Disassembly planning and recovery decision optimization method considering learning effect of workers

Abstract

The invention provides a disassembly planning and recovery decision optimization method considering a worker learning effect, and the method comprises the following steps: S1, considering the influence of a disassembly relation of a part, a disassembly tool, a disassembly direction and tool conversion information on the worker learning effect, correcting the disassembly time, and obtaining the actual disassembly time; s2, on the basis of the actual disassembling time, a recovery decision model based on the total disassembling time, the disassembling profit and the total carbon emission as targets is constructed; and S3, solving the scheduling model by adopting an optimization algorithm to obtain a recovery decision scheme. According to the method, various recovery modes of the product, the worker learning effect and other dynamic factors are effectively modeled and decided, the method adapts to different product characteristics and actual operation scenes, the recovery modes are dynamically matched, the disassembling efficiency and resource recovery benefits of the waste electronic product are remarkably improved, and meanwhile carbon emission is effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of disassembly and recycling planning, and particularly relates to an optimization method for disassembly planning and recycling decision-making considering the learning effect of workers. Background Technique

[0002] The disassembly and recycling of waste electronic products are crucial for improving the lives of people around the world. Reasonable recycling of parts can also maximize the value of waste electronic products, which is in line with the concept of green development. At the same time, in actual disassembly and recycling activities, there are many human activities. When workers perform similar tasks over a period of time, they will gain proficiency through repetition and learning, that is, the learning effect of workers will greatly affect the disassembly efficiency.

[0003] However, most disassembly and recycling enterprises separately conduct disassembly planning or recycling decision-making planning for waste products, ignoring that different recycling methods of parts will lead to different times, incomes, and energy consumption, which are crucial for making reasonable disassembly plans. Therefore, these two types of problems must be combined, that is, collaborative recycling decision-making and disassembly planning, to ensure effective guidance for the actual disassembly and recycling process. Summary of the Invention

[0004] The present invention proposes an optimization method for disassembly planning and recycling decision-making considering the learning effect of workers, so as to solve the technical problem that the existing disassembly planning scheme does not combine the learning effect of workers and different recycling methods, and the finally obtained disassembly planning scheme is difficult to make full use of the learning effect of workers, resulting in the planning scheme not meeting the actual needs.

[0005] To solve the above technical problems, the present invention provides an optimization method for disassembly planning and recycling decision-making considering the learning effect of workers, including the following steps: Step S1: Considering the influence of the disassembly relationship, disassembly tools, disassembly direction, and tool conversion information of parts on the learning effect of workers, correct the disassembly time to obtain the actual disassembly time; Step S2: Based on the actual disassembly time, construct a recycling decision model with the total disassembly time, disassembly profit, and total carbon emissions as the objectives; Step S3: Use an optimization algorithm to solve the scheduling model to obtain a recycling decision-making scheme.

[0006] Preferably, in step S1, the calculation of the actual disassembly time The expression is: ; In the formula, p ik Indicates if the part i At the kAt the position, then p ik = 1, otherwise p ik = 0; L Represents the set of disassembly sequence positions, | L | Represents the maximum disassembly sequence position; t i Represents the part i The basic disassembly time; I Represents the set of total disassembled parts; α kl Indicates that if the k nd position and the l th position have the same disassembly tool, then α kl = 1, otherwise α kl = 0; σ Represents the learning rate.

[0007] Preferably, in step S2, the expression for the total disassembly time is: ; In the formula, y i Indicates that if the disassembled part i is disassembled, then y i = 1, otherwise y i = 0; L ´ = [1, 2, …, | L | - 1]; Represents the tool type used for disassembling the part i and the conversion time of the tool type used for disassembling the part j i o i Represents the tool type used for disassembling the part i ; d k Indicates that if the k th position and the k +1 have the same disassembly direction, then d k = 0, if the direction changes by 90°, then d k = 1, if the direction changes by 180°, then d k = 2; t a Represents the basic direction conversion time.

[0008] Preferably, in step S2, the expression for the disassembly profit is: ; ; ; ; wherein, m is the recycling option number, m = 1 represents reuse, m = 2 represents remanufacturing, m = 3 represents material recycling, m = 4 represents scrapping; r im represents that the recycling option is m for the disassembled parts i income; k im represents that if the recycling option of disassembled part i is m , then k im = 1; c d represents the cost of labor and tools; c s represents the cost of ineffective operation time; c h represents the additional cost of disassembling dangerous tasks; w 1 represents the worker cost coefficient; w 2 represents the tool usage cost coefficient; Cs represents the ineffective unit time cost; Ch represents the additional unit time cost of disassembling dangerous parts; h i represents the harmfulness of disassembling part i ; if part i is a harmful part, then h i = 1, otherwise h i = 0.

