Method for optimizing injection mode of logistics tipping type tray sorting machine

Abstract

The invention relates to a logistics tipping type tray sorting machine injection mode optimization method, and belongs to the technical field of automatic logistics sorting systems. According to the method, the influence of three key control parameters, namely a virtual wireframe control mode, the number of queue conveyors and the running speed of the sorting machine, on the injection efficiency and the energy consumption when goods are injected into the tipping type tray sorting machine is comprehensively considered, and an injection mode optimization problem model of the logistics tipping type tray sorting machine is established; a simulation model for solving the injection mode optimization problem of the logistics tipping type tray sorting machine is constructed, and a DQN model is established based on the simulation model; and designing a dynamic adaptive multi-objective simulation butterfly optimization algorithm to solve the problem model. The method is established according to the actual operation condition of the general logistics tipping type tray sorting machine in various fields, the average cargo injection time can be effectively shortened, the system energy consumption can be balanced, and the operation efficiency of the logistics tipping type tray sorting machine can be improved.

Description

Technical Field

[0001] The invention relates to an optimization method for an injection mode of a logistics tilting tray sorter, and belongs to the technical field of logistics automated sorting. Background Art

[0002] With the development of new generation information technology and automation technology, the logistics automatic sorting system has become one of the core equipment of modern logistics distribution centers. Nowadays, as one of the key technologies of intelligent logistics, automatic sorting technology is widely used in express logistics centers, airport baggage handling systems, e-commerce storage centers, and tobacco, medicine, military and other picking-intensive distribution centers. The core centralized sorting machines used are cross-belt sorters, slider sorters, tilting tray sorters and swing wheel sorters. Through automation technology, efficient and accurate sorting operations can be achieved, the efficiency and accuracy of logistics processing can be improved, and traditional manual sorting can be gradually replaced. Therefore, by optimizing the key control parameters in the logistics automatic sorting system, the logistics automatic sorting system can be better controlled, and the problems of long logistics sorting time and high energy consumption cost of the logistics system can be effectively solved. It has become a research hotspot in the field of logistics automatic sorting system.

[0003] At present, the research of domestic and foreign scholars on the logistics automatic sorting system mainly focuses on the scheduling and allocation of limited resources, optimization of key system processes and integration of multiple information technologies, but lacks research on the impact of the injection stage of the logistics automatic sorting machine on the injection efficiency of goods. Therefore, the present invention proposes an injection mode optimization method for a logistics tilting tray sorting machine, which comprehensively considers the impact of three key control parameters, namely, the virtual wireframe control mode, the number of queue conveyors and the running speed of the sorting machine, on the injection efficiency and energy consumption of the logistics tilting sorting machine. Summary of the invention

[0004] The purpose of the present invention is to propose a method for optimizing the injection mode of a logistics tilting tray sorting machine in view of the shortcomings of existing logistics automatic sorting machine research.

[0005] The technical solution adopted by the present invention to solve the technical problem is: first, a logistics tilt-tray sorting machine injection mode optimization problem model is established; then a simulation model of the logistics tilt-tray sorting machine injection mode optimization problem is constructed, and a DQN model is established based on the simulation model; finally, a dynamic adaptive multi-objective simulation butterfly optimization algorithm is designed to solve the problem model.

[0006] The optimization problem of the injection mode of the logistics tilt-tray sorter is to comprehensively consider the influence of three key control parameters, namely, the virtual wireframe control mode, the number of queue conveyors and the running speed of the sorter, on the injection efficiency and energy consumption of the logistics tilt-tray sorter when the goods are injected into the sorter, and use the Markov decision process to establish the model of the problem as follows:

[0007] With the goal of minimizing the average injection time of goods from the queue conveyor to the tilt-tray sorter at the inlet of each receiving area, the first optimization goal is established as follows:

[0008]

[0009] Where T represents the average injection time of goods from the queue conveyor to the tilt-tray sorter; i∈{1,2,…,n} represents the receiving area number, n represents the number of receiving areas; j∈{1,2,…,l} represents the number of the goods feed port, l represents the number of the goods feed ports; x ij ∈{1,2,…,K ij} represents the number of goods waiting to be injected into the sorter at the goods inlet j of the receiving area i, K ij represents the number of goods successfully injected into the sorter at the goods inlet j of the receiving area i; T xij Indicates the xth position of the cargo feed port j at the receiving area i ij The average injection time of each item from the queue conveyor to the sorter;

