A vehicle route optimization method for waste collection

Abstract

This invention discloses a vehicle routing optimization method for waste collection, comprising constructing a multi-objective cost function integrating fixed deployment, dynamic load-bearing fuel consumption, and processing station variance penalties; rigidly configuring a "fleet size reduction mechanism" as an external iterative global constraint; and relentlessly squeezing idle capacity through rolling probing. The underlying hybrid improved metaheuristic algorithm incorporates a load balancing heuristic factor and seamlessly coordinates with a variable neighborhood descent search operator for node temporal reconstruction. This invention utilizes a collaborative clustering algorithm with capacity overflow blocking to perform macroscopic cluster-level dimensionality reduction on discrete nodes, significantly reducing the overall fleet size, lowering comprehensive energy consumption, and achieving balanced distribution of throughput pressure on sanitation physical processing facilities. It can be applied to vehicle routing optimization in complex waste collection networks encompassing multiple vehicle depots and processing stations, solving the problems of redundant transport resources and unbalanced load on underlying facilities in large-scale road networks.

Description

Technical Field

[0001] This invention relates to the field of route optimization technology, and more specifically to a vehicle route optimization method applied to waste collection. Background Technology

[0002] In modern waste collection and transportation operations, fine-grained optimization of vehicle physical routes is a core element in reducing operating costs and improving facility efficiency. Traditional constrained vehicle routing problems (CVRP) typically focus on continuous path optimization within a single vehicle depot logic. However, real-world waste collection and transportation networks often exhibit complex topologies with multiple vehicle depots and processing stations intertwined, and heavily rely on discontinuous operational logic involving "loading-unloading-reloading." Existing route planning models struggle to resolve such dynamically coupled constraints involving multiple unloading stages, leading to significant deviations between the generated scheduling schemes and the actual physical flow lines.

[0003] Besides the limitations of the path continuity assumption, the optimization strategy widely adopted by existing technologies, which is oriented towards "shortest single distance," is also seriously detached from real-world financial structures and operational considerations. In real-world scenarios, total operating costs not only include fuel consumption related to mileage, but also high depreciation of vehicle fixed assets and labor costs. Simply pursuing the shortest travel distance can easily lead to local optima, and may even cause an imbalance in the workload distribution among processing facilities—some stations are overloaded and queued, while other stations are idle, resulting in high hidden penalty costs.

[0004] Faced with large-scale network nodes, existing methods often struggle to balance computational efficiency and solution quality. To prevent excessive vehicle idleness and ensure balanced workload, dispatched vehicles typically need to meet a "minimum service point coverage constraint," which existing clustering routing techniques lack effective support for. Furthermore, existing heuristic algorithms generally lack built-in optimization designs for "fleet size reduction." For example, adaptive route planning methods based on ant colony clustering often treat the number of available vehicles as a fixed input parameter when applied to fleet optimization. In actual operation, the savings in fixed assets and labor costs from reducing the number of dispatched vehicles far outweigh the fuel cost savings from simply reducing mileage. Existing algorithms lack a feedback mechanism that can proactively detect redundant capacity and forcibly squeeze idle capacity during the optimization process to reduce the size of the dispatched fleet. Summary of the Invention

[0005] The purpose of this invention is to provide a vehicle routing optimization method for waste collection, minimizing the overall comprehensive operating cost of the system while covering all waste collection points, not exceeding vehicle capacity limits, adhering to the network logic of multiple vehicle depots and processing stations, and complying with various municipal operational constraints. This cost model deeply integrates distance-related dynamic fuel costs, labor hour costs calculated based on service and travel time, fixed vehicle usage costs positively correlated with fleet size, and resource imbalance penalty costs for uneven workload distribution and uneven resource allocation at processing stations. This addresses the problems mentioned in the background section.

[0006] The present invention provides the following technical solution: a vehicle route optimization method for waste collection, comprising the following operational steps:

[0007] Step 1: Obtain the geographical and demand data of the waste collection network to be processed, perform data parsing and preprocessing, and simultaneously construct a multi-dimensional network diagram.

[0008] Preferably, the geographical and demand data of the waste collection network to be processed are acquired, analyzed, and preprocessed to construct a multi-dimensional weighted directed graph model containing multiple vehicle originating and returning depots, multiple waste treatment stations with unloading capabilities, and a large number of waste collection points. Three types of node sets are defined: a set of nodes from multiple depots, a set of nodes from multiple treatment stations, and a set of nodes from waste collection points. The expected waste generation, collection priority, geographical coordinates, and minimum service point coverage constraints for each waste collection point are extracted.

