Unmanned vehicle cloud edge collaborative dynamic scheduling optimization method and system

Abstract

The invention discloses an unmanned vehicle cloud edge collaborative dynamic scheduling optimization method and system, and relates to the technical field of unmanned vehicle scheduling, and the method comprises the steps: receiving unmanned vehicle physical information, historical task information, an unmanned vehicle real-time state and an edge node real-time state; integrating the physical information of the unmanned vehicle, the real-time state of the unmanned vehicle and the real-time state of the edge node to generate a dynamic parameter library; the historical tasks are classified based on the historical task information and a dynamic parameter library, task classification results are obtained, and the task classification results comprise light-weight tasks, medium-weight tasks and weight tasks; inputting the dynamic parameter library and the task classification result into a pre-established fission mixed variation crayfish optimization algorithm model, and outputting to obtain a global optimal scheduling scheme by taking the minimization of the weighted sum of the total time delay and the total energy consumption of the system as a target; wherein the global optimal scheduling scheme comprises an unmanned vehicle track adjustment instruction, a task unloading path allocation result and a cloud side resource allocation proportion.

Description

Technical Field

[0001] This invention relates to the field of unmanned vehicle scheduling technology, specifically a dynamic scheduling optimization method and system for unmanned vehicle cloud-edge collaboration. Background Technology

[0002] With the continuous development of vehicle-mounted perception systems and cloud-edge computing technologies, modern autonomous vehicle (RV) swarm scheduling systems typically integrate roadside sensors, vehicle-mounted detectors, edge computing nodes, and cloud data centers to collect real-time status data such as RV location, battery level, and communication quality, as well as business data such as task data volume and latency requirements. Among these, cloud-edge collaborative scheduling has become the mainstream technology for RV swarm management due to its ability to integrate the low-latency computing power of edge nodes with the global optimization capabilities of the cloud. However, due to the dynamic nature of urban ground scenarios and the high mobility of RVs, scheduling systems face multi-dimensional technical challenges: RV movement causes channel gain attenuation and transmission rate fluctuations, and the lack of a unified quantitative description standard for task types means that the processing requirements of lightweight status acquisition tasks and heavy high-definition recognition tasks are not effectively distinguished, making it difficult for traditional scheduling methods to adapt to the processing requirements of differentiated tasks; simultaneously, the coupling relationship between RV driving energy consumption, transmission energy consumption, and computing energy consumption is complex, and a single energy consumption management strategy cannot achieve refined optimization of energy consumption across the entire link, further restricting the endurance of RVs.

[0003] To address the resource constraints in autonomous vehicle (RV) scheduling, traditional methods often employ a two-tier "terminal-cloud" architecture or a three-tier "local-edge-cloud" architecture for task offloading and resource allocation. These approaches attempt to achieve global optimization through the massive computing power of the cloud or reduce task processing latency by relying on edge nodes. However, it's important to note that RVs, as ground-based mobile units, face the dual challenges of limited communication bandwidth and dynamic scenario adaptation. On one hand, urban building obstructions and multipath effects reduce wireless communication bandwidth, making it impossible to support real-time uploading of all task data. Furthermore, the communication link must reserve bandwidth for RV safety control command transmission, further compressing the task data transmission space. On the other hand, traditional architectures lack a "user-vehicle" collaboration mechanism, resulting in a disconnect between task generation and execution. Traditional optimization algorithms are prone to getting trapped in local optima in high-dimensional dynamic environments, failing to achieve joint optimization of RV trajectory planning, task offloading, and cloud-edge resource allocation. Moreover, when the communication connection between the RV and the cloud / edge is interrupted, the scheduling system, entirely reliant on cloud-edge computing power, will malfunction, making it difficult to guarantee the safe operation of the RV. Therefore, dynamic scheduling optimization for RV cloud-edge collaboration becomes a crucial issue. Summary of the Invention

[0004] To address the shortcomings mentioned in the background section, the present invention aims to provide a dynamic scheduling optimization method and system for cloud-edge collaboration of unmanned vehicles.

[0005] Firstly, the objective of this invention can be achieved through the following technical solution: a dynamic scheduling optimization method for cloud-edge collaboration of unmanned vehicles, the method comprising the following steps: Receive physical information of unmanned vehicles, historical task information, real-time status of unmanned vehicles, and real-time status of edge nodes; integrate the physical information of unmanned vehicles, real-time status of unmanned vehicles, and real-time status of edge nodes to generate a dynamic parameter library; The unmanned vehicle's physical information includes vehicle mass, initial battery level, current 2D position, available computing power, and communication link channel gain; the historical task information includes task data volume, maximum acceptable end-to-end latency, computational complexity, and urgency level; the unmanned vehicle's real-time status includes the latest 2D position, remaining battery level, available computing power, and communication link quality; and the edge node's real-time status includes CPU utilization, memory usage, and residual computing power. Historical tasks are classified based on historical task information and a dynamic parameter library to obtain task classification results, which include light tasks, medium tasks, and heavy tasks. The dynamic parameter library and task classification results are input into the pre-established fission hybrid mutation crayfish optimization algorithm model. With the goal of minimizing the weighted sum of the total system delay and total energy consumption, the globally optimal scheduling scheme is output. The globally optimal scheduling scheme includes unmanned vehicle trajectory adjustment instructions, task unloading path allocation results, and cloud-edge resource allocation ratio.

[0006] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the unmanned vehicle's physical information and historical task information constitute a comprehensive parameter package, and the acquisition process is as follows: Unmanned vehicle physical information acquisition: Real-time acquisition of data through sensor nodes. The number of driverless cars in the The physical parameters of the time slot include: vehicle weight Initial battery capacity Current two-dimensional coordinates Available computing power Communication link channel gain Indicates the first The user and the first The number of driverless cars in the The channel gain of a time slot reflects the degree of signal attenuation; Original task information generation: Generate task quadruple ,in For task data volume, For the maximum acceptable latency, For computational complexity, It is classified as an emergency.

[0007] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the process of acquiring the real-time status of the unmanned vehicle includes: The position of the autonomous vehicle is updated using a two-dimensional Cartesian coordinate system. The position update formula is as follows: in: , The first The number of driverless cars in the The x-axis and y-axis coordinates of the time slot; For the first The number of driverless cars in the The travel speed of the time slot; For the first The number of driverless cars in the The direction angle of the time slot; The duration of the time slot; The transmission rate is calculated based on Shannon's formula, which is: in: For the first The driverless car and the first The number of users / edge nodes in the 1st The transmission rate of a time slot; For communication bandwidth; For transmission power; Channel gain; The noise power spectral density; The energy consumption formula for performing lightweight tasks is: in: Calculate energy consumption locally for driverless vehicles; Energy consumption coefficient; The percentage of unloaded tasks for autonomous vehicles; Energy consumption calculation includes driving energy consumption, idling energy consumption, and transmission energy consumption, with the following formulas: Driving power: in: It is the acceleration due to gravity; The rolling resistance coefficient, air density; This refers to the air drag coefficient; For windward area; For transmission efficiency; Driving energy consumption: Idle power consumption: This refers to idle power. Transmission power consumption: , which represents the transmission time, serves as a latency calculation parameter in the autonomous vehicle communication process and is used to update parameters related to energy consumption calculation and scheduling decisions.

