A Flexible Scheduling Method for Charging Autonomous Vehicles Based on Nonlinear Spatiotemporal Impedance and Dynamic MOPSO

By improving the multi-objective particle swarm optimization algorithm and dynamic update mechanism, the adaptive scheduling problem of unmanned charging vehicles in complex road networks is solved, achieving efficient and safe flexible scheduling, reducing accident risk and response delay.

CN123070522APending Publication Date: 2026-11-10HUAIYIN INSTITUTE OF TECHNOLOGY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611044527.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-14
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing mobile charging systems cannot adaptively avoid spatiotemporal barriers such as dynamic congestion and risky restricted areas in complex road networks, resulting in unmanned charging vehicles being incorrectly assigned, increasing the risk of safety accidents, and causing calculation divergence or response timeouts, making it impossible to achieve efficient and safe flexible scheduling.

Method used

A flexible scheduling method for charging unmanned vehicles based on nonlinear spatiotemporal impedance and dynamics MOPSO is adopted. By combining an improved multi-objective particle swarm optimization algorithm with Lyapunov convergence boundary and time-varying learning factor, multiple variable transfer path nodes are enabled, and a continuous dynamic update mechanism and simulated annealing mutation mechanism are introduced to optimize path planning and scheduling.

Benefits of technology

It enables efficient and safe scheduling of unmanned charging vehicles in complex road networks, reduces the risk of high-frequency oscillations, improves the robustness and responsiveness of the algorithm, and enables flexible scheduling in high-concurrency scenarios, avoiding topology deadlock and accident risks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123070522A_ABST
    Figure CN123070522A_ABST
Patent Text Reader

Abstract

This invention relates to the field of unmanned vehicle scheduling, and discloses a flexible scheduling method for charging unmanned vehicles based on nonlinear spatiotemporal impedance and dynamics MOPSO. Urban roads are abstracted as directed graphs carrying nonlinear spatiotemporal impedance, and multiple variable transfer points are initiated through a combination of continuous carriers and discrete manifold mapping. A multidimensional heterogeneous evaluation functional is constructed, incorporating queuing theory arrival time, nonlinear dynamic power consumption, and smooth transfer complexity. The optimization end employs a continuous dynamics multi-objective particle swarm optimization (MOPSO) algorithm with time-varying learning factors, constrained by Lyapunov absolute convergence boundaries, and dynamically allocated global guides using Mahalanobis distance archiving truncation and Softmax annealing mechanisms, outputting a Pareto optimal solution set. A Gaussian fuzzy membership function is introduced, and a comprehensive utility score is calculated using dynamic business weights delivered in real-time from the cloud. This method is suitable for high-concurrency, large-scale scheduling scenarios for urban mobile charging unmanned vehicles.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of unmanned vehicle scheduling, specifically to a flexible scheduling method for charging unmanned vehicles based on nonlinear spatiotemporal impedance and dynamics MOPSO. Background Technology

[0002] In the era of intelligence, the application of autonomous vehicles in industry and transportation is becoming increasingly widespread. With the rapid growth of electric vehicle ownership, the autonomous vehicle charging scheduling industry is also gradually expanding, further addressing various types of electric vehicle charging needs through the proactive assignment of mobile charging autonomous vehicles. In the flexible assignment of mobile charging tasks, the technical challenge and logic lie in achieving vehicle matching and path planning with minimal spatiotemporal impedance while ensuring the absolute safety of battery physical reserves. Simultaneously, the real road network in the physical world requires that the continuous driving trajectory of the vehicle converge to a legal discrete directed graph manifold through conformal mapping, constituting a continuous-discrete hybrid NP-hard scheduling problem with extremely strong physical constraints.

[0003] Most existing technologies rely on static transfer point strategies with fixed topologies, failing to adaptively avoid dynamic congestion and risky restricted areas in complex road networks, and unable to balance the multi-dimensional game of time, energy consumption, and scheduling complexity. The Particle Swarm Optimization (PSO) algorithm used in existing mobile charging systems, lacking mathematical convergence boundaries when processing high-dimensional discrete graph mappings, can lead to speed explosions or topology deadlocks. Speed ​​explosions can cause scheduling servers to crash under massive concurrent requests due to computational divergence, memory overflows, or response timeouts. On the other hand, topology deadlocks can output illegal physical trajectories, causing unmanned charging vehicles to be incorrectly assigned to dead ends, one-way traffic zones, or heavily congested sections, unable to plan their own escape routes. This not only depletes the vehicle's battery through ineffective starts, stops, and detours but also further increases the risk of safety accidents and order delays. Summary of the Invention

[0004] Purpose of the Invention: To address the problems mentioned in the background art, this invention discloses a flexible scheduling method for unmanned charging vehicles based on nonlinear spatiotemporal impedance and dynamics MOPSO. By adopting an improved dynamics multi-objective particle swarm optimization (MOPSO) framework based on a combination of continuous carriers and discrete manifold mapping at the underlying architecture, multiple variable transfer path nodes are enabled. Combined with a second-order continuous dynamic update mechanism that couples a strong contraction factor with strict constraints on Lyapunov absolute convergence boundaries with a time-varying learning factor, the high-frequency oscillation and velocity divergence risk of the algorithm in the complex graph penalty mapping process is reduced, achieving an efficient, robust, and millisecond-level flexible assignment and scheduling scheme for unmanned charging vehicles.