[0009] Step 2: Using a collaborative clustering algorithm with capacity constraints and geographical barrier awareness, the bottom-level garbage collection points are aggregated into multiple macro-collection clusters to perform initial spatial dimensionality reduction allocation for the ultra-large-scale network.

[0010] Preferably, the specific implementation includes: using the optimized K-means++ algorithm, randomly selecting initial seed points in the geographic space as cluster centers, and ensuring the physical distance between seed points is dispersed through probability distribution control; performing constraint classification operations for compatibility and capacity checks; dynamically tracking and accumulating the total demand of already allocated nodes within the cluster when calculating the comprehensive topological distance from each bottom collection point to each cluster center; triggering a capacity overflow blocking mechanism when the system determines that assigning a new node to a specific cluster will cause the total waste demand of the cluster to exceed the preset single-vehicle baseline capacity warning line; forcibly removing the new node from its original target and redistributing it to neighboring clusters with spare capacity within the spatial range according to the distance gradient; after the clustering iteration is completed, treating each cluster as a macroscopic replacement node.

[0011] Preferably, to reduce the dimensionality of solving ultra-large-scale routing problems, a collaborative clustering algorithm with capacity constraints and geographical barrier awareness is adopted to aggregate low-level garbage collection points with similar geographical locations and compatible features into multiple macro-collection clusters. During this process, the total demand within each cluster is dynamically calculated to ensure that it meets the initial allocation logic for single or multiple vehicles without causing severe capacity overflow.

[0012] Step 3: Construct a multi-objective integrated operating cost mathematical model.

[0013] Preferably, a mathematical model is established with the absolute orientation of minimizing the overall comprehensive operating cost. Its objective function covers four dimensions of sub-costs: dynamic fuel and environmental emission costs based on driving distance and dynamic load changes; labor hours and tiered overtime costs based on the total working hours and working hours of drivers; fixed resource deployment costs of vehicles that are linearly positively correlated with the total number of vehicles actually on duty; and resource distribution imbalance penalty costs calculated based on the variance of the receiving load of each waste treatment station.

[0014] Preferably, the multi-objective integrated operating cost mathematical model simultaneously satisfies and is subject to the following constraints:

[0015] Absolute node coverage constraint: ensures that every garbage collection point must be served exactly once;

[0016] Network traffic conservation constraint: After a vehicle reaches any collection node, it must eventually leave that node.

[0017] Dynamic vehicle capacity constraint: The accumulated garbage load inside a vehicle must not exceed the physical limit of that vehicle type, and its internal load variable will be immediately cleared to zero after the vehicle enters the processing station and completes the dumping.

[0018] Minimum number of service nodes constraint: For any vehicle that is officially put into use, the total number of nodes it serves must be greater than or equal to a preset threshold.

[0019] Multi-depot and multi-processing station collaborative closed-loop constraint: dispatched vehicles must and can only depart from a certain available depot, must proceed to a certain processing station to unload when the capacity along the route is saturated, and must finally return to a certain depot to complete the handover closed loop.

[0020] Step 4: Based on the clustering results and the constructed mathematical model, construct an initial feasible solution for the fleet size reduction mechanism.

[0021] Preferably, based on the established mathematical model and clustering results, a forced decreasing fleet size probe algorithm is set in the outer framework of route allocation. During the initial route construction phase, routes are probed with the minimum number of vehicles below the theoretical margin, causing the scheduling logic to squeeze out the idle capacity of all available vehicles. This reduces redundant fleet size and overall manpower requirements by increasing the average load factor. In other words, the fleet size reduction mechanism is configured as a global constraint condition for the algorithm's outer iteration, iteratively reducing from a preset initial margin of vehicles. Through a rolling probe method based on a defect function, the upper limit of the maximum available vehicle number is forcibly reduced generation by generation, forcing subsequent route generation logic to squeeze out redundant idle capacity within the system.

[0022] Step 5: Use a hybrid improved ant colony algorithm to perform global collaborative optimization at both the macro-cluster level and the micro-node level.

[0023] Preferably, the heuristic state transition probability rule is designed by combining travel distance, node demand loading fitness, facility workload balance and minimum service constraint; and in the iterative evolution of the algorithm, a global pheromone update and local pheromone evaporation mechanism based on solution quality penalty is used to guide the ant colony to break out of the limitation of local optimal solutions.