[0008] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the process of acquiring the real-time state of the edge node, including: Receive the The number of driverless cars in the Medium-volume task data transmitted via time slots Parallel computation is performed using edge CPUs, and the processing latency formula is: For edge CPU frequency; The formula for calculating energy consumption is: Medium-volume tasks are executed at the edge; the processing results of medium-volume tasks are output up to the next level. A driverless car; Receive the The number of driverless cars in the Heavy task data transmitted in time slots Data compression is performed using a compression standard determined by system configuration parameters. Redundant pixels are filtered out using a mean filtering algorithm, and preprocessed data is output. The system generates a globally optimal scheduling scheme in the cloud and uses it to guide the adjustment of unmanned vehicle trajectories, selection of task unloading paths, and allocation of cloud and edge resources.

[0009] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the objective function of the pre-established fission hybrid mutation crayfish optimization algorithm model is as follows: in: These are the weighting coefficients; For the total system latency, , For driverless car-user transmission latency, For autonomous vehicle-to-edge transmission latency, For edge-to-cloud transmission latency, To handle latency for autonomous vehicles, To reduce edge processing latency, To reduce latency in the cloud; The total energy consumption of the system. Energy consumption for cloud computing , .

[0010] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the solution process of the fission hybrid mutant crayfish optimization algorithm, comprising the following steps: Population initialization: Each individual is encoded to represent a scheduling scheme that includes the autonomous vehicle trajectory, task offloading ratio, and resource allocation parameters; Population fission: Dividing a population into superior and inferior populations based on the fitness of individuals; Mixed mutation: The superior population is perturbed using the Levy flight strategy, and the inferior population is updated using the logarithmic leader-guided strategy, resulting in the perturbed and updated populations. Gaussian mutation involves perturbing a randomly selected portion of the population with a Gaussian distribution using a preset probability after perturbation and updating. Location update under temperature-food control: Based on temperature parameters related to a preset number of iterations, control individuals to explore or develop, and output the individual with the best fitness. Iterative convergence is achieved, and the scheduling scheme corresponding to the individual with the best fitness is taken as the global optimal scheduling scheme.

[0011] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the population fission partitioning process, comprising: Calculate the fitness value of each individual in the population, and the median fitness value; A population whose individual fitness value is less than the median fitness value is considered a superior population; otherwise, it is considered an inferior population.

[0012] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the fission-hybrid mutation crayfish optimization algorithm is obtained by introducing population fission and hybrid mutation mechanisms based on the Gaussian mutation crayfish optimization algorithm.

[0013] Secondly, in order to achieve the above objectives, this invention discloses a dynamic scheduling and optimization system for unmanned vehicle cloud-edge collaboration, comprising: The data receiving module is used to receive the physical information of the unmanned vehicle, historical task information, real-time status of the unmanned vehicle, and real-time status of edge nodes; it integrates the physical information of the unmanned vehicle, the real-time status of the unmanned vehicle, and the real-time status of edge nodes to generate a dynamic parameter library; The unmanned vehicle's physical information includes vehicle mass, initial battery level, current 2D position, available computing power, and communication link channel gain; the historical task information includes task data volume, maximum acceptable end-to-end latency, computational complexity, and urgency level; the unmanned vehicle's real-time status includes the latest 2D position, remaining battery level, available computing power, and communication link quality; and the edge node's real-time status includes CPU utilization, memory usage, and residual computing power. The task classification module is used to classify historical tasks based on historical task information and a dynamic parameter library to obtain task classification results, wherein the task classification results include light tasks, medium tasks, and heavy tasks. The optimal scheduling module is used to input the dynamic parameter library and task classification results into the pre-established fission hybrid mutation crayfish optimization algorithm model, with the goal of minimizing the weighted sum of the total system delay and total energy consumption, and outputting a globally optimal scheduling scheme; wherein, the globally optimal scheduling scheme includes unmanned vehicle trajectory adjustment instructions, task unloading path allocation results, and cloud-edge resource allocation ratio.

[0014] In another aspect of the present invention, in order to achieve the above-mentioned objective, a terminal device is disclosed, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The memory stores the computer program capable of running on the processor. When the processor loads and executes the computer program, it employs a dynamic scheduling optimization method for unmanned vehicle cloud-edge collaboration as described above.

[0015] The beneficial effects of this invention are: This invention achieves deep collaborative optimization of unmanned vehicle trajectory planning, cross-layer task unloading, and cloud-edge resource allocation by quantifying task description, accurately modeling the physical and logical characteristics of each layer, and improving intelligent optimization algorithms. Ultimately, it achieves the scheduling goal of "low latency, low energy consumption, high security, and high resource utilization". It is applicable to ground mobile scenarios that require dynamic adaptation to the status and task requirements of unmanned vehicles, such as smart city inspection, road emergency response, and environmental monitoring. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the optimized architecture of the present invention; Figure 3 This is a schematic diagram of the fission hybrid mutation COA algorithm of the present invention; Figure 4 This is a schematic diagram of the experimental simulation of the present invention; Figure 5 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0017] 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.

[0018] Example 1: like Figure 1 As shown, a dynamic scheduling optimization method for autonomous vehicle cloud-edge collaboration includes the following steps: S101: Receives unmanned vehicle physical information, historical task information, unmanned vehicle real-time status, and edge node real-time status; integrates unmanned vehicle physical information, unmanned vehicle real-time status, and edge node real-time status to generate a dynamic parameter library; The unmanned vehicle's physical information includes vehicle mass, initial battery level, current 2D position, available computing power, and communication link channel gain; the historical task information includes task data volume, maximum acceptable end-to-end latency, computational complexity, and urgency level; the unmanned vehicle's real-time status includes the latest 2D position, remaining battery level, available computing power, and communication link quality; and the edge node's real-time status includes CPU utilization, memory usage, and residual computing power. Specifically, the physical information and historical mission information of the autonomous vehicle constitute a comprehensive parameter package, and the collection process is as follows: (1) Establish a user terminal information collection and original task generation model: The user terminal consists of user terminals (inspection management platform / logistics scheduling platform / emergency command platform) and sensor nodes (fixed roadside sensors, road surface detectors, user mobile IoT terminals), each time slot (duration) The total simulation time With the number of time slots Calculated Generate a comprehensive parameter package, as follows: Unmanned vehicle physical information collection: Unmanned vehicle physical information collection: Real-time collection of physical information through sensor nodes. The number of driverless cars in the The physical parameters of the time slot include: vehicle weight (Default 1500kg, typical value for electric autonomous vehicles), initial battery capacity (Default 500kWh), Current 2D Coordinates Available computing power (Unit: Hz) Communication link channel gain (No. The user and the first The number of driverless cars in the The channel gain of a time slot reflects the degree of signal attenuation. Original task information generation: Generate task quadruple ,in For task data volume, For the maximum acceptable latency, For computational complexity, Emergency level; The integrated parameter package containing "unmanned vehicle physical information + original mission information" is transmitted to the cloud information integration module.