[0009] Preferably, in step S2, the expression of the total carbon emission is: ; ; ; wherein, E EOL represents the carbon emission of remanufacturing parts after considering the recycling option; when the recycling method is reuse, e m = 1, when the recycling method is remanufacturing, e m= 0.9 when the recycling method is material recycling, e m =0.75 when the recycling method is scrapping, e m = 0.5; E DP represents the carbon emissions generated during the disassembly process; represents the carbon emission equivalent consumed for remanufacturing parts i ; E oi represents the disassembly task i carbon emission equivalent per unit time of using tools; E h represents the additional carbon emission equivalent per unit time for hazardous disassembly tasks.

[0010] Preferably, setting constraints for the recycling decision model includes: Constraint 1: Each part has only one recycling mode, and its expression is: ; Constraint 2: All parts to be reused, remanufactured, and recycled must be disassembled, and its expression is: ; Constraint 3: Hazardous tasks must be disassembled, expressed as: ; Constraint 4: The recycling options for hazardous parts can only be recycling or scrapping, expressed as: ; Constraint 5: The recycling method for parts that do not need to be disassembled is scrapping, expressed as: ; Constraint 6: The precedence constraint is expressed as: ; Constraint 7: The tool type change constraint is expressed as: ; ; ; ; In the formula, I is the total set of disassembled parts; L represents the set of disassembly sequence positions, | L | represents the maximum disassembly sequence position; m is the recycling option number, m = 1 represents reuse, m =2 represents remanufacturing,m = 3 indicates material recycling, m = 4 indicates scrapping; k im Indicates that if the recycling option for disassembling part i is m , then k im = 1; y i Indicates that if disassembling part i is disassembled, then y i = 1, otherwise y i = 0; h i Indicates the harmfulness of disassembling part i . If part i is a harmful part, then h i = 1, otherwise h i = 0; p ik Indicates that if part i is in the k th position, then p ik = 1, otherwise p ik = 0; If disassembling part i is the immediate predecessor part of disassembling part j , then y ij = 1, otherwise y ij = 0; o i Indicates the type of tool used for disassembling part i . The total number of tool types is O ; α kl Indicates that if the disassembling tools at the k th position and the l th position are the same, then α kl = 1, otherwise α kl = 0.

[0011] Preferably, in step S3, the optimization algorithm is solved by using an improved multi-objective genetic algorithm based on reinforcement learning. The solving method includes the following steps: Step S31: Complete the initialization of the population according to the disassembly precedence relationship and the encoding and decoding strategies; Step S32: Use the ratio of the hypervolume indicator HV as the state of reinforcement learning, use several mutation operations as the actions of reinforcement learning for evolution, fuse the evolved population with the population before evolution, adopt the elitist retention strategy, and update the population; Step S33: When the running time of the algorithm reaches the maximum running time, the algorithm terminates and outputs the planning method.

[0012] Preferably, in step S31, the encoding strategy includes: encoding using a three-layer linked list structure, denoted as pop = p 1 , p 2 , p 3 , p 1 = a 1 ,…, a l ,…, a L represents the disassembly sequence, a l is the component at the l th position, and its value represents the component number, L is the position set, i.e., the total number of components; p 2 = b 1 ,…, b l ,…, b L represents the disassembly decision sequence, b l corresponds to the disassembly decision of the component a l , and its value represents whether the component is disassembled; p 3 = c 1 ,…, c l ,…, c L represents p 1 the recycling mode of the corresponding component in, and its value represents the recycling mode number of the component.

[0013] Preferably, in step S31, the decoding strategy includes: traversing the disassembly sequence encoding, determining the disassembly time and disassembly profit of the part according to the corresponding recycling method of the part, and determining the carbon emission equivalent consumed for remanufacturing the part; starting from the second position, i.e., k = 2, traverse the position kThe tool usage of all previous positions. If there is a situation where the same tool is used, there is a learning effect. Calculate the k actual disassembly time of the part corresponding to the i position; t i ' In addition, judge the tool change situation and the part orientation change situation at position k and position k -1, calculate the tool change time and the orientation change time; repeat the above steps until k the maximum sequence length is reached, and the total disassembly time is obtained; calculate the disassembly cost and the carbon emissions released during the disassembly process according to the actual disassembly time; calculate the total disassembly profit and the total carbon emission equivalent, and the decoding is completed.

