% represents the minimum safe charge percentage.

[0012] Coverage constraints ,satisfy: In the formula, Let (i,j) represent the path segment (i,j) between any two workstations in set P, where P is the set of automated guided vehicle (AGV) transport paths. This indicates that there exists at least one position k; This indicates that a charging station has indeed been installed at location k; This represents the distance from path segment (i,j) to the charging station k. It must be less than or equal to the maximum allowable distance d max ;d max This refers to the maximum distance that an automated guided vehicle (AGV) can travel using its remaining battery power without consuming the safety stock (MinCharge%).

[0013] Furthermore, in step S2, The The function is defined as the number of charging stations, i.e. The The function is defined as the maximum distance from all transportation task route segments to their nearest charging station, i.e. In the formula, d ij,k It is the distance from any point on the path (i,j) between workstation i and workstation j to charging pile k.

[0014] The The function is defined as the average utilization rate of charging piles, i.e. In the formula, A is the average charging time of the automated guided vehicle, and T is the total charging time.

[0015] Furthermore, the fitness evaluation in step S4 employs the NSGA-II framework of the non-dominated sorting genetic algorithm. Through non-dominated sorting and crowding calculation, individuals are ranked in multiple objectives to obtain the total crowding distance of each individual in the multi-objective optimization function dimension across multiple non-dominated ranks, and the fitness score is calculated. For individual x, the non-dominant rank Let x be the total crowding distance for individual x. It is the weighting coefficient.

[0016] Furthermore, the selection operation in step S4 adopts a roulette wheel selection method.

[0017] This invention also relates to an automated terminal charging pile site selection optimization system for flexible paths, comprising: The data input module is used to receive the autonomous guided vehicle's operating route scenario, equipment parameters, and algorithm parameters; The model building module is used to establish multi-objective optimization models and constraints. The global optimization module is used to perform population iteration and multi-scenario constraint verification based on genetic algorithms. The local optimization module is used to further optimize the feasible solution through local search and simulated annealing. The solution output module is used to output the optimal charging pile layout solution based on the comprehensive optimization principle.

[0018] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the steps of any of the methods described above.

[0019] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described herein.

[0020] Compared with the prior art, the beneficial effects of the present invention are: This invention constructs a multi-objective optimization model that includes the number of charging piles, coverage, and utilization rate, and uses an evolutionary algorithm to solve it. In a single optimization process, it can simultaneously balance infrastructure costs, service guarantee capabilities, and resource balance to obtain a series of Pareto optimal solutions, providing decision-makers with a variety of layout options that take into account both economy and efficiency.

[0021] To address the limitations of existing methods that are mostly based on fixed paths, this invention introduces a multi-scenario verification mechanism during the optimization process, mandating that candidate layout schemes meet energy constraints and full coverage requirements under all preset AGV operating routes. This "stress test" mechanism ensures that the final scheme remains reliable even when the actual path changes dynamically, effectively avoiding the risk of AGV power outages and shutdowns caused by scenario switching.

[0022] This invention designs a three-layer feasibility control and adaptive repair mechanism. When crossover, mutation, or local optimization produces infeasible solutions that violate constraints, it can quickly locate the breach points and repair the solutions with a minimum-cost strategy (such as key point staking). This not only maintains the feasibility of the population and avoids invalid computation, but also protects superior gene fragments, guides the search towards promising feasible regions, and improves the convergence speed and stability of the algorithm.

[0023] This invention employs a two-stage optimization framework. The first stage uses an evolutionary algorithm for global exploration, while the second stage introduces local search operators such as 2-opt and Or-opt, combined with simulated annealing criteria, for deep optimization. This strategy helps to escape local optima and improve the structure of solutions at a finer scale, thereby obtaining higher-quality and more practical charging pile layout schemes.

