Mobile edge calculation unloading method based on pelican optimization strategy

Abstract

The invention provides a pelican optimization strategy-based mobile edge computing unloading method, and aims to solve the problems of low computing task unloading efficiency and poor system performance in application scenes such as Internet of Things and smart cities. By introducing chaotic mapping and a flight strategy, a traditional pelican optimization algorithm is improved, and the global search ability and the ability of jumping out of a local optimal solution of the algorithm are enhanced. A calculation unloading model of a multi-user multi-edge server is constructed, and delay and energy consumption in the data transmission and calculation process are comprehensively considered. Experimental results show that the method can effectively reduce the time delay and energy consumption of the system, optimize resource allocation and improve the overall performance of the system. Especially in a multi-user and multi-task complex scene, the method disclosed by the invention is excellent in performance, and a new solution is provided for the calculation unloading problem in the mobile edge calculation.

Description

Technical Field

[0001] The present invention relates to the technical field of mobile edge computing (MEC), and specifically to a mobile edge computing offloading method based on an improved Pelican optimization strategy, which aims to improve the performance of a mobile edge computing system and reduce latency and energy consumption by optimizing the computing offloading strategy. Background Art

[0002] With the rapid development of information technology, the demand for real-time data processing and computing power in application scenarios such as smart cities, industrial Internet of Things, and Internet of Vehicles is growing. These applications often involve a large amount of data collection, processing, and real-time analysis, which places extremely high demands on the latency and energy consumption of data processing. The traditional cloud computing model has high latency and high energy consumption due to the need to transmit data over long distances between user devices and remote cloud servers, making it difficult to meet the needs of these applications.

[0003] Mobile edge computing (MEC) is an emerging computing paradigm that effectively shortens the data transmission distance, reduces latency, and reduces network load by deploying computing resources to the edge of the network, close to user devices. However, in the MEC environment, computational offloading is a core issue. Computational offloading refers to transferring the computing tasks of user devices to edge or cloud servers with stronger computing capabilities to improve system performance and user experience. The choice of computational offloading strategy directly affects the system's latency, energy consumption, and resource utilization efficiency. Traditional offloading strategies are often based on simple rules or heuristic methods, which make it difficult to achieve optimal offloading decisions under complex and changing network environments and task requirements.

[0004] Therefore, there is an urgent need for an efficient and stable computational offloading optimization algorithm to adapt to the complex and changeable task requirements and network environment in the MEC environment. To address this problem, this paper proposes a mobile edge computing offloading method based on the improved Pelican optimization strategy, which aims to improve the global search capability and local development capability of the algorithm by introducing innovative mechanisms such as chaos mapping and flight strategy, thereby achieving a more efficient computational offloading strategy and providing strong technical support for application scenarios such as smart cities. Summary of the invention

[0005] The present invention relates to a mobile edge computing offloading method based on an improved Pelican optimization strategy, which aims to solve the problems of complex selection of computing offloading strategies and easy to fall into local optimal solutions in the mobile edge computing (MEC) environment, and provide efficient and stable computing offloading solutions for application scenarios such as smart cities, industrial Internet of Things, and Internet of Vehicles. The core of the present invention is to propose an improved Pelican optimization algorithm (IPOA) and apply it to the optimization of computing offloading strategies for mobile edge computing. The IPOA algorithm significantly improves the global search capability and local development capability of the algorithm by introducing innovative mechanisms such as chaotic mapping and flight strategies, thereby achieving better computing offloading decisions under complex and changeable network environments and task requirements.

[0006] The mobile edge computing offloading method based on the improved Pelican optimization strategy of the present invention mainly includes the following key steps:

[0007] Step 1: Initialize the monitoring device, including the calculation amount and task data volume parameters of the calculation task;

[0008] Step 2: Initialize the computing power, channel bandwidth, and transmission power parameters of the edge server;

[0009] Step 3: Set the relevant parameters of the algorithm, population size, maximum number of iterations, chaos mapping parameters, and flight strategy parameters;

[0010] Step 4: Construct a computation offloading model for multiple users and multiple edge servers, model the monitoring devices and edge servers as nodes, and model the task offloading process as the data transmission and computation process between nodes;

[0011] Step 5: Define the latency model and energy consumption model, and calculate the latency and energy consumption in the local execution and edge offloading cases respectively;

[0012] Step 6: Introduce weight factors, construct a comprehensive optimization objective function, and map latency and energy consumption to the same interval for collaborative optimization;

[0013] Step 7: Generate the initial population using chaotic mapping to increase the diversity of the population and avoid premature convergence of the algorithm;

[0014] Step 8: Calculate the fitness value of the initial population and record the best fitness value and the corresponding individual position;

[0015] Step 9: Randomly generate the location of prey to simulate the behavior of a pelican flying at high altitude to determine the location of prey;

[0016] Step 10: Update the position of the pelican according to the chaotic map and flight strategy, simulating the process of the pelican moving towards the prey;