[0005] This invention discloses a flexible scheduling method for charging unmanned vehicles based on nonlinear spatiotemporal impedance and dynamics MOPSO. The method steps are as follows:

[0006] S1 constructs a spatiotemporal graph network that incorporates slope and elevation, simulates the dynamic vehicle speed under microscopic traffic flow through a stochastic process, initializes the continuous coordinates of the unmanned charging vehicle and maps them to the legal nodes of the spatiotemporal graph network, splices them into a complete path, and obtains the initial particle positions.

[0007] S2 constructs an objective function based on path definition and uses an improved MOPSO algorithm to iteratively evolve the particle swarm: it uses the second-order differential equation of continuous dynamics to perform multi-objective controlled evolution optimization of particle position, introduces a strong contraction factor that satisfies the Lyapunov absolute convergence condition for velocity update, and combines a simulated annealing mutation mechanism based on Boltzmann acceptance probability to determine particle update and escape local topological deadlock.

[0008] S3 stores non-dominated solutions in an external archive and removes overloaded redundant solutions. It dynamically assigns a globally optimal leader to the optimization population and outputs a Pareto front solution set after iteration. The optimal scheduling scheme is selected based on a combination of information entropy and preference dynamic weighting with fuzzy membership scores.

[0009] Furthermore, S2 describes constructing an objective function based on path definition, that is, constructing a heterogeneous nonlinear objective functional tensor for the generated complete path. The objectives include arrival time objective considering queuing theory impedance, energy consumption objective considering nonlinear battery discharge effect, and transfer complexity objective based on continuous smooth approximation.

[0010] Furthermore, the heterogeneous nonlinear objective functional tensor includes an arrival time model. Power consumption model and transit complexity model The arrival time model Specifically:

[0011] Along the generated continuous road node sequence Calculate the sum of the travel time and intersection waiting time for each adjacent road segment. The travel time is the physical length of that road segment. Divide by the dynamic permissible speed of that segment at the current actual moment. The waiting time at an intersection is the expected queuing delay at that node. Add up the two times for all road segments to get the total arrival time of the proposed route.

[0012] The power consumption model Specifically, it is believed that:

[0013] Calculate the baseline energy consumption: For each road segment, calculate the energy consumption per unit distance. Multiply by the road condition attenuation factor for that section Multiply by the road length The summation yields the total base power consumption; a step penalty term is added. If the above-mentioned total basic power consumption exceeds the vehicle's current battery capacity... ,but It can take the value of infinity, otherwise The final power consumption value is the sum of the base power consumption and the penalty.

[0014] The transit complexity model Specifically:

[0015] Preset the maximum allowed number of transit points For each potential transit point location, define a continuous variable. The variable is processed by the Sigmoid function. Mapped to The interval is mapped to a random number that follows a uniform distribution. Comparison: If it is greater than the random number, then the indicator function Count it as 1 if it is not 0 otherwise, for all The above judgments are performed on each potential transit point, and the results are summed to obtain the transit complexity.

[0016] Furthermore, the discretized position and velocity updates of the second-order differential equation of continuous dynamics described in S2 are as follows:

[0017]

[0018]

[0019] In the formula, and Let be the flight velocity vector and spatial position vector of the continuous carrier at the transfer node at the t-th iteration, respectively. It is a random number. For the individual's historical best position, This is the globally optimal guide position.

[0020] Furthermore, the strong contractile factor described in S2 that satisfies the Lyapunov absolute convergence condition... The analytical constraints are: ,in

[0021] In the formula, and These are learning factors for time-varying cognitive and social terms.

[0022] Furthermore, the specific method by which S2 determines particle updates based on the simulated annealing mutation mechanism using Boltzmann acceptance probability is as follows:

[0023] When the individual's historical best position is detected When the number of consecutive pauses exceeds a set threshold, a random perturbation with Levy flight is injected into the current position to generate a new position vector, and the new position vector is determined based on the Boltzmann acceptance probability. Decide whether to accept the new position vector;

[0024] Based on Boltzmann probability The determination logic is as follows: calculate the difference in target degradation energy levels between the old and new locations. ,like Unconditionally accept the new position vector; if Then generate A uniformly random number, only if the random number Accept the inferior solution if it is acceptable, otherwise reject it.

