Fresh product distribution and cold chain maintenance path planning method based on multi-task algorithm
By using Cohen's d value and balance factor to control the knowledge transfer intensity in multi-task optimization algorithm, combined with differential evolution and opposition learning variation strategy, the insufficient knowledge transfer in traditional multi-task optimization algorithm in fresh product distribution and cold chain maintenance path planning is solved. The problem of the problem achieves faster and more efficient optimal solution finding, improving the adaptability and performance of the algorithm.
Patent Information
- Application Number
- CN202410980448.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-22
- Publication Date
- 2024-11-12
AI Technical Summary
When traditional multi-task optimization algorithms deal with fresh product distribution and cold chain maintenance path planning, the knowledge transfer intensity between tasks is insufficient, resulting in a prolonged time to find the optimal solution, and the algorithm performance is not robust enough, making it difficult to adapt to various problems.
Using an improved multi-task algorithm, the population is initialized in the decision space, the knowledge transfer intensity is dynamically controlled by using Cohen's d value and equilibrium factor, combined with differential evolution and opposition learning variation strategies, the adaptive mutation and selection of populations are achieved, and the exploration of algorithms is improved. and development performance.
It effectively accelerates the optimal solution finding time of path planning, improves the performance and adaptability of the algorithm, can better deal with fresh product distribution and cold chain maintenance path planning problems, and shows higher efficiency and stability.
Smart Images

Figure CN118941198A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of multi-objective optimization research in evolutionary intelligent evolutionary computing, and specifically relates to a path planning method for fresh product distribution and cold chain maintenance based on a multi-task algorithm. Background Art
[0002] The problem of planning the distribution path of fresh products is a classic path optimization problem in operations research - the traveling salesman problem (TSP), which aims to determine the optimal access path to minimize the total cost, including time, distance, fuel, etc.; since fresh products are perishable and need to be transported under low temperature conditions, temperature control and cold chain technology in the distribution process are crucial. The maintenance path planning problem of cold chain maintenance belongs to another path optimization problem - the traveling repairman problem (TRP). Cold chain maintenance is usually urgent and requires rapid response and scheduling. When determining the access path, the goal is to minimize the waiting time of all customers. When solving these two problems, it can be found that the two problems have certain similarities, so a multi-task optimization algorithm can be used to solve this type of problem.
[0003] The research on multi-task optimization algorithm uses population optimization search and transfers genetic information between tasks to achieve the purpose of processing multiple tasks simultaneously. The shared knowledge between multiple optimization tasks is used to promote the convergence of each task. Therefore, it has a natural advantage in solving problems such as the access path of fresh product distribution and cold chain maintenance mentioned above. However, when traditional multi-task optimization algorithms solve two problems with low similarity, the transition of knowledge transfer between tasks will lead to negative transfer (that is, the transfer leads to negative effects, slowing down the time it takes for the algorithm to find the optimal solution). Summary of the invention
[0004] The purpose of the present invention is to provide a path planning method for fresh product distribution and cold chain maintenance with a multi-task algorithm, which can achieve adaptive changes in the intensity of knowledge transfer between tasks and accelerate the time to find the optimal solution for the path.
[0005] The technical solution adopted by the present invention is a path planning method for fresh product delivery and cold chain maintenance based on a multi-task algorithm, and the specific steps include:
[0006] S1: Construct the objective function with the goal of minimizing the fresh product delivery path time and the cold chain maintenance path time;
[0007] S2: Use the improved multi-task algorithm to solve the objective function and obtain the optimal path for fresh product distribution and the optimal path for cold chain maintenance.
[0008] The present invention is also characterized in that:
[0009] The objective function of the fresh product distribution path in S1 is:
[0010]
[0011] In formula (1), X i is the coordinate of the ith location, X i+1 is the coordinate of the i+1th location, X1 is the coordinate of the first location, X n is the coordinate of the nth location, T1(X i ,X i+1 ) is from location X i Coordinates to location X i+1 The coordinate time, T2(X1,X n ) is the coordinate from location X1 to location X n The time of the coordinates;
[0012] The objective function of the cold chain maintenance path is:
[0013]
[0014] In formula (2), is the time required for service at the coordinates of the jth location, is the time required for service at the coordinates of the nth location.
