Unmanned sanitation vehicle path planning method and system based on improved NSGA-II

Abstract

The invention relates to the technical field of vehicle path planning, and provides an unmanned sanitation vehicle path planning method and system based on improved NSGA-II, efficient acquisition of path cost is realized by constructing task and road section mapping and a node pair shortest distance matrix, and the path cost is acquired by combining greedy path simulation and an NSGA-II optimization algorithm. A multi-target optimization model with the minimum vehicle deployment number, the minimum system total energy consumption and the minimum uncovered task area proportion as the core is established, in the solving stage, a K-Means clustering initialization strategy is introduced to accelerate convergence, and in combination with dynamic starting point determination and greedy path simulation of a dynamic truncation mechanism, on the premise of not relying on a fixed garage, a multi-target optimization model is established; and performing multi-dimensional optimization by using an improved NSGA-II algorithm. Through the mechanism, the resource investment and operation coverage rate of the unmanned sanitation vehicle can be collaboratively optimized, and the path planning efficiency and the solution feasibility are improved.

Description

Technical Field

[0001] This invention relates to the field of vehicle route planning technology, specifically to a route planning method and system for unmanned sanitation vehicles based on the improved NSGA-II. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] With the rapid development of information technology, artificial intelligence, and autonomous driving technology, smart city construction is increasingly becoming an important means to improve urban governance capabilities and operational efficiency. Unmanned sanitation vehicles, including unmanned sweepers, unmanned water trucks, and unmanned garbage collection vehicles, are various municipal service vehicles capable of autonomous operation. As a core component of the smart sanitation system, they possess advantages such as continuous operation, high energy efficiency, and intelligent scheduling. They have been piloted in several cities, demonstrating significant potential to replace traditional manual labor. In large-scale urban road networks, the complexity and scope of road sanitation tasks increase significantly, involving multiple factors such as operating time windows, range limitations, and uneven task distribution, necessitating scientific and efficient facility deployment and route scheduling strategies. Therefore, systematic optimization research on the facility spatial deployment and route planning of unmanned sanitation vehicles under large-scale road network conditions has significant practical significance and is urgently needed.

[0004] However, existing technologies still have several shortcomings in addressing the aforementioned problems. First, their optimization objectives are too singular, failing to balance resource scheduling rationality with system energy consumption control. Some optimization models focus only on the total distance of vehicle paths or the completion time of operations, neglecting to incorporate the key resource parameter of vehicle quantity into the overall optimization objective. This leads to redundancy risks in vehicle configuration or an inability to effectively reduce energy consumption costs. Second, current optimization solutions lack optimization strategies that can simultaneously consider global exploration and local rationality, easily leading to low-quality or infeasible solutions. Among existing optimization methods, one type emphasizes global search capabilities but suffers from poor solution feasibility, while another emphasizes local greedy decision-making but is prone to getting trapped in local optima, making it difficult to consistently obtain high-quality path planning solutions. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes a path planning method and system for unmanned sanitation vehicles based on an improved NSGA-II algorithm. By constructing a task-road segment mapping and a shortest distance matrix for node pairs, efficient path cost acquisition is achieved. A three-objective optimization model, including vehicle quantity, system energy consumption, and coverage rate, is established by combining greedy path simulation and the NSGA-II optimization algorithm. Furthermore, an improved NSGA-II algorithm based on K-Means clustering initialization and dynamic truncation of coverage threshold is proposed. This method, without relying on a fixed starting point, collaboratively optimizes the number of unmanned sanitation vehicles, system energy consumption, and cleaning area, improving the convergence speed and practicality of the solution in path planning.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: One or more embodiments provide a path planning method for unmanned sanitation vehicles based on the improved NSGA-II, including the following steps: Obtain information on the work tasks and vehicles, discretize the road segments corresponding to the work tasks, establish a mapping between tasks and road segments, calculate the shortest distance between the discretized road nodes, and construct a shortest distance matrix for all node pairs. A multi-objective path planning model is constructed with the optimization objectives of minimizing the number of sanitation vehicles deployed, minimizing the total energy consumption of the system, and minimizing the proportion of uncovered task areas. Based on the spatial location characteristics of road segments, a clustering algorithm is used to generate an initial population for vehicle task allocation. The greedy path simulation is combined with the NSGA-II optimization algorithm to obtain an improved NSGA-II optimization algorithm. Based on the mapping relationship between tasks and road segments and the shortest distance matrix of all nodes, the constructed multi-objective optimization model is solved to obtain the path planning scheme for vehicles to perform tasks.

[0007] One or more embodiments provide a path planning system for unmanned sanitation vehicles based on an improved NSGA-II, including: The preprocessing module is configured to acquire information on the work tasks and work vehicles, discretize the road segments corresponding to the work tasks, establish a mapping between tasks and road segments, calculate the shortest distance between the discretized road nodes, and construct a matrix of shortest distances between all nodes. The model building module is configured to build a multi-objective path planning model with the optimization objectives of minimizing the number of sanitation vehicles deployed, minimizing the total energy consumption of the system, and minimizing the proportion of uncovered task areas. The solution module is configured to generate an initial population for vehicle task allocation using a clustering algorithm based on the spatial location characteristics of road segments; combine greedy path simulation with the NSGA-II optimization algorithm to obtain an improved NSGA-II optimization algorithm; and solve the constructed multi-objective optimization model based on the mapping relationship between tasks and road segments and the shortest distance matrix of all nodes to obtain the path planning scheme for vehicle task execution.

[0008] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the steps in the above-described path planning method for unmanned sanitation vehicles based on the improved NSGA-II.

[0009] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps in the above-described path planning method for unmanned sanitation vehicles based on the improved NSGA-II.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention takes minimizing the number of unmanned sanitation vehicles deployed, the total energy consumption of the system, and the proportion of uncovered task areas as three optimization objectives, avoiding the resource waste or increased energy consumption that may be caused by optimizing a single objective, and improving the overall resource scheduling rationality of the operation system.

[0011] This invention, without relying on fixed infrastructure points, constructs a mapping relationship between tasks and road segments, as well as a shortest distance matrix between node pairs. This enables efficient acquisition of the shortest travel distance between any locations during the path planning phase, improving the computational efficiency of path evaluation and planning. Addressing the problem of blind solution space search and slow convergence caused by traditional random initialization, this invention utilizes clustering to perform spatial clustering initialization on the standardized segmented task road segments. This naturally endows the initial population with good spatial locality, avoiding unnecessary round trips between long-distance tasks and significantly reducing the number of iterations required for algorithm convergence.

[0012] By combining greedy path simulation with the NSGA-II optimization algorithm, the path generation process can simultaneously consider local optimal decision-making and the exploration of the global solution space, thus improving the overall quality of the planning scheme and the effectiveness of the search. The final path planning result has higher feasibility and optimization level, and can effectively adapt to the scheduling needs of unmanned sanitation vehicles in urban road network environments.

[0013] The advantages of the present invention, as well as its additional advantages, will be described in detail in the following specific embodiments. Attached Figure Description

[0014] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute a limitation thereof.