[0019] The process of acquiring the real-time status of autonomous vehicles and edge nodes includes: Establish a driving control and task execution model for autonomous vehicles: Driving control module: The position of the autonomous vehicle is updated using a two-dimensional Cartesian coordinate system. The position update formula is as follows: in: , The first The number of driverless cars in the x-axis and y-axis coordinates of the time slot (unit: m); For the first The number of driverless cars in the Travel speed in time slot (unit: m / s), range of values (Adapted to urban road speed limits); For the first The number of driverless cars in the The direction angle of the time slot (unit: rad), value range ; The duration of the time slot (unit: seconds, default 2.5 seconds); Communication transmission module: The transmission rate is calculated based on Shannon's formula. Considering loss factors unique to ground scenes such as building obstruction and multipath effects, the transmission rate (unit: bps) formula is as follows: in: For the first The driverless car and the first The number of users / edge nodes in the 1st The transmission rate of a time slot (unit: bps); For communication bandwidth (default 1MHz, suitable for short-range terrestrial wireless communication); Transmission power (unit: W, range of values) ); Channel gain; Noise power spectral density (default) (Typical value for terrestrial wireless communication); The module calculates the transmission delay from lightweight task results to the user end and from medium / heavy task data to the edge end based on the transmission rate; Channel gain of communication transmission module The logarithmic distance path loss model is used for calculation, and the formula is as follows: in For reference distance Channel gain at that time (default -30dB). Path loss index (urban densely populated areas) ,suburbs ), For the first The user and the first The straight-line distance between the driverless cars The shadow fading is (following a normal distribution with a mean of 0 and a variance of 8 dB).

[0020] Local task execution module: Only executes lightweight tasks; the energy consumption calculation formula is as follows: in: Calculate the energy consumption (in J) of the autonomous vehicle locally. Energy consumption factor (default) (typical value of automotive processor). The percentage of unloaded tasks for autonomous vehicles (lightweight tasks are executed entirely locally); Energy consumption calculation module: includes driving energy consumption, idling energy consumption, and transmission energy consumption, with the following formulas: ① Driving power: in: The acceleration due to gravity ( ); The rolling resistance coefficient is 0.015 (typical value for asphalt pavement). air density ( ); The drag coefficient is 0.3 (typical value for streamlined autonomous vehicles). For the windward area ( Typical value for small driverless vehicles); Transmission efficiency (0.85, typical value for electric unmanned vehicles); ② Driving energy consumption: ③ Idle energy consumption: This refers to the idle power, which is 150W by default, and the standby power consumption of in-vehicle equipment. ④ Transmission energy consumption: , is the transmission time; Status feedback module: After each time slot ends, the "number" will be sent to the next time slot. The number of driverless cars in the Latest coordinates of the time slot Remaining battery power Available computing power Transmission rate "Package and transmit to the cloud parameter update module."

[0021] Establish an edge task processing and data preprocessing model: The edge includes at least two regional edge nodes (covering a radius of 500m, adapting to the local processing needs of ground mobile scenarios), and each node integrates the following modules: Medium-volume task processing module: receives the first... The number of driverless cars in the Medium-volume task data transmitted via time slots Call the edge CPU (default frequency) The parallel computing is completed, and the processing delay formula is: ( (This refers to the edge CPU frequency); the energy consumption calculation formula is: ( (Medium-volume task edge execution); output the medium-volume task processing results to the first... A driverless car; Heavy task preprocessing module: receives the first The number of driverless cars in the Heavy task data transmitted in time slots Data compression was performed using the H.265 compression standard (compression rate 50%, compressed data size). The preprocessed data is output by filtering out redundant pixels using a mean filtering algorithm. Up to the cloud; Edge status feedback module: After each time slot ends, it will output the "edge node CPU utilization". Memory usage Residual computing power "Package and transmit to the cloud parameter update module."

[0022] S102: Classify historical tasks based on historical task information and dynamic parameter library to obtain task classification results, wherein the task classification results include light tasks, medium tasks and heavy tasks. Establish a cloud-based global optimization and dynamic decision-making model: Information integration module: Receives comprehensive parameter packages from the user terminal, real-time status of the autonomous vehicle, and real-time status from the edge terminal; constructs a dynamic parameter library; the library contains data including: autonomous vehicle parameters ( ), edge parameters ( ), task parameters ( ); Task hierarchical scheduling module: performs initial task screening based on a dynamic parameter library, with the following rules: ① Lightweight tasks: and →Assigned to be executed locally on the autonomous vehicle; ② Medium-volume tasks: and → Assign execution to the edge; ③ Heavy workload: or → Assigned to the “edge preprocessing - cloud processing” link; Constraint verification module: Verifies three types of constraints, with the following formulas: ① Battery constraints for autonomous vehicles: The remaining power should not be less than 10% of the total capacity to avoid power loss; ② Collision avoidance constraints for autonomous vehicles: Among them, the first , The number of driverless cars in the The Euclidean distance of the time slot shall not be less than the minimum collision avoidance distance of 10m; ③ Task completion constraints: in The percentage of tasks that are unloaded. Ensure that all tasks are assigned to autonomous vehicles / edge / cloud; If the constraint is violated, the penalty term will be calculated: in , , This is the penalty coefficient (default value is 1000). Cloud CPU frequency of cloud task processing module default The formula for processing delay of heavy tasks is: The formula for calculating energy consumption is: in The processing results are forwarded to the autonomous vehicle via the edge terminal.

[0023] S103: Input the dynamic parameter library and task classification results into the pre-established fission hybrid mutation crayfish optimization algorithm model, with the goal of minimizing the weighted sum of the total system delay and total energy consumption, and output the globally optimal scheduling scheme; wherein, the globally optimal scheduling scheme includes unmanned vehicle trajectory adjustment instructions, task unloading path allocation results and cloud-edge resource allocation ratio.

[0024] The objective function of the pre-established optimization algorithm model for fission-hybrid mutation crayfish is as follows: in: Weighting coefficient (default 0.5, adjustable for different scenarios; emergency scenarios) Prioritize low latency; logistics scenarios (Prioritize low energy consumption); For the total system latency, ( For driverless car-user transmission latency, For autonomous vehicle-to-edge transmission latency, For edge-to-cloud transmission latency, To handle latency for autonomous vehicles, To reduce edge processing latency, (for cloud processing latency). The total energy consumption of the system. ( Energy consumption for cloud computing , Heavy tasks are executed entirely in the cloud. The solution process of the optimization algorithm for fission-hybrid mutation crayfish includes the following steps: S1 Population Initialization: In the autonomous vehicle-cloud-edge collaborative model, each crayfish individual is coded as a cloud-edge collaborative scheduling scheme, requiring population initialization—randomly generated within the upper and lower bounds of various parameters. Group of candidate solutions ,in Population size, It is the population dimension; Population coding scheme: The variable dimension is 2560, including user-side: task type (light / medium / heavy) and task data volume. Autonomous vehicle terminal: Driving speed Direction angle Task uninstallation ratio Transmission power Edge computing: concurrent task processing, data compression rate, and status feedback cycle; Cloud computing: algorithm iteration parameters and constraint verification thresholds. Encoding vector subdivision: Driving-related: (speed), (Direction angle); Task-related: (Uninstallation ratio) (Transmission power) (Task activation indicator: 1 = Task active, 0 = No task active); Character-related: (Autonomous vehicle role) (Uninstallation path); The initialization formula is: in, Indicates the first The lower bound of dimensionality Indicates the first The upper bound of the dimension, It is a random number between 0 and 1.