[0024] By imposing strict energy constraints (such as minimum power thresholds) and full coverage constraints, this invention first ensures the absolute safety of AGV operation. Building on this, by maximizing average utilization, it promotes the shared deployment of charging stations in key transportation hubs, reducing redundant station deployment and minimizing the lifespan loss caused by frequent shallow charging and discharging of batteries. This achieves synergistic optimization of safety, efficiency, and equipment lifespan at the system level. Attached Figure Description

[0025] Figure 1 A schematic diagram of the site selection and layout for charging piles; Figure 2 This is a comparison diagram of the energy profiles of routes A and B. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the present invention clearer, the implementation methods of the present invention will be described in detail below with reference to specific embodiments.

[0027] The overall process of this invention includes three main stages: model building and parameter setting, global optimization based on evolutionary algorithms, and fine optimization based on local search.

[0028] A dynamic path multi-AGV charging pile location optimization method based on evolutionary computation includes the following steps: S1. Parameter Setting and Data Input Based on the specific conditions of the automated terminal, set the parameters, including: Scenario parameters: Input a set of q different AGV running route scenarios. Each route It contains an ordered sequence of workstation visits. Simultaneously, it includes the dwell time of the input AGV at each workstation i in scenario q. Equipment parameters: Input AGV battery capacity C (unit: energy units), energy consumption rate per unit distance. (Unit: Energy Unit / meter), Charging Power of Charging Station (Unit: Energy Units / Minute), set the safe power threshold MinCharge% (set to 10% in this embodiment). Core algorithm parameters: set the population size N for the genetic algorithm (set to 100 in this embodiment), the maximum number of generations T1 (set to 50 in this embodiment), and the crossover probability P. c(Set to 0.8 in this embodiment), mutation probability P m (Set to 0.05 in this embodiment). Set the maximum number of iterations T2 for the local optimization phase.

[0029] S2. Define the decision space and objective model. 1. Definition of decision variables: Encode the locations of the m candidate workstations into a binary decision vector. X represents a specific charging pile layout scheme, where each... This indicates whether a charging station is placed on the i-th workstation. This indicates that a charging station is deployed at the i-th workstation. This indicates that no charging piles will be deployed. Locations where charging piles cannot be deployed are not considered, and this decision applies to all scenarios. Since there are m locations where charging piles can be deployed, there are 2m solutions in the solution space. However, due to the constraint that charging piles must be deployed at at least one workstation, the number of solutions in the solution space becomes 2m-1.

[0030] 2. Construction of the objective function: Constructing a multi-objective function The aim is to simultaneously (optimize) minimize three objectives.

[0031] All statements written in a multi-objective function (optimizing three objectives simultaneously, the smaller the better) use "-" , "To indicate that the smaller the better."

[0032] a, Number of charging stations (the fewer the better) This reflects the system cost and space usage.

[0033] b、 Accessibility / coverage of charging stations (ensuring that the AGV can charge while working): This measures whether the AGV can easily access charging stations on any transport path. It can be measured by the distance from each transport path to the nearest charging station, and the shorter the distance, the better.

[0034] Formulaic representation: Where (i,j) represents the transportation task between workstations, and d ij,k It is the distance from any point on the path between workstation i and workstation j of the AGV to the charging pile k. It is any point on the path between workstation i and workstation j to each charging station. The minimum reachable distance among the shortest distances, It is all The maximum reachable distance. A smaller value indicates a shorter reachable distance in the worst-case scenario, resulting in better coverage. In minimizing... It maximizes coverage and ensures that all transportation routes have charging support.

[0035] c. Average utilization rate of charging stations (reflecting resource balance) (the higher the better) Wherein, the average charging time of the AGV is A, and the total charging time is T.

[0036] However, since optimization problems are all about minimizing objectives, this embodiment is written as − That is, to maximize utilization.

[0037] 3. Definition of constraints and limitations: a. g1(X): There must be at least one charging station. To avoid an invalid solution where "no workstations are set up".