[0024] Step 6: The cluster-level alternative solution generated by the ant colony algorithm is restored to the specific underlying node level through spatial mapping, realizing local search and path iteration improvement based on variable neighborhood descent.

[0025] Preferably, based on variable neighborhood descent local search and path iteration improvement, the cluster-level alternative solution generated by the ant colony algorithm is restored to the specific underlying node level through spatial mapping. To address potential local path intersections and capacity waste after restoration, a variable neighborhood descent (local search strategy) is applied. Operation operators including single-node relocation, two-node swap, two-arc swap within the path (2-opt), and multi-vehicle path cross-exchange are executed to further reduce system redundancy costs.

[0026] Step 7: Evaluate the overall cost fitness of the current iterative solution, determine whether the cost reduction rate within the continuous iteration cycle meets the preset convergence termination condition, and output the optimal scheduling scheme when the condition is met.

[0027] Preferably, the overall cost fitness of the current iterative solution is evaluated to determine whether the cost reduction rate within the continuous iteration cycle meets the preset convergence termination condition. When the condition is met, the system will output the globally optimal vehicle route planning scheme and a precise fleet configuration list, including the starting depot, service node time sequence, garbage unloading and processing stations along the route, and the final return depot node for each vehicle.

[0028] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0029] (1) This invention represents a fundamental shift from a single distance-oriented approach to a comprehensive operating cost-oriented approach. The optimization model of this invention not only considers the traditional vehicle travel distance, but also incorporates the daily fixed asset depreciation cost of vehicles, the labor cost of drivers and loading / unloading personnel, and the penalty for queuing congestion at processing stations due to uneven facility utilization into the global objective function. This multi-objective evaluation mechanism makes the generated routing and scheduling scheme more economically efficient and socially equitable in actual financial settlement, avoiding the blind spot where local mileage reduction leads to an increase in global costs.

[0030] (2) This invention significantly reduces the total fleet size and total manpower requirements of sanitation departments. By deeply embedding a fleet size reduction mechanism into the initial solution construction and iterative optimization logic, the algorithm can spontaneously find high-load-rate route combinations under strict constraints on the upper limit of available vehicles, effectively squeezing out redundant transport capacity and idle capacity waste within the system. Compared with traditional algorithms that use the number of vehicles as a static input, this invention can deploy fewer vehicles to complete the same garbage collection task while ensuring the same service quality and city-wide coverage, thereby greatly reducing capital expenditure and daily operating manpower expenses.

[0031] (3) This invention effectively solves the problem of efficiently solving large-scale complex road networks. This invention draws on and improves the idea of ​​variable clustering dimensionality reduction in embedded space. Through regional collaborative clustering and intelligent allocation strategies, it reduces the dimensionality of the nondeterministic polynomial time complexity (NP-hard) vehicle routing problem with tens of thousands of nodes. At the same time, by combining the improved ant colony algorithm and the multidimensional variable neighborhood descent local search algorithm, the computation time is significantly shortened while ensuring global optimization capability and solution quality, thus meeting the real-time requirements of daily dynamic scheduling.

[0032] (4) This invention fully satisfies the highly complex real-world constraints of modern daily operations. The solution of this invention naturally supports the complex network logic of joint scheduling of multiple vehicle depots and multiple unloadings of multiple processing stations. At the same time, the model takes into account the special operational lower limit constraint that "each dispatched vehicle must serve at least a specified number of collection points" to prevent vehicle overload or extremely uneven utilization, thereby improving the labor compliance and applicability of the scheduling solution in real-world scenarios. Attached Figure Description

[0033] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0034] Figure 1 This is a schematic diagram of the multi-vehicle depot and multi-processing station waste collection architecture and optimization provided in an embodiment of the present invention;

[0035] Figure 2 This is a schematic diagram of the deconstruction of the multi-objective integrated operating cost mathematical model provided in an embodiment of the present invention;

[0036] Figure 3 This is a schematic diagram of a heuristic path construction decision tree with a fleet size reduction mechanism provided in an embodiment of the present invention;

[0037] Figure 4 This is a schematic diagram of the topology change of the variable neighborhood descent local search (VND) operation operator provided in an embodiment of the present invention. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Combination Figures 1-4 As shown, addressing the technical problems of existing technologies such as using only distance as the optimization objective, neglecting fixed resource costs, lacking a fleet size reduction mechanism, and difficulty in adapting to collaborative scheduling of multiple depots and processing stations, this embodiment of the invention executes the method on a computing device or computing cluster containing a processor and memory. That is, the computing task in this embodiment requires two types of hardware environments: one is a computing device equipped with at least one high-performance central processing unit (CPU), graphics processing unit (GPU), and a relatively large amount of memory; the other is a cloud computing cluster based on a distributed architecture. The system responsible for the computation collects road network topology information and sensor data from IoT trash cans from a dedicated interface. Therefore, this invention specifically provides the following technical solution: a vehicle route optimization method applied to garbage collection, comprising the following operational steps:

[0040] Step 1: Data preprocessing and multidimensional network topology construction.