[0010] With the goal of minimizing the total energy consumption of the Tilt-Tray sorter and related conveying equipment, the second optimization goal is established as follows:

[0011]

[0012] Where, E represents the total energy consumption of the logistics tilt-tray sorter and related conveying equipment; E b represents the energy consumption of the sorting machine; d∈{1,2,…,m} represents the serial number of the relevant conveying equipment, and m represents the number of conveying equipment in the system; E d Indicates the energy consumption of related conveying equipment;

[0013] The state of the automatic sorting process by the Tilt-Tray sorter is defined as a vector containing multiple key control parameters and system performance indicators, and the state space is determined as follows:

[0014] S=[M w ,N c ,v y ,T,E]

[0015] Where S represents the state space; M w Indicates the virtual wireframe control mode; N c Indicates the number of queue conveyors; v y It represents the operating speed of the tilt-tray sorter; T represents the average injection time of goods; E represents the system energy consumption;

[0016] The adjustment of key control parameters is defined as an action, and the action space is determined as follows:

[0017] A=[ΔMw ,ΔN c ,Δv y ]

[0018] Where A represents the action space; ΔM w Indicates changing the virtual wireframe control mode; ΔN c Indicates increasing or decreasing the number of queue conveyors; Δv y Indicates increasing or decreasing the operating speed of the Tilt-Tray sorter;

[0019] The reward function takes into account the system injection efficiency and energy consumption, and is designed as follows:

[0020] R=ω1×T-ω2×|ΔE|

[0021] Where R represents the reward function; ω1 and ω2 are weight coefficients used to balance the importance of injection efficiency and system energy consumption. When the injection efficiency is improved and the system energy consumption fluctuation is small, a larger reward is given;

[0022] At the same time, the constraints of this problem are set as follows:

[0023]

[0024] Where V xij represents the volume of goods successfully injected into the Tilt-Tray sorter at the goods inlet j of the receiving area i; Indicates the minimum volume of cargo allowed to be injected; Indicates the maximum volume of cargo allowed to be injected;

[0025]

[0026] In the formula, represents the weight of goods successfully fed into the Tilt-Tray sorter at the goods inlet j of the receiving area i; Indicates the minimum weight of cargo allowed to be injected; Indicates the maximum weight of cargo allowed to be injected;

[0027]

[0028] In the formula, Indicates the capacity of each queue conveyor that can simultaneously receive goods; N w Indicates the capacity of each virtual wireframe to receive goods at the same time; N R Indicates the capacity of each tray on the Tilt-Tray sorter that can simultaneously receive goods;

[0029]

[0030] In the formula, Indicates that the goods apply for a pallet with a virtual wireframe from the tilt-tray sorter; A pallet with a virtual wireframe representing the actual loading of goods into a Tilt-Tray sorter;

[0031]

[0032] In the formula, Represents the number of empty pallets with virtual wireframes in a Tilt-Tray sorter;

[0033] v d ∈[v dmin ,v dmax ]

[0034] In the formula, v d Indicates the operating speed of the conveying equipment in the tilt-tray sorter; v dmin Indicates the minimum operating speed allowed for the conveying equipment; v dmax Indicates the maximum operating speed allowed for the conveying equipment;

[0035] N c ∈[N cmin ,N cmax ]

[0036] Where N c Indicates the number of queue conveyors; N cmin Indicates the minimum number of queue conveyors allowed to be set; N cmax Indicates the maximum number of queue conveyors allowed to be set;

[0037] T xijmin >T start

[0038] Where, T xijmin Indicates the time when the first cargo starts to be injected into the Tilt-Tray sorter in the simulation model; T start Indicates the start time of the simulation model setting;

[0039] T xijmax <T end

[0040] Where, T xijmax Indicates the time when the last cargo is injected into the tilt-tray sorter in the simulation model; T end Indicates the end time of the simulation model setting.