[0038] b, g2(X): Energy constraint (reachability constraint) Ensure the AGV can safely reach the next charging station on its path from the current charging station, while the remaining battery power is not lower than the minimum battery power threshold MinCharge%. ,like ,but Where: d ij This represents the path distance between workstations i and j when charging piles are placed at workstations i and j; r c The value is the energy consumption per unit distance of the AGV; C is the battery capacity of the AGV; MinCharge% represents the minimum safe charge percentage.

[0039] This constraint ensures that the AGV will not stop midway due to insufficient power.

[0040] c. (Discrete constraints) This is the definition of a binary decision variable, indicating that each workstation can only "place" or "not place" stubs, with no intermediate state.

[0041] d. Charging coverage constraint Ensure that every one-way transportation route has at least one accessible charging station; that is, for each one-way transportation route in the system, there must be a charging station accessible along its route (distance not exceeding d). max Within the specified area, find at least one installed charging station, achieving 100% coverage: In the formula, Let (i,j) represent the path segment (i,j) between any two workstations in set P, where P is the set of AGV transportation path segments between workstations; This indicates that there exists at least one position k; This indicates that a charging station has indeed been installed at location k; This represents the distance from path segment (i,j) to the charging station k. It must be less than or equal to the maximum allowable distance d max ;d max This refers to the maximum distance an AGV can travel using its remaining available power without consuming its safety stock (MinCharge%).

[0042] This method is used to intelligently plan the layout of charging piles for automated guided vehicles (AGVs) in complex workshop environments or automated terminals. It takes into account multiple optimization objectives such as minimizing the number of charging piles, minimizing the total travel distance of AGVs, and maximizing the utilization rate of charging piles, while also meeting constraints such as energy round trip and safe distance.

[0043] S3, Initialize the population N random sequences of individuals, each of length m consisting of combinations of 0s and 1s, are generated as workstation number sequences. These individuals together form the initial population. Each individual sequence is represented as in: Each feasible solution X represents a charging pile layout scheme. Index i represents a unique number of the process, and the value stored at that index position... Should a charging station be installed here? To ensure diversity, heuristic information can be added, such as requiring at least one gene to be 1 to satisfy g1.

[0044] For example: randomly initialize a solution set. X1 matrix: X1={0101000110011010110011010} S4. Global optimization based on genetic algorithm (running in a loop until generation T1 is reached) S4.1 Fitness evaluation: For population P t For each individual X in the simulation, multiple scenarios are performed. Specifically, X is substituted into each route scenario. The simulation demonstrates the process of an AGV traveling along a route and stopping at a workstation equipped with a charging station to recharge. The results are recorded and calculated for this solution in all scenarios. , , The overall fitness score for each individual is calculated using non-dominated ranking and crowding calculation (such as the NSGA-II algorithm).

[0045] NSGA-II is a multi-objective evolutionary algorithm that uses non-dominated sorting to categorize stub placement schemes based on their overall performance, prioritizing individuals that are balanced across "reducing stub count," "improving coverage," and "balanced utilization." Simultaneously, it utilizes congestion calculations to eliminate mediocre schemes with overlapping objectives, ensuring diversity in the solution set across different priorities. This allows for the selection of the optimal layout set that can accommodate various flexible path scenarios during parent-child merging. The NSGA-II algorithm includes: Non-dominated sorting: Traverse all individuals in the population and find the set of individuals that are not dominated by any other individual, and classify them into the first non-dominated layer (Rank = 1). This layer constitutes the Pareto optimal frontier of the current population, representing the set of staking schemes with the best overall performance under multiple objective trade-offs; if two schemes are not mutually dominant, that is, each is dominant on different objectives, they are said to be mutually non-dominated.

[0046] Recursive ranking: Remove individuals with assigned Rank values ​​from the current set to be sorted, repeat the non-dominated sorting on the remaining individuals, identify all schemes that are not dominated by the remaining individuals again, assign them Rank = 2, and continue iterating until all individuals are assigned a Rank.

[0047] Priority ranking: All individuals are arranged in ascending order of their Rank value. Individuals with lower Rank values ​​have better overall performance and the highest priority in evolutionary selection.