[0017] Step 11: In each iteration, compare the fitness of the updated pelican position with the initial position, and use the principle of survival of the fittest to retain the position with better fitness;

[0018] Step 12: After the exploration phase, enter the exploitation phase, simulating the behavior of a pelican spreading its wings on the water to propel itself upward and gather prey into an area that is easy to capture;

[0019] Step 13: Introduce the hyperbolic cotangent random perturbation to balance the exploration and utilization phases of the algorithm and prevent the algorithm from converging to the local optimal solution too early;

[0020] Step 14: Update the position of the pelican and record the best candidate solution in each iteration;

[0021] Step 16: Repeat the exploration phase and the exploitation phase until the maximum number of iterations is reached;

[0022] Step 17: In each iteration, the weight factor is dynamically adjusted according to the updated population position and fitness value to balance the optimization goals of latency and energy consumption;

[0023] Step 18: After reaching the maximum number of iterations, output the best candidate solution in the current iteration, that is, the optimal computing task offloading strategy.

[0024] Furthermore, unlike the traditional random initialization method, the present invention uses chaotic mapping to generate the initial population, increase the diversity of the population, and avoid premature convergence of the algorithm. The introduction of chaotic mapping makes the distribution of the initial population in the search space more uniform, which helps to improve the global search ability of the algorithm. In the population initialization stage, the Pelican Optimization Algorithm (POA) is a population-based algorithm, in which the pelican is a member of the population. According to the lower and upper bounds of the problem, the mathematical description of the initialization of the pelican population is as follows:

[0025] x i,j = l j +rand·(u j -l j ),i=1,2,…,n,j=1,2,…,m#(1)

[0026] where x i,j represents the position of the i-th pelican in the j-th dimension; n represents the population size; m represents the dimension of the problem to be solved; rand represents a random number in the range [0,1]; u j and l j Respectively represent the upper and lower bounds of the j-th dimension;

[0027] In the Pelican Optimization Algorithm (POA), the population matrix is ​​used to represent all members. Each row of the matrix corresponds to a candidate solution, representing an individual in the population, and each column represents a candidate value of the problem variable, which is described as follows:

[0028]

[0029] Where X represents the population matrix; X i represents the i-th pelican;

[0030] In the Pelican Optimization Algorithm (POA), each population member is abstracted as a pelican, representing a candidate solution to a given problem. Each candidate solution is evaluated by the objective function, and the corresponding objective function value can be calculated to measure the quality of the solution. The objective function value of the entire population can be represented by an objective function vector, in which each element corresponds to the objective function value of a pelican, as described below:

[0031]

[0032] Among them, F represents the objective function vector of the population, F i represents the objective function value of the i-th pelican;

[0033] In the exploration phase, the pelican simulates natural hunting behavior by determining the location of the prey and moving toward the target location. The core goal is to avoid falling into the local optimum and ensure comprehensive coverage of the entire search space. The strategy of the pelican moving to the prey location can be expressed as the following mathematical model:

[0034]

[0035] in represents the new state of the i-th pelican in the j-th dimension; I represents a random number equal to 1 or 2; rand represents a random number in the interval [0,1]; p j represents the position of the prey in the jth dimension, F p As its objective function value, parameter I will randomly select a value at each iteration. When the parameter value is 2, it will bring a larger displacement to the members, thereby expanding the search range;

[0036] In the Pelican Optimization Algorithm (POA), if the objective function value of the current position is improved, the new position is accepted. This update is called an effective update to avoid the algorithm entering the non-optimal area. The formula for this process is modeled as follows:

[0037]

[0038] in represents the new position of the i-th pelican; represents the objective function value obtained by the i-th pelican in the new position based on the exploration phase;

[0039] The present invention adopts mapping to perform chaos optimization and increases the population diversity of POA by generating chaotic sequences. The mathematical expression of mapping is as follows:

[0040]

[0041] The mathematical expression after Bernoulli shift transformation is:

[0042] x i+1 =(2x i )mod1#(7)

[0043] After the introduction of chaotic mapping, corresponding optimization changes are made to formula (1) in population initialization and formula (4) in the exploration phase. The random numbers originally in the range [0,1] are replaced by chaotic mapping, and the latest optimization formula is obtained:

[0044] x i,j = l j +tent·(u j -l j ),i=1,2,…,n,j=1,2,…,m#(8)

[0045]

[0046] tent=tent t =(2x t )mod1 x∈[0,1]#(10)

[0047] Where t represents the current number of iterations; t Represents the Tent mixed team sequence generated in the tth iteration.

[0048] Furthermore, when updating the position of the pelican, the present invention not only takes into account the diversity brought by chaotic mapping, but also introduces a flight strategy. By simulating the long-distance flight behavior of the pelican, the search range and step size of the algorithm in the exploration phase are increased, thereby further improving the global search capability of the algorithm.