[0041] In this embodiment, the system of the present invention receives basic road network data and historical and real-time operational demand data provided by the environmental sanitation management terminal, and transforms it into a rigorous mathematical graph theory model. A statically weighted directed graph is defined. , where the vertex set The specific definitions of each set are shown in Table 1.

[0042] Table 1. Set Definition Table

[0043]

[0044] According to the diagram For each node and edge in the system, the system preprocesses and extracts key business parameters in the following dimensions, as shown in Table 2.

[0045] Table 2 Key Business Parameters

[0046]

[0047] At this stage, the system must... Constraints are preprocessed. In real-world scenarios, to prevent excessive pursuit of route intensification in certain areas from leading to severe work-rest imbalances for individual drivers or causing some vehicles to operate under overload conditions for extended periods, a minimum service point constraint must be implemented for vehicles on duty. If the number of service points on a pre-assigned route is less than... If the path is invalid, it is logically deemed invalid and needs to be reassigned.

[0048] Step 2: Customer point clustering based on spatial collaboration and capacity constraints.

[0049] In this embodiment, for node scales of tens of thousands or even hundreds of thousands, traditional metaheuristic algorithms easily fall into the "combinatorial explosion" dilemma of exponential growth, causing the solution time to exceed the upper limit acceptable for real-world scheduling. Therefore, this invention introduces a spatially embedded collaborative clustering strategy to achieve effective dimensionality reduction for ultra-large-scale path problems.

[0050] For example, the specific implementation process is as follows: The system uses an optimized K-means++ algorithm to randomly select initial seed points as cluster centers in the geographic space, and ensures that the physical distances between seed points are as dispersed as possible through probability distribution control. Subsequently, a constraint classification operation for compatibility and capacity checks is performed. When calculating the comprehensive topological distance from each bottom-level collection point to each cluster center, the system dynamically tracks and accumulates the total demand of the already allocated nodes within the cluster. .

[0051] When the system determines that assigning a new node to a specific cluster would cause the total garbage demand of that cluster to exceed the preset single-vehicle baseline capacity warning line (e.g., ,in When the capacity reaches the average capacity of the fleet, the system will trigger a capacity overflow blocking mechanism. At this time, the node will be forcibly removed from its original target and redistributed to a neighboring cluster with available space based on the distance gradient.

[0052] After clustering iterations, the system treats each cluster as a macroscopic substitute node. This clustering abstraction mechanism significantly compresses the original massive network of variable nodes. This dimensionality reduction mechanism keeps the scale of subsequent complex vehicle scheduling solutions within the efficient processing range of the computing cluster, while ensuring the topological rationality of the initial solution.

[0053] Step 3: Construct a multi-objective integrated operating cost mathematical model.

[0054] In this embodiment, most basic models focus only on reducing driving mileage, while this invention, based on the actual financial accounting system, constructs a comprehensive cost function that fully reflects the efficiency of operating expenditures and the utilization of social resources. This invention defines the binary and continuous decision variables required to construct the mathematical model, as shown in Table 3.

[0055] Table 3 Decision Variables Table

[0056]

[0057] For example, the core objective function of this invention The linear weighted combination aimed at minimizing the following four components of cost is expressed as follows:

[0058]

[0059] Among them, the fixed resource deployment cost of vehicles ( This item covers the average daily fixed expenses (including hardware depreciation, fixed insurance, and maintenance amortization) incurred by vehicles that are officially activated and assigned tasks. This indicator is directly linearly correlated with the total number of vehicles in use and is the core reason for the "fleet size reduction" mechanism, representing the fixed resource deployment cost of vehicles. Represented as:

[0060]

[0061] in, This represents the daily fixed asset amortization cost per vehicle. This large base number forces the algorithm engine to minimize the actual number of vehicles dispatched to prevent idle capacity.