[0048] Crowding distance is used for same-level screening: For individuals within the same Rank layer, their crowding distance in the target dimension is further calculated. By measuring the sum of the crowding distances of each individual with its neighboring individuals in all target dimensions, the sparser distribution is prioritized to prevent excessive clustering of the solution set and ensure that the final result has diverse representativeness.

[0049] Crowding distance calculation method: 1. Initialization For all n individuals in the same Rank level, first determine their crowding distance L[ Set all values ​​to 0.

[0050] 2. Calculate iteratively according to the target dimension. For the three objective functions f1, f2, and f3 in the model, perform the following operations respectively: Sorting: Sort the individuals in this layer in ascending order according to the numerical value of the current target dimension.

[0051] Boundary handling: After sorting, individuals at both ends of this dimension (maximum and minimum values) have an infinite crowding distance. This is done to ensure that extreme solutions at the boundary have the highest retention priority during evolution, thereby expanding the search range.

[0052] Accumulate the middle individuals: For individuals ranked in the middle Calculate the distance difference between two adjacent individuals in the current dimension.

[0053] Calculation method: Subtract the target value of the previous individual from the target value of the next individual, then divide by the total range of variation of that target value (maximum value minus minimum value), and sum the resulting normalized distances to L[ ]middle.

[0054] 3. Summary of Results After all target dimensions (f1, f2, f3) have been traversed, the final L[ for each individual is obtained.] [This refers to the distance of its congestion level.]

[0055] Numerical meaning: L[ The larger the value of ], the more "isolated" the individual is in the target space (few surrounding options), and the stronger its representativeness. Therefore, it has a greater advantage in the same-layer screening in S4.1 and the selection in S4.2.

[0056] Calculation formula expression For individuals The total congestion distance under the three objectives can be expressed as: The overall fitness score is: It is a very small weighting coefficient, ensuring that the total crowding distance only plays a decisive role when the Ranks are the same; It is the total congestion distance; S4.2 Selection Operation: Based on the overall fitness score of each individual, a roulette wheel selection method is used to select from the current population P. t When selecting individuals, individuals with higher overall fitness scores are more likely to be selected.

[0057] Roulette wheel selection: The probability of each individual being selected is directly proportional to its overall fitness score as a percentage of the total population's overall fitness score. Individuals with higher overall fitness scores occupy a larger sector on the roulette wheel, thus increasing their probability of being selected by the random number pointer. The specific operation process is as follows: Step 1. Calculate the total overall fitness score: First, calculate the sum of the overall fitness scores of all individuals in the current population. .

[0058] Step 2. Calculate the selection probability: For each individual, calculate the probability of being selected. .

[0059] Step 3. Calculate the cumulative probability: Sum the selection probabilities of all individuals to obtain a cumulative probability distribution. Each individual corresponds to a cumulative probability value in the interval [0,1].

[0060] Step 4. Generate a random number: Generate a random number in the interval [0,1].

[0061] Step 5. Select an individual: Determine which individual's cumulative probability interval the random number falls into; that individual is then selected.

[0062] Step 6. Repeat: Repeat steps 4 and 5 until a sufficient number of individuals are selected to enter the next generation's breeding pool.

[0063] S4.3 Crossover operation: From the current population P t Paired individuals serve as parents P1 and P2, with a crossover rate of... Perform a crossover operation. During the crossover process, new results are generated by swapping the current status of whether charging stations are installed at two different locations. Since the workstation locations are already determined, the crossover process will focus on swapping and optimizing the locations of charging stations.

[0064] S4.4 Mutation Operation: For each offspring generated after crossover, the mutation rate is used to determine the mutation percentage. Perform a swap mutation operation. This operation randomly selects several positions in the individual's encoding array and changes the values ​​at these positions (e.g., from 0 to 1), thus changing whether a charging station is installed at that location.