[0025] Furthermore, as described in S3, based on Mahalanobis generalized congestion degree to eliminate overloaded redundant solutions, the Mahalanobis generalized congestion degree of the corresponding non-dominated solution is calculated using the objective covariance tensor:

[0026] When the external archive size exceeds the maximum limit When calculating the covariance tensor of the target space... , obtain the Crowding distance of each non-dominated solution :

[0027]

[0028] In the formula, represents the target dimension index. , In the first After being sorted by dimension, the spatial position is adjacent to the first. The evaluation values ​​of each solution before and after, The maximum and minimum boundary values ​​for this dimension are determined based on the crowding distance. Remove the crowded solution with the smallest value.

[0029] Furthermore, the scoring logic described in S3, which combines information entropy and dynamic preference weighting with fuzzy membership scoring, is as follows:

[0030] Using Gaussian membership function mapping The first alternative is in the... Individual satisfaction levels on each evaluation indicator, combined with dynamic coupling weights Calculate the overall utility score;

[0031] Dynamic coupling weights The calculation logic is as follows:

[0032] Extract the evaluation matrices of all alternative solutions in the Pareto front solution set on the heterogeneous nonlinear objective functional tensor, normalize the matrices to obtain the probability matrices, and calculate the first... Objective information entropy of each evaluation indicator Calculate the coefficient of variation for each indicator. Then, normalization is performed to obtain the objective information entropy weight. ;

[0033] Define the spatiotemporal window urgency factor for the currently assigned charging task. According to the urgency factor Dynamically adjust arrival time indicators Power consumption model indicators With transit complexity index Initial business weight benchmark This yields a business preference weight vector that changes over time. ; Utilizing the objective information entropy weight For the business preference weight Adaptive correction is performed, and the final dynamic coupling weights are output.

[0034] Beneficial effects:

[0035] 1. This invention abstracts urban roads as directed graphs carrying nonlinear spatiotemporal impedance, and adaptively activates 0 to 3 variable transfer points by combining continuous carriers with discrete manifold mapping. This mechanism improves adaptability to heterogeneous urban topological spaces, not only dynamically avoiding obstacles, but also ensuring the smoothness and safety of driving trajectories at the physical execution level by controlling transfer complexity.

[0036] 2. This invention introduces a strong contraction factor constrained by the Lyapunov convergence boundary and couples it with a time-varying learning factor to form a second-order continuous dynamic update mechanism. This mechanism can effectively absorb the excessive kinetic energy of the algorithm when crossing large road network node spans, further reducing the risk of high-frequency oscillations in the complex graph penalty mapping process, and enabling the mobile charging task to maintain smooth convergence when facing high-concurrency requests.

[0037] 3. This invention introduces a step penalty function for exceeding battery limits. Once the estimated total power consumption exceeds the current battery capacity, high-energy-consuming routes are avoided at the physical level, reducing the risk of the autonomous vehicle breaking down mid-journey. Furthermore, by combining simulated annealing and Mahalanobis distance archiving truncation, the algorithm is endowed with the physical escape capability to penetrate local minimum potential traps, preventing the optimization population from getting stuck in topological deadlock or clustering that causes new congestion.

[0038] 4. In the final scheduling scheme output stage, this invention introduces a Gaussian fuzzy membership function, combining objective information entropy with dynamic business weights issued in real time from the cloud to calculate the comprehensive utility score. This effectively decouples physical road network optimization from top-level business decision-making. Only new weights need to be issued; without recalculating the massive road network, the system can select vehicles and routes with higher comprehensive scores from the Pareto candidate library to execute scheduling tasks, giving unmanned charging vehicles superior flexible response capabilities in high-concurrency, large-scale scenarios. Attached Figure Description

[0039] Figure 1 This is a flowchart of the method of the present invention.

[0040] Figure 2 This is a system flowchart of an embodiment of the present invention;

[0041] Figure 3 This is a comparison diagram of the dynamic stability and anti-divergence of the iterative process in an embodiment of the present invention;

[0042] Figure 4 This is a comparison of scenario switching response latency and the evolution of high-concurrency burst order throughput in an embodiment of the present invention;

[0043] Figure 5 This is a comparison diagram of the distribution of the optimal solution set in an embodiment of the present invention. Detailed Implementation

[0044] 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.

[0045] The method in this embodiment relies on a scheduling system that goes from the bottom-level road network perception to the top-level intelligent dispatching. The system includes, in sequence: a real dynamic road network modeling module, a variable transfer point adaptive mapping module, a multi-dimensional scheduling target evaluation module, a stable multi-objective optimization iteration module, and an intelligent flexible dispatching decision module. In the specific operation of the system, the mapping and correspondence logic between each module and the specific execution steps of this method is as follows: First, the real dynamic road network modeling module executes step S1 to construct a spatiotemporal variational graph network and characterize the micro-dynamic allowable vehicle speed; then, the variable transfer point adaptive mapping module executes step S2 to realize the Riemann manifold projection and path stitching of the transfer pose coordinate carrier; subsequently, the multi-dimensional scheduling target evaluation module executes step S3 to construct a heterogeneous nonlinear objective functional tensor containing time, energy consumption, and transfer complexity; in the core optimization stage, the stable multi-objective optimization iteration module executes steps S4 to S6 in series to complete the dynamic MOPSO controlled evolution optimization, external archive Mahalanobis distance truncation, and global guided dynamic allocation based on Softmax, respectively; finally, the intelligent flexible dispatch decision module executes step S7 to lock the final unmanned charging vehicle dispatch route in the Pareto front solution set by combining objective information entropy and fuzzy membership score.