[0025] S2 Population Fission: Enter the iterative loop and calculate the fitness value of each individual in the population. To obtain the best individual in history The best individual in the current generation And the median number of individuals after the current population fitness ranking. Populations are divided according to fitness: in, Indicates the current individual, This represents the individual in the current population with the median fitness. , These represent individuals from superior and inferior populations, respectively.

[0026] S2.1 Superior Population Mutation: The superior individuals after fission are perturbed by incorporating the Levi flight mechanism to increase their global search performance and optimize global parameters in the cloud (cloud computing power allocation, autonomous vehicle trajectory adjustment); the formula is: in, This represents the currently iterative superior population individual, which undergoes a global search for mutation using the Levy flight perturbation. , (constant), , and It is a random number between 0 and 1.

[0027] S2.2 Inferior Population Mutation: A logarithmic leader-led mechanism is used for inferior populations to optimize the interaction parameters between the edge and the autonomous vehicle (edge ​​concurrency, transmission power) and accelerate convergence; the formula is: in, This refers to individuals in the current generation of inferior populations. (Helix shape parameters) For the range Uniformly distributed random numbers, It is the best individual in the current population. This represents the current iteration number.

[0028] S2.3 Merging Populations Gaussian Mutation: Perform Gaussian mutation on the new population after merging superior and inferior populations; Mutation formula: in, For randomly selected dimension indexes ( ); These are random numbers distributed according to a standard normal distribution. , For the first The upper and lower bounds of the dimension variable; 0.1 is the range scaling factor, which controls the magnitude of variation.

[0029] S3 Temperature-Food Control: This algorithm balances exploration and development through temperature parameter control. The temperature decreases with increasing iterations (adapting to the "global search-local optimization" requirements of autonomous vehicle ground path planning). The formula is: in, For the first The temperature of the next iteration; (Initial temperature, default value); (Total number of iterations, default value); This represents the current iteration number.

[0030] S3.1 Temperature Solution: Individual crayfish randomly seek refuge from the heat or compete for burrows, exploring new cloud-edge collaborative modes; Competition Phase: in, Indicates the current cave location. Represents a random individual in a population; Summer vacation period: in, Each corresponds to the best individual in the current population. and the best individual in history , It is a variable (decreasing as the number of iterations increases). It is a random value between 0 and 1.

[0031] S3.2 Temperature Processing: The current population has favorable living conditions suitable for feeding. Fine-tune the marginal data compression rate and calculate the size of the food. ; Formula for calculating food size: in, It is a random number between 0 and 1. This represents the fitness value of the current individual. This represents the fitness value of the historically best individual. S3.2.1 If Food that is too large needs to be cut, and the population of individuals converges sinusoidally. in, Indicates food value. To dynamically adjust the parameters, the calculation method is as follows: S3.2.2 If The crayfish were feeding normally, and the individual moved to the food location; S4 Iterative convergence: After updating all individuals, update the current best individual in the population. Compared with the best historical individuals When the number of iterations If the objective function value fluctuates by less than 1% for 50 consecutive iterations, stop the iteration and output the optimal scheduling scheme under cloud-edge collaboration (including unmanned vehicle trajectory, task unloading path, and cloud-edge resource allocation).

[0032] The fission-hybrid mutation crayfish optimization algorithm is obtained by introducing population fission and hybrid mutation mechanisms into the Gaussian mutation crayfish optimization algorithm.

[0033] The solution steps based on the Gaussian mutation crayfish optimization algorithm include: 3.1. Initialize parameters, including: maximum number of iterations. Dimension upper limit of solution space Lower bound of solution space Population size , variation threshold ; 3.2. Initialize Population Location: Each crayfish individual represents one cloud-edge collaborative scheduling scheme. The encoding includes the user-side: task type (light / medium / heavy) and task data volume. Autonomous vehicle terminal: Driving speed Direction angle Task uninstallation ratio Transmission power Edge computing: task processing concurrency, data compression rate, and status feedback cycle; Cloud computing: algorithm iteration step size and constraint verification threshold; calculating individual diversity in the population and calculating the optimal individual sequence of the population. ; 3.3. Enter the iterative loop and determine the fitness of the current individual. If the current individual fitness If the performance is good, the fission will be preserved as a superior population (a superior solution for cloud-edge collaboration, with edge task processing rate ≥ 80% and cloud command response latency ≤ 50ms); otherwise, the fission will be preserved as an inferior population (an inferior solution for cloud-edge collaboration, such as edge load ≥ 90% or cloud transmission latency ≥ 100ms). 3.3.1 For superior individuals: Use the Levy flight strategy to perturb superior individuals, focusing on optimizing the global resource allocation parameters in the cloud (the cloud's computing power allocation ratio to the edge). Adjustment range of driverless vehicle trajectory This gives it a more effective global search capability, avoiding getting trapped in local optima; the Lévy flight perturbation formula is: in, (constant), For Levi's flight distribution strategy, ,and , It is a random number between 0 and 1; 3.3.2 For inferior populations: Continue mutation, adopt a logarithmic leader-guided mechanism to accelerate convergence, and focus on correcting the interaction parameters between the edge and the autonomous vehicle (concurrency of edge task processing, autonomous vehicle-edge transmission power). ), accelerates the overall population iteration; the logarithmic leader guide formula is: in, This refers to individuals in the current generation of inferior populations. This represents the current iteration number. (Helix shape parameters) For the range Uniformly distributed random numbers, It is the best individual in the current population; 3.4. Merge superior and inferior populations to obtain a new population, and simultaneously perform Gaussian mutation: focus on perturbing key parameters of cloud-edge collaboration (edge ​​data compression rate, cloud algorithm iteration step size), and based on this, calculate the population temperature value. Determine the range of temperature values ​​and calculate the location of burrows within the population. The Gaussian mutation formula is: in: For randomly selected dimension indexes ( ); These are random numbers distributed according to a standard normal distribution. , For the first The upper and lower bounds of the dimension variable; 0.1 is the range scaling factor; 3.4.1 If the current temperature of the individual crayfish Individual crayfish randomly seek refuge from the heat or compete for burrows, exploring new cloud-edge collaboration solutions and adjusting the deployment positions of edge nodes; the formula for the competition phase is: in, Indicates the current cave location. This represents a random individual in the population; the formula for the summer retreat stage is: in, Each corresponds to the best individual in the current population. and the best individual in history , It is a variable (decreasing as the number of iterations increases). It is a random value between 0 and 1; 3.4.2 If the current temperature of the individual crayfish This indicates that the current population's living conditions are favorable for feeding. The cloud-edge collaborative edge data compression rate has been fine-tuned, and the size of the food has been calculated. The formula for calculating food size is: in, It is a random number between 0 and 1. This represents the fitness value of the current individual. This represents the fitness value of the historically best individual. 3.5. After obtaining the size of the food, if This indicates that the food is too large and needs to be processed. The current individual cuts the food into smaller pieces, and the population exhibits sinusoidal convergence. The sinusoidal convergence formula is: in, Indicates food value. To dynamically adjust the parameters, the calculation method is as follows: ; 3.6. If the food size This indicates that crayfish can eat normally. The current individual moves to the current food location and "eats" the food. The formula is: ; 3.7. After all individuals have been updated, calculate the best position since the last iteration. and the current optimal position of the population Then proceed to the next iteration.