[0065] S4.5 Constraint Verification and Repair: For each new individual (stub placement scheme X) in the offspring population Qt, perform multi-scenario hard constraint detection. The method is: sequentially substitute the stub placement scheme X into each set of preset variable route scenarios. Perform energy trajectory simulation.

[0066] If the test is passed, the individual will be retained; If the detection fails (e.g., in the route) During the simulation, if the remaining power of the AGV falls below the safety threshold MinCharge (e.g., 10%) for the first time before arriving at a certain workstation, the algorithm determines that the layout scheme is unqualified in the current scenario and immediately triggers the adaptive repair mechanism.

[0067] Core logic of the repair mechanism: ① Locating the breach point: The algorithm locates the route. The specific location where the energy is depleted (i.e., the energy level first falls below the threshold); ② Perform repair: Based on the principle of "minimum movement cost", find the nearest feasible location among the candidate workstations near the default point that is not currently equipped with a charging pile, and change the value of its decision variable x from 0 to 1 (i.e., add a charging pile here). ③ Iteration and Verification: After completing the above repairs, resubmit the modified scheme X' to all route scenarios for a complete constraint verification. If a violation is still found under any route, repeat steps ① and ② until the fixed layout scheme X can simultaneously satisfy the energy constraint g2 and the full coverage constraint g3 under all preset route scenarios.

[0068] S4.6 Population Update: The repaired and constrained offspring population Q is updated. t The next generation population is generated by merging with the parent population Pt. .

[0069] S4.7 Iterative Judgment: Determine whether the current evolutionary generation t has reached T1. If not, t = t + 1, return to step S4.1, and start a new round of iterative evolution; if it has, enter the local optimization stage.

[0070] S5. Fine-grained optimization based on local search (running in a loop until T2 generation is reached) S5.1. Use the 2-opt and Or-opt operators to perform local optimization operations on the current solution.

[0071] 2-opt: Select the sequence indices i,j of two pairs of workstations. Swap the installation status of the two pairs of workstations with the charging piles (0 or 1) to form a new path. Calculate the overall fitness score after the change; if it improves, accept the change.

[0072] Or-opt: Select a sequence of three or more charging station locations. Rotate the installation options (0 or 1) to create new installation possibilities. Calculate the overall fitness score after the change; if it improves, accept the change.

[0073] S5.2 Constraint Verification and Repair: For the new scheme generated by local optimization, perform multi-scenario constraint verification and repair as described in step S4.5 again. That is, if the new allocation scheme violates any preset constraint relationship, change the position of the charging pile. If changing the position does not work, increase the number of charging piles appropriately and then optimize.

[0074] Appropriately increasing the number of charging piles means that when a local optimization scheme violates constraints and cannot be resolved by adjusting the location, the system, based on the principle of minimum movement cost, automatically finds the nearest un-deployed workstation near the default point where the AGV's power first falls below the threshold and changes its status from 0 to 1 (i.e., adds a charging pile). Then, it re-substitutes all scenarios for verification and repeats this process until the scheme can simultaneously meet the energy accessibility and full coverage requirements of all route scenarios.

[0075] S5.3. Select the optimal result from the locally optimized results according to the simulated annealing criterion. Then, compare the resulting offspring population Qt, which satisfies the constraints, with the parent population P. t Merge to generate the next generation population .

[0076] Simulated annealing is an intelligent search mechanism that allows algorithms to "probabilistically accept inferior solutions": during local search, when a new layout is inferior to the current one, the algorithm does not discard it directly, but calculates a probability based on the current "temperature" to decide whether to accept it, thus avoiding getting stuck in local optima. As iterations proceed and the temperature gradually decreases, the algorithm shifts from "extensive exploration" in the early stages to "stable convergence" in the later stages, thereby satisfying d... max Under the premise of hard constraints, the globally optimal charging pile layout is obtained.

[0077] S5.4 Iteration Judgment: Determine whether the local optimization iteration has reached T2. If not, let t = t + 1 and return to S5.1 to start a new round of iterative evolution; if it has reached T2, the algorithm terminates and enters the output stage.

