Multi-source solid waste cyclic utilization path optimization method and system

Abstract

The invention belongs to the technical field of multi-source solid waste cyclic utilization, and discloses a multi-source solid waste cyclic utilization path optimization method and system, and the method comprises the steps: solving a multi-target mathematical optimization model of a multi-source solid waste cyclic utilization system through employing a dual-stage multi-target optimization method, and obtaining a multi-source solid waste cyclic utilization optimization scheme set; selecting a preference-based solid waste optimal processing path from the optimization scheme set; in the first stage, an extreme weight vector and a center weight vector are constructed to search a heuristic solution, and in the second stage, further evolution is carried out on the basis of a heuristic solution population, so that the population has better convergence and diversity; in the second stage, deep reinforcement learning is introduced to adaptively optimize parameters of a genetic operator. Meanwhile, the invention also provides a dynamic switching method based on population fitness to avoid unnecessary waste of computing resources. According to the method, the problems of low algorithm convergence speed, difficulty in determining important parameters of genetic operators and the like in practical application are solved, and efficient optimal configuration of path optimization is realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of multi-source solid waste recycling, and more particularly to a multi-source solid waste recycling path optimization method and system. BACKGROUND

[0002] Many solid wastes contain a large amount of recyclable resources, such as waste metals, paper, glass, etc. If solid waste can be effectively recycled, not only can it alleviate the resource shortage situation, but also can reduce the dependence on primary resources. Therefore, how to efficiently and environmentally utilize multi-source solid waste has become an important problem to be solved at present.

[0003] In the process of multi-source solid waste recycling, the optimization of the recycling path is a key link. Reasonable recycling path planning can significantly reduce the processing cost, improve the resource recycling rate and reduce the negative impact on the environment. Multi-source solid waste recycling is usually a complex combination optimization problem that needs to consider multiple optimization objectives such as economic benefit, environmental benefit and social benefit, involving multiple processing links, multiple processing technologies and different constraint conditions (such as processing capacity limit). The existing method mainly considers converting the multi-objective problem into a single-objective problem, and then obtaining an optimal combination through an optimization solver or heuristic algorithm, but such method has the limitation of over-reliance on pre-set preferences and inability to comprehensively consider multiple optimization objectives.

[0004] The core challenge of multi-objective evolutionary algorithm is to ensure the convergence and diversity of the population at the same time. However, in the search process of the algorithm, convergence and diversity are often mutually restrictive. In order to better balance these two core objectives, a two-stage multi-objective optimization method is proposed in the prior art. The two-stage multi-objective optimization method divides the entire optimization process into two continuous stages with different emphases, aiming to decompose the complex optimization task, so as to more efficiently and accurately approximate the Pareto front. In addition, some studies have shown that compared with traditional random initialization methods, adding some high-quality solutions (also known as heuristic solutions) to the randomly initialized population can significantly improve the search efficiency of the optimization algorithm. However, the switching method of the existing two-stage algorithm is mostly dependent on the static process of the algorithm (such as setting the maximum number of iterations, and entering the second stage when the maximum number of iterations is reached), which is easy to cause waste of computing resources; the existing heuristic solution search easily ignores the diversity of the population, leading to the algorithm prematurely falling into a local optimal region, and thus the multi-source solid waste recycling path found is not the optimal scheme. SUMMARY

[0005] To address the above-mentioned deficiencies or improvement needs of existing technologies, this invention provides a method and system for optimizing multi-source solid waste recycling paths. The purpose is to solve the problem that existing methods are prone to getting trapped in local optimum regions during the process of determining multi-source solid waste recycling path optimization schemes, resulting in the determined multi-source solid waste recycling paths not being the optimal schemes.

[0006] To achieve the above objectives, the present invention provides a method for optimizing the recycling pathway of multi-source solid waste, comprising: A two-stage multi-objective optimization method is used to solve the multi-objective mathematical optimization model of the multi-source solid waste recycling system, obtaining a set of optimal solutions for multi-source solid waste recycling; the optimal solid waste treatment path based on preference is selected from the set of optimal solutions; wherein, the two-stage multi-objective optimization method includes: In the first stage, the decision variables of the multi-objective mathematical optimization model are... As individuals within a population, For the first The first type of solid waste adopts the The quality of various solid waste treatment methods; structure Each weight vector includes Extreme weight vectors and a center weight vector ,in, , , ..., , , The number of objectives to be optimized in the multi-objective mathematical optimization model; During the g-th population iteration, from the current population... Select individuals for crossover and mutation to generate new individuals on each weight vector ; Calculate the new individuals separately and the current population Individuals on each weight vector The PBI aggregation function value, if The PBI aggregation function value is less than The PBI aggregation function value will then be As the updated individual on the corresponding weight vector, otherwise discard. and will As individuals updated on the corresponding weight vector, the updated population is obtained; the next iteration is performed based on the updated population until the switching condition is met, and the updated population obtained at this time is used as the heuristic solution population. The heuristic solution population is used as the initial parent population of the second stage for multiple population iterations until the preset number of iterations is reached, and the population obtained in the last iteration is used as the set of optimization schemes for multi-source solid waste recycling.