[0062] Dynamic fuel and environmental emissions costs ( Traditional route models make a rough assumption that distance directly equates to cost, but this cost component breaks that assumption. A vehicle's actual fuel consumption is not solely determined by the distance traveled. It is also determined by the vehicle's real-time load (garbage accumulates continuously in the vehicle, and heavy-duty vehicles consume several times more fuel and emit several times more carbon per kilometer), resulting in dynamic fuel and environmental emission costs. Represented as:

[0063]

[0064] in, and These represent the combined fuel and emissions rates per unit distance under no-load and fully-load conditions, respectively.

[0065] Labor hours and tiered overtime costs This sub-item is linked to the total working hours of drivers and conductors and strictly adheres to labor regulations. Shifts exceeding the standard (e.g., 8 hours per day) will be subject to penalties. Applying punitive tiered overtime rates to labor hours and tiered overtime costs Represented as:

[0066]

[0067] in, For vehicles Total travel time to complete all tasks on the same day (including travel time for all road segments) Service dwell time at each site ).

[0068] The cost of punitive measures for uneven distribution of social resources ( To prevent the scheduling algorithm from directing all vehicles to the same nearby waste treatment station, leading to massive backlogs and hours of waiting while other remote stations remain idle, this invention introduces an operational resource balancing penalty based on the variance of the receiving volume of each treatment station.

[0069]

[0070] in For processing station The total amount of waste to be received in the planning scheme, The expected value (mean) of the amount received by all processing stations. An imbalance penalty coefficient set by the sanitation department based on its actual production capacity.

[0071] In addition, to ensure the absolute feasibility of the scheduling scheme in real-world scenarios, the mathematical model of this invention is also subject to the following multi-dimensional strict constraints:

[0072] Absolute node coverage constraint: ensures that every garbage collection point must be served exactly once.

[0073]

[0074] Network traffic conservation constraint: To maintain route continuity, vehicles must eventually leave any collection node after reaching it.

[0075]

[0076] Dynamic vehicle capacity constraints: The accumulated waste load inside a vehicle must not exceed the physical limit of that vehicle type. And when the vehicle enters the processing station Once the dumping is complete, its internal load variable will be immediately cleared to zero.

[0077]

[0078] Minimum number of service nodes constraint: For any vehicle that is officially put into use (That is, the total number of service points for the vehicle is greater than 0), and the total number of nodes it serves must be greater than or equal to a preset threshold.

[0079]

[0080] Multi-depot and multi-processing station collaborative closed-loop constraint: dispatching vehicles must and can only depart from a certain available depot, must go to a certain processing station to unload when the capacity along the way is saturated, and must finally return to a certain depot to complete the handover closed loop (the depot to which they return may not be the originating depot).

[0081] Step 4: Probe construction logic with built-in fleet size reduction mechanism.

[0082] In this embodiment, most existing route planning algorithms treat the number of available vehicles as a fixed constant. However, this invention encapsulates a layer of proactively perceptive fleet size reduction logic around its algorithm.

[0083] For example, the specific execution method is as follows: Before the main loop starts, the system first calculates the theoretical minimum number of vehicles required, adds a safety redundancy buffer value, and sets this value as the upper limit of the initial number of available vehicles. Then, an exploration was initiated. Before each external iteration of the algorithm, the system used a rolling time window and defect function mechanism to forcibly tighten the vehicle number limit, that is, to reduce the maximum number of available vehicles by one. ).

[0084] The system requires the underlying routing algorithm to find a feasible solution that satisfies capacity and minimum service point constraints in this artificially created, extremely scarce vehicle environment. Due to the objective function... In Cost is given a very high optimization weight. The system will spontaneously squeeze out redundant capacity in all parts of the system by increasing the average load factor of each dispatched vehicle and boldly merging nearby scattered routes at the boundary.

[0085] This exploration will continue until the algorithm is completely unable to perform the specified task. If any valid solution is found under the given conditions (i.e., any combination of routes will cause the capacity hard constraint to be broken), then the system will determine the current... Having reached the physical limit, the system reverted to the number of vehicles successfully verified in the previous stage and solidified it as the optimal theoretical fleet size scheme for this cycle.

[0086] Step 5: Cost-driven hybrid improved ant colony algorithm for global optimization.

[0087] In this embodiment, for the large-scale model constructed above that contains multiple highly nonlinear constraints, the present invention employs a hybrid improved ant colony algorithm to conduct large-scale parallel search on macroscopic clusters and microscopic nodes.