[0078] S6, Output the final solution From the final population, based on the comprehensive optimization principle (first, satisfying all hard constraints; second, prioritizing the scheme with the fewest charging piles f1; if the number of charging piles is the same, then selecting the scheme with the highest comprehensive evaluation of coverage rate f2 and utilization rate f3), the historically optimal number of charging piles and their placement locations recorded throughout the entire evolution process are selected and output. The charging pile planning scheme corresponding to this individual is the final optimization result. This scheme plans a complete, feasible, and highly balanced optimal layout and process allocation scheme at the physical level, which can take into account multiple operating conditions.

[0079] Based on the methods described in the above embodiments, an example is given: Table 1 System parameter assumptions: Table 2 Location Coordinates: Route A: 1-2-3-4-5-6-7-8-9-10-11-12-13-14-15-16-17-18-19-20-1 Route B: 1-2-3-14-15-20-19-18-13-12-11-10-6-5-1 Charging station installation and status at various points: One random scenario: 2, 5, 8, 11, 14, 16, 19, 20 (8 charging stations installed). Table 3 Route A: Table 4 Route B: Both routes A and B meet the requirements, but in route A, the minimum charge level = 24 (when arriving at location 17) ≥ 8 → is safe, but the number of charging and discharging cycles is high, which wastes power, damages battery life, and requires more charging stations, resulting in higher costs.

[0080] After optimization: 1, 5, 10, 15, 19 (5 charging stations installed) Table 5 Route A: Table 6 Route B: Both routes A and B meet the conditions, and the lowest battery level in the route A table is 8 (when the location 15 is reached) ≥ 8 → safe, fewer charging and discharging cycles, and more evenly distributed locations.

[0081] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dynamic path multi-AGV charging pile location optimization method based on evolutionary computation, characterized in that, Includes the following steps: S1. Input contains a set of q different automated guided vehicle (AGV) route scenarios. Each route contains an ordered workstation access sequence and the corresponding AGV dwell time. Input the equipment parameters of the automated guided vehicle, including battery capacity, energy consumption rate per unit distance, safe power threshold MinCharge%, and charging power of the charging pile; set the core parameters of the algorithm; S2. Encode the positions of the m candidate workstations into binary decision vectors. ,in This indicates that a charging station is deployed at the i-th workstation. It indicates that no charging piles will be deployed; a multi-objective optimization function is constructed, the objective function includes minimizing the number of charging piles f1(X), minimizing the farthest unreachable distance to maximize coverage f2(X), and maximizing the average utilization rate of charging piles f3(X); the constraints include: at least one charging pile must be deployed, energy constraint g2: the automated guided vehicle must be able to travel back and forth between any two charging piles without falling below the minimum energy threshold, and full coverage constraint g3: all transportation paths are covered by charging piles; S3. Randomly generate N binary vectors of length m that satisfy at least one charging pile constraint as the initial population; S4. Global optimization based on genetic algorithm, the population is iterated through multiple generations, and fitness score evaluation, selection, crossover and mutation operations are performed in each generation, and energy constraint verification is performed on the offspring individuals in multiple scenarios. If the verification fails, an adaptive repair mechanism is triggered; S5. Based on local search, fine optimization is performed on the solution set obtained by global optimization. The 2-opt or Or-opt local search operator is used to perturb the solution set. The simulated annealing criterion is combined to accept inferior solutions in order to escape local optima. At the same time, multi-scenario constraint verification and repair are performed on the new solutions again. S6. From the feasible solutions that satisfy all hard constraints, prioritize the solution with the fewest number of charging piles. If the number of piles is the same, select the solution with the best overall coverage and utilization rate as the final charging pile layout solution.