[0041] The specific description of constructing a simulation model for the optimization problem of the injection mode of the logistics tilt-tray sorting machine and establishing a DQN model based on the simulation model is as follows:

[0042] In the simulation model, information such as cargo information, model parameters, key control parameters and algorithm parameters are set, and a DQN model is constructed according to the simulation model. A large number of randomly generated initial key control parameter combinations are used for training. According to the Q value output by the DQN strategy network, multiple actions that may obtain high rewards are selected. The corresponding key control parameter combinations are used as part of the initial population of the subsequent simulation optimization algorithm. The other part is generated randomly. The key control parameter combinations represent the population individuals in the algorithm, which can be expressed as e=[M w ,N c ,v y ].

[0043] The specific description of the dynamic adaptive multi-objective simulation butterfly optimization algorithm to solve the problem model is as follows:

[0044] The generated initial population is fed back to the simulation model to obtain the target fitness value. The algorithm controls the simulation model to perform optimization iterations, and the algorithm termination condition determines whether to end the simulation. If the algorithm termination condition is met, the optimal solution set is output and an optimization plan is generated. If the algorithm termination condition is not met, the population is updated and fed back to the simulation model to continue the optimization iteration.

[0045] The steps of dynamic adaptive multi-objective simulation butterfly optimization algorithm are as follows:

[0046] Step 1: Generate the initial population by randomly combining the optimal key control parameter combination in the DQN training model;

[0047] Step 2: Pareto non-dominated sorting, sort the initial population according to the target fitness value, and output an initial Pareto solution set;

[0048] Step 3: Calculate the individual fragrance concentration, calculate the fragrance concentration of each individual corresponding to each target on the Pareto frontier surface according to the target fitness value, select two optimal butterfly individuals according to the corresponding target, and determine the global optimum;

[0049] Step 4: Generate random number rand1 and dynamic switching probability p d Compare and determine the search methods of the remaining butterflies; if rand1 <p d , go to Step5; otherwise go to Step6;

[0050] Step 5: Global search. The current butterfly individual executes the global search mechanism, approaches the global optimal individual to achieve update, and executes the elite retention strategy and neighborhood disturbance strategy for the global optimal individual to achieve the update of the global optimal individual. After each individual is updated, go to Step 9;

[0051] Step 6: Local search: the current butterfly individual executes the local search mechanism to generate a random number rand2 and the local search selection probability q s Compare to determine the local search method; if rand2 s , go to Step7; otherwise go to Step8;

[0052] Step 7: Random exploration: the current butterfly individual conducts random exploration in the neighborhood, and then goes to Step 9 after updating;

[0053] Step 8: Approaching neighboring individuals, the current butterfly individual and the neighboring individuals approach each other, and after updating, go to Step 9;

[0054] Step9: Form a new generation of population;

[0055] Step 10: Pareto non-dominated sorting: perform Pareto non-dominated sorting based on the target fitness values ​​of the individuals in the new generation population and output a current Pareto solution set;

[0056] Step 11: Determine whether the algorithm termination condition has been met. If so, proceed to Step 12; if not, proceed to Step 3 to continue iteration;

[0057] Step 12: Output the Pareto frontier solution set, combine the current Pareto solution set with the initial Pareto solution set to perform Pareto non-dominated sorting, output the Pareto frontier solution set, and the algorithm ends.

[0058] The beneficial effects of the present invention are:

[0059] 1. The optimization problem model of the injection mode of the logistics tilt-tray sorting machine described in the present invention is established based on the actual operating conditions of the general tilt-tray sorting machines in various fields, and can be applied to the automatic sorting systems related to the logistics field;

[0060] 2. The optimization problem model of the injection mode of the logistics tilt-tray sorter established by the present invention comprehensively considers multiple control parameters, and comprehensively considers the effectiveness and performance of the proposed method with multiple optimization objectives. Compared with the method considering a single influencing factor and a single optimization objective, its comprehensiveness is better;

[0061] 3. The simulation model of the problem-solving model proposed in the present invention can meet the injection requirements of the logistics tilt-tray sorting machine according to different combinations of key control parameters;

[0062] 4. The dynamic adaptive multi-objective simulation butterfly optimization algorithm designed by the present invention is adaptively adjusted and improved according to the characteristics of the simulation model, and can obtain a better optimal solution set and provide a variety of optimization schemes. BRIEF DESCRIPTION OF THE DRAWINGS ​

[0063] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described below with reference to the accompanying drawings, in which:

[0064] Figure 1 It is the overall flow chart of the present invention;