[0014] Preferably, in step S32, the method of using the ratio of the hypervolume index as the state of reinforcement learning includes: defining the state as the ratio of the HV value of the first generation to the HV value of the new generation , ∈[0, 1]. According to the result, the state is divided into 20 states, that is S = s (1), s (2), s (3), …, s (19), s (20)], where the interval value is set to 0.05.

[0015] The beneficial effects of the present invention at least include: The present invention globally optimizes the disassembly plan and recycling decision through an algorithm model, comprehensively considers the carbon emission impacts of multiple factors such as energy consumption and recycling modes, provides a more environmentally friendly recycling plan, helps to achieve green recycling and sustainable development, and provides theoretical and technical support for the disassembly and recycling of retired electronic products such as waste smartphones and waste laptop computers; effectively models and makes decisions on various recycling modes, worker learning effects and other dynamic factors of the product, adapts to different product characteristics and actual operation scenarios, dynamically matches the recycling mode, significantly improves the disassembly efficiency and resource recycling benefits of waste electronic products, and effectively reduces carbon emissions at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a schematic flowchart of the method according to an embodiment of the present invention; Figure 2 is a schematic diagram of two-point crossover according to an embodiment of the present invention; Figure 3 is a schematic diagram of preferential retention crossover according to an embodiment of the present invention; Figure 4Schematic diagram of the switching mutation according to an embodiment of the present invention; Figure 5 Schematic diagram of the insertion mutation according to an embodiment of the present invention; Figure 6 Schematic diagram of the priority relationship between parts according to an embodiment of the present invention; Figure 7 Box plot of the genetic algorithm based on reinforcement learning and the comparative algorithm according to an embodiment of the present invention; Figure 8 Iteration diagram of the genetic algorithm based on reinforcement learning and the comparative algorithm according to an embodiment of the present invention; Figure 9 Scatter diagram of the results of the refrigerator disassembly case according to an embodiment of the present invention; Figure 10 Gantt chart of the results of the refrigerator disassembly case according to an embodiment of the present invention. Detailed implementation manners

[0017] Next, in conjunction with the accompanying drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the protection scope of the present invention.

[0018] Before presenting the embodiments, the learning effect of workers and the recycling decision are elaborated: For retired electronic products with high recycling value of their disassembled parts, such as televisions, refrigerators, etc., manual disassembly is mostly used, and the disassembly tools are mostly repeated. When workers use the same tools, regardless of whether the tasks are connected, there will be a learning effect, thus improving the disassembly efficiency and making full use of human resources and tool resources. The difficulty of this problem lies in how to reasonably arrange the order of disassembling parts to maximize the learning effect of workers. The recycling decision problem is defined as choosing the recycling process. For example, recycling options include reuse, remanufacturing, recycling, and landfill disposal. Focusing on evaluating the remaining value of products from multiple dimensions to obtain the optimal recycling decision to maximize the value of waste electronic products, the difficulty of this problem lies in how to reasonably decide the recycling mode of components to achieve the maximum disassembly profit.

[0019] For the convenience of description, the symbols and decision variables used in the embodiments of the present invention are as follows: i , j : Part number, i , j = 0, 1, 2,... I , where I is the total number of disassembled parts; k , l : k , l is the disassembly sequence position, and the position set is L , and the maximum position is | L |. Let the set L ´ = [1, 2, …, | L | - 1]; o , p : tool number, o , p = 0, 1, 2, … O , where O is the total disassembly tool; m : recycling option number, m = 1 indicates reuse, m = 2 indicates remanufacturing, m = 3 indicates material recycling, m = 4 indicates scrapping; I : total disassembly part set; o i : disassembly part i the type of tool used, and the total number of tool types is O ; Y : precedence relation matrix: If disassembly part i is the immediate predecessor part of disassembly part j , then y ij = 1, y ij = 0; h i : the harmfulness of disassembly part i : If part i is a harmful part, then h i = 1, otherwise h i = 0; t i : the basic disassembly time of disassembly part i ; σ : learning rate; t r op : tool o and tool p conversion time; t a : Basic direction conversion time; r im : The recycling option is m disassembled parts i Revenue; : Remanufactured parts i The carbon emission equivalent consumed; e m : To determine the impact of product recycling decisions on carbon emissions, the carbon emissions saved by each option are calculated by share. When the recycling method is reuse, e m = 1. When the recycling method is remanufacturing, e m = 0.9. When the recycling method is material recycling, e m = 0.75. When the recycling method is scrapping, e m = 0.5; Cs : Ineffective unit time cost; Ch : Additional unit time cost for disassembling hazardous parts; w 1 : Worker cost coefficient; w 2 : Tool usage cost coefficient; c d : Cost of using labor and tools; c s : Ineffective operation time cost; c h : Additional cost for disassembling hazardous tasks; E oi : Disassembly task i Carbon emission equivalent per unit time for using tools in the disassembly process; E h : Additional carbon emission equivalent per unit time for disassembling hazardous tasks; E EOL : Carbon emissions of remanufacturing parts after considering the recycling option; E DP : Carbon emissions generated during the disassembly process; y i : Decision variable. If the disassembled part i is disassembled, then y i = 1, otherwise y i = 0; t i ' : The actual disassembly time of the disassembled part i ; d k : If the k th position and the k +1 disassembly directions remain unchanged, then d k = 0. If the direction changes by 90°, then d k = 1. If the direction changes by 180°, then d k = 2; α kl : Decision variable. If the k th position and the l th position use the same disassembly tool, then α kl = 1, otherwise α kl = 0; p ik : If part i is at the k th position, then p ik = 1, otherwise p ik = 0; k im : If the recycling option for disassembling part i is m , then k im = 1, otherwise k im = 0.