[0015] Figure 1 This is a flowchart of the unmanned sanitation vehicle path planning method according to Embodiment 1 of the present invention; Figure 2 This is a flowchart illustrating the solution process of the improved NSGA-II optimization algorithm used in Embodiment 1 of the present invention. Figure 3 This is a schematic diagram of the variable vehicle size chromosome in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of connection cost compensation between non-continuous tasks in Embodiment 1 of the present invention; Figure 5 This is a schematic diagram of the sequential crossing in Embodiment 1 of the present invention; Figure 6 This is a schematic diagram of the mutation operator in Embodiment 1 of the present invention; Figure 7 This is a three-dimensional target space distribution map of the solution results in the experimental example of Embodiment 1 of the present invention; Figure 8 This is a trade-off distribution diagram of vehicle size and system energy consumption in the experimental example of Embodiment 1 of the present invention; Figure 9 This is a specific deployment scheme in the experimental example of Embodiment 1 of the present invention; Figure 10 This is the Pareto solution set distribution diagram of the benchmark NSGA-II algorithm in the experimental example of Embodiment 1 of the present invention; Figure 11 This is a performance comparison of the improved algorithm and the benchmark algorithm in terms of vehicle size and driving mileage in the experimental example of Embodiment 1 of the present invention. Detailed Implementation

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0017] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0018] It should be noted that the terminology used herein is for describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof. It should be noted that, without conflict, the various embodiments and features within those embodiments can be combined with each other. The embodiments will now be described in detail with reference to the accompanying drawings.

[0019] Example 1 In one or more of the technical solutions disclosed in the embodiments, such as Figures 1 to 11 As shown, the path planning method for unmanned sanitation vehicles based on the improved NSGA-II includes the following steps: Step 1: Obtain the work task and work vehicle information, discretize the road segments corresponding to the work task, establish the mapping between tasks and road segments, calculate the shortest distance between the discretized road nodes, and construct the shortest distance matrix between all nodes. Step 2: Construct a multi-objective path planning model with the optimization objectives of minimizing the number of sanitation vehicles deployed, minimizing the total energy consumption of the system, and minimizing the proportion of uncovered task areas; Step 3: Based on the spatial location characteristics of road segments, a clustering algorithm is used to generate an initial population for vehicle task allocation; the greedy path simulation is combined with the NSGA-II optimization algorithm to obtain an improved NSGA-II optimization algorithm; based on the mapping relationship between tasks and road segments and the shortest distance matrix of all nodes, the constructed multi-objective optimization model is solved to obtain the path planning scheme for vehicles to perform tasks.

[0020] In this embodiment, compared with the prior art, without relying on fixed facility points, this implementation method enables efficient acquisition of the shortest travel distance between any location during the path planning stage by constructing the mapping relationship between tasks and road segments and generating the shortest distance matrix between node pairs, thereby improving the computational efficiency of path evaluation and planning. This implementation method uses minimizing the number of unmanned sanitation vehicles deployed, the total system energy consumption, and the uncovered task area as three optimization objectives. By introducing the uncovered rate as an optimization objective, it avoids the problem of deploying unmanned sanitation vehicles in a very small number of difficult-to-cover areas, significantly reducing vehicle deployment while ensuring the completion of the main work areas. Through the K-Means clustering initialization strategy, the initial solution possesses spatial locality. By integrating greedy path simulation with a dynamic stopping mechanism and the NSGA-II optimization algorithm, it can simultaneously consider local optimal decision-making and the exploration capability of the global solution space during the path generation process, improving the overall quality of the planning scheme and the effectiveness of the search. The final path planning result has higher feasibility and optimization level, and can effectively adapt to the scheduling needs of unmanned sanitation vehicles in urban road network environments.

[0021] In step 1, the task to be acquired can be a sweeping task or a watering task. The road segment corresponding to the task is relatively long. In order to achieve orderly planning, the road segment is divided, and the task can be segmented, thereby improving the granularity of the planning. Correspondingly, there are unmanned sweeping vehicles, watering trucks, etc. In this embodiment, an unmanned sweeping vehicle is used as an example for explanation.

[0022] Step 1, the method for establishing the mapping between tasks and road segments, includes the following steps: Step 1.1, Road network discretization: Divide the road into multiple road segments according to a set length, and use the two endpoints of each road segment as nodes; Specifically, the road arc segments are divided into fixed lengths. Divide the road into equal-length segments and then perform geometric polylines on each road. Interpolation sampling is performed based on the cumulative arc length to generate a segmented node sequence. The segment length between adjacent nodes is approximately: Form a global node set. and its coordinate matrix ; Specifically, the geometric polylines of each road are interpolated and sampled according to the cumulative arc length to generate a segmented node sequence. The segment length between adjacent nodes is approximately the set step size. All generated nodes are then incorporated into the global node set to construct a node coordinate matrix.

[0023] For example, if the total length of the road polygon is 1000 meters and the fixed segment length is 100 meters, then the node positions are determined by interpolation at the cumulative arc lengths of 100 meters, 200 meters...900 meters, and finally 10 segments of equal length are obtained, each segment being about 100 meters long; Step 1.2: Construct a task-endpoint mapping table: Define the adjacent node pairs after segmentation in Step 1.1 as the smallest task unit, assign a unique task number to each smallest task unit, and record the global node numbers corresponding to both ends of the task segment to form an endpoint-task mapping table. After segmentation, adjacent node pairs Considered the smallest cleaning task unit (task section). To facilitate subsequent indexing and querying within the shortest distance matrix of all node pairs, the globally unique IDs of the aforementioned node pairs in the road network are extracted and recorded as endpoint ID pairs. This forms an endpoint mapping table, which is used for subsequent vehicle origin determination, connection cost completion, and return cost calculation.

[0024] Specifically, regarding cleaning operations, the operation time... In one embodiment, estimation is based on road segment area and sweeping parameters: ; in, Indicates the area of ​​the road segment. Indicates the width of a single cleaning pass. Indicates the vehicle's speed; Furthermore, step 1 also includes constructing the shortest distance matrix between all nodes in the urban road network for subsequent optimization calculations; the process of constructing the shortest distance matrix between all nodes includes the following steps: Step 1.3: Construct an undirected graph: Based on the adjacent node pairs after segmentation in Step 1.1 Constructing an undirected edge set By taking the side length as the geometric length between the two nodes, we obtain the undirected weighted adjacency matrix adjMatrix; An undirected weighted adjacency matrix adjMatrix is square array ( (total number of road network nodes), the first node in the matrix Line number Column element values Represents a node With nodes The direct connectivity distance between nodes. and If they are directly adjacent, then Its geometric distance; if not directly adjacent or unreachable, then Set to infinity; diagonal elements Set it to 0.

[0025] Step 1.4: Calculate and cache the shortest distance matrix for all node pairs: Construct a graph object based on the undirected weighted adjacency matrix adjMatrix, calculate the shortest path length between all node pairs as the matrix element values, and obtain the shortest distance matrix for all node pairs. The shortest distance matrix stores the shortest path distance between any two points. Specifically, an initial distance matrix is ​​constructed based on adjMatrix. : ; Calculate the shortest distance between all nodes to construct the shortest distance matrix between all node pairs: ; in, This indicates that nodes are allowed. When acting as an intermediate node, the node To the node The shortest path length. The final matrix obtained after traversing all nodes. This is the shortest distance matrix for all node pairs, where the elements are... Represents a node and nodes The shortest connectivity distance along the road network topology.

[0026] To avoid repeated calculations during algorithm execution, the system employs a persistent caching mechanism: storing the calculated matrix... Serialization and storage are performed as binary files, using the .mat format in the experiment. When the program starts and detects the existence of the cached file and its match with the current road network fingerprint, it directly loads the matrix as the basis for subsequent individual evaluations. This caching mechanism allows for the evaluation of the connection costs of a large number of individuals during evolutionary iterations. The table lookup implementation reduces computational overhead and improves project operability.

[0027] In this embodiment, by segmenting the complex road network structure into models and pre-calculating the shortest distance matrix between all node pairs, the connection cost between any two points can be quickly obtained through efficient table lookup in the subsequent evaluation stage. This method not only effectively avoids redundant calculations but also significantly reduces the computational overhead required during large-scale evolutionary iterations, thereby improving the overall algorithm's operating efficiency and response speed.