[0007] Furthermore, the heuristic solution population is described as including: The second phase During the next population iteration, simulated binary crossover and polynomial mutation operations are performed on each individual in the parent population to generate the offspring population. The offspring population and the parent population are merged and elite individuals are selected to form the next generation population. Let the next generation population be used as the parent population, and Repeat the above steps for the next population iteration, until the preset number of iterations is repeated.

[0008] Furthermore, in each iteration of the second phase, the crossover probability in the binary crossover operation is simulated. and cross index And the mutation probability in polynomial mutation operations. and variation index The optimization is achieved using deep reinforcement learning; where, during the deep reinforcement learning optimization process, the state of the input agent is... Actions output by the intelligent agent and corresponding rewards They are respectively:

[0009]

[0010]

[0011] in, It is a population The hypervolume index is used to characterize the assessment of population convergence; the larger the value, the better the population convergence. It is a population Spacing and breadth indicators are used to characterize the assessment of population diversity; the smaller the value, the better the population diversity. Indicating the second stage The population of the nth iteration is used to characterize the nth iteration. The quality of the treatment method used for each type of solid waste obtained after the next iteration; It is a bonus factor that improves convergence and rewards. It is a bonus factor that enhances the diversity of rewards.

[0012] Furthermore, and The calculation method is as follows:

[0013] In the formula, As a preset reference point, express The volume of a set in a dimensional target space. For population Any individual in the group, Indicates population There exists an individual No more than in all targets , This represents taking the union of sets;

[0014] In the formula, Representing the solution To its most recent solution Euclidean distance, solution Indicates population One of the individuals, ; and These represent the current population at the [number]th [time]. l The maximum and minimum values ​​on each target. This is the initial population size.

[0015] Furthermore, the switching condition is: compared with the population in the g-th iteration. Recent The average rate of change of the convergence index of the generation population is less than the preset switching threshold. The convergence index is as follows:

[0016] In the formula, The initial population size is set. and They represent the current population. In the l The maximum and minimum values ​​on each target. Indicates population The first in Individual, express In the l The target value for each objective is calculated based on the objective function of the multi-objective mathematical optimization model.

[0017] Furthermore, the optimization objective of the multi-objective mathematical optimization model is to minimize economic costs and environmental impact; The objective function of the multi-objective mathematical optimization model includes the economic cost objective function. and environmental impact objective function The economic cost objective function and the environmental impact objective function They are respectively:

[0018]

[0019] In the formula, It is the first The first type of solid waste adopts the The unit processing cost of each processing method It is the first The one-time construction cost of this treatment method It is the first The service life of this treatment method The discount rate is... The number of days in a year. It is the first The first type of solid waste adopts the The unit carbon emissions processed by this type of treatment method; Operational constraints include:

[0020]

[0021]

[0022] In the formula, It is the first The total amount of each type of solid waste It refers to the number of types of solid waste. It is the first The maximum processing capacity of this processing method It refers to the number of processing methods; It is the upper limit for each decision variable.

[0023] Furthermore, the PBI aggregation function value for each individual in the population is calculated using the following formula. :

[0024]

[0025]

[0026] in, Represents an individual to weight vector The projection distance, Represents an individual to weight vector vertical distance, Let be the objective function of the multi-objective mathematical optimization model. , Represents an individual a target value of the ideal point; a target value of the ideal point, is an adjustable parameter; take or , , denotes the 2-norm of a vector.

[0027] The application further provides a multi-source solid waste recycling path optimization system, comprising a computer readable storage medium and a processor; The computer readable storage medium is used for storing executable instructions; The processor is used for reading the executable instructions stored in the computer readable storage medium to execute the multi-source solid waste recycling path optimization method.