[0034] The optimization algorithm for crayfish based on fission and hybrid mutation introduces population fission and hybrid mutation mechanisms, enabling the algorithm to escape local convergence in later stages. This includes defining superior and inferior fission populations: in, Indicates the current individual, This represents the individual in the current population with the median fitness. , These represent individuals from superior and inferior populations, respectively. The worst individual in the current population is identified by comparing the fitness of current individuals. and median fitness value To classify populations.

[0035] The fission-hybrid mutation optimization algorithm for crayfish introduces population fission and hybrid mutation mechanisms, enabling the algorithm to escape local convergence in later stages, including the population hybrid mutation mechanism: Superior population mixing and variation: in, This represents the current iteration of superior individuals in the population, which undergoes a global search for mutation using the Lévy flight perturbation; and , (constant), , and It is a random number between 0 and 1; Inferior population mixing and mutation: using a logarithmic leader-guided mechanism to accelerate convergence: in, This refers to individuals in the current generation of inferior populations. (Helix shape parameters) For the range Uniformly distributed random numbers, It is the best individual in the current population. This represents the current iteration number; Gaussian mutation after merging superior and inferior populations: Mutation probability: 15%, meaning that each individual in the population has a 15% probability of triggering mutation, ensuring that the algorithm retains its global search capability and avoiding suboptimal scheduling schemes due to local optima on the ground path (such as local road congestion); Mutation Dimension: 5% of the total dimensions, that is, 128 dimensions are randomly selected from the 2560-dimensional encoding for mutation (e.g., randomly adjusting the speed of the autonomous vehicle in some time slots). Uninstallation ratio ), balancing "exploration" (new path search) and "development" (existing path optimization); Mutation formula: in: For randomly selected dimension indexes ( ); For standard normal distribution random numbers ( This generates Gaussian distribution perturbations; , For the first Upper and lower bounds of dimensional variables (such as direction angle) of , Transmission power of , ); 0.1 is the range scaling factor, which controls the mutation amplitude (avoiding sudden changes in speed from 1m / s to 15m / s, ensuring smooth ground driving); the Gaussian mutation operation is applied independently to each individual in the population in each iteration, breaking local optima through random perturbation (such as energy redundancy of fixed ground routes), and enhancing the robustness of the algorithm; Furthermore, the fission-hybrid crayfish optimization algorithm is constructed based on the Gaussian crayfish optimization algorithm: The Gaussian mutation crayfish optimization algorithm is derived from a swarm intelligence optimization model. It simulates the foraging, burrow competition, and temperature response behaviors of crayfish in the natural environment to search for continuous high-dimensional optimization problems. The Gaussian mutation mechanism is used to introduce random perturbations during the individual position update process to improve the algorithm's search diversity in complex solution space and its ability to escape local optima.

[0036] Based on this, the present invention combines the high-dimensional, strong-constraint and time-varying characteristics of the cloud-edge collaborative dynamic scheduling scenario of unmanned vehicles, and makes targeted improvements to the Gaussian mutant crayfish optimization algorithm. It introduces a population fission mechanism and a hybrid mutation mechanism to construct a fission hybrid mutant crayfish optimization algorithm, which can adapt to the joint optimization requirements of unmanned vehicle trajectory planning, task unloading and cloud-edge resource allocation.

[0037] Specifically, the population fission mechanism divides the current population into superior and inferior populations based on the median fitness of individuals. The superior population focuses on enhancing global search capabilities to optimize the allocation ratio of cloud resources and the overall trajectory adjustment parameters of the unmanned vehicle. The inferior population focuses on improving local convergence performance to correct local scheduling variables such as transmission power and concurrent processing parameters between the unmanned vehicle and edge nodes, thereby avoiding getting trapped in local optima while ensuring convergence speed.

[0038] The hybrid mutation mechanism is based on the Gaussian mutation crayfish optimization algorithm, combining the Lévy flight perturbation with the logarithmic leader guidance strategy: the Lévy flight perturbation is introduced to the superior population to expand the search range, the logarithmic leader guidance is introduced to the inferior population to accelerate convergence, and the Gaussian mutation operation is uniformly applied after the superior and inferior populations are merged, thus forming a multi-level, multi-scale search strategy.

[0039] In the unmanned vehicle cloud-edge collaborative scheduling application of the present invention, the input parameters of the fission hybrid mutant crayfish optimization algorithm include at least: unmanned vehicle physical parameters, unmanned vehicle real-time status parameters, edge node real-time status parameters, task data volume, task latency constraints, computational complexity, and urgency level; the optimization output results of the algorithm include at least: unmanned vehicle trajectory adjustment parameters, task unloading path allocation results, and the computing power resource allocation ratio between the cloud and edge nodes.

[0040] Through the above methods, the fission hybrid mutation crayfish optimization algorithm can achieve joint optimization of the total system latency and total energy consumption in the dynamic scheduling scenario of unmanned vehicle cloud-edge collaboration, thereby effectively solving the problem that the existing unmanned vehicle scheduling scheme is prone to getting trapped in local optima and is difficult to simultaneously take into account real-time performance and energy consumption control.