[0028] In step 2, an optimization model is constructed with multiple objectives: minimizing the number of unmanned sanitation vehicles deployed, minimizing the total system energy consumption, and minimizing the proportion of uncovered task areas. Its core objective is to achieve comprehensive optimization of facility spatial layout and vehicle route planning while meeting the requirements of operation time, vehicle range, and road connectivity, thereby achieving a balance between operation coverage, vehicle configuration costs, and operating energy consumption.

[0029] Primary objective: Minimize the number of unmanned sanitation vehicles; Number of vehicles Treated as a dynamic decision variable, it is adaptively determined by the NSGA-II optimization algorithm during the search process. In the chromosome encoding stage, each chromosome code contains road allocation and path order information, and also implicitly contains the boundary positions of task partitioning; the task sequence is divided into multiple segments using delimiters, with each non-empty segment corresponding to one work vehicle. At this point, the number of vehicles is determined by the number of non-empty task groups in the chromosome decoding result. Therefore, the first optimization objective is defined as: ; in, Indicates the maximum number of vehicles allowed by chromosome encoding (determined by the number of delimiters); For binary decision variables, when the first... When a vehicle task group is not empty (i.e., the vehicle has been actually assigned a task) ,otherwise The goal is to minimize the number of vehicles required while meeting sweeping coverage and operation time constraints, thereby reducing the system's fixed costs.

[0030] Second objective: Minimize system energy consumption; Given a fixed number of vehicles, the second objective focuses on the energy efficiency of the system. Since the energy consumption of unmanned sanitation vehicles, such as sweepers, is primarily determined by their travel distance, the energy minimization objective can be equivalently transformed into minimizing the total system travel distance. The formula for the second objective is: ; in, This represents the total number of vehicles currently in use under the current plan; Operating travel distance : refers to the The effective working distance of a vehicle for all assigned tasks is calculated using the following formula: ; in, The work area of ​​the task section. This refers to the vehicle's effective sweeping width. The actual operating mileage is calculated here using the ratio of area to width. For the first The task set for the vehicle.

[0031] Initial empty run distance : refers to the The distance the vehicle travels from the starting point to the first task starting point; the calculation formula is: ; In the greedy strategy of this embodiment, It can be selected as the starting point of the first task. The overlapping or closest nodes, where the distance approaches 0.

[0032] Continuous empty run distance between missions : refers to the The cumulative empty driving distance of a vehicle from the start of the next discontinuous task after completing the previous task is calculated using the following formula: ; in, Indicates the first The total number of tasks assigned to each vehicle. Indicates the first The first car The endpoint node number of each task segment. Indicates the first The first car The starting node number of each task segment; Return empty driving distance : refers to the The distance the vehicle travels back to its initial deployment node after completing its last task is calculated using the following formula: ; in, Indicates vehicle The node number of the last task segment, Indicates the first The initial deployment node of the vehicle, i.e., the dynamically selected starting point; Third objective: Minimize the percentage of uncovered task area; The excessive resource consumption of urban fringe areas or long-tail tasks transforms coverage into a third optimization objective. This objective aims to minimize the area of ​​unexecuted tasks (i.e., coverage gaps), allowing the algorithm to explore "incomplete but efficient" solutions during the search process. Third Objective Function The ratio of the total area of ​​road segments not assigned to any vehicles to the total area of ​​the road network requiring work is given by the formula: ; in, This indicates the total area of ​​the assigned road segments actually covered by all activated vehicles in the current dispatching plan; This represents the total area of ​​all road sections to be cleaned within the target work area; By minimizing The algorithm drives the population to evolve towards higher coverage. However, due to... and There is a competitive relationship between the number of vehicles. To eliminate a small number of uncovered areas, it is often necessary to add several vehicles. The NSGA-II algorithm will automatically weigh the two and identify those marginal tasks that are extremely costly but have very low benefits. It will keep them in the uncovered set, thus preserving a high-coverage (e.g., 80%) and low-cost high-quality compromise solution on the Pareto front, rather than forcibly pursuing 100% coverage and wasting resources.

[0033] Finally, the multi-objective optimization model can be expressed as: ; in, Focus on the macro-level configuration efficiency of the system (number of vehicles), and Focus on micro-level operational efficiency (path energy consumption); Pay attention to gaps in task coverage to demonstrate a flexible trade-off between task completion and resource investment.

[0034] There is a clear trade-off among the three objectives: reducing the number of vehicles may lead to increased workload and energy consumption per vehicle, as well as a decrease in coverage, while pursuing maximum coverage usually requires deploying more vehicles or increasing mileage. To obtain a balanced solution among the three, this embodiment adopts an improved NSGA-II algorithm and combines it with a greedy path planning module based on a coverage threshold for local optimization search, ultimately obtaining a Pareto optimal solution set for the relationship between vehicle number, energy consumption, and coverage, providing multi-level decision support for the deployment and operation of unmanned urban cleaning operations.

[0035] Optionally, the solution constraints for the multi-objective optimization model include: runtime constraints, single-cycle endurance and dynamic charging constraints, task allocation uniqueness, and job path connectivity constraints. To ensure the feasibility and safety of the unmanned sanitation vehicle scheduling and route planning scheme, the model incorporates conditions such as running time, single-use range and dynamic charging constraints, task allocation uniqueness, and path connectivity to ensure that the optimization results can reflect the actual operation logic of urban road cleaning.

[0036] 1) Operating time constraints: Constraining the operating time of unmanned sanitation vehicles within a set time period; Taking an unmanned street sweeper as an example, considering the city's nighttime cleaning policy, the unmanned street sweeper only operates during low-traffic periods at night. Let the nighttime operation window be... Then each vehicle The start and end times satisfy: ; Meanwhile, the total operating time of a single vehicle must not exceed the length of the operating window: ; This constraint ensures that unmanned cleaning operations avoid impacting traffic efficiency and citizens' travel during peak hours, and complies with the nighttime concentrated operation strategy in the sanitation operation system.

[0037] 2) Single-charge range and dynamic charging constraints: The single-charge operating time of the unmanned sanitation vehicle shall not exceed the maximum power supply time threshold of the vehicle battery. When the upper limit of the working time is reached, the vehicle shall be charged. Taking unmanned sweepers as an example, their continuous operation time is limited by battery capacity and cannot exceed a certain threshold. When the cumulative driving and cleaning time reaches the upper limit, the vehicle must go to a charging point to recharge. Definition: ; in, Indicates the first The total time consumed by an unmanned sanitation vehicle in a single continuous operation cycle (from departure from the charging point until it must be recharged); Indicates the first The total empty driving time of a vehicle in a single cycle; Indicates the first The cumulative effective working time of a vehicle in a single cycle; This refers to the single continuous running time.

[0038] To ensure that the energy consumption and operating time of each vehicle are within the time window, the cumulative operating time after charging should meet the following requirements: ; in, Indicates the first The total cumulative time occupied by the unmanned sanitation vehicles during the entire designated operating period (such as all night); Indicates the first The car in The cumulative empty running time in each work cycle (including the time from the starting point / charging point to the task point, and the transfer time between tasks); Indicates the first The car in The cumulative effective cleaning time in each work cycle; Indicates the first The car in The charging time required for each work cycle; This indicates the total length of the nighttime operation time window.

[0039] The task assignment uniqueness constraint means that each task segment is cleaned by one and only one sanitation vehicle. Taking an unmanned street sweeper as an example, to ensure the integrity of the road sweeping task, all target road segments are assigned to one and only one sweeper. Let the set of road segments be denoted as . The vehicle assembly is denoted as Introducing binary decision variables To indicate road sections Whether by vehicle Cleaning; This embodiment introduces the coverage rate as an optimization objective, allowing some non-critical road segments to be uncovered. The task allocation uniqueness constraint ensures that a road segment is either cleaned by a certain vehicle (value 1) or belongs to an uncovered road segment (value 0), and is included in the objective function. However, it will not be cleaned repeatedly, and its value is greater than 1.