[0028] The application further provides a computer readable storage medium, which has a computer program stored thereon, and the program is executed by a processor to realize the multi-source solid waste recycling path optimization method.

[0029] The application further provides a computer program product, which comprises a computer program, and when the computer program is run on a computer, the computer program makes the computer execute the multi-source solid waste recycling path optimization method.

[0030] Overall, the above technical scheme conceived by the application can achieve the following beneficial effects: (1) The application adopts a two-stage multi-objective optimization method to optimize the multi-source solid waste recycling path, and introduces heuristic solution (i.e., the parent population in the initial iteration of the second stage) search, which can better balance the convergence and diversity, two core objectives of multi-objective optimization. Specifically, in the first stage, heuristic solution (representing a better multi-source solid waste recycling path) search is mainly performed, and by constructing extreme weight vectors and a center weight vector, in the search process of the heuristic solution, extreme weight vectors (extreme weight vectors) located on the boundary of the Pareto front correspond to the optimization model The central weight vector uniformly sets the weight on each target, corresponds to the central region of the Pareto front, so that the optimization process can search in the direction of the corresponding weight vector, uniformly cover different regions in the target space, avoid the heuristic solution from being too dense and concentrated in a certain region of the target space, and make the population quickly converge to the Pareto front or the surrounding region, so that the high-quality heuristic solution population can provide diversified position information for the second-stage offspring population when the second-stage population is guided by the heuristic solution population, to guide the offspring population to search for optimal solutions in different regions of the target space, maintain good diversity of solutions, avoid the population from falling into a local optimal region, and further converge and cover the Pareto front, so as to obtain an optimal population with good convergence and diversity, which is beneficial to find a better multi-source solid waste recycling path.

[0031] (2) Further, considering that the dependence of the simulated binary crossover and polynomial mutation genetic operators on parameters makes them inefficient in solving complex optimization problems, and the tuning of optimal parameters of the operators is also a difficult problem in practical applications, the present application introduces reinforcement learning into the multi-source solid waste recycling path optimization, adopts an adaptive parameter control mechanism based on deep reinforcement learning in the second stage, dynamically adjusts the four important parameters of the simulated binary crossover operator and the polynomial mutation operator, namely the crossover probability, the mutation probability, the crossover index and the mutation index, avoids the problems of premature convergence or low search efficiency of the algorithm caused by improper parameter setting, improves the search ability and convergence speed of the algorithm in the multi-source solid waste recycling path optimization problem, and can find a better treatment path scheme.

[0032] (3) As a preferred, the deep reinforcement learning algorithm DDPG is adopted, and continuous parameter combinations are generated in real time according to the state information of the current population, so as to realize continuous adjustment of the parameters of the genetic operators. Compared with the existing method of dynamically adjusting the parameters of the optimization algorithm by using Q-learning and DQN algorithm, the method of the present application can fully exert the performance of reinforcement learning in the multi-source solid waste treatment path optimization problem, and is more beneficial to find a better multi-source solid waste recycling path optimization scheme.

[0033] (4) As a preferred, the designed SS index can measure the coverage and uniformity of the solution on the Pareto front at the same time, provide a more comprehensive diversity evaluation, and be beneficial to find a better treatment path scheme.

[0034] (5) Further, in order to solve the problem of waste of computing resources caused by the static switching method of the existing two-stage method, the application designs a dynamic switching method based on population fitness. When the average change rate of the convergence degree index of the last k generations of the current iteration population is less than the preset switching threshold, the method will switch from the first stage to the second stage, preventing the algorithm from stagnating in the first stage after obtaining a high-quality heuristic solution and avoiding unnecessary waste of computing resources.

[0035] Overall, the two-stage optimization method of the multi-source solid waste recycling path of the application uses a weight vector guided heuristic solution search, which not only significantly improves the convergence speed of the heuristic solution population, but also maintains good population diversity, avoiding the provision of too concentrated heuristic solution information that leads the algorithm to fall into a local optimal region. Through the dynamic switching method based on population fitness, the convergence of the heuristic solution population is effectively evaluated, and the stage switching is dynamically controlled to avoid wasting too much computing resources. The adaptive parameter control mechanism based on deep reinforcement learning in the second stage dynamically adjusts the operator parameters through deep reinforcement learning technology, which can optimize the key parameter settings of the operator in real time according to the running state of the algorithm, avoiding premature convergence or low search efficiency of the algorithm caused by improper parameter settings, significantly improving the search ability and convergence speed of the algorithm in the multi-source solid waste recycling path optimization problem, and finding a better treatment path scheme. The method of the application can adaptively and dynamically adjust according to the real-time changes of various factors in the multi-source solid waste recycling process and the running state of the algorithm, has stronger dynamic adaptability, ensures the accuracy and effectiveness of the optimization result, and improves the efficiency of multi-objective optimization. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 is a multi-source solid waste recycling path optimization method schematic diagram proposed in the embodiment of the application; Figure 2 is a second stage optimization schematic diagram of the two-stage multi-objective optimization method proposed in the embodiment of the application; Figure 3 is a multi-source solid waste recycling path optimization system schematic diagram of the application. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical scheme and advantages of the application more clear, the application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the application and not to limit the application. In addition, the technical features involved in each embodiment of the application described below can be combined with each other as long as they do not conflict with each other.