2. The method as described in claim 1, characterized in that, The specific operation of constraint verification and repair is as follows: For the offspring individuals generated through crossover and mutation, they are sequentially substituted into each set of route scenarios to perform energy trajectory simulation and constraint checks; if the energy reachability constraint or full coverage constraint is violated under any route, an adaptive repair mechanism is triggered: ① Locating the breach point: The algorithm locates the route. The exact location at which the battery level first falls below the minimum battery threshold; ② Remediation: Based on the principle of minimum movement cost, among the candidate workstations near the default point, find the nearest feasible location that is not currently equipped with a charging pile, and change the value of its decision variable x from 0 to 1, that is, add a charging pile there; ③ Iteration and verification: After completing the above repairs, the modified scheme X' is substituted into all route scenarios for complete constraint verification. If a violation is still found under any route, steps ① and ② are repeated until the fixed layout scheme X can simultaneously satisfy the energy constraint g2 and the full coverage constraint g3 under all preset route scenarios. The 2-opt local search operation is as follows: randomly select two workstation indices i and j, and exchange their charging pile layout status x. i With x j If the new solution satisfies the constraints and the fitness score is improved or accepted according to the simulated annealing probability, then update the current solution; The Or-opt local search operation involves selecting three or more consecutive workstation subsequences, rotating their charging pile deployment status, generating new solutions, and evaluating their feasibility and fitness scores.

3. The method as described in claim 1, characterized in that, The core parameters of the algorithm in step S1 include population size N, maximum number of generations T1, and crossover probability P. c Probability of mutation P m .

4. The method as described in claim 1, characterized in that, The energy constraint g2(X) in step S2 is expressed as: the path distance between any two adjacent charging piles i and j ,satisfy: ,like ,but In the formula, d ij This represents the path distance between workstations i and j when charging piles are placed at workstations i and j; r c C represents the energy consumption per unit distance of the automated guided vehicle; C represents the battery capacity of the automated guided vehicle; MinCharge% represents the minimum safe charge percentage. Coverage constraints ,satisfy: In the formula, Let (i,j) represent the path segment (i,j) between any two workstations in set P, where P is the set of automated guided vehicle (AGV) transport paths. This indicates that there exists at least one position k; This indicates that a charging station has indeed been installed at location k; This represents the distance from path segment (i,j) to the charging station k. It must be less than or equal to the maximum allowable distance d max ;d max This refers to the maximum distance that an automated guided vehicle (AGV) can travel using its remaining battery power without consuming the safety stock (MinCharge%).

5. The method according to claim 1, characterized in that, In step S2, The The function is defined as the number of charging stations, i.e. The The function is defined as the maximum distance from all transportation task route segments to their nearest charging station, i.e. In the formula, d ij,k It is the distance from any point on the path (i,j) between workstation i and workstation j to charging pile k; The The function is defined as the average utilization rate of charging piles, i.e. In the formula, A is the average charging time of the automated guided vehicle, and T is the total charging time.

6. The method as described in claim 1, characterized in that, The fitness score evaluation in step S4 adopts the NSGA-II framework of the non-dominated sorting genetic algorithm. Individuals are ranked in multiple objectives through non-dominated sorting and crowding calculation, obtaining the total crowding distance of each individual in the multi-objective optimization function dimension across multiple non-dominated ranks, and then the fitness score is calculated. For individual x, the non-dominant rank Let x be the total crowding distance for individual x. It is the weighting coefficient.

7. The method as described in claim 1, characterized in that, The selection operation in step S4 adopts a roulette wheel selection method.

8. A dynamic path multi-AGV charging pile location optimization system based on evolutionary computation, implementing the method as described in any one of claims 1 to 7, characterized in that, include: The data input module is used to receive the autonomous guided vehicle's operating route scenario, equipment parameters, and algorithm parameters; The model building module is used to establish multi-objective optimization models and constraints. The global optimization module is used to perform population iteration and multi-scenario constraint verification based on genetic algorithms. The local optimization module is used to further optimize the feasible solution through local search and simulated annealing. The solution output module is used to output the optimal charging pile layout solution based on the comprehensive optimization principle.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images