[0065] Figure 2 It is a schematic diagram of goods injection into the sorting machine for optimizing the injection mode of the logistics tilt-tray sorting machine proposed by the present invention;

[0066] Figure 3 This is a flow chart of constructing a DQN model based on a simulation model according to the present invention;

[0067] Figure 4 This is a flow chart of the dynamic adaptive multi-objective simulation butterfly optimization algorithm proposed in the present invention. DETAILED DESCRIPTION

[0068] Example 1: Figure 1-4 As shown in the figure, a method for optimizing the injection mode of a logistics tilt-tray sorting machine is proposed. First, a model of the optimization problem of the injection mode of a logistics tilt-tray sorting machine is established; then a simulation model of the optimization problem of the injection mode of the logistics tilt-tray sorting machine is constructed, and a DQN model is established based on the simulation model; finally, a dynamic adaptive multi-objective simulation butterfly optimization algorithm is designed to solve the problem model.

[0069] Aiming at the optimization problem of the injection mode of the logistics tilt-tray sorter, the influence of three key control parameters, namely the virtual wireframe control mode, the number of queue conveyors and the running speed of the sorter, on the injection efficiency and energy consumption of the logistics tilt-tray sorter is comprehensively considered, and the Markov decision process is used to establish the model of the problem as follows:

[0070] With the goal of minimizing the average injection time of goods from the queue conveyor to the tilt-tray sorter at the inlet of each receiving area, the first optimization goal is established as follows:

[0071]

[0072] Where T represents the average injection time of goods from the queue conveyor to the tilt-tray sorter; i∈{1,2,…,n} represents the receiving area number, n represents the number of receiving areas; j∈{1,2,…,l} represents the number of the goods feed port, l represents the number of the goods feed ports; x ij ∈{1,2,…,K ij} represents the number of goods waiting to be injected into the sorter at the goods inlet j of the receiving area i, K ij represents the number of goods successfully injected into the sorter at the goods inlet j of the receiving area i; T xijIndicates the xth position of the cargo feed port j at the receiving area i ij The average injection time of each item from the queue conveyor to the sorter;

[0073] With the goal of minimizing the total energy consumption of the Tilt-Tray sorter and related conveying equipment, the second optimization goal is established as follows:

[0074]

[0075] Where, E represents the total energy consumption of the logistics tilt-tray sorter and related conveying equipment; E b represents the energy consumption of the sorting machine; d∈{1,2,…,m} represents the serial number of the relevant conveying equipment, and m represents the number of conveying equipment in the system; E d Indicates the energy consumption of related conveying equipment;

[0076] The state of the automatic sorting process by the Tilt-Tray sorter is defined as a vector containing multiple key control parameters and system performance indicators, and the state space is determined as follows:

[0077] S=[M w ,N c ,v y ,T,E]

[0078] Where S represents the state space; M w Indicates the virtual wireframe control mode; N c Indicates the number of queue conveyors; v y It represents the operating speed of the tilt-tray sorter; T represents the average injection time of goods; E represents the system energy consumption;

[0079] The adjustment of key control parameters is defined as an action, and the action space is determined as follows:

[0080] A=[ΔM w ,ΔN c ,Δv y ]

[0081] Where A represents the action space; ΔM w Indicates changing the virtual wireframe control mode; ΔN c Indicates increasing or decreasing the number of queue conveyors; Δv y Indicates increasing or decreasing the operating speed of the Tilt-Tray sorter;

[0082] The reward function takes into account the system injection efficiency and energy consumption, and is designed as follows:

[0083] R=ω1×T-ω2×|ΔE|

[0084] Where R represents the reward function; ω1 and ω2 are weight coefficients used to balance the importance of injection efficiency and system energy consumption. When the injection efficiency is improved and the system energy consumption fluctuation is small, a larger reward is given;

[0085] At the same time, the constraints of this problem are set as follows:

[0086]

[0087] In the formula, represents the volume of goods successfully injected into the Tilt-Tray sorter at the goods inlet j of the receiving area i; Indicates the minimum volume of cargo allowed to be injected; Indicates the maximum volume of cargo allowed to be injected;

[0088]