[0038] Embodiment 1 As Figures 1-2As shown, the embodiment of the application provides a multi-source solid waste recycling path optimization method, comprising: (1) From the aspects of economic cost and environmental impact, a multi-source solid waste recycling system multi-objective mathematical optimization model is established, wherein the objective functions include an economic cost objective function and an environmental impact objective function, and both are minimum objective functions.

[0039] (2) Decision variables, i.e. population individuals, are set to represent a solution of an optimization problem (in the application, a recycling path of multi-source solid waste, including the quality of the treatment method adopted for each solid waste), and a two-stage multi-objective optimization method guided by heuristic solutions is used to optimize the recycling path of multi-source solid waste, wherein in the first stage, the population is responsible for the rapid search of heuristic solutions (characterizing a better recycling path of multi-source solid waste); in the second stage, the population is responsible for further convergence and extension on the basis of the heuristic solutions to obtain better convergence and diversity; and a dynamic switching method based on the fitness of the population is used for stage switching.

[0040] (3) The second stage uses a parameter adaptive control mechanism based on deep reinforcement learning to dynamically adjust the operator parameters, and the agent obtains the population state information in the evolution process (iteration process) by interacting with the environment, generates action output through the Actor network, and obtains genetic operator parameters; on this basis, the population evolves to the next generation, and returns the new population state information of the agent and the immediate reward signal. After each iteration, the experience is collected into the experience replay buffer.

[0041] (4) Every d iterations, a batch of experiences are taken from the experience replay buffer to train the Critic network and the Actor network of the agent, and the target Critic network and the target Actor network are parameter-soft-updated at regular intervals; a noise-based exploration strategy is introduced in the network training process to achieve a balance between exploration and utilization; the trained DDPG algorithm (deep reinforcement learning algorithm) is used to adaptively adjust the genetic operator parameters to achieve efficient multi-objective optimization of the multi-source solid waste treatment path.

[0042] As a specific implementation, the economic cost minimization objective function and the environmental impact minimization objective function are as follows:

[0043]

[0044] wherein, wherein is the economic cost objective function, including the treatment cost and the construction cost, is the treatment cost of the i-th solid waste, is the construction cost of the i-th solid waste, ​A unit processing cost of a processing mode, is the first The quality of the solid waste using the first processing mode (decision variable), is the one-time construction cost of the first processing mode, is the service life of the first processing mode, is the discount rate, is the number of days in a year; is the environmental impact objective function, which is composed of processing carbon emissions, is the unit processing carbon emission of the first solid waste using the first processing mode. In other embodiments, the objective function can also be modeled according to actual needs, such as resource recovery rate, etc. Correspondingly, the target dimension can be other values.

[0045] In terms of running constraints, the embodiment of the present application considers solid waste mass balance constraints, processing capacity constraints of each processing mode, and non-negative constraints of decision variables, etc. to establish a multi-source solid waste recycling system multi-objective mathematical optimization model. Specifically, the mass balance constraint wherein is the total amount of the first type of solid waste, is the number of solid waste types; the processing capacity constraint wherein is the maximum processing capacity of the first processing mode, is the number of processing modes; the decision variable constraint wherein is the upper limit of each decision variable.