[0046] like Figure 1-2 As shown, a flexible scheduling method for charging unmanned vehicles based on nonlinear spatiotemporal impedance and dynamics MOPSO is disclosed. The method steps are as follows:

[0047] S1. Construct a spatiotemporal variational graph network, introduce road segment slope and elevation tensors, and combine stochastic processes to characterize the dynamic vehicle speed under microscopic traffic flow.

[0048] First, construct a spatiotemporal variational directed graph manifold. To accurately depict terrain undulations, the system extracts high-precision data. Figure 3 3D coordinates generate continuous elevation field and along the arc length of the road Taking the partial derivative, we obtain the instantaneous slope angle:

[0049]

[0050] In the formula, It is a two-dimensional elevation gradient vector. This is the unit tangent vector of the road centerline.

[0051] Instantaneous allowable vehicle speed to capture micro-disturbances in traffic flow Modeled as an Ornstein-Uhlenbeck stochastic differential equation:

[0052]

[0053] In the formula, The mean regression rate, For the macroscopic expected vehicle speed, Traffic noise intensity, This is the standard Wiener process.

[0054] S2. Initialize the continuous coordinate carrier representing the intermediate pose of the unmanned charging vehicle, and converge it to the legal physical nodes of the spatiotemporal variational graph network through the Riemann manifold projection operator based on geodesic distance, splice them to generate a complete path, and obtain the initial particle position.

[0055] The multi-objective evolution algorithm generates coordinate carriers representing intermediate poses for particles in continuous space. To ensure its legitimacy, the system employs a Riemannian manifold projection operator based on geodesic distance. Force it to converge to a legitimate physical node. :

[0056]

[0057] In the formula, For the Riemannian metric tensor, The terrain crossing penalty coefficient is used, and the integral term represents the cumulative terrain elevation gradient along the route. When the algorithm calculates a virtual optimal coordinate in continuous space, it is forced to "attach" the actual road geodesic to the real reachable intersection, ensuring that the trajectory generated by the algorithm is absolutely drivable in the physical world.

[0058] S3. Construct a heterogeneous nonlinear objective functional tensor for the generated complete path. The objectives include at least: arrival time objective considering queuing theory impedance, energy consumption objective considering nonlinear battery discharge effect, and transfer complexity objective based on continuous smooth approximation.

[0059] Heterogeneous nonlinear objective functional tensor includes arrival time model Power consumption model and transit complexity model The specific formula is as follows:

[0060] (1) True arrival time model with nodal impedance for:

[0061]

[0062] Along the generated continuous road node sequence Calculate the sum of the travel time and intersection waiting time for each adjacent road segment. The travel time is the physical length of that road segment. Divide by the dynamic permissible speed of that segment at the current actual moment. The waiting time at an intersection is the expected queuing delay at that node. Add up the two times for all road segments to get the total arrival time of the proposed route.

[0063] (2) Heterogeneous road condition coupled power consumption model for:

[0064]

[0065] Calculate the baseline energy consumption: For each road segment, calculate the energy consumption per unit distance. Multiply by the road condition attenuation factor for that section Multiply by the road length The summation yields the total base power consumption; a step penalty term is added. If the above-mentioned total basic power consumption exceeds the vehicle's current battery capacity... ,but It can take the value of infinity, otherwise The final power consumption value is the sum of the base power consumption and the penalty.

[0066] The unmanned charging vehicle itself is a heavy energy carrier. The amplification effect of uphill resistance on power consumption was dynamically considered. More importantly, the penalty for exceeding limits was taken into account. In our operations, a dispatch vehicle breaking down en route is an extremely high-risk incident. Once the total power consumption is estimated... Greater than the battery's physical margin The penalty is infinitely large, directly vetoing this "dangerous route" from the perspective of physical laws.

[0067] (3) Transfer complexity model based on continuous smooth approximation for:

[0068]

[0069] Preset the maximum allowed number of transit points For each potential transit point location, define a continuous variable. The variable is processed by the Sigmoid function. Mapped to The interval is mapped to a random number that follows a uniform distribution. Comparison: If it is greater than the random number, then the indicator function Count it as 1 if it is not 0 otherwise, for all The above judgments are performed on each potential transit point, and the results are summed to obtain the transit complexity.

[0070] The more turning points (or major bends in a route) there are, the more frequently vehicles need to decelerate, steer, and accelerate again. This not only increases mechanical wear but also significantly raises the risk of accidents at intersections. By controlling the complexity of transfers, the smoothness and safety of the driving trajectory are ensured at the physical execution level.

[0071] In the formula, For the first The trolley is in The scheduling scheme for the next iteration. This is the generated sequence of continuous road nodes. This represents the total number of nodes included. For adjacent road nodes, The physical length of the road segment between nodes. For dynamically permissible vehicle speed, Impedance for expected waiting time at intersection nodes; Based on energy consumption per unit distance, This is the road condition attenuation coefficient. This is a step penalty function for exceeding the battery limit, when the total power consumption exceeds the current battery capacity. hour ,on the contrary ; This represents the maximum number of transit points. For continuous switching carrier variables, For indicator functions, These are random numbers that follow a uniform distribution.

[0072] S4. The improved MOPSO algorithm is used to iteratively evolve the particle swarm: the particle position is optimized by multi-objective controlled evolution using the second-order differential equation of continuous dynamics. The velocity update introduces a strong contraction factor that satisfies the Lyapunov absolute convergence condition, and combines the simulated annealing mutation mechanism based on Boltzmann acceptance probability to determine the particle update and escape local topological deadlock.

[0073] The discretized position and velocity update model of the second-order differential equation of continuous dynamics is as follows:

[0074]

[0075]

[0076] In the formula, and Let be the flight velocity vector and spatial position vector of the continuous carrier at the transfer node at the t-th iteration, respectively. It is a random number. For the individual's historical best position, The optimal location for the global guide;

[0077] Strong contractile factor satisfying Lyapunov's absolute convergence condition The analytical constraints are: ,in

[0078] In the formula, and These are learning factors for time-varying cognitive and social terms. Shrinkage factor. This is equivalent to a basic "intelligent ABS anti-lock braking system." When the algorithm attempts to traverse vast road network nodes, It will forcibly absorb excessive kinetic energy to ensure smooth convergence of the optimization process and prevent the server from crashing due to speed divergence when scheduling massive road networks concurrently.

[0079] The simulated annealing mutation mechanism is as follows: when the individual's historical best position is detected... When the number of consecutive pauses exceeds a set threshold, a random perturbation with Lévy flight is injected into the current position to generate a new position vector, and this vector is based on the Boltzmann acceptance probability. Decide whether to accept the new position vector:

[0080]

[0081] In the formula, The target is the energy level difference that degrades. It is a constant. This is the annealing temperature that decreases exponentially with the number of iterations.

[0082] Based on Boltzmann acceptance probability The specific decision-making logic for whether to accept the new position vector is as follows: First, calculate the target degradation energy level difference between the new position vector generated after the injection of random perturbation and the original individual's historical best position. .like If the overall performance of the new position is better than or equal to that of the current position, then the calculation result of the formula is limited by the extremum function. The system unconditionally accepts the new position vector; if If the new position is a disadvantageous solution, then the acceptance probability between 0 and 1 is calculated according to the Boltzmann probability formula, and a random number following a uniform distribution is generated within the same interval. The acceptance probability is determined if and only if this random number is less than or equal to... If the system accepts the new, disadvantageous position vector briefly, it will reject it otherwise.

[0083] When a vehicle gets stuck in a one-way street trap or a dead end (local optimum) while planning its route, the system allows it to... The probability of briefly accepting a disadvantageous route that "seems to be a detour" gives the algorithm the opportunity to re-enter the main road (globally optimal), thus endowing it with the physical escape ability to penetrate the topological barriers of complex road networks.

[0084] S5. Store the non-dominated solution set generated by the evolution into an external archive, use the target covariance tensor to calculate the Mahalanobis generalized crowding degree of the non-dominated solution, and truncate and remove the overloaded archive accordingly.

[0085] The generalized congestion level of Mahalanobis is calculated as follows: when the external archive capacity exceeds the maximum limit... When calculating the covariance tensor of the target space... And obtain the first by the following formula Crowding distance of each non-dominated solution :

[0086]

[0087] In the formula, represents the target dimension index. , In the first After being sorted by dimension, the spatial position is adjacent to the first. The evaluation values ​​of each solution before and after, These are the maximum and minimum boundary values ​​for this dimension. The system is based on... Remove the crowded solution with the smallest value.

[0088] S6. To avoid the loss of superior genes, based on the Mahalanobis generalized crowding degree and the Shannon information entropy maximization criterion, the Softmax distribution probability is used to dynamically assign the global optimal leader to the optimization population.

[0089] The specific calculation method for the Softmax distribution probability is as follows: The following formula is used to calculate the probability of the first... One non-dominated solution is selected as the global optimal guide. probability :

[0090]

[0091] In the formula, For the first The probability that a non-dominated solution is selected as the global wizard. This represents the actual size of the current external archive's solution set. This represents the annealing temperature coefficient.

[0092] S7. When the maximum number of iterations is reached, output the Pareto front solution set, generate dynamic coupling weights by combining objective information entropy and business preferences, calculate the comprehensive utility score of each scheme through Gaussian fuzzy membership function, and finally select the unmanned charging vehicle and route corresponding to the highest score to perform the scheduling task.

[0093] The comprehensive utility score of each scheme is calculated, and the specific logic includes: using a Gaussian membership function to map the first... The first alternative is in the... Individual satisfaction with each evaluation indicator:

[0094]

[0095] In the formula, The platform's desired ideal target extreme value, To tolerate variance; combined with dynamically coupled weights Calculate the overall utility score:

[0096]

[0097] In the formula, To meet the normalization requirements for dynamic service weights, the weights are assigned by the scheduling system based on the real-time urgency of the services. The specific logic includes the following steps:

[0098] Extract the evaluation matrices of all alternative solutions in the Pareto front solution set on three evaluation metrics, normalize the matrices to obtain the probability matrix, and calculate the probability matrix according to the following formula. Objective information entropy of each evaluation indicator :

[0099]

[0100] In the formula, This represents the actual size of the current external archive's solution set. For the first The first scheme is in the Normalized satisfaction probability for each indicator; calculate the difference coefficient for each indicator according to the following formula. Then, normalization is performed to obtain the objective information entropy weight. :

[0101]

[0102] Construct a business preference weighting function based on spatiotemporal window constraints: Define the spatiotemporal window urgency factor of the current charging assignment task. The calculation formula is as follows:

[0103]

[0104] In the formula, The latest expected deadline for the current order. The physical real-time time at which the current cloud-based dispatch system initiates the order dispatch. The emergency time threshold constant set for the system; based on the urgency factor. Dynamically adjust arrival time indicators Power consumption model indicators With transit complexity index Initial business weight benchmark This yields a business preference weight vector that changes over time. .

[0105] Perform matrix coupling and normalized composition: utilize objective information entropy weights Weight of business preferences Adaptive correction is performed, and the final dynamic coupling weight is output using the following formula.

[0106] To further illustrate the underlying operational logic of this embodiment, we now extract the scheduling system's operation at iteration t=50 for a specific unmanned charging vehicle ( One of the alternative routes (paths) Local numerical execution slices for multidimensional heterogeneous functional evaluation and dynamic position update.

[0107] Basic physical parameter settings

[0108] (1) Driverless vehicle status: small car Current battery remaining Baseline power consumption rate Road condition attenuation coefficient .

[0109] (2) Current evaluation path It consists of three physical nodes: starting point A → intersection B → ending point C.

[0110] Section 1 (AB):

[0111] length Uphill, instantaneous slope angle Road conditions and vehicle speed

[0112] Intersection (B): Queuing impedance obtained from a spatiotemporal variational graph network (i.e., 0.033h).

[0113] Section 2 (BC): Length Downhill, instantaneous slope angle Road conditions and vehicle speed

[0114] Numerical derivation of multidimensional heterogeneous objective functionals

[0115] The system performs a three-dimensional objective functional calculation on the splicing path:

[0116] (1) Arrival Time Model Evaluate

[0117] The system calculates the sum of pure travel time and queuing impedance:

[0118]

[0119] The estimated arrival time is approximately 11 minutes.

[0120] Power consumption model and step hard penalty Evaluate:

[0121] The system calculates the nonlinear power consumption of each road segment due to the terrain elevation tensor:

[0122] Power consumption of uphill section AB:

[0123] Power consumption (kinetic energy recovery) on downhill section BC:

[0124] Estimated total power consumption:

[0125] Triggering hard penalty judgment: due to Strictly less than the battery's physical margin Step penalty function This route is deemed safe in terms of physical energy consumption, with a target value of [missing information]. .

[0126] Lyapunov's strong contractile factor-based intervention

[0127] After evaluation, the particle needs to fly towards the optimal position. To prevent the continuous dynamics MOPSO from experiencing velocity explosion (i.e., flying out of the real road network) when updating the position, the system invokes a strong contraction factor constrained by the Lyapunov absolute convergence condition. .

[0128] Assume the cognitive learning factor in the current iteration Social learning factor ,but The system calculates the underlying anti-lock braking coefficient. : .

[0129] Calculation results show that in this position update iteration, the algorithm forcibly reduces the particle synthesis burst velocity to 64.1% of its original value. Through dynamic damping calculations, the algorithm ensures smooth convergence of the autonomous vehicle scheduling algorithm when facing extremely congested high-dimensional road networks, preventing server memory overflow or system crashes.