[0020] Example 1 As Figure 1 shown, the embodiment of the present invention provides an optimization method for disassembly planning and recycling decision-making considering the learning effect of workers, including the following steps: Step S1: Consider the influence of the disassembly relationship, disassembly tool, disassembly direction, and tool conversion information of parts on the learning effect of workers, correct the disassembly time, and obtain the actual disassembly time.

[0021] Specifically, in this embodiment, the expression for the actual disassembly time is: .

[0022] Step S2: Based on the actual disassembly time, construct a recycling decision-making model with the total disassembly time, disassembly profit, and total carbon emissions as the objectives.

[0023] Specifically, minimize the disassembly time, maximize the disassembly profit, and minimize the carbon emission equivalent. The expression is: .

[0024] Sub-goal 1: Minimize the disassembly time, expressed as: .

[0025] Sub-goal 2: Maximize the disassembly profit, where c d represents the cost of labor and tools, c s represents the cost of ineffective operation time, c h represents the additional cost of disassembling dangerous tasks. The formula is expressed as: ; ; ; .

[0026] Sub-goal 3: Minimize the carbon emission equivalent, where E EOL represents the carbon emissions of remanufacturing parts after considering the recycling options, E DP represents the carbon emissions generated during the disassembly process. The formula can be expressed as: ; ; .

[0027] In this embodiment, in order to make the solution more in line with the actual situation, the following constraints are also set.

[0028] Constraint 1: Each type of part has only one recycling mode, and its expression is: ; Constraint 2: All parts to be reused, remanufactured, and recycled must be disassembled, and its expression is: ; Constraint 3: Dangerous tasks must be disassembled, expressed as: ; Constraint 4: The recycling options for dangerous parts can only be recycling or scrapping, expressed as: ; Constraint 5: The recycling method for parts that do not need to be disassembled is scrapping, expressed as: ; Constraint 6: The precedence constraint is expressed as: ; Constraint 7: The tool type change constraint is expressed as: ; ; ; .

[0029] Step S3: Use an optimization algorithm to solve the scheduling model and obtain a recycling decision plan.

[0030] The optimization algorithms in this embodiment include, but are not limited to, genetic algorithms, particle swarm optimization algorithms, differential evolution algorithms, simulated annealing algorithms, and dung beetle optimization algorithms.

[0031] Embodiment 2 The collaborative recycling decision and disassembly planning problem is defined as an NP-hard problem. Existing solution methods tend to use metaheuristic-based approximation algorithms because they can solve large-scale complex problems and find near-optimal solutions within a reasonable computation time. However, the metaheuristic algorithms under certain strategies have poor adaptability and cannot dynamically adjust the optimization strategy according to the iteration of the algorithm, and are prone to falling into local optima.

[0032] Therefore, based on Embodiment 1, this embodiment constructs an encoding and decoding strategy based on disassembly constraints according to the disassembly characteristics of waste electronic products, and uses an improved multi-objective genetic algorithm based on reinforcement learning to replace the traditional optimization algorithm in Step S3, which mainly includes the following steps: Step S31: Initialize the algorithm parameters and set the learning rate; Step S32: Complete the population initialization according to the disassembly precedence relationship and the encoding and decoding strategies; Step S33: Perform corresponding crossover and mutation operations according to the state to which the population belongs; Step S34: Merge the evolved population with the pre-evolution population, adopt the elite retention strategy, and update the population; Step S35: When the running time of the algorithm reaches the maximum running time, the algorithm terminates and outputs the planning method.