[0046] In the embodiment of the present application, the heuristic solution search method is: The decomposition idea of the improved MOEA / D algorithm (decomposition-based multi-objective evolutionary algorithm) is used to decompose the multi-objective optimization problem into several single-objective sub-problems for heuristic solution search. First, a set of uniformly distributed weight vectors is constructed. For a multi-objective optimization problem with objectives, the embodiment of the present application considers constructing weight vectors, including extreme weight vectors ( ) and a central weight vector : Extreme weight vector: Weight vector 1:

[0047] Weight vector 2:

[0048]

[0049] Weight vector :

[0050] One central weight vector: Weight vector :

[0051] in, Each extreme weight corresponds to in The optimal solution for each different objective, with the number of elements in each extreme weight vector being... There is one and only one element that is 1, and all the other elements are 0; the central weight vector sets the same weight for each target. In the first stage of the two-stage multi-objective optimization method, a weight vector guides the randomly initialized solution to evolve towards a heuristic solution. Compared to the traditional MOEA / D method, this embodiment of the invention only employs... A high-quality search for heuristic solutions can be achieved with one extreme weight vector and one central weight vector, which also improves the search efficiency on these weight vectors in the first stage.

[0052] The two-stage multi-objective optimization method in this embodiment of the invention, in the first stage, a heuristic solution search is performed under the guidance of a weight vector: firstly, the population is randomly initialized, and each individual in the population represents a decision variable. That is, the first The first type of solid waste adopts the The quality of solid waste treatment methods; in each iteration, simulated binary crossover and polynomial mutation are used to generate offspring individuals, based on each weight vector. ( The multi-objective optimization problem is transformed into a series of single-objective subproblems through the PBI aggregation method. Specifically, in the g-th iteration, for each weight vector, from the current population... The selected individuals are generated using simulated binary crossover and polynomial mutation to produce new offspring individuals on each weight vector. Calculate the population separately Old individuals on each weight vector PBI aggregation function value and new individuals PBI aggregation function value The population is updated based on the PBI aggregation function values ​​of the old and new individuals: if Then use the corresponding weight vector The new individual replaces the corresponding weight vector in the population. Otherwise, retain the corresponding weight vector in the population. For old individuals, discard the corresponding weight vector. The new individuals are calculated using the following formula:

[0053]

[0054]

[0055] in, Solution (population) (one of the entities) to the weight vector The projection distance, Solution to weight vector vertical distance, Let be the objective function. , Solution The target value; This represents the target value at the ideal point. It is an adjustable parameter, with an empirically determined value; Pick or . This represents the 2-norm of a vector, also known as the Euclidean norm.

[0056] The next iteration is performed based on the updated population until the switching conditions are met, thus entering the second stage of evolution. To address the problem of wasted computational resources caused by existing static switching methods that rely solely on algorithm processes (such as setting a maximum number of iterations, and entering the second stage when the maximum number of iterations is reached), this invention proposes a dynamic switching method based on population fitness. When the conditions for dynamic switching are met in the first stage of iteration, the second stage of evolutionary process begins.

[0057] Specifically, the dynamic switching method based on population fitness is as follows: The first stage of the two-stage optimization method of this invention focuses more on the convergence performance of the population. Better convergence provides a higher quality initial parent population for the second stage. Therefore, in this embodiment, a convergence index is used as the fitness of the population to evaluate the convergence during the population evolution process and dynamically adjust the timing of the switch from the first stage to the second stage. The convergence index is defined as:

[0058] in The initial population size is set. For the target dimension, and These represent the current population at the [number]th [time]. l The maximum and minimum values ​​on each target. Indicates population The first in Individual, express In the l The target value on each objective, according to the objective function get.

[0059] Based on the above convergence index, the dynamic switching condition in this embodiment of the invention is: when the population in the g-th iteration... recent The average rate of change of the convergence index of the generation is less than the preset switching threshold. If the algorithm fails, it will switch from the first stage to the second stage; otherwise, the algorithm will continue to execute the first stage.

[0060] As a specific implementation method, this dynamic switching condition can be expressed as:

[0061] in The switching threshold is a very small constant, determined empirically. This indicates the current iteration number in the first phase. Indicates the closest to the population in the g-th iteration. The next iteration.

[0062] After the switching conditions are met, the second stage begins. At this stage, the solution obtained in the first stage (each individual in the population) is used as a heuristic solution. During the evolutionary process in the second stage, information on high-quality individuals from the heuristic solution population is used to generate offspring populations through simulated binary crossover and polynomial mutation operations. The offspring populations are then merged with the parent populations and selected using the SPEA2 environment selection method. Elite individuals form the next generation population, and this process iterates until the termination condition of evolution (the preset number of evolutions, Gen) is met. The population from the last iteration is output as the optimal Pareto solution set. Each Pareto optimal solution is a feasible optimal solution for multi-objective optimization of multi-source solid waste recycling paths, thus obtaining the optimal configuration scheme set for multi-source solid waste recycling. From the obtained optimal configuration scheme set, the preferred optimal solid waste treatment path is selected.