[0089] In the formula, represents the weight of goods successfully fed into the Tilt-Tray sorter at the goods inlet j of the receiving area i; Indicates the minimum weight of cargo allowed to be injected; Indicates the maximum weight of cargo allowed to be injected;

[0090]

[0091] In the formula, Indicates the capacity of each queue conveyor that can simultaneously receive goods; N w Indicates the capacity of each virtual wireframe to receive goods at the same time; N R Indicates the capacity of each tray on the Tilt-Tray sorter that can simultaneously receive goods;

[0092]

[0093] In the formula, Indicates that the goods apply for a pallet with a virtual wireframe from the tilt-tray sorter; A pallet with a virtual wireframe representing the actual loading of goods into a Tilt-Tray sorter;

[0094]

[0095] In the formula, Represents the number of empty pallets with virtual wireframes in a Tilt-Tray sorter;

[0096] v d ∈[v dmin ,v dmax ]

[0097] In the formula, v d Indicates the operating speed of the conveying equipment in the tilt-tray sorter; vdmin Indicates the minimum operating speed allowed for the conveying equipment; v dmax Indicates the maximum operating speed allowed for the conveying equipment;

[0098] N c ∈[N cmin ,N cmax ]

[0099] Where N c Indicates the number of queue conveyors; N cmin Indicates the minimum number of queue conveyors allowed to be set; N cmax Indicates the maximum number of queue conveyors allowed to be set;

[0100]

[0101] In the formula, Indicates the time when the first cargo starts to be injected into the Tilt-Tray sorter in the simulation model; T start Indicates the start time of the simulation model setting;

[0102]

[0103] In the formula, Indicates the time when the last cargo is injected into the tilt-tray sorter in the simulation model; T end Indicates the end time set by the simulation model; this formula ensures that the time it takes for the last cargo in the simulation model to be injected into the tilt-tray sorter is less than the end time set by the simulation model;

[0104] The key control parameters considered in the optimization problem of the injection mode of the logistics tilt-tray sorter are described as follows:

[0105] (1) Virtual wireframe control method: M w

[0106] ①Specify virtual wireframe mode;

[0107] ②Mixed specified and random virtual wireframe modes;

[0108] ③Random virtual wireframe mode.

[0109] (2) Number of queue conveyors: N c

[0110] (3) Sorting machine running speed: v y

[0111] The specific description of constructing a simulation model for the optimization problem of the injection mode of the logistics tilt-tray sorting machine and establishing a DQN model based on the simulation model is as follows:

[0112] In the simulation model, information such as cargo information, model parameters, key control parameters and algorithm parameters are set, and a DQN model is constructed according to the simulation model. A large number of randomly generated initial key control parameter combinations are used for training. According to the Q value output by the DQN strategy network, multiple actions that may obtain high rewards are selected. The corresponding key control parameter combinations are used as part of the initial population of the subsequent simulation optimization algorithm. The other part is generated randomly. The key control parameter combinations represent the population individuals in the algorithm, which can be expressed as e=[M w ,N c ,v y ];

[0113] When establishing a DQN model, a deep neural network is constructed. The number of nodes in the input layer is consistent with the dimension of the state space, and the number of nodes in the output layer is consistent with the dimension of the action space. The hidden layer can adopt a multi-layer perceptron structure, with 64 neurons in the first layer and 32 neurons in the second layer. An experience replay buffer is set up to store the experience tuples (s, a, r, s′) of the interaction between the agent and the environment, where s is the current state, a is the action taken, r is the reward obtained, and s′ is the next state. The ε-greedy strategy is used to select actions. In the early stages of training, exploration is usually performed with a larger probability ε, that is, actions are randomly selected; as training progresses, ε gradually decreases, and the optimal action is selected more based on the Q value output by the policy network;

[0114] The specific description of the dynamic adaptive multi-objective simulation butterfly optimization algorithm to solve the problem model is as follows:

[0115] Feed the generated initial population back to the simulation model to obtain the target fitness value. The algorithm controls the simulation model to perform optimization iterations, and the algorithm termination condition determines whether to end the simulation. If the algorithm termination condition is met, the optimal solution set is output and an optimization plan is generated. If the algorithm termination condition is not met, the population is updated and fed back to the simulation model to continue the optimization iteration.