[0088] For example, in the dynamic evolution of the initial scheduling solution, a virtual ant is used to represent a garbage truck in the scheduling state. When the vehicle Located at the current node When that happens, it chooses to go to the next unvisited neighboring node. The probability rule is determined by the following heuristic state transition formula:

[0089]

[0090] In this core transfer formula, Indicates the connection node With nodes The current concentration of virtual pheromones deposited along the path edge represents the group's historical experience. The system defines a comprehensive heuristic factor representing the relationship between fusion distance and fuel consumption characteristics as the reciprocal of operating cost. This will prompt ants to prioritize adjacent road segments with shorter travel times and lower fuel consumption when choosing local routes.

[0091] This is one of the core original creations of this invention: the resource and dynamic loading balancing heuristic factor. This factor evaluates nodes in real time. Will the anticipated amount of waste prompt current vehicle usage? The cumulative load rate smoothly and quickly enters the optimal economic operating range (set at 85% to 95% of capacity). Meanwhile, if the node... It belongs to a waste treatment station. The system will then retrieve the current cumulative unloading task volume allocated to that processing station. If it finds that the station is approaching the overload threshold, it will immediately reduce the weight coefficient of that factor, forcing the ants to bypass the congested processing station and instead allocate resources to other idle facilities, thereby continuously maintaining a dynamic balance of the global load. Parameters It is an adjustable sensitivity weight index used to regulate the game relationship between pheromones and various heuristic factors in decision-making.

[0092] To avoid the premature convergence and stagnation problems that easily occur in the later stages of traditional ant colony algorithms, this invention incorporates a dual-track pheromone update defense system. On one hand, it performs a local evaporation operation: whenever an ant crosses an edge... The system immediately performs a slight pheromone concentration reduction and evaporation operation on the local edge, thereby proactively reducing the probability of subsequent ants blindly following, greatly expanding the breadth of exploration of the diversity of the learning space. On the other hand, a global reward and punishment mechanism is implemented: after the entire ant colony completes a round of full network traversal, the system rewards and punishes the ants that synthesized the objective function in this round. The elite solution with the lowest value applies a high positive pheromone reward to the path it covers. It's worth emphasizing that the system introduces a strict veto penalty logic: if a generated solution is found to have a vehicle route service point count below a mandatory threshold... The system not only deprives it of the right to obtain global rewards, but also imposes additional punitive decay on the inferior paths it has taken, thereby quickly filtering and eliminating these non-compliant and discarded schemes at the pheromone level.

[0093] Step 6: Optimize the underlying local search based on multidimensional variable neighborhood descent (VND).

[0094] In this embodiment, after the hybrid ant colony algorithm generates a preliminary global skeleton scheme based on macroscopic substitution clusters, the system initiates a mapping and restoration mechanism to meticulously expand the coarse routing network down to the lowest level of a single garbage bin node. Given the inherent stochastic limitations of metaheuristic algorithms, the expanded microscopic solutions inevitably exhibit entanglement and detours in local street layout details. Therefore, it is necessary to introduce variable neighborhood descent to conduct extremely rigorous deep and fine-grained adjustments.

[0095] For example, the present invention designs a four-layer progressive neighborhood structure operator, which eliminates every redundant cost without violating global constraints. The four-layer progressive neighborhood structure operator includes:

[0096] Single-node relocation operator (Relocate): The system identifies all high-time-consuming nodes and attempts to relocate them from their original inefficient routes. It was completely separated from the middle and tentatively inserted into a route that was closer in distance. This operation is based on the premise that system simulations prove the insertion will absolutely not exceed the route. The maximum physical capacity of a vehicle.

[0097] Two-node swap operator (Swap): Between two parallel routes belonging to different vehicles, the system finds two collection points with similar demand levels but whose geographical locations overlap due to initial clustering defects, and performs an equal swap operation to straighten out their respective driving trajectories.

[0098] 2-opt path internal respin operator: For trajectory knots and overlaps caused by detours within a single vehicle, the system logically cuts off two intersecting edges and reverses the temporal sequence of a segment of nodes in the middle, stitching them together with new non-intersecting edges, thus purely compressing the path. (Fuel costs) and reduce runtime.