[0063] In this embodiment of the invention, the second stage uses a parameter adaptive control mechanism based on deep reinforcement learning to dynamically adjust the parameters of the simulated binary crossover operator and the polynomial mutation operator. The agent obtains population state information during the evolution process by interacting with the environment, generates action output through its Actor network, and obtains genetic operator parameters. On this basis, the population evolves to the next generation and returns new population state information and immediate reward signals to the agent.

[0064] In this embodiment of the invention, the optimization process of the second-stage algorithm for the multi-source solid waste recycling problem is constructed as an environment, and in the second stage... In each iteration, the optimal state of the population is obtained. The agent's Actor network receives the current state as input and outputs action vectors. This allows us to obtain the crossover probabilities in the simulated binary crossover operator and the polynomial mutation operator. Probability of mutation Cross-index Variation index Four key parameters. After obtaining the next generation population through the SPEA2 environment selection method, the agent interacts with the environment to obtain the optimal state of the new population. And calculate the reward signal. After each iteration, experience is collected. The tuples are placed into the experience replay buffer. In this embodiment of the invention, the size of the experience replay buffer is set to... .

[0065] Every *d* iterations, a batch of experience tuples is retrieved from the experience replay buffer for training the Critic and Actor networks, with periodic soft updates to the parameters of the target Critic and Actor networks. A noise-based exploration strategy is introduced during network training to achieve a trade-off between exploration and utilization. The trained DDPG algorithm (the trained agent) adaptively adjusts the parameters of the simulated binary crossover and polynomial mutation operators, achieving efficient multi-objective optimization of multi-source solid waste recycling paths. During the training of the Critic and Actor networks, the parameters of the target Critic and Actor networks are soft-updated after each learning interval.

[0066] In this embodiment of the invention, the Critic network learns its network parameters by minimizing the TD error. The Critic network uses a commonly used loss function, and the Actor network updates using policy gradients, which will not be elaborated further here. For the learning of the target network, a soft update method is used.

[0067]

[0068] wherein, are parameters of the Critic target network, are parameters of the Actor target network, are parameters of the Critic network, are parameters of the Actor network, is a soft update rate, usually 0.005. The soft update of the target network makes its parameters slowly close to the main network, which helps the training stability.

[0069] Specifically, in reinforcement learning, the state, action and immediate reward signal are set as: The state is set as wherein is the hyper-volume indicator of the population is the spread and diversity indicator between populations, both are continuous variables, which focus on the convergence and diversity of the population respectively, represents the population of the th iteration, which is used to represent the quality of the treatment method used for each solid waste after the th iteration; the action is set as wherein , are the crossover probability and crossover index parameters of the simulated binary crossover operator respectively, , are the mutation probability and mutation index parameters of the polynomial mutation respectively, which have an important influence on the performance of the operator, and are both continuous variables; the immediate reward signal is set as wherein is the gain factor of the convergence improvement reward, is the gain factor of the diversity improvement reward, which is empirically valued. In the embodiments of the present application, the HV indicator is represented as:

[0070]

[0071] wherein is a preset reference point, represents the volume of the set in the dimensional target space, is any individual in the population represents that there is an individual in the population does not exceed on all targets, ​​​represents the union. The larger the value of the HV indicator, the better the convergence of the population, i.e. the better the performance of the multi-objective optimization method.

[0072] In order to measure the coverage and uniformity of the solution on the Pareto front simultaneously, the SS indicator designed in the embodiment of the present application is:

[0073] wherein represents the solution to the nearest solution of the Euclidean distance, represents the average value of all nearest neighbor distances, the solution represents one individual in the population ; and respectively represent the maximum value and the minimum value of the current population on the first l target and the second target. is the number of solutions in the solution set, which is consistent with the initially set population size. The smaller the value of the SS indicator, the better the population diversity of the multi-objective optimization method, i.e. the better the performance of the multi-objective optimization method.

[0074] Embodiment 2 As shown in Figure 3 , the embodiment of the present application provides a multi-source solid waste recycling path optimization system, comprising: a model construction module, configured to receive economic cost data, environmental impact data and operation information, analyze the economic cost and environmental pollution of the solid waste recycling process, and establish a multi-source solid waste recycling path multi-objective optimization model. The multi-objective optimization model includes an economic benefit minimization objective function, an environmental impact minimization objective function and necessary operation constraint conditions.