[0040] ; 4) Operation path connectivity constraint: vehicles have reachable paths between their assigned task operation segments, and vehicles return to their starting point after completing their tasks. Taking unmanned sweeping vehicles as an example, to ensure that the sweeping sequence of each unmanned sweeping vehicle is executable on the road network, it is required that the vehicles have connected and reachable paths between their assigned sweeping segment sequences. Let... Represents nodes in the road network To the node The shortest path distance (if unreachable, then) ).make and These represent the start and end point numbers of the road segment, respectively. An ordered sequence of tasks for vehicles requires that adjacent cleaning segments be reachable: ; in, Indicates vehicle The first task in the sequence The node number of the end point of each task segment. Indicates vehicle The first task in the sequence The node number of the starting point of each task segment. If two segments are not directly connected, the distance of the connecting segment to be traversed is included in the vehicle's empty travel distance. The vehicle must return to its starting point after completing the task. This constraint ensures the topological continuity and closure of the path.

[0041] The constraint system and objective optimization framework constructed in this embodiment together form a realistic framework for unmanned vehicle path optimization. Specifically, the running time and range constraints control the pace of operations and energy consumption safety, ensuring efficient vehicle operation within limited time periods and energy limits; the task allocation uniqueness constraint effectively prevents multiple vehicles from repeatedly cleaning the same road segment, avoiding resource waste; the coverage rate objective function drives the algorithm to find the optimal balance between resource input and operational coverage; and path connectivity and return constraints ensure the spatial accessibility and closed-loop nature of the cleaning path. Through comprehensive modeling of the above constraints, the model not only possesses a feasible solution space at the theoretical level but also achieves efficient, closed-loop scheduling and task execution in actual urban road network environments.

[0042] The core challenge of resource allocation and route planning for unmanned sanitation vehicles lies in the significant coupling relationship between the number of vehicles, starting point location, and operation route. Furthermore, constraints such as endurance and operation window must be met while balancing coverage and cost. To address these challenges, this embodiment constructs an improved NSGA-II optimization algorithm based on clustering initialization and objective evolution to achieve coordinated optimization of vehicle deployment and route planning.

[0043] In step 3, the greedy path simulation based on the coverage threshold is combined with the improved NSGA-II optimization algorithm by introducing clustering initialization, resulting in the improved NSGA-II optimization algorithm. This improved algorithm is then used to solve the constructed multi-objective optimization model. The method for solving the constructed multi-objective optimization model to obtain the path planning scheme for vehicle task execution includes the following steps: Step 31: Initialization of K-Means clustering: Vehicle task groups are encoded by merging task numbers with delimiters, and chromosomes are generated by combining spatial clustering heuristics with random mechanisms to construct the initial population; Step 32: Perform chromosome decoding and task grouping: Analyze the gene sequence in each chromosome, identify the location of the boundary markers, and divide the chromosome into segments. Each subsequence corresponds to a list of task segments for a vehicle. The chromosome is converted into a task assignment for the vehicle, which is then used as the decoding result. The gene segments between the delimiters represent the sequence of task road segment numbers assigned to a particular vehicle. By decoding, the long chromosome is segmented into multiple subsequences to determine the number of vehicles activated in the current scheme and the initial task list for each vehicle.

[0044] Step 33: Based on the decoded results, dynamically determine the initial deployment node for each vehicle, and perform a greedy path simulation with dynamic truncation. Based on the pre-calculated shortest distance matrix of all nodes, fill in the connection cost to generate the vehicle path. Step 34: For the path obtained for each chromosome, calculate the fitness based on the multi-objective optimization model to obtain the number of vehicles, total travel distance and the proportion of uncovered task area. Based on the constraints of the multi-objective optimization model, perform constraint verification. For individuals that pass the constraint verification, perform fast non-dominated sorting and crowding calculation based on the fitness calculation results.

[0045] Step 35, Evolution and Elite Preservation: Perform selection, crossover, mutation, and environmental selection to generate new chromosomes, execute step 32, and proceed to the next iteration until the iteration stopping condition is met, obtaining the Pareto solution set; The solution method described in this embodiment employs a greedy strategy to construct the vehicle's driving trajectory. This path algorithm utilizes a locally optimal heuristic strategy, abandoning the strategy of pre-setting fixed garages and instead adopting a task-oriented starting point determination mechanism. The starting point of each sweeper is dynamically determined; this strategy couples the path's starting point with the task location, significantly reducing the initial empty driving distance.

[0046] Furthermore, to ensure the superiority and reliability of the results in a single run, this embodiment pre-executed orthogonal optimization experiments for the parameters. Key parameters of the algorithm, including population size, crossover probability, mutation probability, and maximum number of iterations, were fine-tuned using orthogonal experimental methods. By analyzing the signal-to-noise ratio and the mean of the objective function under different parameter level combinations, an optimal set of parameters was determined. Based on this set of optimal parameters, the algorithm can stably converge to a high-quality Pareto front in a single execution without inefficient repeated runs, meeting the stringent requirements for computational efficiency in engineering applications.

[0047] In step 31, variable vehicle scale chromosome encoding: vehicle task groups are encoded using task number fusion delimiters to obtain each chromosome, such as... Figure 3 As shown, the specific encoding methods include the following: Step 311: Use non-zero gene values ​​to represent road task numbers, with values ​​ranging from 1 to n; Step 312: Set delimiters to divide the task sequence into multiple vehicle task segments. Use the K-Means algorithm to perform spatial clustering of all tasks based on the center coordinates of the vehicle task segments to generate pre-grouping of tasks with spatial proximity. Encode task segments belonging to the same cluster as the task sequence of the same vehicle to obtain the encoded chromosome, which serves as the first part of the individuals. Among them, the task segments of the same type of cluster are spatially adjacent; Step 313: Randomly shuffle the task sequence and randomly insert delimiters to assign vehicle task segments to vehicles, generating the second part of individuals. The first part of individuals and the second part of individuals constitute the initial population. In this embodiment, some individuals in the initial population are directly derived from the clustering results, while the remaining individuals are generated by random shuffling. This ensures population diversity while giving the initial solution good spatial locality, significantly accelerating algorithm convergence.

[0048] To improve the algorithm's convergence speed, a hybrid initialization strategy is adopted. Some individuals are generated using K-Means-based spatial clustering: K-Means clustering is performed based on the center point coordinates of task segments, encoding task segments belonging to the same cluster as the task sequence of the same vehicle, thus giving the initial solution good spatial locality, ensuring that the work segments of each unmanned sanitation vehicle are adjacent to each other. The other individuals are generated randomly: the task sequences are randomly shuffled and random delimiters are inserted to ensure population diversity. This hybrid strategy ensures that the initial population contains both high-quality "seeds" with geographical proximity and retains the randomness of the global search.

[0049] Optionally, the delimiter can be a specific character such as " "" can also be numbers, letters, or combinations of characters, numbers, or letters; In this embodiment, the value 0 is used as a delimiter for vehicle operation tasks, dividing the task sequence into multiple vehicle task segments, such as... Figure 3 As shown, for example, chromosomes: [3,8,2,0,4,5,7,0,9,6,1]; The chromosomes described above represent the task segments for the three vehicles as [3,8,2], [4,5,7], and [9,6,1], respectively, with the number of vehicles equal to the number of delimiters plus 1. This encoding method in this embodiment expresses the number of vehicles and task grouping in the same chromosome, enabling joint optimization during evolution and outputting the Pareto tradeoff between the number of vehicles and the total cost.