[0049] Furthermore, in each iteration, the fitness of the updated pelican position is compared with the initial position, and the principle of survival of the fittest is adopted to retain the position with better fitness. First, the population position and speed are initialized in combination with mapping, and the wandering process of random step length and direction is simulated. In the exploration stage, iteration is performed according to the position update formula. The fitness of the updated pelican position is compared with the initial position in each iteration, and the principle of survival of the fittest is adopted to ensure that the position with better fitness is retained. The update formula for introducing flight is as follows:

[0050]

[0051] in represents the new position of the i-th pelican after the exploration phase ends; ω represents a scaling factor used to adjust the influence of the Levy flight step length, with a value of 0.8; Levy(γ) represents the step length of the Levy flight, γ represents the exponential parameter of the Levy flight step length, with a value of 1.5; u and v both represent random numbers that conform to the normal distribution, and the specific values ​​are obtained by formula (13); Γ represents the gamma function, which is used to calculate the distribution parameters of the random step length.

[0052] Furthermore, in the utilization phase, the present invention introduces random disturbances generated by the hyperbolic cotangent function to balance the exploration and utilization phases of the algorithm. The balance, symmetry and rapid decay characteristics of the hyperbolic cotangent function enable the algorithm to maintain a certain degree of exploration during the search process and avoid premature convergence to the local optimal solution. The goal of this phase is to make full use of the existing information and deeply optimize the solution. From a mathematical perspective, the algorithm must check points near the pelican's position in order to converge to a better solution. Therefore, the behavior of the pelican in the hunting process is modeled, which can be described as:

[0053]

[0054] in represents the new state of the i-th pelican in the j-th dimension based on the utilization phase; R represents a scaling factor of 0.2; rand represents a random number in the range [0,1]; R·(1-t / T) represents x i,j The domain radius, t represents the iteration counter, and T represents the maximum number of iterations;

[0055] The parameter represents the radius of the local search of the population members in their surrounding areas in order to obtain a better solution. In the initial stage, the parameter value is large, so the search range of each member is wide. As the number of iterations of the algorithm increases, the parameter will gradually decrease, shrinking the neighborhood search range of the members. In this way, the area where each member is located can be scanned more accurately with a smaller step size, so that POA can converge to a solution closer to (or even equal to) the global optimal solution.

[0056] At this stage, the effective update strategy is also used to improve the pelican position, using the formula described as:

[0057]

[0058] in represents the new position of the i-th pelican; represents the objective function value obtained by the i-th pelican in the new position based on the utilization phase;

[0059] Introducing the hyperbolic cotangent random perturbation into formula (14) in the POA utilization phase results in a new optimization formula:

[0060]

[0061] Where F represents the hyperbolic cotangent random perturbation; csch represents the hyperbolic cotangent function, and the specific mapping relationship is obtained by formula (18); ε represents the scaling factor for adjusting the parameter range, and its value is 0.8; Represents a random number in the range [0,1].

[0062] The advantages and positive effects of the present invention are:

[0063] The present invention mainly designs a mobile edge computing offloading method based on the improved Pelican optimization strategy. In this method, the problem that the traditional centralized computing mode cannot meet the actual application needs is mainly studied. In order to improve the performance of the algorithm, the present invention improves the Pelican optimization algorithm, introduces chaotic mapping and flight mechanism, thereby increasing the diversity of the initial population, preventing the algorithm from premature convergence, and enhancing the algorithm's ability to jump out of the local optimal solution. After combining chaotic mapping and flight mechanism, the improved algorithm achieves a good balance between global search and local optimization, significantly improving search efficiency and optimization accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a real-time video analysis system model;

[0065] Figure 2 It is a graph showing the relationship between latency and number of iterations;

[0066] Figure 3 It is a graph showing the relationship between energy consumption and the number of iterations;

[0067] Figure 4 It is a graph showing the relationship between fitness and the number of iterations;

[0068] Figure 5 It is a graph showing the relationship between latency and the number of user devices;

[0069] Figure 6 It is a graph showing the relationship between energy consumption and the number of user devices;

[0070] Figure 7 It is a graph showing the relationship between fitness and the number of user devices;

[0071] Figure 8 This is a graph showing the relationship between latency and the number of edge servers;

[0072] Fig. 9 It is a graph showing the relationship between energy consumption and the number of edge servers;

[0073] Fig.10 It is a graph showing the relationship between fitness and the number of edge servers;

[0074] Fig.11 It is a flow chart of the mobile edge computing offloading method based on the improved Pelican optimization strategy of the present invention. DETAILED DESCRIPTION

[0075] See attached Fig.11 , this embodiment is based on the mobile edge computing offloading method of improving the Pelican optimization strategy, and mainly includes the following key steps:

[0076] This offloading model focuses on the collaborative optimization of latency and energy consumption. The latency calculation specifically includes the data transmission latency and the task computing latency. After the monitoring device collects the video data, it can choose to compute locally or transmit the task data to the edge server for computing via wireless transmission. When choosing to compute locally, there is no data transmission latency, and only the local computing latency T needs to be considered. loc When computing on the edge server, the data transmission delay T tran Calculate the delay T with the edge server serv The energy consumption calculation specifically includes data transmission energy consumption and task computing energy consumption, which is also divided into two cases: local offloading and edge offloading. When choosing local computing, there is no data transmission energy consumption, and only the local computing energy consumption E needs to be considered. loc When computing on the edge server, since the computing energy consumption of the edge server is generally borne by an external third party, only the data transmission energy consumption E needs to be considered. tran The model architecture is shown in the attached Figure 1 .