[0015] The specific steps of S2 are:
[0016] S201: In the improved multi-task algorithm, individuals carry two strategy solutions, which correspond to the solution of the fresh product distribution path optimization problem and the solution of the cold chain maintenance path optimization problem. Multiple individuals form a population and the population is initialized in the decision space;
[0017] S202: Set the number of population iterations, calculate the average standard deviation and average mean of each individual on minf1 and minf2 based on the statistical distribution characteristics of the population, and then calculate the balance factor through the average standard deviation and average mean to ensure the adaptability and balance of minf1 and minf2 in the multi-task optimization process. Finally, use the effect size measurement method to obtain Cohen's d value to quantify the standardized mean difference of the performance difference between minf1 and minf2;
[0018] S203: The individuals of the initial population begin to evolve iteratively. According to the knowledge transfer strength between minf1 and minf2, the population is mutated using a mixed mutation strategy to obtain a child population, and then a crossover operation is performed on the child population to obtain a subsequent generation population.
[0019] S204: Offspring selection: the offspring population is merged with the initial population, and a mixed offspring selection strategy is adopted to select an excellent offspring population;
[0020] S205: quantify the similarity between minf1 and minf2 through the effect size Cohen's d value and the balance factor, and then adaptively change the migration intensity. After the adaptive change of the migration intensity is completed, if the number of iterations is reached, enter S206, if not, enter S202;
[0021] S206: Decode the individuals in the excellent offspring population and select the optimal solution of the path among the individuals.
[0022] The average mean of individuals in S202 on minf1 and minf2 is:
[0023]
[0024] In formula (3), t is the number of iterations, c is the first correlation coefficient, is the average mean of individuals in the tth generation;
[0025] Expressed as:
[0026]
[0027] In formula (4), x i (t) is the i-th individual of the t-th generation, and N is the number of individuals;
[0028] The average standard deviation of individuals on minf1 and minf2 is:
[0029] Std(t+1)=(1-c)Std(t-1)+c(Std(t)+ε) (5)
[0030] In formula (5), Std(t) is the average standard deviation of individuals in the tth generation, and ε is the second correlation coefficient;
[0031] Std(t) is expressed as:
[0032]
[0033] The balance factor is expressed as:
[0034]
[0035] In formula (7), T is the number of objective functions, Std(i) is the average standard deviation of the i-th objective function;
[0036] Cohen's d value is expressed as:
[0037]
[0038] In formula (8), and They represent the means of the population positions corresponding to minf1 and minf2 respectively, and n represents the size of the population.
[0039] The hybrid coding strategy in S203 includes a differential evolution algorithm and an improved anti-learning mutation strategy, which is expressed as:
[0040]
[0041] In formula (9), F is the variation factor, x r1 and x r2 For from and x i (t) Individuals randomly selected from a population with the same skill factor, x′ r1 and x′ r2 For from and x i (t) Individuals randomly selected from a population with different skill factors, rand is a random number in [0, 1].
[0042] In S205, the similarity between minf1 and minf2 is quantified by the effect size Cohen's d value and the balance factor, expressed as:
[0043] RE=e -(lg(1+Se)) ×(1-bf) (10)
[0044] In formula (10), RE is the similarity between minf1 and minf2,
[0045] The specific contents of the adaptive change of migration intensity in S205 are as follows:
[0046] If all individuals in the excellent offspring population have not undergone knowledge transfer, the knowledge transfer intensity is:
[0047] rpm t =min{rmp+θ×RE×(1-ER),1} (11)
[0048] In formula (11), rmp is the initial knowledge transfer intensity, rmp t is the knowledge transfer intensity of the tth generation, θ is 0.2, and ER is the evolutionary success rate of the offspring;
[0049] ER is expressed as:
[0050]
[0051] In formula (12), αp is the number of offspring individuals that perform better than individuals in the initial population;
[0052] If there are offspring in the excellent offspring population that are produced through knowledge transfer, the knowledge transfer intensity is:
[0053]
[0054] The beneficial effects of the present invention are:
[0055] The present invention adopts a path planning method for fresh product distribution and cold chain maintenance based on a multi-task algorithm. The core of the method is to dynamically control the knowledge transfer intensity in the multi-task evolutionary algorithm by analyzing the distribution of the population in the decision space during the evolution process, and successively using a dual-strategy offspring selection mechanism and an RMP update strategy based on the effect size Cohen's d value; in the offspring selection mechanism, a balance factor is used to judge the stage of population evolution, and the exploratory nature of the algorithm is emphasized in the early stage of evolution, and the algorithm performance is improved by increasing the diversity of the population. Similarly, the development performance of the algorithm is emphasized in the late stage of evolution; in the update of the knowledge transfer intensity, the effect size Cohen's d value is used to reflect the degree of association between the two sample groups, and the balance factor is combined to dynamically quantify the similarity of the two problems; furthermore, an improved opposing learning strategy is proposed to improve the performance of the algorithm, making it more robust and adaptable to more problems; the EMT-ATS proposed by the present invention can more effectively solve the planning problems of fresh product distribution paths and cold chain maintenance paths than the existing algorithms. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1 It is a flow chart of the path planning method for fresh product delivery and cold chain maintenance of the multi-task algorithm of the present invention;
[0057] Figure 2 It is the access route map of the EMT-ATS in the present invention used in the TSPLIB-ch130 test to simulate the fresh product distribution path optimization problem;
[0058] Figure 3 It is the access route map of the comparison algorithm MFDE in the present invention used in the TSPLIB-ch130 test to simulate the fresh product distribution path optimization problem;
[0059] Figure 4 It is the access route map of the comparison algorithm MFDE-OBT in the present invention used in the TSPLIB-ch130 test to simulate the fresh product distribution path optimization problem;
[0060] Figure 5 It is the access route map of the comparison algorithm MFEA in the present invention used in the TSPLIB-ch130 test to simulate the fresh product distribution path optimization problem;
[0061] Figure 6 It is the access route map of the EMT-ATS in the present invention used in the TSPLIB-ch130 test to simulate the cold chain maintenance path optimization problem;
[0062] Figure 7It is the access route map of the comparative algorithm MFDE in the present invention used in the TSPLIB-ch130 test to simulate the cold chain maintenance path optimization problem;
[0063] Figure 8 It is the access route map of the comparative algorithm MFDE-OBT in the present invention used in the TSPLIB-ch130 test to simulate the cold chain maintenance path optimization problem;
[0064] Fig. 9 It is the access route map of the comparative algorithm MFEA in the present invention used in the TSPLIB-ch130 test to simulate the cold chain maintenance path optimization problem;
[0065] Fig.10 It is a schematic diagram of decoding the solution to the example problem in the unified decision space in the present invention. DETAILED DESCRIPTION
[0066] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.
[0067] Embodiment 1:
[0068] like Figure 1 As shown, the path planning method for fresh product delivery and cold chain maintenance based on the multi-task algorithm includes the following specific steps:
[0069] S1: Construct the objective function with the goal of minimizing the fresh product delivery path time and the cold chain maintenance path time;
[0070] S2: Use the improved multi-task algorithm to solve the objective function and obtain the optimal path for fresh product distribution and the optimal path for cold chain maintenance.
[0071] The objective function of the fresh product distribution path in S1 is:
[0072]
[0073] In formula (1), X i is the coordinate of the ith location, X i+1 is the coordinate of the i+1th location, X1 is the coordinate of the first location, X n is the coordinate of the nth location, T1(X i ,X i+1 ) is from location X i Coordinates to location X i+1 The coordinate time, T2(X1,X n ) is the coordinate from location X1 to location X n The time of the coordinates;
[0074] The objective function of the cold chain maintenance path is:
[0075]
[0076] In formula (2), is the time required for service at the coordinates of the jth location, is the time required for service at the coordinates of the nth location.
[0077] The specific steps of S2 are:
[0078] S201: Initialize the population in the decision space, where each individual in the population carries two strategy solutions, corresponding to the solution to the fresh product distribution path optimization problem and the solution to the cold chain maintenance path optimization problem;
[0079] S202: Set the number of population iterations, calculate the average standard deviation and average mean of each individual on minf1 and minf2 based on the statistical distribution characteristics of the population, and then calculate the balance factor through the average standard deviation and average mean to ensure the adaptability and balance of minf1 and minf2 in the multi-task optimization process. Finally, use the effect size measurement method to obtain Cohen's d value to quantify the standardized mean difference of the performance difference between minf1 and minf2;
[0080] S203: The individuals of the initial population begin to evolve iteratively. According to the knowledge transfer strength between minf1 and minf2, the population is mutated using a mixed mutation strategy to obtain a child population, and then a crossover operation is performed on the child population to obtain a subsequent generation population.