[0130] Combination Figure 3The comparison diagram of iterative process dynamic stability and anti-divergence illustrates the underlying anti-divergence effect after introducing the Lyapunov strong contraction factor. In the traditional single-objective PSO algorithm, due to the lack of continuous dynamic constraints, the root mean square velocity (RMS) of the particle swarm is prone to exceeding the limit and causing severe oscillatory divergence in the later stages of optimization iteration (around the convergence critical point of the 45th iteration). Figure 3 As shown by the orange curve in the middle), this caused the system's computing power to collapse. The continuous dynamics MOPSO algorithm in this embodiment (…) Figure 3 The blue-marked curve effectively absorbs excess kinetic energy through a strong contraction factor, ensuring that the particle swarm velocity smoothly and strictly converges to a stable convergence region throughout the entire iteration cycle. Figure 3 The green shaded area in the middle ensures the absolute stability of the underlying optimization process from a physical mechanism perspective.

[0131] Combination Figure 4 The comparison of scenario switching response latency and the evolution of throughput under high concurrency burst orders shown in the diagram verify the efficient response capability of the top-level intelligent flexible order dispatching decision module in this embodiment. Figure 4 As shown in the left-hand bar chart, in scenarios involving instantaneous weight switching such as "emergency rescue," traditional algorithms suffer from system dispatch response delays of tens of thousands of milliseconds due to the need to recalculate massive amounts of road network data. However, this embodiment, benefiting from the effective decoupling of underlying road network optimization and top-level business decision-making, only requires reconstructing dynamically coupled weights using objective information entropy and business preferences in the cloud, combined with Gaussian fuzzy membership degree scoring, to achieve millisecond-level (e.g., 2.4ms) rapid redistribution, resulting in an order-of-magnitude improvement in response speed. Furthermore, as... Figure 4 As shown in the graph on the right, in a simulated high-concurrency burst order scenario, traditional algorithms are prone to order dispatching stagnation due to computing power bottlenecks and deadlocks, while this embodiment can smoothly process nearly a hundred orders within 60 seconds, demonstrating extremely strong high-concurrency throughput.

[0132] Combination Figure 5 The Pareto Front diagram, showing the distribution of the optimal solution set, visually demonstrates the spatial search and escape capabilities of this embodiment under a multi-dimensional scheduling objective game. The horizontal axis in the diagram represents the estimated arrival time of the route. The vertical axis represents the expected power consumption. The light-colored scatter points represent the feasible solution space of random traversal. It can be seen that the traditional single-objective PSO algorithm (orange scatter points) suffers from a lack of topological penetration, causing a large number of the optimization population to get trapped in local minima. In contrast, this embodiment successfully escapes local deadlock by relying on a simulated annealing mutation mechanism based on Boltzmann probability and Mahalanobis distance truncation. Its output Pareto front solution set (blue lines and scatter points) is not only widely and evenly distributed, but also achieves deep approximation towards the lower left corner (i.e., the direction of optimal time and energy consumption). The scheduling system can use a flexible sliding window to weigh the "extreme time-sensitive point" and the "extreme energy-saving point" based on the real-time urgency factor, thereby accurately assigning the optimal scheduling route.

[0133] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent transformations or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A flexible scheduling method for unmanned charging vehicles based on nonlinear spatiotemporal impedance and dynamics MOPSO, characterized in that, The steps are as follows: S1 constructs a spatiotemporal graph network that incorporates slope and elevation, simulates the dynamic vehicle speed under microscopic traffic flow through a stochastic process, initializes the continuous coordinates of the unmanned charging vehicle and maps them to the legal nodes of the spatiotemporal graph network, splices them into a complete path, and obtains the initial particle positions. S2 constructs an objective function based on path definition and uses an improved MOPSO algorithm to iteratively evolve the particle swarm: it uses the second-order differential equation of continuous dynamics to perform multi-objective controlled evolution optimization of particle position, introduces a strong contraction factor that satisfies the Lyapunov absolute convergence condition for velocity update, and combines a simulated annealing mutation mechanism based on Boltzmann acceptance probability to determine particle update and escape local topological deadlock. S3 stores non-dominated solutions in an external archive and removes overloaded redundant solutions. It dynamically assigns a globally optimal leader to the optimization population and outputs a Pareto front solution set after iteration. The optimal scheduling scheme is selected based on a combination of information entropy and preference dynamic weighting with fuzzy membership scores.

2. The flexible scheduling method for charging unmanned vehicles based on nonlinear spatiotemporal impedance and dynamics MOPSO as described in claim 1, characterized in that, The objective function described in S2 is based on the path definition. That is, a heterogeneous nonlinear objective functional tensor is constructed for the generated complete path. The objectives include the arrival time objective considering queuing theory impedance, the energy consumption objective considering the nonlinear discharge effect of the battery, and the transfer complexity objective based on continuous smooth approximation.