[0033] The specific coding process is as follows: Step S3211: In the coding strategy, a three-layer linked list structure is used for coding, denoted as pop = p 1 , p 2 , p 3 , p 1 = a 1 ,…, a l ,…, a L represents the disassembly sequence, a l is the component at the l th position, and its value represents the component number. L is the position set, i.e., the total number of components. p 2 = b 1 ,…, b l ,…, b L represents the disassembly decision sequence, b l corresponding to the disassembly decision of the component a l , and its value represents whether the component is disassembled. p 3 = c 1 ,…, c l ,…, c L represents p 1 the recycling mode of the corresponding component in, and its value represents the recycling mode number of the component.

[0034] Step S3212: The disassembly sequence p 1 uses positive integer coding, and each number represents a disassembled part. The specific steps are as follows: ① Randomly select a part with all 0s in the columns of the matrix Y from the candidate parts, such as part 1; ② Use the selected part i as the part at the current position k of the sequence, and remove the part i from the candidate parts; ③ Update the matrix Yto release the constraints related to the parts i i.e., y ij = 0, y ji = 0 I For example, if i = 1, then y 1j = 0, otherwise y j1 = 0; ④ Allocate the parts at the next position, let k ← k + 1; ⑤ Repeat the above steps until all parts are allocated and a disassembly sequence p 1 is obtained.

[0035] Step S3213: Disassembly decision sequence p 2 is composed of variables 0 - 1, where 0 indicates that the part has not been disassembled and 1 indicates that the part has been disassembled. The conditions for the disassembly decision sequence p 2 are as follows: ① The disassembly decision variables of the hazardous parts and all their immediate predecessors are 1; ② The disassembly decision variables need to satisfy the precedence constraints, i.e., the disassembly decision variables of all immediate predecessors of the parts with disassembly decision variables of 1 are all 1; ③ The disassembly decision variables of other parts are randomly selected between 0 and 1.

[0036] Step S3214: Recycling mode decision sequence p 3 is composed of variables 1 - 4, where 1 indicates that the recycling mode is reuse, 2 indicates that the recycling mode is remanufacturing, 3 indicates that the recycling mode is recycling and reuse, and 4 indicates that the recycling mode is scrapping. The disassembly mode decision sequence p 3 needs to satisfy conditions similar to the above, including: ① Each part corresponds to a unique disassembly mode decision variable; ② The recycling mode decision variable of the hazardous parts is 3 or 4; ③ The recycling mode of the parts with decision variables of 0 is 4; ④ The recycling mode decision variables of other parts are randomly selected between 1 and 4.

[0037] After obtaining a feasible individual solution, it is necessary to decode it into a disassembly plan and calculate its fitness value. The disassembly sequences obtained by initialization all belong to complete disassembly sequences. First, decode them into partial disassembly sequences, that is, delete the parts with disassembly decision variables of 0 in the second - layer encoding, and only perform the disassembly operations with disassembly decisions of 1. In this way, the complete disassembly sequence will be decoded into a partial disassembly sequence, and the actual sequence length is K .

[0038] The specific decoding process is as follows: ① Traverse the actual disassembly sequence, determine the disassembly time and disassembly profit of the parts according to the corresponding recycling method of the parts, and determine the carbon emission equivalent consumed by reproducing the parts; ② Start from the second position, that is, k =2, traverse position k The tool usage of all previous positions. If the same tool is used, there is a learning effect. Calculate the k Position corresponding parts i Actual disassembly time t i ' In addition, the position k and location k -1 Tool replacement and part direction change, calculate tool replacement time and direction change time; ③ k ← k +1; ④ Repeat steps 2 to 3 until k = K , and obtain the total disassembly time; ⑤ Calculate the disassembly cost and the carbon emissions released during the disassembly process based on the actual disassembly time; ⑥ Calculate the total disassembly profit and the total carbon emissions equivalent, and the decoding is completed.

[0039] In this embodiment, the state of reinforcement learning is defined as the ratio of the HV value of the first generation to the HV value of the new generation. , ∈[0,1], different ratios, different states, according to The result divides the state into 20 states, namely S =[ s (1), s (2), s (3),…, s (19), s (20)], where The interval value is set to 0.05, for example: if ∈[0, 0.05], s = s (1); ∈[0.05, 0.10], s = s (2), and so on.