[0081] S204: Offspring selection: the offspring population is merged with the initial population, and a mixed offspring selection strategy is adopted to select an excellent offspring population;
[0082] S205: quantify the similarity between minf1 and minf2 through the effect size Cohen's d value and the balance factor, and then adaptively change the migration intensity. After the adaptive change of the migration intensity is completed, if the number of iterations is reached, enter S206, if not, enter S202;
[0083] S206: Decode the individuals in the excellent offspring population and select the optimal solution of the path among the individuals.
[0084] The average mean of individuals in S202 on minf1 and minf2 is:
[0085]
[0086] In formula (3), t is the number of iterations, c is the first correlation coefficient, is the average mean of individuals in the tth generation;
[0087] Expressed as:
[0088]
[0089] In formula (4), x i (t) is the i-th individual of the t-th generation, and N is the number of individuals;
[0090] The average standard deviation of individuals on minf1 and minf2 is:
[0091] Std(t+1)=(1-c)Std(t-1)+c(Std(t)+ε) (5)
[0092] In formula (5), Std(t) is the average standard deviation of individuals in the tth generation, and ε is the second correlation coefficient;
[0093] Std(t) is expressed as:
[0094]
[0095] The balance factor is expressed as:
[0096]
[0097] In formula (7), T is the number of objective functions, Std(i) is the average standard deviation of the i-th objective function;
[0098] Cohen's d value is expressed as:
[0099]
[0100] In formula (8), and They represent the means of the population positions corresponding to minf1 and minf2 respectively, and n represents the size of the population.
[0101] The hybrid coding strategy in S203 includes a differential evolution algorithm and an improved anti-learning mutation strategy, which is expressed as:
[0102]
[0103] In formula (9), F is the variation factor, x r1 and x r2 For from and x i (t) Individuals randomly selected from a population with the same skill factor, x′ r1 and x′ r2 For from and x i (t) Individuals randomly selected from a population with different skill factors, rand is a random number in [0, 1].
[0104] In S205, the similarity between minf1 and minf2 is quantified by the effect size Cohen's d value and the balance factor, expressed as:
[0105] RE=e -(lg(1+Se)) ×(1-bf) (10)
[0106] In formula (10), RE is the similarity between minf1 and minf2,
[0107] The specific contents of the adaptive change of migration intensity in S205 are as follows:
[0108] If all individuals in the excellent offspring population have not undergone knowledge transfer, the knowledge transfer intensity is:
[0109] rpm t =min{rmp+θ×RE×(1-ER),1} (11)
[0110] In formula (11), rmp is the initial knowledge transfer intensity, rmp t is the knowledge transfer intensity of the tth generation, θ is 0.2, and ER is the evolutionary success rate of the offspring;
[0111] ER is expressed as:
[0112]
[0113] In formula (12), αp is the number of offspring individuals that perform better than individuals in the initial population.
[0114] If there are offspring in the excellent offspring population that are produced through knowledge transfer, the knowledge transfer intensity is:
[0115]
[0116] Embodiment 2:
[0117] A path planning method for fresh product delivery and cold chain maintenance based on a multi-task algorithm, the specific steps include:
[0118] S1: Construct the objective function with the shortest fresh product delivery route and cold chain maintenance route as the goal;
[0119] S2: Use the improved multi-task algorithm to solve the objective function and obtain the optimal path for fresh product distribution and cold chain maintenance;
[0120] The objective function of the fresh product distribution path in S1 is:
[0121]
[0122] In formula (1), Xi is the coordinate of the ith location, X i+1 is the coordinate of the i+1th location, X1 is the coordinate of the first location, X n is the coordinate of the nth location, T1(X i ,X i+1 ) is from location X i Coordinates to location X i+1 The coordinate time, T2(X1,X n ) is the coordinate from location X1 to location X n The time of the coordinates;
[0123] The objective function of the cold chain maintenance path is:
[0124]
[0125] In formula (2), is the time required for service at the coordinates of the jth location, is the time required for service at the coordinates of the nth location.
[0126] The specific steps of S2 are:
[0127] S201: In the improved multi-task algorithm, individuals carry two strategy solutions, which correspond to the solution of the fresh product distribution path optimization problem and the solution of the cold chain maintenance path optimization problem. Multiple individuals form a population and the population is initialized in the decision space;
[0128] S202: Set the number of population iterations, calculate the average standard deviation and average mean of each individual on minf1 and minf2 based on the statistical distribution characteristics of the population, and then calculate the balance factor through the average standard deviation and average mean to ensure the adaptability and balance of minf1 and minf2 in the multi-task optimization process. Finally, use the effect size measurement method to obtain Cohen's d value to quantify the standardized mean difference of the performance difference between minf1 and minf2;
[0129] S203: The individuals of the initial population begin to evolve iteratively. According to the knowledge transfer strength between minf1 and minf2, the population is mutated using a mixed mutation strategy to obtain a child population, and then a crossover operation is performed on the child population to obtain a subsequent generation population.