[0041] Specifically, the present invention will be further illustrated below through embodiments: I. Experimental Environment Configuration: Software environment: This embodiment uses MATLAB R2020b as the simulation development tool. Relying on its numerical calculation, matrix operation and visualization functions, it realizes the construction of the cloud-edge collaborative model of unmanned vehicles, the coding and solving of the COA algorithm, and the plotting of the convergence curve and result analysis. Simulation scenario setting; This embodiment focuses on a smart city ground inspection scenario, constructing a 1000m×1000m rectangular service area. The scenario covers the entire chain from "user task input - autonomous vehicle path planning - task unloading decision - cloud-edge collaborative processing," simulating typical physical and communication interference on urban roads to ensure the realism and universality of the experimental scenario. The specific scenario settings are as follows; The four-layer architecture deployment nodes are as follows: User side: Simulate 10 groups of user input task sources, corresponding to 10 fixed demand points (5 street light inspection points, 3 manhole cover inspection points, and 2 traffic signal light inspection points), randomly distributed in the service area; On the unmanned vehicle side: 8 inspection unmanned vehicles are initially distributed at the edge of the area (coordinates are (50,50), (50,950), (950,50), (950,950), (50,500), (950,500), (500,50), (500,950)); the initial speed is 5m / s, the initial direction angle is random, and the basic path is planned according to the principle of "full coverage and no missed inspection"; real-time status parameters (current two-dimensional coordinates, remaining power, available computing power, communication link quality) are fed back to the user terminal, and the speed ([1,15]m / s) and direction angle ([0,2π]rad) are dynamically adjusted based on the user task priority; Edge: Two regional edge nodes are logically deployed in the center of the business district (coordinates (500, 250)) and the center of the park (coordinates (500, 750)); the coverage radius of a single node is 500m, the CPU frequency is 5e8Hz, the data compression rate is 50%, the status feedback cycle is 2.5s, and it receives user task input and unmanned vehicle status information, adapting to the needs of "nearby processing" of medium-volume tasks and pre-processing heavy tasks; Cloud: The logic is deployed outside the service area and connected to the edge nodes through a high-speed wired link. The CPU frequency is 1e12Hz, the initial temperature is 100℃, and the total number of iterations is 1000. It summarizes the user-end task parameters and the status data of the unmanned vehicle and the edge terminal to achieve global scheduling optimization. Scene interference simulation: Communication interference: In the high-rise area of ​​the business district (coordinates (300-700, 100-400)), the channel gain of the unmanned vehicle-edge node is reduced by 20%; in rainy weather, the transmission rate of the edge-cloud fluctuates by ±15% (simulated multipath effect).

[0042] Physical losses: The unmanned vehicle travels on an asphalt road surface (rolling resistance coefficient 0.015), with an ambient wind speed of 3 m / s (air resistance coefficient 0.3), simulating real-world driving energy consumption losses; Dynamic constraints: Minimum collision avoidance distance for autonomous vehicles is 10m, and maximum battery capacity is 500kWh (1.8×10). 9 J), to ensure that the experiment conforms to the actual engineering constraints.

[0043] System parameter and algorithm parameter settings; The parameter configuration in this embodiment strictly matches the four-layer sub-model of "user-autonomous vehicle-edge-cloud", clearly defining the core parameters and COA algorithm parameters of each layer to ensure the rationality of the simulation, as shown in Tables 1 and 2 below: Table 1 System Parameters Table 2 COA Algorithm Parameters and Task Parameters The implementation steps of the optimization algorithm for fission-hybrid mutation crayfish are as follows: Based on the above parameter configuration, this embodiment implements the fission-hybrid mutation COA algorithm using MATLAB R2020b encoding to solve the optimal scheme for cloud-edge collaborative scheduling of unmanned vehicles. The specific steps are as follows: Step 1: Population Initialization: 1. Coding rule design: Each individual crayfish corresponds to a complete "four-layer collaborative scheduling scheme", with a coding dimension of 2560 dimensions. The coding vector structure is classified into "driving-related - task-related - role-related", as follows: Driving-related (640 dimensions): 8 autonomous vehicles × 40 time slots × 2 parameters (speed) Direction angle ),in m / s, rad; Task-related (1200 dimensions): 10 user task sources × 40 time slots × 3 parameters (offload ratio) Transmission power Task activation indicator ),in , W, (0 = no task input, 1 = task input); Role-related (720 dimensions): 8 autonomous vehicles × 40 time slots × 1 role parameter ( 1 = Edge computing collaboration, 2 = Local service, 3 = Cloud collaboration) + 10 user task sources × 40 time slots × 1 unload path parameter ( 1 = driverless car, 2 = regional edge, 3 = cloud center). 2. Population Initialization: Randomly generate 50 candidate solutions (population) within the upper and lower bounds of each parameter. The initialization formula is: in, For the first The first individual Dimensional parameter values, , The first Upper and lower bounds of the dimensional parameter, A random number that is uniformly distributed between 0 and 1.

[0044] Step 2: Population fitness calculation and fission; 1. Using "minimizing the total system cost" as the fitness index, the function formula is: in: The total system latency (in seconds) includes user task transmission latency. With processing latency ( (To calculate node frequency, dynamically switch according to unloading path). Let J be the total system energy consumption, and J be the energy consumption of the autonomous vehicle. Idle energy consumption Task transmission power consumption Calculate energy consumption sum; These are the weighting coefficients. To constrain violations and penalties.

[0045] 2. Calculation of constraints and penalties: Constraints: ① Battery constraints for autonomous vehicles: ; ② Collision avoidance constraints: m; ③ Task completion constraints: (Ensure that all user input tasks are assigned for processing).

[0046] Penalty formula: in This ensures that the algorithm prioritizes meeting engineering constraints.

[0047] 3. Population fission operation: Calculate the fitness value of all individuals, using the median fitness value. To divide the population into boundaries: in As a superior individual, Individuals belonging to an inferior population. The best individual in the current population. As the worst individual in the current population, fission enables a division of labor where "superior populations enhance global search and inferior populations accelerate convergence".

[0048] Step 3: Mixed mutation operation: 1. Superior Population Mutation (Levi Flight Perturbation): Adds a Levy flight perturbation to individuals in the superior population to enhance global search capabilities. The formula is: in , , The random numbers are based on a standard normal distribution, and the focus is on optimizing the global resource allocation parameters in the cloud and the adjustment range of the autonomous vehicle trajectory.

[0049] 2. Inferior Population Variation (Logarithmic Leader Guidance): For inferior populations, a logarithmic leader guidance mechanism is used to accelerate convergence. The formula is: in This represents the current iteration number. (Helix shape parameters) To ensure uniform distribution of random numbers, the interaction parameters between the edge device and the autonomous vehicle device (such as the number of concurrent edge task processing and transmission power) are carefully corrected.

[0050] 3. Gaussian Mutation of Merged Populations: Gaussian mutation is performed on the new population after merging superior and inferior populations to break local optima. The formula is: in For randomly selected dimension indexes ( ), 0.1 is the range scaling factor, which controls the mutation range; in this embodiment, 128 dimensions (5% of the total dimensions) are randomly selected for mutation, including core parameters such as vehicle speed, transmission power, and edge compression rate.

[0051] Step 4: Temperature-Food Control Mechanism 1. Temperature Update: The temperature decreases with each iteration to balance the algorithm's "exploration" (high temperature) and "development" (low temperature) capabilities. The formula is: in For the first The temperature of the next iteration The initial temperature, This represents the total number of iterations.

[0052] 2. Temperature branching processing: like Individual crayfish randomly engage in either "competitive burrowing" or "seeking refuge from the heat" behavior. ① Competitive Caves: in This is the current cave location. For random individuals in the population; ② Escaping the summer heat: in As the best individual in history, It decreases linearly with the number of iterations.

[0053] like Calculate food size And perform the feeding procedure: ① Food size calculation: ② If (Excessive food supply): Population sinusoidal convergence in For food location, To dynamically adjust parameters; ③ If (Appropriate food): Individuals directly "eat". Step 5: Iterative convergence and output of the optimal solution; 1. Convergence criteria: Iteration stops when any of the following conditions are met: the number of iterations reaches a certain threshold. The objective function value fluctuates by less than 1% over 50 consecutive iterations.