[0077] 1. System initialization:

[0078] Step 1: Initialize the monitoring device, including the calculation amount of the computing task, the amount of task data and other parameters, and initialize the edge server's computing power, channel bandwidth, transmission power and other parameters. There are N monitoring devices, each of which corresponds to a series of computing tasks, then these monitoring devices can be described as Tasks = {Task 1 ,Task 2 ,...,Task N In addition, suppose there are M edge servers in total, that is, S = {S 1 ,S 2 ,...,S M}, since this model uses a complete offloading strategy, for all monitoring devices, their computing tasks can be chosen to be executed locally or offloaded to any edge server for processing. There are a total of M+1 offloading locations, that is, P = {P0 ,P 1 ,P 2 ,...,P M}, where P 0 Indicates local execution. The offloading location of all computing tasks can be described as X = {X 1 ,X 2 ,...,X M |X i ∈P,1≤i≤M}, see Table 1 for specific parameter settings.

[0079] Table 1 Experimental equipment and unloading environment parameter settings

[0080]

[0081]

[0082] Step 2: Set the relevant parameters of the algorithm.

[0083] Table 2 Algorithm parameter settings

[0084]

[0085] 2. Uninstall model construction:

[0086] Step 1: Construct a computation offloading model for multi-user and multi-edge servers, model the monitoring devices and edge servers as nodes, and model the task offloading process as the data transmission and computation process between nodes.

[0087] Step 2: Define the latency model and energy consumption model, and calculate the latency and energy consumption in the two cases of local execution and edge offloading. The latency calculation can be divided into two cases for analysis: local execution and edge offloading. When the task is chosen to be calculated locally, no data transmission is involved, so only the local calculation delay needs to be considered:

[0088]

[0089] Where D i Represents the computing requirements of the ith monitoring device, i.e., Task i Data size (bit); K i Indicates the computing power of the i-th device (rpm / bit); f i Indicates the CPU frequency (Hz) of the i-th device.

[0090] When the task is chosen to be calculated on the edge server, the return delay when the user offloads to the edge server can be ignored due to the small scale of the calculation result. Therefore, only the data transmission delay and the edge server calculation delay need to be considered. The first is the transmission delay:

[0091]

[0092] Where T ij represents the transmission delay of the i-th task to transmit the computing task data to the j-th edge server through wireless transmission; v ij represents the corresponding transmission rate, which is obtained by formula (2).

[0093] The second is the edge server computing latency:

[0094]

[0095] Where D i represents the computing requirement of the ith monitoring device (bit); K ij represents the computing power (rpm / bit) provided by the jth edge server to the ith device; f j represents the CPU frequency (Hz) of the jth edge server.

[0096] Therefore, when the task is selected to be unloaded at the edge, the total delay T mec Expressed as transmission delay T tran Compared with edge computing latency T serv The sum of:

[0097] T mec =T tran +T serv #(4)

[0098] The calculation of energy consumption can also be divided into two cases: local offloading and edge offloading. When the task is chosen to be calculated locally, there is no data transmission problem, so only the local computing energy consumption needs to be considered:

[0099]

[0100] Where P loc Indicates the power of the monitoring device; T loc represents the computing delay of the monitoring device, which is obtained by formula (3.3); α represents the energy consumption coefficient related to chip design, which is a constant; f i represents the CPU cycle frequency (Hz) of the ith monitoring device; V i The operating voltage of the i-th monitoring device.

[0101] When the task is chosen to offload computing on the edge server, the computing energy consumption of the edge server is usually borne by the service provider such as the operator. Since the user has paid for the server equipment and other related usage fees, there is usually no need to consider this part of the energy consumption, and only the energy consumption during the data transmission process needs to be paid attention to:

[0102]

[0103] Where T tran represents the data transmission delay, which is obtained by Formula 6.

[0104] Step 3: Introduce weight factors, construct a comprehensive optimization objective function, and map latency and energy consumption to the same interval for collaborative optimization. This model considers the comprehensive optimization of latency and energy consumption, and unloads all tasks to reasonable locations to minimize latency and energy consumption. However, since latency and energy consumption units are different and not in the same order of magnitude, they cannot be directly added and compared. Therefore, this model adopts an interval mapping method and uses the mapping function g(x) to map the two variables of latency and energy consumption to the same interval for collaborative optimization:

[0105]

[0106] In addition, in order to flexibly adjust the importance of latency and energy consumption, this model introduces a weight factor β to dynamically configure the attention ratio of latency and energy consumption (β∈[0,1]), and obtains the final fitness function, that is, the optimization objective function of this model:

[0107] fitness i =βg(T i )+(1-β)g(E i )#(8)

[0108] Among them fitness i represents the fitness of the i-th task; T i Represents its total delay; E i Represents its total energy consumption.