[0130] S204: Offspring selection: the offspring population is merged with the initial population, and a mixed offspring selection strategy is adopted to select an excellent offspring population;
[0131] S205: quantify the similarity between minf1 and minf2 through the effect size Cohen's d value and the balance factor, and then adaptively change the migration intensity. After the adaptive change of the migration intensity is completed, if the number of iterations is reached, enter S206, if not, enter S202;
[0132] S206: Decode the individuals in the excellent offspring population and select the optimal solution of the path among the individuals.
[0133] The average mean of individuals in S202 on minf1 and minf2 is:
[0134]
[0135] In formula (3), t is the number of iterations, c is the first correlation coefficient, which is 0.5 here. is the average mean of individuals in the tth generation;
[0136] Expressed as:
[0137]
[0138] In formula (4), x i (t) is the i-th individual of the t-th generation, and N is the number of individuals;
[0139] The average standard deviation of individuals on minf1 and minf2 is:
[0140] Std(t+1)=(1-c)Std(t-1)+c(Std(t)+ε) (5)
[0141] In formula (5), Std(t) is the average standard deviation of individuals in the tth generation, and ε is the second correlation coefficient, which is 1e -6 ;
[0142] Std(t) is expressed as:
[0143]
[0144] The balance factor is expressed as:
[0145]
[0146] In formula (7), T is the number of objective functions, Std(i) is the average standard deviation of the i-th objective function;
[0147] Cohen's d value is expressed as:
[0148]
[0149] In formula (8), and They represent the means of the population positions corresponding to minf1 and minf2 respectively, and n represents the size of the population.
[0150] The hybrid coding strategy in S203 includes a differential evolution algorithm and an improved anti-learning mutation strategy, which is expressed as:
[0151]
[0152] In formula (9), F is the variation factor, x r1 and x r2 For from and x i (t) Individuals randomly selected from a population with the same skill factor, x′ r1 and x′ r2 For from and x i (t) Individuals randomly selected from a population with different skill factors, rand is a random number in [0, 1].
[0153] After all individuals have gone through the mutation stage, some individuals will undergo recombination through the binomial crossover process, thereby generating offspring with genetic diversity. Binomial crossover is a commonly used crossover method, which is expressed as:
[0154]
[0155] In formula (14), Cr is the control crossover probability, which is randomly selected from [0, 1], and j rand is a random dimension in the range [1, Dmax], v i,j For x i,j The mutation vector, x i,j is the initial population of the jth dimension
[0156] The specific content of S204 is: the initial population generates a offspring population through genetic operations, including mutation and crossover. After the offspring population is evaluated by the fitness function, it is merged with the initial population to form an intermediate population. The intermediate population is composed of the union of the initial population and the offspring population.
[0157] After obtaining the intermediate population, a specific selection mechanism is used to select N individuals from it to form the next generation population. The specific implementation strategy is:
[0158] All individuals in the intermediate population are sorted by scalar fitness, and then the first f% individuals are selected as candidates for the next generation population according to the elite selection strategy. The selected individuals are excluded from the intermediate population S, and then candidates are selected from the remaining individuals according to the roulette selection strategy, and then they are merged as the excellent offspring population.
[0159] In S205, the effect size Cohen's d value and the balance factor are used to quantify the similarity between minf1 and minf2 through the formula, expressed as:
[0160] RE=e -(lg(1+Se)) ×(1-bf) (10)
[0161] The specific contents of adaptive changes in migration intensity are as follows:
[0162] If all individuals in the excellent offspring population have not undergone knowledge transfer, the knowledge transfer intensity is:
[0163] rpm t =min{rmp+θ×RE×(1-ER),1} (11)
[0164] In formula (11), rmp is the initial knowledge transfer intensity, rmp t is the knowledge transfer intensity of the tth generation, θ is 0.2, and ER is the evolutionary success rate of the offspring;
[0165] ER is expressed as:
[0166]
[0167] In formula (12), αp is the number of offspring individuals that perform better than individuals in the initial population, and n is the population size.