[0075] A two-stage optimization algorithm module is used to design a two-stage multi-objective optimization algorithm based on the established multi-source solid waste recycling path multi-objective optimization model, the decision variable is designed as the mass of each solid waste allocated to different treatment methods, the decision variable is used as an individual in the population, and a new population is generated through the operations of simulated binary crossover and polynomial mutation operators. In the first stage, the decomposition idea of the MOEA / D algorithm is used to guide the search of the heuristic solution population using the weight vector. The stage switching is performed through the dynamic switching method based on the population fitness. In the second stage, an adaptive parameter control mechanism based on deep reinforcement learning is used to dynamically adjust the operator parameters, and the next generation population is selected through the SPEA2 environmental selection method after the new population and the previous generation population are combined, so as to maintain the diversity of the algorithm and cope with the irregular Pareto frontier problem. This iteration is continued until the allowed maximum number of iterations is reached, and the optimal Pareto solution set is obtained by optimization, wherein each Pareto optimal solution is a feasible optimal scheme of multi-source solid waste recycling path multi-objective optimization. Receive external decision preference information, and select a preferred solid waste optimal treatment path from the Pareto optimal solution set obtained by the multi-objective optimization algorithm.

[0076] An experience replay module is used to collect experience tuples in the population evolution process Put into the experience replay buffer In the experience replay buffer, the agent takes out a batch of experience tuples for training and learning of the Actor network and the Critic network in the DDPG algorithm each time the network parameters are updated.

[0077] An adaptive parameter control module is used to design a DDPG reinforcement learning algorithm, the values of the population on the HV and SS indicators are used as continuous states, the crossover probability, the crossover index, the mutation probability and the mutation index constitute a continuous action space. The reward is set as the negative value of the weighted sum of the average values of the population on each target, and the sign indicates that the minimization target is converted into a maximization reward. In each iteration process, the DDPG algorithm generates an action vector, and a reasonable range of genetic operator parameters is generated through mapping. During the training and learning process of the Critic network and the Actor network, the parameters of the target Critic network and the target Actor network are updated softly after each learning interval.

[0078] The related technical solutions are the same as those of Embodiment 1, which will not be repeated here.

[0079] Embodiment 3 The embodiment of the present application provides a multi-source solid waste recycling path optimization system, which comprises a memory and a processor, the memory stores a computer program, and the processor realizes the steps of the multi-source solid waste recycling path optimization method in Embodiment 1 when executing the computer program.

[0080] The related technical solutions are the same as above and will not be repeated here.

[0081] Embodiment 4 The embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the steps of the multi-source solid waste recycling path optimization method in the above embodiment 1.

[0082] Specifically, the memory can include a high-speed random access memory, and can also include a non-volatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid state memory devices.

[0083] The related technical solutions are the same as above and will not be repeated here.

[0084] Embodiment 5 The embodiment of the present application provides a computer program product, which comprises a computer program, and when the computer program runs on a computer, the computer program makes the computer execute the steps of the multi-source solid waste recycling path optimization method in the above embodiment 1.

[0085] The related technical solutions are the same as above and will not be repeated here.

[0086] Those skilled in the art can easily understand that the above description is only a preferred embodiment of the present application, and is not intended to limit the present application, and any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A multi-source solid waste recycling path optimization method, characterized in that, The method comprises the following steps: A multi-source solid waste recycling optimization scheme set is obtained by solving a multi-objective mathematical optimization model of a multi-source solid waste recycling system by using a two-stage multi-objective optimization method; An optimal solid waste treatment path based on preferences is selected from the optimization scheme set; the two-stage multi-objective optimization method comprises the following steps: In the first stage, the decision variables of the multi-objective mathematical optimization model are determined as population individuals, for the first type of solid waste using the first type of solid waste treatment; a weight vector is constructed including an extreme weight vector and a central weight vector , wherein , , , , is the number of objectives to be optimized of the multi-objective mathematical optimization model; During the g-th population iteration, from the current population... Select individuals for crossover and mutation to generate new individuals on each weight vector ; Calculate the new individuals separately and the current population Individuals on each weight vector The PBI aggregation function value, if The PBI aggregation function value is less than The PBI aggregation function value will then be As the updated individual on the corresponding weight vector, otherwise discard. and will As individuals updated on the corresponding weight vector, the updated population is obtained; the next iteration is performed based on the updated population until the switching condition is met, and the updated population obtained at this time is used as the heuristic solution population. The heuristic solution population is used as the initial parent population of the second stage for multiple population iterations until the preset number of iterations is reached, and the population obtained in the last iteration is used as the set of optimization schemes for multi-source solid waste recycling.