[0109] Therefore, the final offloading model can be described as:

[0110]

[0111] Among them, Fitness represents the average fitness of all offloaded tasks. The smaller the fitness value, the less system resources are consumed. In this way, the original problem of finding the lowest latency and energy consumption is transformed into the problem of finding the minimum average fitness.

[0112] 3. Population initialization:

[0113] Chaotic mapping is used to generate the initial population, increase the diversity of the population, avoid premature convergence of the algorithm, calculate the fitness value of the initial population, and record the best fitness value and the corresponding individual position. The present invention uses Tent mapping for chaos optimization to increase the population diversity of POA by generating chaotic sequences. The mathematical expression of Tent mapping is as follows:

[0114]

[0115] The mathematical expression after Bernoulli shift transformation is:

[0116] x i+1 =(2x i )mod1#(11)

[0117] After the introduction of Tent chaotic mapping, the formula in population initialization is optimized accordingly to obtain the latest optimization formula:

[0118] x i,j = l j +tent·(u j -l j ),i=1,2,…,n,j=1,2,…,m#(12)

[0119] tent=tent t =(2x t )mod1 x∈[0,1]#(13)

[0120] Where t represents the current number of iterations; t represents the Tent mixed team sequence generated in the tth iteration, where x i,j represents the position of the i-th pelican in the j-th dimension; n represents the population size; m represents the dimension of the problem to be solved; u j and l j denote the upper and lower bounds of the j-th dimension respectively.

[0121] In the Pelican Optimization Algorithm (POA), the population matrix is ​​used to represent all members. Each row of the matrix corresponds to a candidate solution, representing an individual in the population, and each column represents a candidate value for the problem variable. The specific description is as follows:

[0122]

[0123] Where X represents the population matrix; X i represents the i-th pelican.

[0124] In the Pelican Optimization Algorithm (POA), each population member is abstracted as a pelican, representing a candidate solution to a given problem. Each candidate solution is evaluated by the objective function, and the corresponding objective function value can be calculated to measure the quality of the solution. The objective function value of the entire population can be represented by an objective function vector, in which each element corresponds to the objective function value of a pelican, as described below:

[0125]

[0126] Among them, F represents the objective function vector of the population, F irepresents the objective function value of the i-th pelican.

[0127] 4. Exploration Phase:

[0128] The position of prey is randomly generated, simulating the behavior of pelicans in determining the position of prey while flying at high altitude. The position of the pelican is updated according to the chaotic map and flight strategy, simulating the process of the pelican moving towards the prey position. In each iteration, the fitness of the updated pelican position is compared with the initial position, and the principle of survival of the fittest is adopted to retain the position with better fitness. After the introduction of the Tent chaotic map, the formula in the exploration phase is optimized accordingly to obtain the latest optimization formula:

[0129]

[0130] tent=tent t =(2x t )mod1 x∈[0,1]#(17)

[0131] Where t represents the current number of iterations; t represents the Tent mixed team sequence generated in the tth iteration, where represents the new state of the i-th pelican in the j-th dimension; I represents a random number equal to 1 or 2; rand represents a random number in the interval [0,1]; p j represents the position of the prey in the jth dimension, F p is the objective function value. Parameter I will be randomly selected at each iteration. When the parameter value is 2, it will bring a larger displacement to the members, thereby expanding the search range.

[0132] In the Pelican Optimization Algorithm (POA), if the objective function value of the current position is improved, the new position is accepted. This update is called an effective update to avoid the algorithm entering a non-optimal area. The formula for this process is modeled as follows:

[0133]

[0134] in represents the new position of the i-th pelican; It represents the objective function value obtained by the i-th pelican in the new position based on the exploration phase.

[0135] 5. Utilization stage:

[0136] Step 1: Enter the utilization phase, simulating the behavior of a pelican spreading its wings on the water surface to propel itself upward and gather prey to an area that is easy to capture. The behavior of the pelican in the hunting process is modeled, which can be described as:

[0137]

[0138] in represents the new state of the i-th pelican in the j-th dimension based on the utilization phase; R represents a scaling factor of 0.2; rand represents a random number in the range [0,1]; R·(1-t / T) represents x i,j The radius of the domain, t represents the iteration counter, and T represents the maximum number of iterations.

[0139] The parameter R·(1-t / T) represents the radius of the local search of the population members in their surrounding areas in order to obtain a better solution. In the initial stage, the parameter value is large, so the search range of each member is wide. As the number of iterations of the algorithm increases, the parameter R·(1-t / T) will gradually decrease, causing the neighborhood search range of the members to shrink. This allows the area where each member is located to be scanned more accurately with a smaller step size, allowing the POA to converge to a solution that is closer to (or even equal to) the global optimal solution.

[0140] At this stage, the effective update strategy is also used to improve the pelican position. The formula used is described as:

[0141]

[0142] in represents the new position of the i-th pelican; It represents the objective function value obtained by the i-th pelican in the new position based on the utilization phase.