[0054] 2. Optimal solution extraction: Output the scheduling scheme corresponding to the historical best individual. The core content includes: Autonomous vehicle trajectory: The two-dimensional coordinates of the autonomous vehicle in each time slot (calculated by speed and orientation angle) are fed back to the user terminal for task execution path; Task processing decision: Each user inputs the task's unloading path (local / autonomous vehicle / edge / cloud) and unloading ratio, specifying the task execution node; Resource allocation: transmission power of autonomous vehicles, computing power allocation of edge nodes, and cloud computing power scheduling ratio; Status feedback instructions: Real-time location, battery level, task completion progress, and other parameters pushed by the autonomous vehicle to the user terminal.

[0055] Experimental Results and Analysis: This embodiment completes the simulation through 1000 algorithm iterations. The output optimal scheduling scheme satisfies all constraints. The core performance indicators and cloud-edge collaboration effects are as follows: Table 3 Core Performance Indicators Cloud-edge collaboration effect: 1. Role distribution of unmanned vehicles: Edge computing collaboration: Local service: Cloud collaboration = 128:96:64, with balanced role division and no overload of any single role; 2. User task unloading path distribution: Local:Autonomous Vehicle:Regional Edge:Cloud = 144:96:128:96. Lightweight tasks 92% choose local / autonomous vehicle processing, medium-weight tasks 88% choose edge processing, and heavyweight tasks 100% choose "edge preprocessing - cloud processing", matching the processing needs of user task types. 3. Edge node load: The CPU utilization of the two edge nodes is 65% and 62% respectively, and the memory utilization is 58% and 61% respectively. There is no overload or idle resources.

[0056] Algorithm convergence verification: 1. The total cost decreases monotonically with the number of iterations and tends to stabilize after 800 iterations (fluctuation < 0.5%), proving that the algorithm converges without oscillations; 2. The total latency of a single time slot user task decreased from the initial 3.22s to 0.38s, and the total energy consumption of a single time slot decreased from 285632.17J to 203267.89J, verifying the algorithm's ability to coordinate the optimization of latency and energy consumption; 3. Compared with traditional PSO and GA algorithms, the fission hybrid mutation COA algorithm of this invention improves the convergence speed by 30%, improves the quality of the optimal solution by 15%, and avoids the problem of local optima.

[0057] Implementation Conclusion: This embodiment fully reproduces the "unmanned vehicle cloud-edge collaborative scheduling optimization method based on fission hybrid mutation COA algorithm" through MATLAB simulation. Experimental results show that: 1. The four-layer collaborative model of "user-autonomous vehicle-edge-cloud" constructed in this invention can accurately map the interaction characteristics between user task input and autonomous vehicle status feedback, and the optimization results deviate from the real scene by less than 10%; 2. The improved fission-hybrid mutation COA algorithm can efficiently solve high-dimensional dynamic scheduling problems and achieve joint optimization of unmanned vehicle trajectory planning, user task offloading and resource allocation; 3. This method can effectively reduce the total system cost by 18.5%, optimize energy consumption distribution, and meet the real-time requirements of user tasks. It is applicable to ground unmanned vehicle cluster scenarios such as smart city inspection and urban logistics distribution, and has high practical value and scalability.

[0058] Furthermore, the user-end task input types of this invention can be flexibly expanded (such as logistics delivery orders and emergency response instructions), and the time slot duration can be adaptively adjusted according to task density (shortened to 1.5s for high-density tasks and extended to 5s for low-density tasks), further improving scenario adaptability and providing a general solution for cloud-edge collaborative scheduling of unmanned vehicles in different ground mobile scenarios.

[0059] Example 2: To achieve the above objective, such as Figure 5 As shown, based on Embodiment 1, this invention discloses a dynamic scheduling optimization system for unmanned vehicle cloud-edge collaboration, comprising: Data receiving module 11 is used to receive unmanned vehicle physical information, historical task information, unmanned vehicle real-time status and edge node real-time status; integrate unmanned vehicle physical information, unmanned vehicle real-time status and edge node real-time status to generate dynamic parameter library; The unmanned vehicle's physical information includes vehicle mass, initial battery level, current 2D position, available computing power, and communication link channel gain; the historical task information includes task data volume, maximum acceptable end-to-end latency, computational complexity, and urgency level; the unmanned vehicle's real-time status includes the latest 2D position, remaining battery level, available computing power, and communication link quality; and the edge node's real-time status includes CPU utilization, memory usage, and residual computing power. The task classification module 12 is used to classify historical tasks based on historical task information and dynamic parameter library to obtain task classification results, wherein the task classification results include light tasks, medium tasks and heavy tasks. The optimal scheduling module 13 is used to input the dynamic parameter library and task classification results into the pre-established fission hybrid mutation crayfish optimization algorithm model, with the goal of minimizing the weighted sum of the total system delay and total energy consumption, and outputting a globally optimal scheduling scheme; wherein, the globally optimal scheduling scheme includes unmanned vehicle trajectory adjustment instructions, task unloading path allocation results and cloud-edge resource allocation ratio.

[0060] Based on the same inventive concept, this invention also provides a computer device, comprising: one or more processors, and a memory for storing one or more computer programs; the programs include program instructions, and the processor executes the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, used to implement one or more instructions, specifically for loading and executing one or more instructions stored in a computer storage medium to implement the above-described method.

[0061] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, performs the above-described method. This storage medium can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the present invention, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0062] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0063] The foregoing has shown and described the basic principles, main features, and advantages of this disclosure. Those skilled in the art should understand that this disclosure is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this disclosure. Various changes and modifications can be made to this disclosure without departing from its spirit and scope, and all such changes and modifications fall within the scope of this disclosure as claimed.

Claims

1. A dynamic scheduling optimization method for cloud-edge collaboration of unmanned vehicles, characterized in that, The method includes the following steps: Receive physical information of unmanned vehicles, historical task information, real-time status of unmanned vehicles, and real-time status of edge nodes; integrate the physical information of unmanned vehicles, real-time status of unmanned vehicles, and real-time status of edge nodes to generate a dynamic parameter library; The unmanned vehicle's physical information includes vehicle mass, initial battery level, current 2D position, available computing power, and communication link channel gain; the historical task information includes task data volume, maximum acceptable end-to-end latency, computational complexity, and urgency level; the unmanned vehicle's real-time status includes the latest 2D position, remaining battery level, available computing power, and communication link quality; and the edge node's real-time status includes CPU utilization, memory usage, and residual computing power. Historical tasks are classified based on historical task information and a dynamic parameter library to obtain task classification results, which include light tasks, medium tasks, and heavy tasks. The dynamic parameter library and task classification results are input into the pre-established fission hybrid mutation crayfish optimization algorithm model. With the goal of minimizing the weighted sum of the total system delay and total energy consumption, the globally optimal scheduling scheme is output. The globally optimal scheduling scheme includes unmanned vehicle trajectory adjustment instructions, task unloading path allocation results, and cloud-edge resource allocation ratio.