[0168] If there are offspring in the excellent offspring population that are produced through knowledge transfer, the knowledge transfer intensity is:
[0169]
[0170] Through this adaptive update strategy, the algorithm is able to dynamically adjust the intensity of knowledge transfer to balance the co-evolution and independent search among different task populations, thereby improving the overall performance of multi-task optimization.
[0171] The specific content of S206 is: After the above iterative evolution process, a set of optimal solutions for minf1 and minf2 will be obtained. However, since these solutions are generated in a unified search space, the range is [0, 1], they may not directly conform to the standard form of the specific path optimization problem, so they need to be decoded.
[0172] Since the algorithm uses a unified representation method, each solution is encoded with a random key between 0 and 1. Since the solution to the fresh product distribution path and cold chain maintenance path optimization problem should be a set of non-repeated integers, the real number is replaced with the code. Fig. 9, shows a simple example of the permutation rule. For the solution x = (0.42, 0.85, 0.61, 0.53, 0.36), all dimension values of the solution x are sorted in ascending order. Then its permutation solution x′ = (2, 5, 4, 3, 1) starts from the second city, visits the fifth, fourth, third and first cities, and then returns to the second city.
[0173] Embodiment 3:
[0174] The simulation experiment of the present invention selects (EMT-ATS) and compares with MFDE, MFEA, and MFDE-OBT, and randomly selects seven groups of problems from TSPLIB.
[0175] like Figure 2-9 As shown, the ch130 test problem is extracted from TSPLIB, and EMT-ATS and the comparison algorithms MFDE, MFDE-OBT and MFEA are compared to obtain their respective access roadmaps.
[0176] Table 1 shows 7 test problems randomly selected from the TSPLIB test set, which are used to test the comparative experiments of the present invention and other three algorithms on the fresh food delivery and cold chain maintenance problems.
[0177] Table 1
[0178]
[0179]
[0180] Through experiments, we can get the access route of EMT-ATS in the ch130 problem in the fresh food delivery path optimization problem test set TSPLIB:
[0181] 92→59→90→62→115→26→122→29→10→32→1→129→108→80→103→41→65→38→86→74→77→127→100→75→24→123→84→3→91→112→126→6→88→124→104→89→81→56→44→51→71→85→83→13→28→20→60→121→98→113→4→31→50→22→97→49→68→5→39→37→18→48→61→19→105→4 3→109→70→72→58→2→33→45→114→25→106→130→128→63→94→82→119→15→93→57→30→16→67→78→27→102→125→12→55→54→23→21→76→87→47→101→9→111→17→79→99→35→14→118→40→34→36→7→66→96→46→8→52→110→107→11→117→116→69→42→73→95→53→120→64.
[0182] As shown in the above route, the fresh product delivery route simulated this time starts from city 92, passes through all cities and finally arrives at city 64, and the final total cost is the lowest.
[0183] The following is the access route for EMT-ATS in the ch130 problem in the cold chain maintenance path optimization problem test set TSPLIB:
[0184] 67→98→111→112→26→113→19→123→101→84→65→88→29→73→82→74→87→40→6→81→92→7→11→105→57→61→63→41→78→125→34→104→46→73→71→2→48→19→15→99→34→108→8→72→49→27→106→127→35→39→18→87→103→109→16→76→110→1→17→27→79→18→14→83→123→ 56→116→3→54→114→130→80→58→94→60→107→128→45→53→61→12→119→43→31→42→70→20→22→118→126→96→12→23→72→126→9→28→38→4→47→58→49→102→37→124→86→25→5→97→44→21→14→30→1→77→24→90→36→48→89→69→75→108→64→55→120→93→122→10→115.
[0185] As shown in the above route, the cold chain maintenance path simulated this time starts from city 67, passes through all cities, and finally arrives at city 115. In the end, the total waiting time for all customers is the shortest.
Claims
1. A multi-task algorithm for path planning of fresh product delivery and cold chain maintenance, characterized in that: The specific steps include; S1: Construct the objective function with the goal of minimizing the fresh product delivery path time and the cold chain maintenance path time; S2: Use the improved multi-task algorithm to solve the objective function and obtain the optimal path for fresh product distribution and the optimal path for cold chain maintenance.