[0143] Step 2: Introduce the hyperbolic cotangent random perturbation to balance the exploration and utilization phases of the algorithm, avoid the algorithm from converging to the local optimal solution too early, update the position of the pelican, and record the best candidate solution in each iteration. Introduce the hyperbolic cotangent random perturbation in formula (19) of the POA utilization phase to obtain a new optimization formula:

[0144]

[0145]

[0146] Where F represents the hyperbolic cotangent random perturbation; csch represents the hyperbolic cotangent function, and the specific mapping relationship is obtained by formula (23); ε represents the scaling factor for adjusting the parameter range, and its value is 0.8; Represents a random number in the range [0,1], which is different from rand in formula (21). The two have independent values ​​and do not affect each other.

[0147] 6. Iterative Optimization:

[0148] The exploration phase and the utilization phase are repeated until the maximum number of iterations is reached. In each iteration, the weight factor is dynamically adjusted according to the updated population position and fitness value to balance the optimization goals of latency and energy consumption.

[0149] The present invention proposes a new computational offloading algorithm. To verify the performance of the algorithm, a simulation experiment and a real-scene experiment are proposed. The parameters required for the experiment are shown in Tables 1 and 2.

[0150] First, simulation experiments are tested and analyzed to explore a multi-user multi-edge server unloading solution based on the improved Pelican optimization strategy. In this experimental scenario, all tasks adopt a complete unloading strategy to unload data to the optimal location to optimize the system's latency and energy consumption. This paper uses smart city real-time video analysis as an application scenario, and uses MATLAB and Python to conduct four groups of simulation experiments, respectively. The sparrow optimization algorithm (SSA), whale optimization algorithm (WOA), gray wolf optimization algorithm (GWO), quantum particle swarm algorithm (QPSO), pelican optimization algorithm (POA) and the improved pelican optimization algorithm (IPOA) proposed by this invention are compared.

[0151] 1) Experiment 1: This experiment compares the changes of each algorithm in terms of latency, energy consumption and fitness as the number of iterations increases. The number of monitoring devices is set to 50, the number of edge servers is set to 10, and the task data volume is 3MB. Figure 2 The delay varies with the number of iterations. Figure 3 The energy consumption varies with the number of iterations. Figure 4 This is a graph showing the relationship between fitness and the number of iterations.

[0152] The results show that after the introduction of Tent chaotic mapping and Levy flight strategy, IPOA performs well in delay optimization and can quickly reduce the delay in early iterations. In contrast, although POA performs well in early iterations, its optimization ability is gradually surpassed by IPOA in the later stages; QPSO, due to its limitations in global search, does not work well in delay optimization and fails to achieve significant improvements. IPOA also performs well in energy consumption optimization. IPOA can achieve lower energy consumption levels in fewer iterations and ultimately outperforms other algorithms. SSA and POA perform relatively close in this indicator, but neither can achieve the optimization effect of IPOA. In terms of fitness optimization, IPOA maintains a stable optimization trend throughout the iteration process, and the fitness value that converges in the end is better than other algorithms. Although SSA and POA also show certain capabilities in fitness optimization, the gap with IPOA gradually widens as the number of iterations increases. QPSO performs poorly in fitness optimization due to its tendency to fall into local optimality.

[0153] 2) Experiment 2: This experiment compares the changes of each algorithm in terms of latency, energy consumption and fitness during the increase of iteration number. The number of edge servers is set to 10, the task data volume is 3MB, and the number of user devices increases proportionally from 30. Figure 5 The relationship between latency and the number of user devices is shown in the figure. Figure 6 The energy consumption varies with the number of user devices. Figure 7 This is a graph showing the relationship between fitness and the number of user devices.

[0154] The experimental results show that in terms of latency, IPOA and SSA perform well. These two algorithms can maintain low and stable latency when facing an increase in the number of user devices. The latency of QPSO increases significantly with the increase in the number of devices. In terms of energy consumption, IPOA and SSA also perform well, especially when the number of devices increases, the increase in energy consumption is small. The energy consumption of QPSO and WOA increases significantly with the increase in the number of devices. In terms of fitness, IPOA always maintains the optimal value, and the fitness fluctuates less when the number of user devices increases, showing its strong adaptability in a multi-user environment. Although GWO and WOA perform well in some cases, their fitness begins to decline when facing a higher number of devices.

[0155] 3) Experiment 3: This experiment compares the changes of each algorithm in the process of increasing the number of iterations from the three aspects of latency, energy consumption and fitness. The number of user devices is set to 50, the task data volume is 3MB, and the edge servers are increased proportionally from 10. Figure 8 The relationship between latency and the number of edge servers is shown in the figure. Fig. 9 The energy consumption varies with the number of edge servers. Fig.10 This is a graph showing how fitness changes with the number of edge servers.