[0050] In this embodiment, a variable vehicle size encoding mechanism with delimiters is introduced to achieve an organic and unified expression of the number of vehicles and task grouping. This encoding scheme can not only efficiently represent fleet configurations of different sizes, but also directly generate Pareto trade-off solutions between the number of vehicles and the total travel distance, providing a more intuitive and efficient solution for path optimization problems.

[0051] Furthermore, delimiters are inserted into the task sequence to generate or reconstruct chromosomes. The delimiter insertion must satisfy the following conditions: (1) The delimiter does not appear at the beginning of the sequence; (2) The delimiter is not continuous, that is, "0,0" does not appear as the final output state; (3) Each vehicle task segment is not empty; (4) If the end of the chromosome after insertion is a delimiter, remove the end delimiter to ensure that the end of the output chromosome is not a delimiter, such as 0.

[0052] For a given task sequence Number of delimiters Based on the above conditions in the location set Random selection Insert a delimiter 0 at each unique position; this ensures that the output chromosome is in the standard "task + delimiter" format.

[0053] Step 32, Decoding Segmentation: Read the position of the delimiter 0 in the chromosome as the segmentation point, and add the segmentation boundary at the beginning and end to obtain the beginning and end endpoints of the road segment, thus dividing the chromosome into segments. Each subsequence corresponds to a list of task segments for a vehicle, converting the chromosome into task assignments for vehicles. Step 32 determines which vehicle will perform which task(s), but does not specify the path each vehicle should take to execute its assigned task. Furthermore, the process involves identifying uncovered tasks: after decoding, all unassigned task segments are identified. Unlike traditional forced allocation mechanisms, this embodiment does not force the creation of new vehicles to cover these remaining tasks; instead, it incorporates the total area of ​​these uncovered segments into the third optimization objective. The NSGA-II algorithm automatically weighs the trade-offs between "increasing the number of vehicles to cover more areas" and "accepting a small number of uncovered areas to save costs".

[0054] In this embodiment, by introducing an undercoverage optimization objective, a flexible coverage decision system is constructed. This method effectively avoids the diminishing marginal utility trap of disproportionately increasing vehicle investment to cover a very small number of long-tail tasks. As a result, it can output a series of Pareto optimal solutions that achieve the best balance between high coverage and low resource investment, significantly enhancing the overall competitiveness of the scheduling scheme in terms of economy and practicality.

[0055] In step 33, a connection cost completion and task-oriented starting point mechanism based on the shortest distance matrix is ​​executed. Based on the decoded result, the endpoint of the vehicle's first task is used as the starting point of the vehicle path. A greedy path simulation is performed, and the connection cost is completed based on the pre-calculated shortest distance matrix of all nodes to generate the vehicle path. This method includes the following steps: Step 331: Dynamically determine the vehicle starting point: With the goal of completing all task segments with the shortest total travel distance, determine one of the endpoint nodes of the first task segment as the starting point, i.e., the initial deployment node of each vehicle, through the shortest distance matrix of all nodes. Specifically, instead of using a fixed garage as the vehicle's starting point, the starting point is determined by the first task segment in the vehicle's task list.

[0056] For example, for vehicles Take its first task section Read endpoint node pairs from the endpoint mapping table Using both endpoints as candidate starting points, perform the same path cost evaluation process as follows: Step 3311: Using the pre-computed shortest distance matrix of all nodes, calculate the connection cost between vehicles in adjacent task segments; Step 3312: During the greedy selection of the next task, accumulate the total distance traveled by the vehicle, which may include the connection distance, the distance traveled on the work section, and the return distance; Starting from the endpoint node of the first task, steps 3311 to 3312 are executed to select the endpoint with the smaller cumulative total driving distance as the starting point. This realizes a non-fixed starting mechanism where the starting point changes with the task.

[0057] In cleaning tasks, existing path planning relies on fixed starting points, lacking flexibility and resulting in low efficiency in assessing traffic costs. Traditional methods often use fixed garages or central stations as vehicle starting points, ignoring the spatial distribution characteristics of the cleaning task itself. This makes it difficult to adapt to scenarios with multiple flexible starting points, and the path assessment process requires repeated calls to the shortest path algorithm, resulting in low computational efficiency and difficulty in scaling to large-scale road networks. This embodiment adopts a task-oriented starting point setting mechanism, which does not rely on fixed garage location assumptions. This allows the starting point of the vehicle to be flexibly adjusted according to the needs of the specific cleaning task, thereby effectively reducing vehicle empty mileage and resource waste, further optimizing the number of unmanned cleaning vehicles required in the entire system, and improving operational efficiency and resource utilization.

[0058] Step 332: Start from the determined starting point and execute the task. With the goal of minimizing the connection distance, iteratively and greedily select the next task until all tasks have determined the vehicle path and the connection cost is completed. The connection cost is the total distance the vehicle travels, which is the sum of the distances the vehicle travels for each task. The vehicle employs a greedy algorithm when performing tasks: from the set of incomplete tasks, it selects the node that is closest to the current node. The task segment with the shortest connection distance will be selected as the next task. (This refers to the process of selecting candidate task segments.) Its endpoints are We obtained this by looking up the table: ; Pick The connection cost is used as the link cost from the vehicle to the task segment, and the corresponding endpoint is used as the destination node. Add to the vehicle's travel distance and Calculate and accumulate to the moving time; then follow the formula Accumulate cleaning time into working time. The above "connection cost compensation" only uses the shortest distance matrix. It is implemented by looking up a table, without the need to run the shortest path algorithm online, and is suitable for rapid evaluation of large-scale populations.

[0059] Furthermore, connection cost completion can be achieved using a connection cost completion based on the shortest distance matrix. Specifically, based on the decoded task sequence, the connection cost completion is performed using the initial deployment node of the vehicle. Using the starting point of the vehicle path, a greedy path simulation is performed. Based on the pre-calculated shortest distance matrix of all node pairs, the empty driving paths connecting adjacent tasks are completed.

[0060] Furthermore, a dynamic truncation mechanism is introduced during the greedy path simulation process: During the simulation, the vehicle's cumulative operating time and energy consumption are calculated in real time. If completing the next task is anticipated to cause the vehicle to exceed its maximum range threshold, the simulation will be conducted accordingly. If the task is not completed within the specified time window, the vehicle's task assignment will be immediately terminated, forcing it to return to the starting point. Any remaining unexecuted task segments in the vehicle sequence will be marked as uncovered areas, and their total area will be included in the third optimization objective. .

[0061] Introducing a coverage threshold stopping mechanism during the greedy decoding process effectively solves the "long tail effect" problem of traditional greedy algorithms forcibly adding vehicles to cover the last few remote tasks, further ensuring the economic feasibility and rationality of the generated vehicle scheduling scheme.

[0062] like Figure 4 As shown, the shortest distance from node 1 to node 5 is D(1,5)=120; The shortest distance from node 1 to node 6 is D(1,6)=80.

[0063] Since node 6 is closer to the current node, the system determines node 6 as the starting endpoint for the vehicle to connect to the task segment, and calculates the connection cost as 80 accordingly.

[0064] In step 35, based on the calculated fitness results, chromosomes with fitness values ​​that meet the requirements are selected as the selected individuals; In step 35, the crossover process is performed to generate new individuals (chromosomes). Specifically, for the parent individuals undergoing the crossover process, the delimiters between the two parent individuals are removed, the task order is cross-combined, and then delimiters are inserted according to the vehicle quantity ratio in the original individual scheme to obtain new candidate schemes. Specifically, for the parent generation, after removing the delimiter, the execution order of the task sequence is interleaved, and then the delimiter is inserted back according to the original ratio; Specifically, one particular implementation of the crossover process to generate new individuals is, for example... Figure 5 As shown, the interleaving is performed in the task sequence space after removing the delimiters, including the following steps: Step 351: Remove delimiters from the parent chromosomes: Delete all delimiters from the two parent chromosomes to obtain two parent task sequences. and ; Step 352, Sequential Crossover: Perform sequential crossover on the task sequence and randomly generate crossover regions. ,Will As an intermediate segment of offspring; Step 353, then from Extract tasks in the original order to form inserted fragments. The inserted fragments are spliced ​​together in the intermediate fragments to generate the child task. ; Step 354: Count the number of delimiters between the two parent generations: and ; The number of offspring chromosome delimiters is generated as follows: the average number of parent chromosome delimiters is added to a perturbation of a set magnitude and then rounded to the nearest integer. ; Step 355, Delimiter Insertion Back: Call the delimiter insertion function. Insert the corresponding number of delimiters into the child task sequence to obtain the child chromosome. The number of vehicles is... .