[0040] The actions of reinforcement learning are composed of four local search strategies, four actions consisting of two crossover operators and two mutation operators. The crossover operation includes two-point mapping crossover and priority preservation crossover, and the mutation operation includes exchange mutation and insertion mutation. Figure 2 As shown, priority is given to retaining crossovers such as Figure 3 As shown, the exchange mutation is Figure 4 As shown, insertion mutations such asFigure 5 as shown

[0041] When updating the population, the Pareto dominance relationship is used to describe the quality of solutions. The better solutions are called Pareto non-dominated rank solutions or non-inferior solutions. During the iteration process, the non-dominated solutions of each iteration are retained and stored in an external archive. After the current iteration is completed, they are compared with the non-dominated solutions in the external archive. The non-dominated solutions are retained and the dominated solutions are removed. When updating the population, adding the non-dominated solutions in the external archive to the population can accelerate the convergence of the algorithm. To keep the population size of the algorithm unchanged, when the number of non-dominated solutions in the external archive is less than the population size, individuals need to be added to the population to ensure that the population size remains the same.

[0042] Example 3 Taking the disassembly of a refrigerator as an example, a collaborative planning problem of disassembly planning and recycling decision-making considering the learning effect is constructed to analyze the application performance of the method of the present invention in actual engineering cases. Combining the structure and main components of the refrigerator, the disassembly parts are re-divided. The specific disassembly part information is shown in Tables 1 and 2, with a total of 66 disassembly parts.

[0043] Table 1

[0044] Table 2

[0045] The precedence relationship between parts is as Figure 6 shown. The hazardous parts are 7, 14, 33, 41, and 52. The disassembly tools are numbered 1 to 5, representing the hand, electric screwdriver 1, wire cutter, cutting machine, and electric screwdriver 2 respectively. The unit time costs of the disassembly tools are 0, 0.20, 0.10, 0.30, and 0.20 yuan respectively, and the unit time energy consumptions of the disassembly tools are 0, 1.5, 0.5, 2.0, and 1.5 kW respectively. Eh =0.125 CO 2 equivalent, assuming that the carbon emission equivalent released by reproducing a part E i new =1CO 2 equivalent. w 1 =0.06, w 2 =0.04, the unit time cost of invalid operations Cs =0.04, the unit time of disassembling hazardous tasks Ch= 0.006. The value of the components or materials obtained from each disassembled part was evaluated with reference to the market value of the reference components. The orientation information of the disassembled parts is as follows: parts 1 to 43 are on the front; parts 44 to 60 are on the back; parts 61 to 66 are on the bottom.

[0046] Three classical algorithms, the Strength Pareto Evolutionary Algorithm (SPEA), the Hypervolume Estimation Algorithm for Multi-objective Optimization (HypE), and the Pareto-based Non-dominated Sorting Genetic Algorithm 2 (PESA2), were compared with the improved genetic algorithm based on reinforcement learning (GAQL) provided by the present invention. The population size and the number of iterations of the comparison algorithms were set to 200 and 500 respectively. The crossover and mutation probabilities in the relevant algorithms were also set to 0.85 and 0.15 respectively. Each algorithm was independently run 10 times, and all the results were screened for non-dominated solutions, and their objective values were used as the true Pareto front. The results were normalized, and the hypervolume (HV), the inverted generational distance (IGD), and the spread / coverage metric (Spread) were used to evaluate the non-dominated solution sets of the algorithms. The means and 95% confidence intervals of the four algorithms on the three metrics are as Figure 7 shown.

[0047] Figure 7 It can be seen from the comparison that the confidence interval span of GAQL on the HV metric is slightly larger than that of SPEA, but smaller than that of PESA2 and HypE, and there is no overlap with the confidence intervals of SPEA and HypE; on the IGD metric, the confidence interval span of GAQL is smaller than that of the three comparison algorithms, and there is also no overlap with the confidence intervals of SPEA and HypE; on the Spread metric, the confidence interval span of GAQL is slightly larger than that of PESA2, but smaller than that of SPEA and HypE. There are no outliers in all algorithms on the HV index, there is 1 outlier in PESA2 on the IGD index, and 3 outliers in PESA2 on the Spread index. The comparison shows that on the three metrics, the average value of GAQL is better than the average values of the three comparison algorithms, indicating that the convergence and dispersion of GAQL are better than those of the comparison algorithms.

[0048] Taking the HV metric as an example, the convergence of different algorithms during the iteration process was analyzed. The result data when each algorithm obtained the maximum HV metric was taken to draw the iteration graph, as Figure 8 shown. It can be seen from the comparison that the final HV value of GAQL is better than the final HV values of the three comparison algorithms. At the initial stage of the algorithm iteration, the initial solution quality of GAQL is better, and GAQL can still be improved after 350 s, indicating that reinforcement learning expands the optimization ability of the algorithm. The performance ranking of the four algorithms from best to worst is GAQL > PESA2 > HypE > SPEA.

[0049] Further analyze the relationship between these three objectives. Run the GAQL algorithm ten times and plot the 370 non-dominated solutions obtained after running ten times in a scatter plot matrix, as Figure 9 shown, which shows the scatter plot of the corresponding objective values of these solutions and illustrates the correlation between individual objectives. It can be seen that f 1 and f 2 have a strong positive correlation, that is, an increase in disassembly time will lead to an increase in profit. The reason is that different disassembly times and recovery values of different components in the product correspond to different recycling modes. When the mode is 1, the disassembly time of the components increases, and the recovery value of the components also increases. It should be noted that f 1 and f 3 also have a positive correlation, which means that in some cases, the two objectives of improving efficiency and reducing carbon emissions are not absolutely contradictory. In other words, solutions with good disassembly efficiency are likely to be energy-saving records and marks the optimal disassembly solutions achieved under the three objectives S 1 、 S 2 、 S 3 As Figure 10 shown, the vertical line frame in the figure represents the tool change time plus the direction change time, and the horizontal line frame represents dangerous parts. Such diverse options allow decision-makers to make choices according to different situations in this case. For example, if time-critical efficiency is the priority, option S 1 can be selected. If profit is the main consideration, solution S 2 may be preferred. If environmental priority is crucial, option S 3 might be the best choice. For a more balanced approach, the decision-maker can consider all 3 options.

[0050] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. Only the preferred embodiments of the present invention are expressed. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the present invention. As long as the combinations of these technical features do not conflict, they should be considered as within the scope described in this specification.

[0051] It should be noted that, for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the appended claims.

Claims

1. A disassembly planning and recycling decision optimization method considering worker learning effect, characterized by: The following steps are involved: Step S1: Considering the influence of the disassembly relationship of parts, disassembly tools, disassembly direction and tool conversion information on the worker's learning effect, the disassembly time is corrected to obtain the actual disassembly time; Step S2: Based on the actual disassembly time, construct a recycling decision model based on total disassembly time, disassembly profit and total carbon emissions as targets; Step S3: using an optimization algorithm to solve the scheduling model and obtain a recycling decision plan.

2. The method for optimizing disassembly planning and recycling decision-making considering worker learning effect according to claim 1, characterized in that: In step S1, the actual disassembly time is calculated The expression is: ； In the formula, p ik If the parts i In the k In the position, p ik =1, otherwise p ik =0; L represents the disassembly sequence position set, | L | indicates the maximum disassembly sequence position; t i Indicates parts i Basic disassembly time; I Represents the total disassembled parts collection; α kl If k Position and l The disassembly tools for each position are the same. α kl =1, otherwise α kl =0; σ Represents the learning rate.

3. The method for optimizing disassembly planning and recycling decision-making considering worker learning effect according to claim 2, characterized in that: In step S2, the total disassembly time is expressed as: ； In the formula, y i If you disassemble the parts i If disassembled, y i =1, otherwise y i =0; L ´=[1,2,…,| L |-1]; Indicates disassembled parts i Types of tools used and parts disassembled j The conversion time of the tool type used, o i Indicates disassembled parts i The type of tools used; d k If k Position and k +1's disassembly direction remains unchanged, then d k =0, if the direction changes by 90°, then d k =1, if the direction changes by 180°, then d k =2; t a Indicates the cardinal direction transition time.

4. The method for optimizing disassembly planning and recycling decision-making considering worker learning effect according to claim 3 is characterized by: In step S2, the expression of the disassembly profit is: ； ； ； ； In the formula, m is the recycling option number, m =1 means reuse, m =2 means remanufacturing, m =3 means material recycling, m =4 means scrapped; r im Indicates the recycling option is m Disassembled parts i income; k im If the recycling option of dismantled part i is m ,but k im =1; c d Indicates the cost of manpower and tools used; c s Indicates the time cost of ineffective operation; c h It indicates the additional cost of dismantling dangerous tasks; w 1 represents the worker cost coefficient; w 2 represents the tool use cost coefficient; Cs represents the ineffective unit time cost; Ch It represents the additional unit time cost of removing dangerous parts; h i Indicates disassembled parts i Hazards, if parts i If it is a harmful part, h i =1, otherwise h i =0.

5. The method for optimizing disassembly planning and recycling decision-making considering worker learning effect according to claim 4, characterized in that: In step S2, the expression of the total carbon emissions is: ； ； ； In the formula, E EOL represents the carbon emissions of reproducing the component after taking into account recycling options; when the recycling method is reuse, e m = 1, when the recycling method is remanufacturing, e m = 0.9, when the recycling method is material recycling, e m =0.75, when the recycling method is scrapping, e m = 0.5; E DP Indicates the carbon emissions generated during the dismantling process; Indicates remanufactured parts i Carbon emissions equivalent consumed; E oi Indicates disassembly task i Carbon emissions per unit time of using the tool; E h Represents the carbon emission equivalent per unit time of dismantling dangerous tasks.