[0156] In Experiment 3, as the number of edge servers increases, each algorithm shows different trends in latency, energy consumption, and fitness, but the average latency, average energy consumption, and average fitness of each algorithm do not decrease linearly. From the analysis of the offloading model and experimental parameters, different edge server CPU clock frequencies will affect the results of this experiment.

[0157] In order to simulate the offload calculation in complex scenarios, this experiment randomly sets the clock frequency of the edge server CPU between 4 and 8 GHz. If two sets of data do not conform to the linear law, the main reason is that the average values ​​of the randomly set CPU clock frequencies of the edge servers in the two sets of experiments are quite different. From the above experiments, it can be seen that the QPSO algorithm performs the worst in terms of latency and energy consumption. As the number of servers increases, the performance decreases significantly, indicating that its adaptability to resource expansion is weak. The performance of the GWO and WOA algorithms in terms of latency and energy consumption is relatively stable, but they fail to achieve the optimal fitness. The SSA algorithm performs well in latency and energy consumption, but is slightly inferior to IPOA in terms of fitness. In contrast, the IPOA algorithm performs well in latency and energy consumption, and can effectively reduce latency and maintain a low energy consumption level.

Claims

1. A mobile edge computing offloading method based on Pelican optimization strategy, characterized in that The method mainly includes the following steps: Step 1: Initialize the monitoring device, including the calculation amount and task data volume parameters of the calculation task; Step 2: Initialize the computing power, channel bandwidth, and transmission power parameters of the edge server; Step 3: Set the relevant parameters of the algorithm, population size, maximum number of iterations, chaos mapping parameters, and flight strategy parameters; Step 4: Construct a computation offloading model for multiple users and multiple edge servers, model the monitoring devices and edge servers as nodes, and model the task offloading process as the data transmission and computation process between nodes; Step 5: Define the latency model and energy consumption model, and calculate the latency and energy consumption in the local execution and edge offloading cases respectively; Step 6: Introduce weight factors, construct a comprehensive optimization objective function, and map latency and energy consumption to the same interval for collaborative optimization; Step 7: Generate the initial population using chaotic mapping to increase the diversity of the population and avoid premature convergence of the algorithm; Step 8: Calculate the fitness value of the initial population and record the best fitness value and the corresponding individual position; Step 9: Randomly generate the location of prey to simulate the behavior of a pelican flying at high altitude to determine the location of prey; Step 10: Update the position of the pelican according to the chaotic map and flight strategy, simulating the process of the pelican moving towards the prey; Step 11: In each iteration, compare the fitness of the updated pelican position with the initial position, and use the principle of survival of the fittest to retain the position with better fitness; Step 12: After the exploration phase, enter the exploitation phase, simulating the behavior of a pelican spreading its wings on the water to propel itself upward and gather prey into an area that is easy to capture; Step 13: Introduce the hyperbolic cotangent random perturbation to balance the exploration and utilization phases of the algorithm and prevent the algorithm from converging to the local optimal solution too early; Step 14: Update the position of the pelican and record the best candidate solution in each iteration; Step 16: Repeat the exploration phase and the exploitation phase until the maximum number of iterations is reached; Step 17: In each iteration, the weight factor is dynamically adjusted according to the updated population position and fitness value to balance the optimization goals of latency and energy consumption; Step 18: After reaching the maximum number of iterations, output the best candidate solution in the current iteration, that is, the optimal computing task offloading strategy.

2. A mobile edge computing offloading method based on the Pelican optimization strategy as claimed in claim 1, characterized in that: Chaotic mapping is used to generate the initial population, increase the diversity of the population, and avoid premature convergence of the algorithm. The introduction of chaotic mapping makes the distribution of the initial population in the search space more uniform, which helps to improve the global search ability of the algorithm. In the population initialization stage, the Pelican Optimization Algorithm (POA) is a population-based algorithm, in which pelicans are members of the population. According to the lower and upper bounds of the problem, the mathematical description of the initialization of the pelican population is as follows: x i,j =l j +rand·(u j -l j ),i=1,2,…,n,j=1,2,…,m#(1) where x i,j represents the position of the i-th pelican in the j-th dimension; n represents the population size; m represents the dimension of the problem to be solved; rand represents a random number in the range [0,1]; u j and l j Respectively represent the upper and lower bounds of the j-th dimension; In the Pelican Optimization Algorithm (POA), the population matrix is ​​used to represent all members. Each row of the matrix corresponds to a candidate solution, representing an individual in the population, and each column represents a candidate value for the problem variable. The specific description is as follows: Where X represents the population matrix; X i represents the i-th pelican; In the Pelican Optimization Algorithm (POA), each population member is abstracted as a pelican, representing a candidate solution to a given problem. Each candidate solution is evaluated through the objective function, and the corresponding objective function value is calculated to measure the quality of the solution. The objective function value of the entire population can be represented by an objective function vector, in which each element corresponds to the objective function value of a pelican, as described below: Among them, F represents the objective function vector of the population, F i represents the objective function value of the i-th pelican; In the exploration phase, the pelican simulates natural hunting behavior by determining the location of the prey and moving toward the target location. The core goal is to avoid falling into the local optimum and ensure comprehensive coverage of the entire search space. The strategy of the pelican moving to the prey location can be expressed as the following mathematical model: in represents the new state of the i-th pelican in the j-th dimension; I represents a random number equal to 1 or 2; rand represents a random number in the interval [0,1]; p j represents the position of the prey in the jth dimension, F p As its objective function value, parameter I will randomly select a value at each iteration. When the parameter value is 2, it will bring a larger displacement to the members, thereby expanding the search range. Therefore, parameter I will affect the exploration ability of POA; In the Pelican Optimization Algorithm (POA), if the objective function value of the current position is improved, the new position is accepted. This update is called an effective update to avoid the algorithm entering the non-optimal area. The formula for this process is modeled as follows: in represents the new position of the i-th pelican; represents the objective function value obtained by the i-th pelican in the new position based on the exploration phase; Mapping is used for chaos optimization to increase the population diversity of POA by generating chaotic sequences. The mathematical expression of mapping is as follows: The mathematical expression after Bernoulli shift transformation is: x i+1 =(2x i )mod1#(7) After the introduction of chaotic mapping, corresponding optimization changes are made to formula (1) in population initialization and formula (4) in the exploration phase. The random numbers originally in the range [0,1] are replaced by chaotic mapping, and the latest optimization formula is obtained: x i,j =l j +tent·(u j -l j ),i=1,2,…,n,j=1,2,…,m#(8) tent tent t =(2x t )mod1 x∈[0,1]#(10) Where t represents the current number of iterations; t Represents the Tent mixed team sequence generated in the tth iteration.