2. The multi-source solid waste recycling path optimization method according to claim 1, characterized in that, The heuristic solution population is used as an initial parent population of the second stage for multiple population iterations, which comprises the following steps: In the second stage, the first population is iterated Simulated binary crossover and polynomial mutation operations are applied to each individual in the parent population to generate a child population, which is combined with the parent population and filtered to select elite individuals to form the next generation. Let the next generation population be used as the parent population, and Repeat the above steps for the next population iteration, until the preset number of iterations is repeated.

3. The multi-source solid waste recycling path optimization method according to claim 2, characterized in that, In each iteration of the second stage, simulate the crossover probability in the binary crossover operation and the crossover index , and the mutation probability in the polynomial mutation operation and the mutation index are obtained by deep reinforcement learning optimization; wherein, in the deep reinforcement learning optimization process, the state of the input agent , the action output by the agent and the corresponding reward are respectively: wherein, is a hypervolume indicator of the population for characterizing the evaluation of the convergence of the population, the greater the value the better the convergence of the population; is a distance and spread indicator of the population for characterizing the evaluation of the diversity of the population, the smaller the value the better the diversity of the population; denotes the population of the -th iteration in the second phase, for characterizing the quality of the treatment adopted for each solid waste after the -th iteration; is a gain factor of the convergence improvement reward, is a gain factor of the diversity improvement reward.

4. The multi-source solid waste recycling path optimization method according to claim 3, characterized in that, and The calculation is as follows: wherein is a predetermined reference point, denotes the volume of a set in the target space, is a population of any one individual, denotes a population of one individual does not exceed , denotes the union; wherein represents the solution to its nearest solution in the Euclidean distance, solution represents an individual in the population ; ; and represent the maximum and minimum values of the current population on the l th objective, respectively, is the initial set population size.

5. The multi-source solid waste recycling path optimization method according to any one of claims 1-4, characterized in that, The switching condition is that a convergence index of the population of the gth iteration is less than a preset switching threshold Recently The convergence index of the population of the gth iteration is less than a preset switching threshold ; wherein the convergence index is wherein is the initial set population size, and denote the current population maximum and minimum values on the l th objective, denotes the th individual in the population th objective, denotes the objective value on the l th objective, calculated from the objective function of the multi-objective mathematical optimization model.

6. The multi-source solid waste recycling path optimization method according to any one of claims 1-4, characterized in that, The optimization objective of the multi-objective mathematical optimization model is to minimize economic cost and environmental impact; The objective function of the multi-objective mathematical optimization model includes an economic cost objective function and an environmental impact objective function , wherein the economic cost objective function and the environmental impact objective function are respectively: In the formula, is the unit treatment cost of the first kind of solid waste using the first kind of treatment method, is the one-time construction cost of the first kind of treatment method, is the service life of the first kind of treatment method, is the discount rate, is the number of days in a year, is the unit treatment carbon emission of the first kind of solid waste using the first kind of treatment method; The operation constraints comprise the following steps: In the formula, It is the first The total amount of each type of solid waste It refers to the number of types of solid waste. It is the first The maximum processing capacity of this processing method It refers to the number of processing methods; It is the upper limit for each decision variable.

7. The multi-source solid waste recycling path optimization method according to claim 6, characterized in that, The PBI aggregation function value for each individual in the population is calculated using the following equation : wherein denotes the individual to the weight vector , denotes the individual to the weight vector , is the objective function of the multi-objective mathematical optimization model, , denotes the objective value of the individual ; denotes the objective value of the ideal point, is a tunable parameter; takes the value or , , denotes the 2-norm of a vector.

8. A multi-source solid waste recycling path optimization system, characterized in that, The computer readable storage medium comprises a computer readable storage medium and a processor; The computer readable storage medium is used to store executable instructions; The processor is used to read the executable instructions stored in the computer readable storage medium to execute the multi-source solid waste recycling path optimization method according to any one of claims 1-7.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the multi-source solid waste recycling path optimization method according to any one of claims 1-7.

10. A computer program product, characterised in that, The computer program is executed on a computer to enable the computer to execute the multi-source solid waste recycling path optimization method according to any one of claims 1-7.

Images