[0116] The steps of dynamic adaptive multi-objective simulation butterfly optimization algorithm are as follows:

[0117] Step 1: Generate the initial population by randomly combining the optimal key control parameter combination in the DQN training model;

[0118] Step 2: Pareto non-dominated sorting, sort the initial population according to the target fitness value, and output an initial Pareto solution set;

[0119] Step 3: Calculate the individual fragrance concentration, calculate the fragrance concentration of each individual corresponding to each target on the Pareto frontier surface according to the target fitness value, select two optimal butterfly individuals according to the corresponding target, and determine the global optimum;

[0120] The individual fragrance concentration is calculated by the following formula:

[0121] f=c·I α

[0122] In the formula, f represents the scent function; c is a constant, which represents the butterfly's ability to perceive scent; I represents the stimulus intensity, which is related to the individual fitness value; a is a power exponent, which is used to simulate the rate at which scent decays with distance;

[0123] On the Pareto frontier, select the two individuals with the strongest fragrance concentrations of T and E as the global optimal individuals;

[0124] Step 4: Generate random number rand1 and dynamic switching probability p d Compare and determine the search methods of the remaining butterflies; if rand1 <p d , go to Step5; otherwise go to Step6;

[0125] The dynamic switching probability p is determined by the following formula: d :

[0126] p d =p s -ω·(R max -R) / R max

[0127] In the formula, p d represents the dynamic switching probability; p s represents the static switching probability; ω represents the adaptive weight factor; R represents the current number of iterations; R max Indicates the maximum number of iterations;

[0128] Step 5: Global search. The current butterfly individual executes the global search mechanism, approaches the global optimal individual to achieve update, and executes the elite retention strategy and neighborhood disturbance strategy for the global optimal individual to achieve the update of the global optimal individual. After each individual is updated, go to Step 9;

[0129] Step 6: Local search: the current butterfly individual executes the local search mechanism to generate a random number rand2 and the local search selection probability q s Compare to determine the local search method. s , go to Step7; otherwise go to Step8;

[0130] The local search probability q is determined by the following formula: s :

[0131]

[0132] In the formula, q s ​represents the probability of local search selection; β1 and β2 represent the probability of local search selection in the early and late stages of algorithm search, respectively; R represents the current number of iterations; R max Indicates the maximum number of iterations;

[0133] Step 7: Random exploration: the current butterfly individual conducts random exploration in the neighborhood, and then goes to Step 9 after updating;

[0134] The current individual randomly changes a parameter value that is different from the original parameter;

[0135] Step 8: Approaching neighboring individuals, the current butterfly individual and the neighboring individuals approach each other, and after updating, go to Step 9;

[0136] Two adjacent individuals randomly select the same parameters to exchange parameter values;

[0137] Step9: Form a new generation of population;

[0138] Step 10: Pareto non-dominated sorting: perform Pareto non-dominated sorting based on the target fitness values ​​of the individuals in the new generation population and output a current Pareto solution set;

[0139] Step 11: Determine whether the algorithm termination condition has been met. If so, proceed to Step 12; if not, proceed to Step 3 to continue iteration;

[0140] Step 12: Output the Pareto frontier solution set, combine the current Pareto solution set with the initial Pareto solution set to perform Pareto non-dominated sorting, output the Pareto frontier solution set, and the algorithm ends.

[0141] Problem setting: The research object is a logistics automatic sorting system with 6 receiving areas, 16 cargo inlets in each receiving area, and 1 tilting tray sorter; a one-day sorting task with a total of 31,781 cargo at peak time; virtual wireframe control method M w Select from designated virtual wireframe mode, mixed designated, random virtual wireframe mode, and random virtual wireframe mode; the number of queue conveyors N c The value is in [1,5], unit: piece; sorting machine running speed v y The value is in [1.4,3], unit: m / s;

[0142] Algorithm parameter settings: determined by orthogonal experiments, where the population size S = 80, the static switching probability p s =0.6, maximum number of iterations R max =100,;

[0143] Table 1 gives the Pareto frontier solution set obtained by this method under the peak scale;

[0144] Table 1 Pareto frontier solutions obtained under peak scale

[0145] Target value 1 2 3 4 5 6 7 8 9 <![CDATA[F1(s)]]> 3.278 3.365 2.765 3.956 4.007 4.598 4.687 4.865 4.946 <![CDATA[F2(kW·h)]]> 1994.43 1954.06 1864.40 1763.85 1746.32 1699.49 1687.05 1595.79 1594.46

[0146] The specific embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for optimizing the injection mode of a logistics tilt-tray sorting machine, characterized in that: Firstly, a model of optimization problem of injection mode of logistics tilt-tray sorting machine is established; then, a simulation model of the optimization problem of injection mode of logistics tilt-tray sorting machine is constructed, and a DQN model is established based on the simulation model; finally, a dynamic adaptive multi-objective simulation butterfly optimization algorithm is designed to solve the problem model.