3. A mobile edge computing offloading method based on the Pelican optimization strategy as claimed in claim 1, characterized in that: When updating the pelican's position, not only the diversity brought by chaotic mapping is taken into account, but also a flight strategy is introduced. By simulating the long-distance flight behavior of the pelican, the algorithm's search range and step size in the exploration phase are increased, further improving the algorithm's global search capability.

4. A mobile edge computing offloading method based on the Pelican optimization strategy as claimed in claim 1, characterized in that: In each iteration, the fitness of the updated pelican position is compared with the initial position, and the principle of survival of the fittest is adopted to retain the position with better fitness. First, the population position and speed are initialized in combination with mapping, and the wandering process of random step length and direction is simulated. Then, in the exploration phase, iteration is performed according to the position update formula. In each iteration, the fitness of the updated pelican position is compared with the initial position, and the principle of survival of the fittest is adopted to ensure that the position with better fitness is retained. The update formula for introducing flight is as follows: in represents the new position of the i-th pelican after the exploration phase ends; ω represents a scaling factor used to adjust the influence of the Levy flight step length, with a value of 0.8; Levy(γ) represents the step length of the Levy flight, γ represents the exponential parameter of the Levy flight step length, with a value of 1.5; u and v both represent random numbers that conform to the normal distribution, and the specific values ​​are obtained by formula (13); Γ represents the gamma function, which is used to calculate the distribution parameters of the random step length.

5. A mobile edge computing offloading method based on the Pelican optimization strategy as claimed in claim 1, characterized in that: In the exploitation phase, random disturbances generated by the hyperbolic cotangent function are introduced to balance the exploration and exploitation phases of the algorithm. The balance, symmetry and rapid decay characteristics of the hyperbolic cotangent function enable the algorithm to maintain a certain degree of exploration during the search process and avoid converging to the local optimal solution too early. The goal of this phase is to make full use of the existing information and deeply optimize the solution. From a mathematical perspective, the algorithm must check points near the pelican's position in order to converge to a better solution. Therefore, the behavior of the pelican in the hunting process is modeled as follows: in represents the new state of the i-th pelican in the j-th dimension based on the utilization phase; R represents a scaling factor of 0.2; rand represents a random number in the range [0,1]; R·(1-t / T) represents x i,j The domain radius, t represents the iteration counter, and T represents the maximum number of iterations; The parameter represents the radius of the local search of the population members in their surrounding areas in order to obtain a better solution. In the initial stage, the parameter value is large, so the search range of each member is wide. As the number of iterations of the algorithm increases, the parameter will gradually decrease, shrinking the neighborhood search range of the members. In this way, the area where each member is located can be scanned more accurately with a smaller step size, so that POA can converge to a solution that is closer to or even equal to the global optimal solution. At this stage, the effective update strategy is also used to improve the pelican position, using the formula described as: in represents the new position of the i-th pelican; represents the objective function value obtained by the i-th pelican in the new position based on the utilization phase; Introducing the hyperbolic cotangent random perturbation into formula (14) in the POA utilization phase results in a new optimization formula: Where F represents the hyperbolic cotangent random perturbation; csch represents the hyperbolic cotangent function, and the specific mapping relationship is obtained by formula (18); ε represents the scaling factor for adjusting the parameter range, and its value is 0.8; Represents a random number in the range [0,1], which is different from rand in formula (16). The two have independent values ​​and do not affect each other.

Images