2. The dynamic scheduling optimization method for unmanned vehicle cloud-edge collaboration according to claim 1, characterized in that, The physical information and historical mission information of the unmanned vehicle constitute a comprehensive parameter package, and the collection process is as follows: Unmanned vehicle physical information acquisition: Real-time acquisition of data through sensor nodes. The number of driverless cars in the The physical parameters of the time slot include: vehicle weight Initial battery capacity Current two-dimensional coordinates Available computing power Communication link channel gain Indicates the first The user and the first The number of driverless cars in the The channel gain of a time slot reflects the degree of signal attenuation; Original task information generation: Generate task quadruple ,in For task data volume, For the maximum acceptable latency, For computational complexity, It is classified as an emergency.

3. The dynamic scheduling optimization method for unmanned vehicle cloud-edge collaboration according to claim 1, characterized in that, The process of acquiring the real-time status of the unmanned vehicle includes: The position of the autonomous vehicle is updated using a two-dimensional Cartesian coordinate system. The position update formula is as follows: in: , The first The number of driverless cars in the The x-axis and y-axis coordinates of the time slot; For the first The number of driverless cars in the The travel speed of the time slot; For the first The number of driverless cars in the The direction angle of the time slot; The duration of the time slot; The transmission rate is calculated based on Shannon's formula, which is: in: For the first The driverless car and the first The number of users / edge nodes in the 1st The transmission rate of a time slot; For communication bandwidth; For transmission power; Channel gain; The noise power spectral density; The energy consumption formula for performing lightweight tasks is: in: Calculate energy consumption locally for driverless vehicles; Energy consumption coefficient; The percentage of unloaded tasks for autonomous vehicles; Energy consumption calculation includes driving energy consumption, idling energy consumption, and transmission energy consumption, with the following formulas: Driving power: in: It is the acceleration due to gravity; The rolling resistance coefficient, air density; This refers to the air drag coefficient; For windward area; For transmission efficiency; Driving energy consumption: Idle power consumption: This refers to idle power. Transmission power consumption: , which represents the transmission time, serves as a latency calculation parameter in the autonomous vehicle communication process and is used to update parameters related to energy consumption calculation and scheduling decisions.

4. The dynamic scheduling optimization method for unmanned vehicle cloud-edge collaboration according to claim 1, characterized in that, The process of acquiring the real-time status of the edge nodes includes: Receive the The number of driverless cars in the Medium-volume task data transmitted via time slots Parallel computation is performed using edge CPUs, and the processing latency formula is: For edge CPU frequency; The formula for calculating energy consumption is: Medium-volume tasks are executed at the edge; the processing results of medium-volume tasks are output to the next level. A driverless car; Receive the The number of driverless cars in the Heavy task data transmitted in time slots Data compression is performed using a compression standard determined by system configuration parameters. Redundant pixels are filtered out using a mean filtering algorithm, and preprocessed data is output. The system generates a globally optimal scheduling scheme in the cloud and uses it to guide the adjustment of unmanned vehicle trajectories, selection of task unloading paths, and allocation of cloud-edge resources.

5. The dynamic scheduling optimization method for unmanned vehicle cloud-edge collaboration according to claim 1, characterized in that, The objective function of the pre-established fission-hybrid mutation crayfish optimization algorithm model is as follows: in: These are the weighting coefficients; For the total system latency, , For driverless vehicle-to-user transmission latency, For autonomous vehicle-to-edge transmission latency, For edge-to-cloud transmission latency, To handle latency for autonomous vehicles, To reduce edge processing latency, To reduce latency in the cloud; The total energy consumption of the system. Energy consumption for cloud computing , .

6. The dynamic scheduling optimization method for unmanned vehicle cloud-edge collaboration according to claim 5, characterized in that, The solution process of the fission hybrid mutation crayfish optimization algorithm includes the following steps: Population initialization: Each individual is encoded to represent a scheduling scheme that includes the autonomous vehicle trajectory, task offloading ratio, and resource allocation parameters; Population fission: Dividing a population into superior and inferior populations based on the fitness of individuals; Mixed mutation: The superior population is perturbed using the Levy flight strategy, and the inferior population is updated using the logarithmic leader-guided strategy, resulting in the perturbed and updated populations. Gaussian mutation involves perturbing a randomly selected portion of the population with a Gaussian distribution using a preset probability after perturbation and updating. Location update under temperature-food control: Based on temperature parameters related to a preset number of iterations, control individuals to explore or develop, and output the individual with the best fitness. Iterative convergence is achieved, and the scheduling scheme corresponding to the individual with the best fitness is taken as the global optimal scheduling scheme.

7. The dynamic scheduling optimization method for unmanned vehicle cloud-edge collaboration according to claim 6, characterized in that, The process of population fission includes: Calculate the fitness value of each individual in the population, and the median fitness value; A population whose individual fitness value is less than the median fitness value is considered a superior population; otherwise, it is considered an inferior population.

8. The dynamic scheduling optimization method for unmanned vehicle cloud-edge collaboration according to claim 7, characterized in that, The fission-hybrid mutation crayfish optimization algorithm is obtained by introducing population fission and hybrid mutation mechanisms based on the Gaussian mutation crayfish optimization algorithm.

9. A dynamic scheduling optimization system for unmanned vehicle cloud-edge collaboration, employing the dynamic scheduling optimization method for unmanned vehicle cloud-edge collaboration as described in any one of claims 1 to 8, characterized in that, include: The data receiving module is used to receive physical information of the unmanned vehicle, historical task information, real-time status of the unmanned vehicle, and real-time status of edge nodes. The system integrates the physical information of autonomous vehicles, their real-time status, and the real-time status of edge nodes to generate a dynamic parameter library. The unmanned vehicle's physical information includes vehicle mass, initial battery level, current 2D position, available computing power, and communication link channel gain; the historical task information includes task data volume, maximum acceptable end-to-end latency, computational complexity, and urgency level; the unmanned vehicle's real-time status includes the latest 2D position, remaining battery level, available computing power, and communication link quality; and the edge node's real-time status includes CPU utilization, memory usage, and residual computing power. The task classification module is used to classify historical tasks based on historical task information and a dynamic parameter library to obtain task classification results, wherein the task classification results include light tasks, medium tasks, and heavy tasks. The optimal scheduling module is used to input the dynamic parameter library and task classification results into the pre-established fission hybrid mutation crayfish optimization algorithm model, with the goal of minimizing the weighted sum of the total system delay and total energy consumption, and outputting a globally optimal scheduling scheme; wherein, the globally optimal scheduling scheme includes unmanned vehicle trajectory adjustment instructions, task unloading path allocation results, and cloud-edge resource allocation ratio.

10. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, The memory stores a computer program that can run on the processor. When the processor loads and executes the computer program, it employs a dynamic scheduling optimization method for unmanned vehicle cloud-edge collaboration as described in any one of claims 1 to 8.

Images