[0065] The above method enables the synchronous evolution of task order and vehicle number (demarcation number) during the crossover process, thereby achieving adaptive changes in vehicle size as the evolution process progresses.

[0066] In step 35, the mutation process includes task exchange mutation and delimiter mutation, such as Figure 6 As shown, the specific process is as follows: Step 3501, Task Exchange Mutation: For the individual to be mutated, randomly select two locations from all non-delimiter task gene positions and exchange their task numbers to disrupt the task execution order. Step 3502, Delimiter Mutation: Insert a delimiter at a random position or randomly delete an existing delimiter to change the number of vehicles, where the number of vehicles = the number of delimiters + 1.

[0067] The mutation process in step 35 includes task exchange mutation that swaps the order of task execution, and delimiter mutation that changes the number of vehicles by inserting or deleting delimiters.

[0068] To maintain the stability of the output coding structure and suppress duplicate task genes, a recoding stabilization process is performed after mutation. The specific process includes: first, extracting all non-delimiter (0) task genes, and then performing stable deduplication according to their order of appearance to obtain the task sequence. Count the number of delimiters and limit it to an allowed range, which is no less than 0 and no greater than the vehicle limit minus 1; then call... Reinsert the delimiter 0 and output the mutated chromosome in standard format.

[0069] This embodiment features a carefully designed evolutionary operation. The algorithm employs a repairable sequential crossover operator, which effectively ensures no task duplication or omission during crossover operations. It also automatically restores vehicle demarcation markers that may have been damaged by the crossover operation, thus maintaining the encoding validity of the decoding process and further enhancing the algorithm's robustness. Regarding mutation operations, this algorithm introduces two strategies: positional mutation and boundary mutation. Positional mutation primarily achieves fine-tuning of local paths by swapping the order of tasks in the task sequence; boundary mutation dynamically adjusts the number of vehicles by inserting or deleting vehicle demarcation markers, thus flexibly responding to the needs of vehicles of different sizes. This two-layer mutation strategy not only significantly expands the algorithm's search space but also effectively enhances its sensitivity to changes in vehicle size, thereby improving the algorithm's adaptability and solution efficiency in complex scenarios.

[0070] Furthermore, based on each solution, a specific deployment plan for sanitation vehicles can be generated, including the number of vehicles, route planning, and the layout of starting points and charging points. The hardware implementation of the sanitation vehicle's working area is completed by laying out the garage location of the vehicles and the location of the charging points according to the optimal solution. A further technical solution, in order to select representative deployment schemes from the Pareto solution set generated by the improved NSGA-II algorithm, can be to use the entropy-weighted TOPSIS comprehensive evaluation method to rank and select the best candidate schemes in multiple dimensions. The specific steps are as follows: (1) Construct the initial decision matrix; obtain the solutions in the Pareto solution set and perform deduplication, and construct the original decision matrix based on the evaluation index corresponding to the objective value of each solution; Specifically, after completing a preset iteration (e.g., 100 iterations), a non-dominated solution set with a population size of 80 is obtained. All individuals satisfying the constraints within this non-dominated solution set are deduplicated, resulting in 68 remaining solutions. Each candidate solution corresponds to three evaluation metrics: number of vehicles, total system energy consumption (total travel distance), and task non-coverage rate. Since these metrics are all cost-based indicators where lower values ​​are better, the original decision matrix is ​​constructed as follows: ; (2) Indicator homogenization and standardization: The range transformation method is used to transform the evaluation indicators into benefit-type indicators (i.e., the larger the better), and a homogenization matrix is ​​obtained. The conversion formula is: ; (3) For the benefit-type indicators in the homogenization matrix, the entropy weight method is used to calculate the adaptive weights. ; The adaptive weight determination method based on information entropy (entropy weight method) automatically determines the weight of each indicator by calculating the degree of dispersion of each indicator. This avoids the subjectivity of manually setting weights. Specifically, it includes: calculating the... The first indicator The proportion coefficient of each scheme Calculate the index entropy value and the coefficient of difference Ultimately, the entropy weight is obtained. The greater the dispersion of the indicator data, the greater its corresponding weight. The larger it is, the calculation formula is as follows: ; ; ; (4) Construct a weighted decision matrix, and use the obtained entropy weights to perform weighted calculations on the homogenization matrix to obtain the weighted matrix: .

[0071] (5) Calculate the closeness and determine the typical solution: Select the optimal value of each index in the weighted matrix to form the positive ideal solution. The worst values ​​of each indicator are selected to form the negative ideal solution. Calculate the distance from each candidate solution to the positive ideal solution. and distance to the negative ideal solution Calculate proximity .

[0072] ; ; ; Proximity The larger the value, the closer the solution is to the positive ideal solution and the farther it is from the negative ideal solution, indicating better overall performance. The 68 candidate solutions are then sorted according to... Sort from largest to smallest, and select... The largest solution serves as a typical example and can be used for subsequent spatial layout, vehicle path and charging point linkage demonstration.

[0073] To illustrate the effectiveness and implementation process of the method described in this embodiment, actual experiments were conducted for verification, and the details are as follows; To systematically evaluate the impact of road network unit division accuracy on the site selection and deployment optimization of unmanned sweepers, comparative experiments were conducted with road network unit lengths set to 1000m, 500m, 250m, and 100m, while keeping other algorithm parameters fixed (see Table 1). Experimental results show that as the road network unit length is refined from 1000m to 100m, the number of road network nodes surges from 797 to 3836, and the number of task road segments increases from 1032 to 4037. This exponential expansion of the solution space directly leads to a significant increase in computation time. At a 100m granularity, although the lowest uncovered rate (0.66%) was achieved, the computation time exceeded 5 hours, making it difficult to meet the engineering requirements for "response speed" in daily sanitation operations. At 1000m and 500m granularities: although the computation speed is fast, the coarse granularity makes it difficult to accurately characterize the topological features of complex urban road networks, easily leading to a disconnect between planned paths and actual terrain. At a granularity of 250m, a key phenomenon was observed in the experiment: the minimum uncovered rate was 3.33%, the highest among the four experimental groups. This seemingly "poor" metric actually strongly demonstrates the effectiveness of the third objective of this invention: given that the 250m granularity provides sufficient spatial modeling accuracy, the algorithm does not force 100% coverage when facing more fragmented and complex task distributions. Instead, the algorithm intelligently identifies these 3.33% of tasks as "high-cost, low-return tasks" (i.e., covering these scattered tasks may require additional vehicle deployment or a significant increase in empty driving mileage). Therefore, the improved NSGA-II algorithm utilizes a soft constraint mechanism to abandon these 3.33% of inefficient work areas in exchange for achieving a globally optimal balance between the number of vehicles deployed and the total system energy consumption.

[0074] Table 1. Experimental parameters;

[0075] Taking into account both the feasibility of the algorithm and the optimization accuracy, subsequent studies adopted 250m as the default partitioning accuracy.