[0099] Cross-exchange operator across vehicle domains and multiple stations: This is a macro-level operator specifically designed for multi-processor station architectures. When the system detects a vehicle... and They went to the processing station respectively and During the uninstallation process, the system will The sequential task sequence before and after uninstallation, and The corresponding sequence undergoes large-scale segmentation and transposition. This approach directly impacts and undermines the objective function. (Resource imbalance penalty) smoothly transfers the waste load that was originally over-concentrated at a single processing station to other processing stations, just like a flood diversion.

[0100] For example, each fine-tuning attempt triggered by the above operator will be put back into the huge and complex fitness function F by the system to complete the recalculation of the global cost. The system will only accept and solidify the transformation operation, provided that the feedback result can prove that the overall comprehensive cost has indeed decreased. This logic is entirely based on cost control.

[0101] Step 7: Convergence determination and optimal scheduling scheme system output.

[0102] In this embodiment, the system of the present invention monitors continuous If the global optimal solution data of the algorithm iteration generation shows that the improvement of the fitness function is lower than the set small threshold, or the calculation time has reached the warning line of the longest calculation time allowed by the daily early shift scheduling, the system will automatically terminate the iterative optimization process.

[0103] Finally, the system outputs a globally optimal vehicle route planning scheme and fleet configuration combination asset report to the sanitation dispatch terminal after multiple fine-tuning. The output data structure is no longer just a few simple route maps, but includes: the number of vehicles dispatched after precise reduction of the minimum fleet size and the corresponding heterogeneous vehicle allocation list; the departure time of each activated vehicle from the depot accurate to the minute, and the strict time sequence of instructions for serving each garbage bin and street node; the dynamic expected load change curve of each vehicle along the route; the on-time dumping queuing time node upon arrival at the designated processing facility; and the crucial estimated total operating cost report (which lists the estimated total fuel consumption pre-settlement, manpower attendance and overtime costs, facility resource balance index and other key operational analysis indicators).

[0104] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0105] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A vehicle route optimization method applied to waste collection, characterized in that: The following steps are included: Step 1: Obtain the geographical and demand data of the waste collection network to be processed, perform data parsing and preprocessing, and simultaneously construct a multi-dimensional network diagram; Step 2: Using a collaborative clustering algorithm with capacity constraints and geographical barrier awareness, the bottom-level garbage collection points are aggregated into multiple macro-collection clusters to perform initial spatial dimensionality reduction allocation for the ultra-large-scale network; Step 3: Construct a multi-objective integrated operating cost mathematical model; Step 4: Based on the clustering results and the constructed mathematical model, construct an initial feasible solution for the fleet size reduction mechanism; The initial feasible solution for constructing the fleet size reduction mechanism includes: configuring the fleet size reduction mechanism as a global constraint condition for the outer layer iteration of the algorithm, starting from the preset initial number of vehicles to iteratively reduce the fleet size, and forcibly reducing the upper limit of the maximum number of available vehicles generation by generation through a rolling detection method based on the defect function, thereby forcing the subsequent route generation logic to squeeze out the redundant idle capacity in the system. Step 5: Use a hybrid improved ant colony algorithm to perform global collaborative optimization at both the macro-cluster level and the micro-node level; In the dynamic evolution of the initial scheduling solution, a virtual ant is used to represent a garbage truck in the scheduling state; When the vehicle Located at the current node When that happens, it chooses to go to the next unvisited neighboring node. The probability rule is determined by the heuristic state transition formula; The heuristic state transition formula synchronously introduces a resource and dynamic loading balancing heuristic factor to evaluate neighboring nodes in real time. Will the anticipated amount of waste prompt current vehicle usage? The cumulative load rate is smooth and falls within the optimal economic operating range; If adjacent nodes As it belongs to the waste treatment station, the resource and dynamic loading balance heuristic factor retrieves the current cumulative unloading task volume allocated to the treatment station. If it is found that it is close to the overload red line, the weight coefficient of the factor is immediately reduced, so that the ants bypass the congested treatment station and go to other idle facilities. The global collaborative optimization includes: using a hybrid improved ant colony algorithm combined with a fleet size reduction mechanism to perform a path search combining macro clusters and micro nodes; when calculating the transfer probability, introducing the traditional pheromone and path distance reciprocal heuristic factor, and adding a dynamic resource state heuristic mechanism that comprehensively considers the vehicle's cumulative load saturation and the corresponding unloading facility workload balance factor to generate an initial route scheduling scheme. Step 6: The cluster-level alternative solution generated by the ant colony algorithm is restored to the specific underlying node level through spatial mapping, realizing local search and path iteration improvement based on variable neighborhood descent; It also includes four progressive neighborhood structure operators: single-node relocation operator, two-node swap operator, 2-opt path despin operator, and cross-vehicle domain multi-site cross-swapping operator; Step 7: Evaluate the overall cost fitness of the current iterative solution, determine whether the cost reduction rate within the continuous iteration cycle meets the preset convergence termination condition, and output the optimal scheduling scheme when the condition is met.