2. The method for optimizing the injection mode of a logistics tilt-tray sorter according to claim 1, characterized in that: The optimization problem of the injection mode of the logistics tilt-tray sorter is to comprehensively consider the influence of three key control parameters, namely, the virtual wireframe control mode, the number of queue conveyors and the running speed of the sorter, on the injection efficiency and energy consumption of the logistics tilt-tray sorter when the goods are injected into the sorter, and use the Markov decision process to establish the model of the problem as follows: With the goal of minimizing the average injection time of goods from the queue conveyor to the tilt-tray sorter at the inlet of each receiving area, the first optimization goal is established as follows: Where T represents the average injection time of goods from the queue conveyor to the tilt-tray sorter; i∈{1,2,…,n} represents the receiving area number, n represents the number of receiving areas; j∈{1,2,…,l} represents the number of the goods feed port, l represents the number of the goods feed ports; x ij ∈{1,2,…,K ij } represents the number of goods waiting to be injected into the sorter at the goods inlet j of the receiving area i, K ij Indicates the number of goods successfully injected into the sorter at the goods inlet j of the receiving area i; Indicates the xth position of the cargo feed port j at the receiving area i ij The average injection time of each item from the queue conveyor to the sorter; With the goal of minimizing the total energy consumption of the Tilt-Tray sorter and related conveying equipment, the second optimization goal is established as follows: Where, E represents the total energy consumption of the logistics tilt-tray sorter and related conveying equipment; E b represents the energy consumption of the sorting machine; d∈{1,2,…,m} represents the serial number of the relevant conveying equipment, and m represents the number of conveying equipment in the system; E d Indicates the energy consumption of related conveying equipment; The state of the automatic sorting process by the Tilt-Tray sorter is defined as a vector containing multiple key control parameters and system performance indicators, and the state space is determined as follows: S=[M w ,N c ,v y ,T,E] Where S represents the state space; M w Indicates the virtual wireframe control mode; N c Indicates the number of queue conveyors; v y It represents the operating speed of the tilt-tray sorter; T represents the average injection time of goods; E represents the system energy consumption; The adjustment of key control parameters is defined as an action, and the action space is determined as follows: A=[ΔM w ,ΔN c ,Δv y ] Where A represents the action space; ΔM w Indicates changing the virtual wireframe control mode; ΔN c Indicates increasing or decreasing the number of queue conveyors; Δv y Indicates increasing or decreasing the operating speed of the Tilt-Tray sorter; The reward function takes into account the system injection efficiency and energy consumption, and is designed as follows: R=ω1×T-ω2×|ΔE| Where R represents the reward function; ω1 and ω2 are weight coefficients used to balance the importance of injection efficiency and system energy consumption. When the injection efficiency is improved and the system energy consumption fluctuation is small, a larger reward is given; At the same time, the constraints of this problem are set as follows: In the formula, represents the volume of goods successfully injected into the Tilt-Tray sorter at the goods inlet j of the receiving area i; Indicates the minimum volume of cargo allowed to be injected; Indicates the maximum volume of cargo allowed to be injected; In the formula, represents the weight of goods successfully fed into the Tilt-Tray sorter at the goods inlet j of the receiving area i; Indicates the minimum weight of cargo allowed to be injected; Indicates the maximum weight of cargo allowed to be injected; In the formula, Indicates the capacity of each queue conveyor that can simultaneously receive goods; N w Indicates the capacity of each virtual wireframe to receive goods at the same time; N R Indicates the capacity of each tray on the Tilt-Tray sorter that can simultaneously receive goods; In the formula, Indicates that the goods apply for a pallet with a virtual wireframe from the tilt-tray sorter; A pallet with a virtual wireframe representing the actual loading of goods into a Tilt-Tray sorter; In the formula, Represents the number of empty pallets with virtual wireframes in a Tilt-Tray sorter; v d ∈[v dmin ,v dmax ] In the formula, v d Indicates the operating speed of the conveying equipment in the tilt-tray sorter; v dmin Indicates the minimum operating speed allowed for the conveying equipment; v dmax Indicates the maximum operating speed allowed for the conveying equipment; N c ∈[N cmin ,N cmax ] Where N c Indicates the number of queue conveyors; N cmin Indicates the minimum number of queue conveyors allowed to be set; N cmax Indicates the maximum number of queue conveyors allowed to be set; In the formula, Indicates the time when the first cargo starts to be fed into the Tilt-Tray sorter in the simulation model; T start Indicates the start time of the simulation model setting; In the formula, Indicates the time when the last cargo is injected into the tilt-tray sorter in the simulation model; T end Indicates the end time of the simulation model setting.