[0076] Regarding algorithm stability, to further analyze the impact of NSGA-II parameters on optimization performance, nine representative parameter combinations were constructed from four dimensions: population size, number of iterations, crossover probability, and mutation probability (see Table 2). Multiple independent experiments were conducted on each parameter combination to reduce the interference of randomness. The parameters were comprehensively evaluated from three dimensions: convergence quality, solution set diversity, and computation time.

[0077] Table 2. Orthogonal experiment parameters;

[0078] A comprehensive analysis of the results of the nine orthogonal experiments in Table 1 shows that the fourth set of experimental parameters (population size 80, number of iterations 100, mutation probability 0.10, crossover probability 0.95) exhibits the best overall performance. This set of parameters has the lowest IGD index (0.01868), indicating the best algorithm convergence; simultaneously, it has the highest HV index (0.69585), indicating the richest diversity of the solution set. Although its computation time (approximately 72 minutes) is not the shortest, it achieves the best optimization quality within an acceptable timeframe. Therefore, all subsequent optimization solutions in this embodiment will use the fourth set of parameter combinations.

[0079] like Figure 7 As shown, the experimental results of Embodiment 1 of the present invention present the distribution of the Pareto optimal solution set obtained by the improved NSGA-Ⅱ algorithm in the form of a three-dimensional scatter plot. Figure 7 The image shows the number of unmanned cleaning vehicles deployed. Total system energy consumption ( ) and the percentage of uncovered task area ( The complex trade-offs among these three competing objectives, with the increase in coverage ( The reduction in the number of axles is usually accompanied by a decrease in the number of vehicles deployed ( (axle) or total system mileage ( The upward trend of the axis provides an intuitive quantitative basis for actual sanitation deployment. The continuous and uniform scatter distribution in three-dimensional space indicates that the algorithm can stably output feasible solutions with varying coverage and cost ratios while meeting constraints such as endurance and operation window. This not only verifies the convergence of the model when handling large-scale tasks, but also provides a solid data foundation for subsequent screening of "typical deployment solutions" using the entropy weight-TOPSIS method.

[0080] like Figure 8 As shown, the method of Embodiment 1 of the present invention demonstrates the trade-off between vehicle size and energy consumption in the Pareto solution set through a two-dimensional scatter plot combined with color mapping. Figure 8 The horizontal axis represents the number of vehicles ( The vertical axis represents the total energy consumption of the system. The shades of the scatter dots represent the coverage area. The results intuitively reveal the coupling characteristics among the three objectives: within the optimization range of 38-54 vehicles, the algorithm can flexibly adjust the energy consumption level of the route according to the coverage requirement. This multidimensional distribution characteristic not only proves the improved algorithm's ability to accurately lock onto the high-quality solution space, but also provides more flexible and refined decision support for sanitation scheduling under different cost budgets.

[0081] like Figure 9 As shown, the spatial visualization results of typical deployment schemes selected using the entropy weight-TOPSIS method are presented. Figure 9 The blue sections represent covered work areas, while the orange sections represent strategically abandoned uncovered areas. Circles indicate the dynamic deployment points of vehicles. With 39 vehicles deployed and an coverage rate of approximately 17.9%, this solution controlled the total system mileage to 1819.75 km. This intuitive road network mapping not only verifies the effective implementation of the algorithm's "non-fixed starting point" deployment mechanism in complex topologies but also clearly demonstrates how the system intelligently abandons high-cost edge tasks (orange sections) to achieve global optimization of overall resource allocation (number of vehicles and energy consumption), proving the high engineering feasibility and economic rationality of the planning results.

[0082] To verify the superiority of the improved NSGA-II algorithm in solving the path planning problem for unmanned sanitation vehicles in large-scale road networks, the standard NSGA-II algorithm was selected as the benchmark for comparative experiments. Considering that the standard algorithm lacks heuristic strategies and has a slow convergence speed, to ensure the rigor and persuasiveness of the comparison, this experiment used the optimal parameter combinations of the two algorithms respectively. The improved algorithm used a population size of 80, 100 iterations, and a road network discretization threshold of 250m. The benchmark algorithm set a population size of 150 and 300 iterations. Both algorithms were run under the same hard constraints: a maximum single operation time of 360 minutes, a maximum total operation time of 540 minutes, and a maximum allowable coverage rate of 20%.

[0083] Figure 10 This is the Pareto solution set distribution diagram of the benchmark NSGA-II algorithm in the experimental example of Embodiment 1 of the present invention. Figure 10 As can be seen, the scattered points are significantly clustered in areas with 200 to 230 vehicles. The total energy consumption of the system is approximately to ( The high-cost range of this distribution characteristic directly reveals that traditional algorithms, lacking spatial clustering and greedy strategies, are prone to falling into local optima and struggle to find economically feasible low-resource-investment solutions in large-scale road networks. This, in turn, strongly supports the significant technical advantages of the improved algorithm in locking in a high-quality solution space and improving solution efficiency.

[0084] Figure 11 The diagram visually illustrates the Pareto front distributions of the improved algorithm (blue sphere) and the baseline NSGA-II algorithm (red sphere) across the two objective dimensions of vehicle number (F1) and total travel distance (F2). The comparison results show that the improved algorithm significantly outperforms the baseline algorithm in three aspects: First, it exhibits a significant dominance advantage. The solution set of the improved algorithm is generally distributed to the lower left of the baseline algorithm, meaning that the improved algorithm achieves better solutions across all objective dimensions. With the same vehicle size, the total path mileage planned by the improved algorithm is significantly shorter; conversely, under the same mileage constraint, it requires a significantly reduced number of vehicles. Second, it has stronger solution space search capabilities. The improved algorithm can search for high-quality solutions within the 40-60 vehicle range, while the baseline algorithm's solutions mainly remain in the high-cost range of over 200 vehicles. This indicates that the improved algorithm can effectively escape local optima and discover extremely low-cost scheduling schemes. Third, there is a huge difference in convergence efficiency. Although the experiment provided the baseline algorithm with 300 generations of evolutionary resources, the quality of its final solution set is still far lower than that of the improved algorithm, which only evolved for 100 generations. This fully demonstrates that the greedy fusion repair decoding strategy and multi-vehicle collaborative mechanism proposed in this study have extremely high search efficiency when dealing with complex constraints.

[0085] Example 2 Based on Example 1, this example provides a path planning system for unmanned sanitation vehicles based on the improved NSGA-II, including: The preprocessing module is configured to acquire information on the work tasks and work vehicles, discretize the road segments corresponding to the work tasks, establish a mapping between tasks and road segments, calculate the shortest distance between the discretized road nodes, and construct a matrix of shortest distances between all nodes. The model building module is configured to build a multi-objective path planning model with the optimization objectives of minimizing the number of sanitation vehicles deployed, minimizing the total energy consumption of the system, and minimizing the proportion of uncovered task areas. The solution module is configured to generate an initial population for vehicle task allocation using a clustering algorithm based on the spatial location characteristics of road segments; combine greedy path simulation with the NSGA-II optimization algorithm to obtain an improved NSGA-II optimization algorithm; and solve the constructed multi-objective optimization model based on the mapping relationship between tasks and road segments and the shortest distance matrix of all nodes to obtain the path planning scheme for vehicle task execution.

[0086] It should be noted that each module in this embodiment corresponds one-to-one with each step in embodiment 1, and their specific implementation process is the same, so it will not be repeated here.

[0087] Example 3 Based on Embodiment 1, this embodiment provides an electronic device, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, they complete the steps in the unmanned sanitation vehicle path planning method based on the improved NSGA-II described in Embodiment 1.

[0088] Example 4 Based on Embodiment 1, this embodiment provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, they complete the steps in the unmanned sanitation vehicle path planning method based on the improved NSGA-II described in Embodiment 1.