2. The vehicle route optimization method for waste collection according to claim 1, characterized in that: Step 2 includes the following steps: The optimized K-means++ algorithm is used to randomly select initial seed points in the geographic space as cluster centers, and the physical distance between seed points is dispersed through probability distribution control. Perform constraint classification operations for compatibility and capacity checks; When calculating the comprehensive topological distance from each bottom collection point to each cluster center, the total demand of the already allocated nodes within the cluster is dynamically tracked and accumulated. When the system determines that assigning a new node to a specific cluster would cause the total garbage demand of that cluster to exceed the preset single-vehicle baseline capacity warning line, the capacity overflow blocking mechanism is triggered. The new node is forcibly removed from its original target and redistributed to a neighboring cluster with available space according to the distance gradient. After the clustering iteration is completed, each cluster is regarded as a macroscopic substitute node.

3. The vehicle route optimization method for waste collection according to claim 2, characterized in that: The comprehensive operating costs in step 3 include: fixed vehicle resource deployment costs, dynamic fuel and environmental emission costs, labor hours and tiered overtime costs, and penalty costs for unbalanced resource distribution.

4. The vehicle route optimization method for waste collection according to claim 3, characterized in that: In dynamic fuel and environmental emission costs: the costs are dynamically adjusted based on the instantaneous garbage load of the current road segment where the vehicle is located, and quantified by a defined fuel consumption coefficient function; In the resource imbalance penalty cost: When simulating scheduling, the algorithm counts and calculates the total garbage load of each garbage unloading and processing station in the network that has been assigned a task in real time. It extracts the statistical variance of this data and assigns a weight factor that is dynamically adjusted based on the total cost of the road network. The larger the variance value, the more the system will over-concentrate the unloading task on a very small number of stations, and the penalty cost will increase accordingly. In the fitness evaluation stage, the algorithm is forced to distribute the unloading task evenly to each processing station.

5. The vehicle route optimization method for waste collection according to claim 4, characterized in that: The multi-objective integrated operating cost mathematical model constructed in step 3 simultaneously satisfies and is subject to the following constraints: Absolute node coverage constraint: ensures that every garbage collection point must be served exactly once; Network traffic conservation constraint: After a vehicle reaches any collection node, it must eventually leave that node. Dynamic vehicle capacity constraint: The accumulated garbage load inside a vehicle must not exceed the vehicle's physical limit, and its internal load variable will be immediately cleared to zero after the vehicle enters the processing station and completes the dumping. Minimum number of service nodes constraint: For any vehicle that is officially put into use, the total number of nodes it serves must be greater than or equal to a preset threshold. Multi-depot and multi-processing station collaborative closed-loop constraint: dispatched vehicles must and can only depart from a certain available depot, must proceed to a certain processing station to unload when the capacity along the route is saturated, and must finally return to a certain depot to complete the handover closed loop.

6. The vehicle route optimization method for waste collection according to claim 5, characterized in that: The hybrid improved ant colony algorithm also includes a cooperative local pheromone evaporation mechanism and a globally optimized pheromone update mechanism with rule veto power; for discarded inferior solutions with short total travel distances but which are judged to have violated the lower limit constraint of "minimum number of service nodes" in the legality verification process, an additional severe punitive concentration decay deduction operation is performed on the core violation road segments traversed by the inferior solution, and invalid path features are cut off in the initial stage of algorithm iteration.

7. The vehicle route optimization method for waste collection according to claim 6, characterized in that: The cross-vehicle domain multi-site cross-exchange operator also includes: under the absolute premise of strictly meeting the maximum physical capacity limit of all vehicles and the open service time window of each site after the operation is executed; performing exchange operations on the path node sequence at the entry and exit of different operation paths at the processing station; performing large-segment modular swapping and seamless reassembly of the path node sequence of different operation vehicles before unloading at the intermediate processing station and after unloading and leaving the processing station, reducing unnecessary long-distance empty round-trip routes caused by the rough allocation in the early stage, and directly intervening and coordinating the difference in throughput workload between major waste treatment facilities from a macro perspective.

Images