3. The method for optimizing the injection mode of a logistics tilt-tray sorter according to claim 1, characterized in that: The specific description of constructing a simulation model for the optimization problem of the injection mode of the logistics tilt-tray sorting machine and establishing a DQN model based on the simulation model is as follows: In the simulation model, information such as cargo information, model parameters, key control parameters and algorithm parameters are set, and a DQN model is constructed according to the simulation model. A large number of randomly generated initial key control parameter combinations are used for training. According to the Q value output by the DQN strategy network, multiple actions that may obtain high rewards are selected. The corresponding key control parameter combinations are used as part of the initial population of the subsequent simulation optimization algorithm. The other part is generated randomly. The key control parameter combinations represent the population individuals in the algorithm, which can be expressed as e=[M w ,N c ,v y ].

4. The method for optimizing the injection mode of a logistics tilt-tray sorter according to claim 1, characterized in that: The specific description of the dynamic adaptive multi-objective simulation butterfly optimization algorithm to solve the problem model is as follows: Feed the generated initial population back to the simulation model to obtain the target fitness value. The algorithm controls the simulation model to perform optimization iterations, and the algorithm termination condition determines whether to end the simulation. If the algorithm termination condition is met, the optimal solution set is output and an optimization plan is generated. If the algorithm termination condition is not met, the population is updated and fed back to the simulation model to continue the optimization iteration. The steps of dynamic adaptive multi-objective simulation butterfly optimization algorithm are as follows: Step 1: Generate the initial population by randomly combining the optimal key control parameter combination in the DQN training model; Step 2: Pareto non-dominated sorting, sort the initial population according to the target fitness value, and output an initial Pareto solution set; Step 3: Calculate the individual fragrance concentration, calculate the fragrance concentration of each individual corresponding to each target on the Pareto frontier surface according to the target fitness value, select two optimal butterfly individuals according to the corresponding target, and determine the global optimum; Step 4: Generate random number rand1 and dynamic switching probability p d Compare and determine the search methods of the remaining butterflies; if rand1 <p d , go to Step5; otherwise go to Step6; Step 5: Global search. The current butterfly individual executes the global search mechanism, approaches the global optimal individual to achieve update, and executes the elite retention strategy and neighborhood disturbance strategy for the global optimal individual to achieve the update of the global optimal individual. After each individual is updated, go to Step 9; Step 6: Local search: the current butterfly individual executes the local search mechanism to generate a random number rand2 and the local search selection probability q s Compare to determine the local search method; if rand2 s , go to Step7; otherwise go to Step8;​ Step 7: Random exploration: the current butterfly individual conducts random exploration in the neighborhood, and then goes to Step 9 after updating; Step 8: Approaching neighboring individuals, the current butterfly individual and the neighboring individuals approach each other, and after updating, go to Step 9; Step9: Form a new generation of population; Step 10: Pareto non-dominated sorting: perform Pareto non-dominated sorting based on the target fitness values ​​of the individuals in the new generation population and output a current Pareto solution set; Step 11: Determine whether the algorithm termination condition has been met. If so, proceed to Step 12; if not, proceed to Step 3 to continue iteration; Step 12: Output the Pareto frontier solution set, combine the current Pareto solution set with the initial Pareto solution set to perform Pareto non-dominated sorting, output the Pareto frontier solution set, and the algorithm ends.

Images