3. The flexible scheduling method for charging unmanned vehicles based on nonlinear spatiotemporal impedance and dynamics MOPSO as described in claim 2, characterized in that, The heterogeneous nonlinear objective functional tensor includes an arrival time model. Power consumption model and transit complexity model The arrival time model Specifically: Along the generated continuous road node sequence Calculate the sum of the travel time and intersection waiting time for each adjacent road segment. The travel time is the physical length of that road segment. Divide by the dynamic permissible speed of that segment at the current actual moment. The waiting time at an intersection is the expected queuing delay at that node. Add up the two times for all road segments to get the total arrival time of the proposed route. The power consumption model Specifically, it is believed that: Calculate the baseline energy consumption: For each road segment, calculate the energy consumption per unit distance. Multiply by the road condition attenuation factor for that section Multiply by the road length The summation yields the total base power consumption; a step penalty term is added. If the above-mentioned total basic power consumption exceeds the vehicle's current battery capacity... ,but It can take the value of infinity, otherwise The final power consumption value is the sum of the base power consumption and the penalty. The transit complexity model Specifically: Preset the maximum allowed number of transit points For each potential transit point location, define a continuous variable. The variable is processed by the Sigmoid function. Mapped to The interval is mapped to a random number that follows a uniform distribution. Comparison: If it is greater than the random number, then the indicator function Count it as 1 if it is not 0 otherwise, for all The above judgments are performed on each potential transit point, and the results are summed to obtain the transit complexity.

4. The flexible scheduling method for charging unmanned vehicles based on nonlinear spatiotemporal impedance and dynamics MOPSO as described in claim 1, characterized in that, The discretized position and velocity updates of the second-order differential equation of the continuous dynamics described in S2 are as follows: ; ; In the formula, and Let be the flight velocity vector and spatial position vector of the continuous carrier at the transfer node at the t-th iteration, respectively. It is a random number. For the individual's historical best position, This is the globally optimal guide position.

5. The flexible scheduling method for charging unmanned vehicles based on nonlinear spatiotemporal impedance and dynamic MOPSO as described in claim 1, characterized in that, The strong contractile factor that satisfies the Lyapunov absolute convergence condition, as described in S2. The analytical constraints are: ,in ; In the formula, and These are learning factors for time-varying cognitive and social terms.

6. The flexible scheduling method for charging unmanned vehicles based on nonlinear spatiotemporal impedance and dynamics MOPSO as described in claim 1, characterized in that, The specific method by which S2 determines particle updates based on the simulated annealing mutation mechanism of Boltzmann acceptance probability is as follows: When the individual's historical best position is detected When the number of consecutive pauses exceeds a set threshold, a random perturbation with Levy flight is injected into the current position to generate a new position vector, and the new position vector is determined based on the Boltzmann acceptance probability. Decide whether to accept the new position vector: Based on Boltzmann probability The determination logic is as follows: calculate the difference in target degradation energy levels between the old and new locations. ,like Unconditionally accept the new position vector; if Then generate A uniformly random number, only if the random number... Accept the inferior solution if it is acceptable, otherwise reject it.

7. The flexible scheduling method for charging unmanned vehicles based on nonlinear spatiotemporal impedance and dynamic MOPSO as described in claim 1, characterized in that, S3 describes the process of eliminating overloaded redundant solutions based on Mahalanobis generalized congestion, and then using the objective covariance tensor to calculate the Mahalanobis generalized congestion degree of the corresponding non-dominated solution: When the external archive size exceeds the maximum limit When calculating the covariance tensor of the target space... , obtain the Crowding distance of each non-dominated solution : ; In the formula, represents the target dimension index. , In the first After being sorted by dimension, the spatial position is adjacent to the first. The evaluation values ​​of each solution before and after, The maximum and minimum boundary values ​​for this dimension are determined based on the crowding distance. Eliminate the crowded solution with the smallest value.

8. The flexible scheduling method for unmanned charging vehicles based on nonlinear spatiotemporal impedance and dynamics MOPSO as described in claim 1, characterized in that, The scoring logic described in S3, which combines information entropy and dynamic weighted preference with fuzzy membership scoring, is as follows: Using Gaussian membership function mapping The first alternative is in the... Individual satisfaction levels on each evaluation indicator, combined with dynamic coupling weights Calculate the overall utility score; Dynamic coupling weights The calculation logic is as follows: Extract the evaluation matrices of all alternative solutions in the Pareto front solution set on the heterogeneous nonlinear objective functional tensor, normalize the matrices to obtain the probability matrices, and calculate the first... Objective information entropy of each evaluation indicator Calculate the coefficient of variation for each indicator. Then, normalization is performed to obtain the objective information entropy weight. ; Define the spatiotemporal window urgency factor for the currently assigned charging task. According to the urgency factor Dynamically adjust arrival time indicators Power consumption model indicators With transit complexity index Initial business weight benchmark This yields a business preference weight vector that changes over time. ; Using the objective information entropy weight For the business preference weight Adaptive correction is performed, and the final dynamic coupling weights are output.