6. The method for optimizing disassembly planning and recycling decision-making considering worker learning effect according to claim 1, characterized in that: Setting constraints for the recycling decision model includes: Constraint 1: Each part has only one recycling mode, which is expressed as: ； Constraint 2: All parts to be reused, remanufactured, and recycled must be disassembled, which is expressed as: ； Constraint 3: Dangerous tasks must be disassembled, expressed as: ； Constraint 4: The recycling options for hazardous parts can only be recycling or scrapping, expressed as: ； Constraint 5: The recycling method of parts that do not need to be disassembled is scrapping, which is expressed as: ； Constraint 6: The precedence constraint is expressed as: ； Constraint 7: Tool type change constraint is expressed as: ； ； ； ； In the formula, I For total disassembly of parts collection; L represents the disassembly sequence position set, | L | indicates the maximum disassembly sequence position; m is the recycling option number, m =1 means reuse, m =2 means remanufacturing, m =3 means material recycling, m =4 means scrapped; k im If the recycling option of dismantled part i is m ,but k im =1; y i If you disassemble the parts i If disassembled, y i =1, otherwise y i =0; h i Indicates disassembled parts i Hazards, if parts i If it is a harmful part, h i =1, otherwise h i =0; p ik If the parts i In the k In the position, p ik =1, otherwise p ik =0; if disassembled parts i Disassembled parts j The preceding part of y ij =1, otherwise y ij =0; o i Indicates disassembled parts i The types of tools used, the total number of tool types is O ; α kl If k Position and l The disassembly tools for each position are the same. α kl =1, otherwise α kl =0.

7. The method for optimizing disassembly planning and recycling decision-making considering worker learning effect according to claim 1, characterized in that: In step S3, the optimization algorithm adopts an improved multi-objective genetic algorithm based on reinforcement learning to solve, and the solution method includes the following steps: Step S31: completing population initialization according to the disassembly priority relationship and encoding and decoding strategies; Step S32: using the ratio of the hypervolume index HV as the state of reinforcement learning, using several mutation operations as actions of reinforcement learning to evolve, merging the evolved population with the population before evolution, adopting an elite retention strategy, and updating the population; Step S33: When the algorithm running time reaches the maximum running time, the algorithm terminates and outputs the planning method.

8. The method for optimizing disassembly planning and recycling decision-making considering worker learning effect according to claim 7, characterized in that: In step S31, the encoding strategy includes: using a three-layer linked list structure for encoding, denoted as pop =[ p 1, p 2, p 3], p 1=[ a 1,…, a l ,…, a L ] indicates the disassembly sequence, a l For the l The value of the component at the position represents the component number. L is the position set, i.e. the total number of parts; p 2=[ b 1,…, b l ,…, b L ] represents the disassembly decision sequence, b l Corresponding parts a l The disassembly decision of, whose value represents whether the component is disassembled; p 3=[ c 1,…, c l ,…, c L ]express p 1 corresponds to the recycling mode of the parts, and its value represents the recycling mode number of the parts.

9. The method for optimizing disassembly planning and recycling decision-making considering worker learning effect according to claim 8, characterized in that: In step S31, the decoding strategy includes: traversing the disassembly sequence code, determining the disassembly time and disassembly profit of the parts according to the corresponding recycling method of the parts, and determining the carbon emission equivalent consumed by reproducing the parts; starting from the second position, that is, k =2, traverse position k The tool usage of all previous positions. If the same tool is used, there is a learning effect. Calculate the k Position corresponding parts i Actual disassembly time t i ' In addition, the position k and location k -1 tool change situation and part direction change situation, calculate tool change time and direction change time; repeat the above steps until k The maximum sequence length is reached and the total disassembly time is obtained; the disassembly cost and the carbon emissions released during the disassembly process are calculated based on the actual disassembly time; the total disassembly profit and the total carbon emissions equivalent are calculated, and the decoding is completed.

10. The method for optimizing disassembly planning and recycling decision-making considering worker learning effect according to claim 7, characterized in that: In step S32, the method of using the ratio of the hypervolume index as the state of the reinforcement learning includes: defining the state as the ratio of the HV value of the first generation to the HV value of the new generation , ∈[0,1], according to The result divides the state into 20 states, namely S =[ s (1), s (2), s (3),…, s (19), s (20)], where The interval value is set to 0.05.

Images