[0089] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0090] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A path planning method for unmanned sanitation vehicles based on the improved NSGA-II, characterized in that, Includes the following steps: Obtain information on the work tasks and vehicles, discretize the road segments corresponding to the work tasks, establish a mapping between tasks and road segments, calculate the shortest distance between the discretized road nodes, and construct a shortest distance matrix for all node pairs. A multi-objective path planning model is constructed with the optimization objectives of minimizing the number of sanitation vehicles deployed, minimizing the total energy consumption of the system, and minimizing the proportion of uncovered task areas. Based on the spatial location characteristics of road segments, a clustering algorithm is used to generate an initial population for vehicle task allocation. The greedy path simulation is combined with the NSGA-II optimization algorithm to obtain an improved NSGA-II optimization algorithm. Based on the mapping relationship between tasks and road segments and the shortest distance matrix of all nodes, the constructed multi-objective optimization model is solved to obtain the path planning scheme for vehicles to perform tasks.

2. The unmanned sanitation vehicle path planning method based on the improved NSGA-II as described in claim 1, characterized in that: The method for establishing a mapping between tasks and road segments includes the following steps: The road is divided into multiple road segments according to a set length, and the two ends of each road segment are used as nodes; The adjacent node pairs after segmentation are defined as the smallest task units. Each smallest task unit is assigned a unique task number, and the global node numbers corresponding to both ends of the task segment are recorded to form an endpoint-task mapping table.

3. The unmanned sanitation vehicle path planning method based on the improved NSGA-II as described in claim 1, characterized in that: The process of constructing the shortest distance matrix for all node pairs includes the following steps: Construct an undirected edge set based on the adjacent node pairs after segmentation. By taking the side length as the geometric length between the two nodes, we obtain an undirected weighted adjacency matrix; A graph object is constructed based on an undirected weighted adjacency matrix. The shortest path length between all node pairs is calculated as the matrix element value to obtain the shortest distance matrix of all node pairs.

4. The unmanned sanitation vehicle path planning method based on the improved NSGA-II as described in claim 1, characterized in that: The constraints for solving the multi-objective optimization model include: running time constraints, single-trip range and dynamic charging constraints, task allocation uniqueness constraints, and operation path connectivity constraints; wherein, the task allocation uniqueness constraint means that any task segment can be assigned to at most one vehicle, and unassigned task segments are allowed.

5. The unmanned sanitation vehicle path planning method based on the improved NSGA-II as described in claim 1, characterized in that: The method for solving the constructed multi-objective optimization model to obtain the path planning scheme for vehicle task execution includes the following steps: Step 31: Encode vehicle task groups by merging task numbers with delimiters, generate chromosomes by combining spatial clustering heuristics with random mechanisms, and construct the initial population; Step 32: Analyze the gene sequence in each chromosome, identify the location of the boundary markers, and divide the chromosome into segments. Each subsequence corresponds to a list of task segments for a vehicle. The chromosome is converted into a task assignment for the vehicle, which is then used as the decoding result. Step 33: Based on the decoded results, dynamically determine the initial deployment node for each vehicle, and perform a greedy path simulation with dynamic truncation. Based on the pre-calculated shortest distance matrix of all nodes, fill in the connection cost to generate the vehicle path. Step 34: For the vehicle paths corresponding to each chromosome, calculate the fitness based on the multi-objective optimization model to obtain the number of vehicles, total travel distance and the proportion of uncovered task area. Based on the constraints of the multi-objective optimization model, perform constraint verification. For individuals that pass the constraint verification, perform fast non-dominated sorting and crowding calculation based on the fitness calculation results. Step 35: Perform selection, crossover, mutation, and environmental selection to generate new chromosomes, then execute step 32 for the next iteration until the iteration stopping condition is met, and obtain the Pareto solution set.

6. The unmanned sanitation vehicle path planning method based on the improved NSGA-II as described in claim 1, characterized in that: Vehicle task groups are encoded by merging task numbers with delimiters to obtain each chromosome. The specific encoding methods include the following: Road task numbers are represented by non-zero gene values; Set a delimiter to divide the task sequence into multiple vehicle task segments. Use the K-Means algorithm to spatially cluster all tasks based on the center coordinates of the vehicle task segments to generate pre-grouping of tasks with spatial proximity. Encode task segments belonging to the same cluster into the task sequence of the same vehicle to obtain the encoded chromosome, which serves as the first part of the individuals. The task sequence is randomly shuffled and a delimiter is randomly inserted. The vehicle task segments are assigned to the vehicles, generating a second group of individuals. The first group of individuals and the second group of individuals constitute the initial population. Delimiters are inserted into the task sequence to generate or reconstruct chromosomes. Delimiter insertion must meet the following conditions: The delimiter does not appear at the beginning of the sequence; The delimiter is not continuous; Each vehicle task segment is not empty; If the end of the chromosome after insertion is a delimiter, remove the delimiter to ensure that the end of the output chromosome is not a delimiter.

7. The unmanned sanitation vehicle path planning method based on the improved NSGA-II as described in claim 1, characterized in that: Based on the decoded results, the initial deployment node for each vehicle is dynamically determined, and a greedy path simulation with dynamic truncation is performed. The method for generating vehicle paths, based on the pre-calculated shortest distance matrix of all nodes, involves plucking connection costs to complete the path. This includes the following steps: With the goal of completing all task segments with the shortest total travel distance, the starting point is determined by using the shortest distance matrix of all nodes to determine one of the endpoint nodes of the first task segment, which is the initial deployment node of each vehicle. The greedy selection and dynamic truncation are performed, which means starting from a determined starting point, executing tasks with the goal of minimizing the connection distance, iterating greedily to determine the next task, until all tasks have determined the vehicle path and the connection cost is filled. During the selection process, the vehicle's cumulative driving time is checked in real time. If adding the next task would cause the driving time threshold to be exceeded, the task allocation for that vehicle will be stopped and the vehicle will be forced to return to the starting point. The remaining unassigned tasks will be marked as uncovered. Alternatively, a crossover process can be performed to generate new chromosomes. Specifically, for the parent individuals undergoing the crossover process, the delimiters between the two parent individuals are removed, the task order is cross-combined, and then delimiters are inserted according to the proportion of vehicles in the original individual scheme to obtain new candidate schemes. Alternatively, the mutation process may include task exchange mutation and delimiter mutation: Task swap mutation: For the individual to be mutated, two locations are randomly selected from all non-delimiter task gene positions and their task numbers are swapped to disrupt the task execution order. Delimiter mutation: Insert a delimiter at a random position or randomly delete an existing delimiter to change the number of vehicles.

8. A path planning system for unmanned sanitation vehicles based on the improved NSGA-II, characterized in that, include: The preprocessing module is configured to acquire information on the work tasks and work vehicles, discretize the road segments corresponding to the work tasks, establish a mapping between tasks and road segments, calculate the shortest distance between the discretized road nodes, and construct a matrix of shortest distances between all nodes. The model building module is configured to build a multi-objective path planning model with the optimization objectives of minimizing the number of sanitation vehicles deployed, minimizing the total energy consumption of the system, and minimizing the proportion of uncovered task areas. The solution module is configured to generate an initial population for vehicle task allocation using a clustering algorithm based on the spatial location characteristics of road segments; combine greedy path simulation with the NSGA-II optimization algorithm to obtain an improved NSGA-II optimization algorithm; and solve the constructed multi-objective optimization model based on the mapping relationship between tasks and road segments and the shortest distance matrix of all nodes to obtain the path planning scheme for vehicle task execution.

9. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the processor executes the computer instructions, it performs the steps in the path planning method for unmanned sanitation vehicles based on the improved NSGA-II as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, complete the steps in the path planning method for unmanned sanitation vehicles based on the improved NSGA-Ⅱ as described in any one of claims 1-7.

Images