2. The path planning method for fresh product delivery and cold chain maintenance based on the multi-task algorithm of claim 1 is characterized in that: The objective function of the fresh product distribution path in S1 is: In formula (1), X i is the coordinate of the ith location, X i+1 is the coordinate of the i+1th location, X1 is the coordinate of the first location, X n is the coordinate of the nth location, T1(X i ,X i+1 ) is from location X i Coordinates to location X i+1 The coordinate time, T2(X1,X n ) is the coordinate from location X1 to location X n The time of the coordinates; The objective function of the cold chain maintenance path is: In formula (2), is the time required for service at the coordinates of the jth location, is the time required for service at the coordinates of the nth location.
3. The path planning method for fresh product delivery and cold chain maintenance based on the multi-task algorithm according to claim 2 is characterized in that: The specific steps of S2 are: S201: In the improved multi-task algorithm, individuals carry two strategy solutions, which correspond to the solution of the fresh product distribution path optimization problem and the solution of the cold chain maintenance path optimization problem. Multiple individuals form a population and the population is initialized in the decision space; S202: Set the number of population iterations, calculate the average standard deviation and average mean of each individual on minf1 and minf2 based on the statistical distribution characteristics of the population, and then calculate the balance factor through the average standard deviation and average mean to ensure the adaptability and balance of minf1 and minf2 in the multi-task optimization process. Finally, use the effect size measurement method to obtain Cohen's d value to quantify the standardized mean difference of the performance difference between minf1 and minf2; S203: The individuals of the initial population begin to evolve iteratively. According to the knowledge transfer strength between minf1 and minf2, the population is mutated using a mixed mutation strategy to obtain a child population, and then a crossover operation is performed on the child population to obtain a subsequent generation population. S204: Offspring selection: the offspring population is merged with the initial population, and a mixed offspring selection strategy is adopted to select an excellent offspring population; S205: quantify the similarity between minf1 and minf2 through the effect size Cohen's d value and the balance factor, and then adaptively change the migration intensity. After the adaptive change of the migration intensity is completed, if the number of iterations is reached, enter S206, if not, enter S202; S206: Decode the individuals in the excellent offspring population and select the optimal solution of the path among the individuals.
4. The path planning method for fresh product delivery and cold chain maintenance based on the multi-task algorithm according to claim 3 is characterized in that: The average mean of individuals in S202 on minf1 and minf2 is: In formula (3), t is the number of iterations, c is the first correlation coefficient, is the average mean of individuals in the tth generation; Expressed as: In formula (4), x i (t) is the i-th individual of the t-th generation, and N is the number of individuals; The average standard deviation of individuals on minf1 and minf2 is: Std(t+1)=(1-c)Std(t-1)+c(Std(t)+ε) (5) In formula (5), Std(t) is the average standard deviation of individuals in the tth generation, and ε is the second correlation coefficient; Std(t) is expressed as: The balance factor is expressed as: In formula (7), T is the number of objective functions, Std(i) is the average standard deviation of the i-th objective function; Cohen's d value is expressed as: In formula (8), and They represent the means of the population positions corresponding to minf1 and minf2 respectively, and n represents the size of the population.
5. The path planning method for fresh product delivery and cold chain maintenance based on the multi-task algorithm of claim 4 is characterized in that: The hybrid coding strategy in S203 includes a differential evolution algorithm and an improved anti-learning mutation strategy, which is expressed as: In formula (9), F is the variation factor, x r1 and x r2 is an individual randomly selected from the population with the same skill factor as xi(t), x′ r1 and x′ r2 For from and x i (t) Individuals randomly selected from a population with different skill factors, rand is a random number in [0, 1].
6. The path planning method for fresh product delivery and cold chain maintenance based on the multi-task algorithm of claim 5 is characterized in that: In S205, the similarity between minf1 and minf2 is quantified by the effect size Cohen's d value and the balance factor, expressed as: RE=e -(lg(1+Se)) ×(1-bf) (10) In formula (10), RE is the similarity between minf1 and minf2, The specific contents of the adaptive change of migration intensity in S205 are as follows: If all individuals in the excellent offspring population have not undergone knowledge transfer, the knowledge transfer intensity is: rmp t =min{rmp+θ×RE×(1-ER),1} (11) In formula (11), rmp is the initial knowledge transfer intensity, rmp t is the knowledge transfer intensity of the tth generation, θ is 0.2, and ER is the evolutionary success rate of the offspring; ER is expressed as: In formula (12), αp is the number of offspring individuals that perform better than individuals in the initial population; If there are offspring in the excellent offspring population that are produced through knowledge transfer, the knowledge transfer intensity is:
