Dense warehousing operation optimization system

Abstract

The invention discloses a dense warehousing operation optimization system, which takes storage efficiency, warehouse gravity center and goods allocation density as targets, optimizes warehouse goods allocation layout according to the initial goods quantity of a warehouse, reasonably allocates access points in orders, and effectively avoids collision during multi-vehicle collaborative operation while ensuring the safety and high-efficiency storage of the warehouse. The task allocation between the shuttle vehicle and the elevator is reasonably scheduled by taking the minimum order completion time and energy consumption as the optimization target, and the efficient utilization of resources and the balanced distribution of operation loads are ensured, so that the phenomenon that a single vehicle is excessively busy or idle is reduced; according to the method, the shortest path and vehicle avoidance are taken as targets, an intelligent algorithm is adopted to plan an optimal driving route for each vehicle, a lengthy path is avoided, the collision risk and the congestion phenomenon between the vehicles are reduced, and a dynamic avoidance strategy of multi-vehicle collaborative operation is considered in path planning so as to ensure that the vehicles can efficiently complete tasks in a dense operation environment.

Description

Technical Field

[0001] The present invention belongs to the technical field of logistics, and more specifically, relates to a dense warehousing operation system. Background Art

[0002] The dense warehousing system consists of high-density three-dimensional shelves, shuttle cars, elevators and conveyor roller beds. The system structure is as Figure 1 shown. The storage locations in the warehouse and the tracks are arranged in a grid pattern. The shuttle cars can travel bidirectionally within the tracks, cooperate with the elevators to achieve cross-level transportation of goods, and can also achieve cross-aisle operations on the same level, so as to complete the tasks of storing and retrieving goods at any target storage location. The shuttle cars use a pallet structure to carry goods, run to the bottom of the pallet, and move the pallet and the goods as a whole in the horizontal direction. The vertical transfer of goods is completed by the elevator, and the elevator is installed at the end positions of the horizontal and vertical tracks (). The intersection of the first layer of the shelf storage and the elevator is the entrance / exit (I / O port) of the pallet. There is a conveyor roller bed beside the I / O port for the transfer of goods in and out, and the inbound and outbound operations are completed.

[0003] With the continuous increase in the storage locations in the dense warehouse, unreasonable storage location allocation will lead to problems such as low warehousing operation efficiency, low warehouse safety, and traffic jams of shuttle cars caused by concentrated storage locations. Secondly, in the multi-vehicle system when dealing with multiple inbound and outbound tasks, problems of unreasonable scheduling often occur. Especially in the collaborative operation of shuttle cars and elevators, if the scheduling is improper, it may lead to an extension of the order completion time. In addition, the operation frequency of the shuttle cars may be uneven among multiple tasks, which will reduce the efficiency of the overall system. Finally, in the environment of multiple shuttle cars operating on the same level, unreasonable path planning will cause a series of problems, such as too long operation time, long paths, and traffic jams of the vehicle flow, and may even lead to vehicle collisions, seriously affecting the operation efficiency of the system. Due to the limited warehousing space and the complexity of tasks, the path selection and avoidance strategies of shuttle cars in a narrow space are particularly important. Frequent congestion and avoidance not only increase the operation cost of the system, but also affect the timely completion of orders. Summary of the Invention

[0004] The purpose of the present invention is to propose an optimized dense warehousing operation system, which optimizes the operations of the dense warehousing system from three aspects: storage location allocation, multi-vehicle scheduling, and shuttle car path planning, so as to achieve the effect of improving the overall operation efficiency of the system.

[0005] The present invention is implemented by adopting the following technical solutions:

[0006] Propose an optimized dense warehousing operation system, including:

[0007] The storage location optimization unit is used to establish a storage location optimization target model with the shelf center of gravity, inbound and outbound efficiency, and storage density as the goals, and allocate orders to the storage locations in the intensive warehousing system based on the goods optimization target model;

[0008] The scheduling optimization unit is used to establish a shuttle - elevator scheduling model with the minimum order completion time and minimum energy consumption as the optimization goals, and solve the model based on the improved genetic algorithm to obtain the optimal scheduling plan;

[0009] The shuttle path planning unit is used to establish a path planning model with the shortest path and vehicle avoidance as the goals, and solve the model based on the A* algorithm to guide the decision - making of the DQN algorithm to obtain the optimal operation path of the shuttle.

[0010] In some embodiments of the present invention, the storage location optimization target model is:

[0011]

[0012] Among them, is the shelf center of gravity target function; m xyz represents the weight of the goods placed in the storage location (x rows, y columns, z layers), z represents the number of layers where the goods are stored, and L0 represents the height of each layer of the shelf;

[0013] is the target function of storage density, where R i represents the turnover rate of the i - th good, and R m is the average value of the turnover rates of all goods;

[0014] is the target function of inbound and outbound efficiency, where x i 、y i 、z i represent the coordinates of the i - th good, v x 、v y 、v z represent the driving speeds of the shuttle in the x, y, and z directions when performing the handling operation, and R i represents the turnover rate of the i - th good.

[0015] In some embodiments of the present invention, the shuttle - elevator scheduling model includes:

[0016] The minimum time model: f = ω1×minT makespan +ω2×minE t ; In the formula, ω1 and ω2 are the weighting coefficients of the minimum time and the minimum energy consumption; among them, is the horizontal movement time of the shuttle from the starting position to the elevator or the target position; is the waiting time of the shuttle for the elevator; For the vertical movement time of the elevator; For the horizontal movement time of the shuttle car from the elevator to the target position;

[0017] Minimum energy consumption model: E t = E St + E Lt ; where, E St = ∑E S is the total energy consumption of the shuttle car, and E Lt = ∑E L is the total energy consumption of the elevator.

[0018] In some embodiments of the present invention, obtaining the optimal scheduling scheme based on the improved genetic algorithm to solve the model includes:

[0019] Initializing the shelf state: Initialize the storage state of the shelf. If there is goods on the shelf, record the shelf state as 1, otherwise record it as 0;

[0020] Population initialization and coding: According to the working characteristics of the shuttle car and the elevator, design a chromosome with a three-layer structure; The first layer is the operation sequence coding of the shuttle car, with positive integers for inbound operations and negative integers for outbound operations; The second layer is the operation sequence coding of the shuttle car, and the operation tasks are randomly assigned according to the number of shuttle cars; The third layer is the operation sequence of the elevator, sorted according to the tasks executed by the shuttle cars with layer-changing requirements, and calculated based on the current position of the elevator and the operation relationship between the first layer and the second layer;

[0021] Chromosome decoding: Construct a chromosome with the operation information and operation sequence of the shuttle car and the elevator;

[0022] Fitness calculation: With the goal of the shortest time and the minimum energy consumption for the system to complete the task, use this goal as the fitness;

[0023] Genetic operators: Include selection, crossover, and mutation; In the selection design, the tournament selection method is adopted. First, randomly select a certain number of individuals in the population, compare the fitness of these individuals, and select the individual with the highest fitness to enter the next generation; In the crossover design, the two-point crossover method is adopted. Select two crossover points and exchange the chromosome segments between these two points, and set the crossover probability as p c , when the random probability p i of each group of chromosomes < p c perform the crossover operation; In the mutation operation, the adaptive mutation method is adopted, and the mutation rate is dynamically adjusted at different stages of the genetic algorithm. As the number of iterations increases, the mutation rate is reduced; The adaptive mutation rate calculation is implemented using , where P m1 represents the initial mutation rate, P m2 represents the final mutation rate, t represents the current number of iterations, and G represents the total number of iterations.

[0024] In some embodiments of the present invention, a priority scheduling mechanism is introduced into the A* algorithm, including:

[0025] Priority weight adjustment: When calculating the g value, a priority factor is introduced to adjust the path cost of vehicles with different priorities; g represents the path cost from the starting point to the current node, and p i represents the priority factor, and g′ represents the path cost after priority adjustment;

[0026] Processing when priorities are the same: If the priorities of two vehicles are the same, the order is determined by the distance h to the target location, and the vehicle closer to the target location has priority to pass.

[0027] In some embodiments of the present invention, a dynamic avoidance mechanism is introduced into the A* algorithm, including:

[0028] Introduce a real-time detection and path update mechanism. When expanding each node, detect whether the current node may overlap with the paths of other vehicles; if there is an overlap, increase the cost of the current node, prompting the A* algorithm to select other paths, mark the paths or expected positions of other vehicles as "temporary obstacles", and increase the g values of these nodes, so that the A* algorithm does not tend to select these paths.

[0029] In some embodiments of the present invention, local path adjustment and replanning are introduced into the A* algorithm, including:

[0030] Local path adjustment: When performing A* path planning each time, the vehicle only executes part of the path, and then replans the remaining path according to the latest position information of other vehicles;

[0031] Path replanning trigger: When the vehicle enters a dead end or all paths are occupied by dynamic obstacles, pause the vehicle and wait for the environment to change or actively trigger path replanning.

[0032] In some embodiments of the present invention, the A* guided DQN algorithm includes:

[0033] Action selection: At each step of training, the agent selects the action planned by the A* algorithm according to the current state, rather than relying entirely on random exploration;

[0034] Experience replay: The paths generated by the A* algorithm are used as the experience data of the DQN and stored in the experience replay pool; these paths can provide high-quality samples for the training of the DQN and help the agent learn effective strategies faster;

[0035] Dynamic adjustment: During the training process, as the agent's learning gradually deepens, gradually reduce the dependence on the A* algorithm and instead rely more on the strategies learned by the DQN itself.

[0036] In some embodiments of the present invention, an exploration rate ε is designed in the DQN algorithm to determine the probability of using a neural network for decision-making, and 1 - ε represents the probability of using the A* algorithm.

[0037] In some embodiments of the present invention, the DQN algorithm guided by A* path planning includes two neural networks: a current value network for calculating state values and a target value network for evaluating state actions; the parameters of the target value network are derived from the current value network, and after a certain number of trainings, the parameters of the current value network are copied to the target value network.

[0038] Compared with the prior art, the advantages and positive effects of the present invention are as follows: The present invention proposes to optimize the warehouse location layout according to the initial quantity of goods in the warehouse with the goals of storage efficiency, warehouse center of gravity, and location density, and reasonably allocate the access points in the order, effectively avoiding collisions during multi-vehicle collaborative operations while ensuring the safety and high-efficiency storage of the warehouse; with the goal of minimizing the order completion time and energy consumption, reasonably scheduling the task allocation between the shuttle cars and the elevators to ensure the efficient use of resources and the balanced distribution of the operation load, thereby reducing the phenomenon of excessive busyness or idleness of a single vehicle; with the goals of the shortest path and vehicle avoidance, using intelligent algorithms to plan the optimal driving route for each vehicle, avoiding long paths, reducing the risk of collisions and congestion between vehicles, and considering the dynamic avoidance strategy for multi-vehicle collaborative operations in path planning to ensure that vehicles can efficiently complete tasks in a dense operation environment; through the above optimizations, the present invention provides strong technical support for the efficient operation of shuttle cars in complex warehousing environments, ensuring the smooth operation of inbound and outbound operations and improving the overall operation efficiency of the dense warehousing system.

[0039] After reading the detailed description of the embodiments of the present invention in conjunction with the accompanying drawings, other features and advantages of the present invention will become clearer. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0041] Figure 1 Schematic top view structure of a dense warehousing system;

[0042] Figure 2 Schematic diagram of the scheduling optimization process of shuttle cars - elevators in the present invention;

[0043] Figure 3 Schematic diagram of the path planning process of shuttle cars in the present invention;

[0044] Figure 4 Schematic diagram of solving the shuttle-lift scheduling model based on the improved genetic algorithm in the present invention;

[0045] Figure 5 Chromosome exemplified in the improved genetic algorithm of the present invention;

[0046] Figure 6 Schematic diagram of the process of solving the optimal path of the shuttle vehicle using the A* algorithm in the present invention;

[0047] Figure 7 Schematic diagram of the change process of the exploration rate in guiding the DQN algorithm using the A* algorithm in the present invention;

[0048] Figure 8 Structure diagram of guiding the DQN algorithm based on the A* path planning in the present invention;

[0049] Figure 9 Schematic diagram of the training process of the shuttle vehicle in the embodiment of the present invention;

[0050] Figure 10 Schematic diagram of the system structure of the intensive warehousing operation optimization system proposed by the present invention. Detailed implementation manners

[0051] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0052] The intensive warehousing system in the embodiment of the present invention includes two elevators, and its set is L = {l1, l2}, and m shuttle vehicles, and its set is S = {s1, s2,..., s m}. The system needs to complete n inbound / outbound order tasks, and the order set is O = {o1,..., o j ,..., o n}, and each order o j includes three attributes: order type (o j > 0 indicates inbound, o j < 0 indicates outbound), coordinate data (coords(o j ) = (x j , y j )) and storage layer number (z j ∈ {1,..., Z max}). The shuttle vehicle task set is B = {B1, B2,..., B m}, where Bi is the task sequence of the i-th shuttle, expressed as: Where E i is the number of tasks for the i-th shuttle. During the task execution, if the shuttle needs to move across floors, it needs to use the elevator to complete the floor change operation. The task set of the elevator is A = {a1, a2, ..., a k}, where k is the total number of times all shuttles use the elevator during the mission.

[0053] Each task Corresponding order o j The execution includes two operations: inbound and outbound. For inbound orders, the shuttle goes from the current location to the inbound point and transports the goods from the inbound point to the storage location specified by the order; for outbound orders, the shuttle goes from the current location to the storage location specified by the order and transports the goods from the storage location to the outbound point.

[0054] When the shuttle performs a task on a certain floor of a dense warehouse, an optimal path is planned for the shuttle from the starting position to the target position, avoiding obstacles while complying with the layout and movement constraints of the warehouse. m The path of each shuttle vehicle to complete the task set is P = {P1, P2, ..., P m}.

[0055] For this system, the present invention establishes a mathematical model with the shortest job completion time as the goal. In order to accurately describe and define the problem, the following assumptions are made:

[0056] (1) The initial positions of the elevator and shuttle are both at or on the bottom of the shelf;

[0057] (2) The shuttle vehicle is responsible for at most one cargo during its operation;

[0058] (3) The hoist is responsible for at most one shuttle car when performing an operation task;

[0059] (4) The hoist is set at the left end of the transverse track;

[0060] (5) The acceleration (deceleration) of the equipment should be considered when the shuttle and the elevator are running;

[0061] (6) Tasks are issued to make the goods in / out location known.

[0062] 1. Establish a warehouse model.

[0063] According to the operation process of the system of the present invention, when the shuttle car carrying goods drives out of the elevator, reaches the layer where the task is located, travels horizontally to the goods storage location, and after completing the task, returns to the buffer area on this layer, and waits for the elevator to be idle before completing the subsequent tasks.

[0064] 1. Establish a mathematical model for optimizing storage locations.

[0065] As the number of storage locations in the dense library continues to increase, unreasonable storage location allocation will lead to problems such as low storage operation efficiency, low warehouse safety, and traffic jams of shuttle cars caused by concentrated storage locations. The present invention reduces the problems of shuttle car operation through storage location optimization.

[0066] Aiming at the problems of unreasonable storage in the warehouse, low safety and efficiency, a multi-objective optimization model with the center of gravity of the shelf, the inbound and outbound efficiency, and the storage density as the objectives is established, and an objective function is established according to the optimization objectives.

[0067] The present invention takes the in-line placed shelf as the research object, including row A of shelves. Each row of shelves has storage locations of B columns and C layers, including several parallel aisles. The row closest to the entrance and exit is recorded as row 1, the column closest to the shelf exit is recorded as column 1, and the bottom layer of the shelf is the first layer. Therefore, the storage location coordinates of the x-th row, y-th column, and z-th layer are recorded as (x, y, z), where x ∈ {1, 2,..., A}, y ∈ {1, 2,..., B}, and z ∈ {1, 2,..., C}. From this, the coordinates H of the i-th item can be obtained i (x i , y i , z i ).

[0068] (1) The objective function of the center of gravity of the shelf:

[0069] For a single row of shelves, when different goods are placed on different layers of the shelf, the center of gravity height of the shelf will change. Therefore, in order to ensure the stability of the shelf during use, when carrying out the storage location allocation work, goods with a larger mass should be placed on the storage locations of lower layers to reduce the center of gravity of the shelf. Thus, the optimization objective function for the stability of the shelf is obtained as:

[0070]

[0071] In the formula, m xyz represents the weight of the goods placed in the storage location (x-th row, y-th column, z-th layer), z represents the layer where the goods are stored, and L0 represents the height of each layer of the shelf.

[0072] (2) The objective function of the storage density:

[0073] In warehouse management, the turnover rate is a key indicator for measuring the frequency of goods inbound and outbound. Specifically, the turnover rate refers to the number of times a certain good is received and shipped out within a certain period. A higher turnover rate means that the good stays in the warehouse for a shorter time and moves in and out frequently, while a lower turnover rate indicates that the good stays in the warehouse for a longer time. Therefore, when optimizing the storage location under the uniform distribution strategy, considering the turnover rate of goods can effectively avoid uneven resource allocation and shuttle car congestion problems, thereby improving the overall warehouse efficiency. Based on this, by quantifying the difference between the sum of the turnover rates of adjacent two rows and the overall average turnover rate, the objective function for optimizing the storage density is obtained as follows:

[0074]

[0075] In the formula, R i represents the turnover rate of the i-th good, and R m is the average value of the turnover rates of all goods.

[0076] (3) Objective function for inbound and outbound efficiency:

[0077] In the present invention, the inbound efficiency of the shuttle car is related to the distance from the pick-up station to the target storage location. Therefore, the allocation of storage locations will directly affect the efficiency of the entire warehouse. Goods with a high turnover rate indicate that they need to be transported frequently. Therefore, these goods should be closer to the entrance / exit or in a more accessible location in the warehouse to reduce handling time and costs. Thus, the objective function for optimizing the inbound and outbound efficiency is obtained as follows:

[0078]

[0079] In the formula, x i , y i , z i represent the coordinates of the i-th good, v x , v y , v z represent the traveling speeds of the shuttle car in the x, y, and z directions when performing the handling operation, and R i represents the turnover rate of the i-th good.

[0080] 2. Establish a mathematical model for the shuttle car.

[0081] (1) Distance calculation:

[0082] According to the Figure 1 shown method to establish a coordinate system, then the coordinates of task o j in the shelf are coords(o j ) = (x j , y j , z j ), where x j is the column where task o j is located; yj is the row where the task is located; z j is the layer where the task is located. For the inbound, empty-load, and outbound operations divided for a single inbound and outbound task, record the inbound and outbound task of a single compound operation in the system as o j , o j >0 indicates inbound, o j <0 indicates outbound.

[0083] The distances for inbound and outbound operations, that is, the shuttle car starts from the current position and reaches the storage location for inbound, or the shuttle car starts from the outbound position and reaches the elevator position on that layer. Since the specific driving route of the shuttle car is calculated by the path planning module, the distance is obtained by calculating the straight-line distance between two points. The distance expression is:

[0084] d h = |x i -x j | × X L + |y i -y j | × Y L ;

[0085] In the formula, (x i , y i ) and (x j , y j ) are the coordinates of the current position and the end position of the trolley respectively; X L , Y L are the length and width of a single storage bin.

[0086] For the empty-load operation, that is, when the shuttle car travels from the inbound storage location to the outbound storage location, its driving route is divided into 2 cases:

[0087] (a) The inbound and outbound storage locations are on the same layer, that is, z1 = z2.

[0088] The driving route of the shuttle car from the inbound storage location to the outbound storage location, and its distance expression is

[0089] d h = |x i -x j | × X L + |y i -y j | × Y L ;

[0090] (b) The inbound and outbound storage locations are on different layers, that is, z1 ≠ z2.

[0091] Divide the driving route of the shuttle car for the empty-load operation into 2 parts: the driving distance of the shuttle car from the inbound storage location to the buffer area on that layer At this time, the shuttle car must wait for the elevator to be idle, and then transport the shuttle car to the layer where the outbound storage location is located. The shuttle car then moves to the buffer area on this layer; the driving distance of the shuttle car from the outbound buffer area to the outbound storage location The expressions are as follows:

[0092]

[0093] (2) Time calculation:

[0094] Calculate the time required for the shuttle car during horizontal movement. The formula is:

[0095] T S = d horizontal / v s ; In the formula, d horizontal is the horizontal movement distance, and v s is the driving speed of the shuttle car.

[0096] (3) Energy consumption calculation:

[0097] In order to minimize the energy consumption for completing the order, it is necessary to consider the energy consumption in different states in path planning and scheduling. The energy consumption calculation formula for the shuttle car is:

[0098] E Se = d horizontal × e Se ;

[0099] E Sl = d horizontal × e Sl ;

[0100] In the formula, E Se , E Sl are the no-load energy consumption and load energy consumption of the shuttle car respectively, d horizontal is the horizontal movement distance, e Se is the no-load energy consumption per unit length of shuttle car movement, and e Sl is the load energy consumption per unit length of shuttle car movement.

[0101] 3. Establish a mathematical model for the elevator.

[0102] When the task is issued, the elevator operates according to the task set of the shuttle car. At this time, the elevator has three actions: inbound action, the elevator moves the shuttle car from the first layer to the target layer of the current task of the car; outbound action, the elevator moves the shuttle car from the current layer to the first layer; no-load action, the elevator starts from the current layer and reaches the target layer with a need for layer change.

[0103] (1) Distance calculation: d v = |z i - z j | × ZL ;

[0104] In the formula, z i is the current floor number of the elevator, z j is the target floor number of the next task of the elevator, and Z L is the height of each floor.

[0105] (2) Time calculation:

[0106] Calculate the time required for the elevator to move in the vertical direction. The formula is T L = d vertical / v L ; In the formula, d vertical is the vertical moving distance, and v L is the traveling speed of the elevator.

[0107] (3) Energy consumption calculation:

[0108] E Le = d vertical × e Le ;

[0109] E Ll = d vertical × e Ll ;

[0110] In the formula, E Le , E Ll are the no-load energy consumption and load energy consumption of the elevator respectively, d vertical is the vertical moving distance, e Le is the no-load energy consumption per unit length of the elevator's movement, and e Ll is the load energy consumption per unit length of the elevator's movement.

[0111] 4. Establish a minimum-time mathematical model.

[0112] (1) Decision variables:

[0113] The order allocation variable x ij , if the order o j is allocated to the shuttle car s i then x ij = 1, otherwise x ij = 0; The start time of the shuttle car task The time (in seconds) when the shuttle car s i starts to execute the order o j ; The completion time of the shuttle car task The time when the shuttle car s i completes the order o j ; The start time of the elevator task The shuttle car s i uses the elevator lk Start time; Completion time of the elevator task Shuttle car s i Use elevator l k Completion time.

[0114] (2) Auxiliary variables:

[0115] Available time of the shuttle car Ti a , Shuttle car s i Time when it can be used again after completing the current task; Available time of the elevator Elevator l k Time when it can be used again after completing the current task.

[0116] (3) Constraints:

[0117] a) Must be assigned to and only assigned to one shuttle car:

[0118] b) For multiple shuttle cars, after their tasks are executed in sequence, the start time of the subsequent task is not earlier than the completion time of the previous task: And x i,j = x i,j+1 = 1;

[0119] c) The relationship between the start and completion times of executing an order is:

[0120] d) The available time is equal to the completion time of the last task: In the formula, E i Is the last task of shuttle car s i ;

[0121] e) When multiple vehicles need to use the same elevator, conflicts need to be avoided, and the elevator processes tasks according to the first-come, first-served principle; for any two tasks using the same elevator, it is necessary to satisfy:

[0122] Then

[0123] f) The completion time of a task is related to the start time and vertical movement time:

[0124] (4) Time calculation:

[0125] Total travel time of the shuttle car executing order o j , including horizontal and vertical movement times, and possible waiting times, and its expression is:

[0126] In the formula, is the horizontal movement time of the shuttle vehicle from the starting position to the elevator or the target position; is the waiting time of the shuttle vehicle for the elevator; is the vertical movement time of the elevator; is the horizontal movement time of the shuttle vehicle from the elevator to the target position.

[0127] The waiting time for the shuttle vehicle to wait for the elevator is:

[0128] Minimizing the total completion time is:

[0129] 5. Establish a mathematical model for minimum energy consumption.

[0130] The total energy consumption of the shuttle vehicle is: E St = ∑E S ; where E S Select the corresponding energy consumption calculation formula according to the movement state of the trolley in this section.

[0131] The total energy consumption of the elevator is: E Lt = ∑E L ; where E L Select the corresponding energy consumption calculation formula according to the movement state of the elevator in this section.

[0132] The total energy consumption of the system is the sum of the energy consumptions of the shuttle vehicle and the elevator operating in different states, and its formula is: E t = E St + E Lt .

[0133] 6. Establish an objective function.

[0134] (1) Objective function for storage location allocation:

[0135] Considering the inbound and outbound efficiency of turnover, the optimization objective function of shelf stability and shelf density, design a multi-objective optimization model:

[0136]

[0137] In this model, the present invention uses the weight method to solve the multi-objective optimization problem in storage location optimization. The weight method solves the problem by assigning a weight to each objective and then combining all weighted objectives into a single optimization objective. The formula of the weight method is:

[0138] where ω i ≥ 0, and

[0139] Assign corresponding weight coefficients to the three objectives, then the objective function can be simplified to:

[0140]

[0141] (2) Scheduling optimization objective function.

[0142] Considering multiple optimizations, the objective function is calculated using the weighted method:

[0143] f = ω1 × minT makespan + ω2 × minE t ; where ω1 and ω2 are the weighting coefficients for minimum time and minimum energy consumption.

[0144] II. Optimization process design of scheduling and path planning.

[0145] This system adopts a composite operation mode. The operations of the shuttle vehicle include inbound, empty-load, and outbound operations. To improve the system efficiency, when the system receives an inbound order, the shuttle vehicle first goes to the inbound cargo location. After completing the task, it does not immediately return to the I / O point but directly goes to the outbound cargo location. After completing the pick-up task, it then returns to the I / O point.

[0146] Specifically, as Figure 2 shown, when the system receives a scheduling instruction, it will request the shuttle vehicle service. The shuttle vehicle confirms whether it needs to use the elevator according to the scheduling instruction. If needed, the shuttle vehicle will move to the elevator, and the elevator will transport it to the first floor. Subsequently, it receives the goods from the conveyor roller path, and the elevator transports the fully loaded shuttle vehicle to the floor where the inbound task is located to complete the goods storage operation.

[0147] After completing the inbound operation, the elevator and the shuttle vehicle are in the empty-load state. If the outbound task and the inbound task are on the same floor, the shuttle vehicle does not need to return to the elevator but directly goes to the outbound cargo location through the shortest path to perform the outbound operation, and the empty-load time is zero at this time. If the outbound task is on a different floor, the shuttle vehicle will first move to the buffer area on that floor to wait, and then be transferred to the floor where the task is located by the elevator. After the shuttle vehicle leaves the elevator, the elevator completes the empty-load operation, and the shuttle vehicle then moves to the outbound cargo location to complete the outbound operation.

[0148] In the outbound operation, after the shuttle vehicle picks up the goods, it transports them to the buffer area on that floor to wait for the elevator service, and then transports the goods to the I / O port and releases the goods on the conveyor roller path to complete the outbound operation. When the outbound task is completed, the system will judge whether there are subsequent tasks. If there are no new tasks, the shuttle vehicle will take the elevator back to the bottom floor and enter the standby state to prepare for the next scheduling.

[0149] After the scheduling optimization is completed, the system will output data of the optimal scheduling plan, including the order sequence, the task set of the shuttle vehicle, and the selection of the elevator. When the shuttle vehicle performs the inbound and outbound operations with the output of the scheduling optimization as the initial data, there are three path planning scenarios: inbound operation, no-load operation, and outbound operation.

[0150] (1) Inbound operation: The path planning starts from the buffer area on the layer where the task is located and ends at the location where the goods are stored.

[0151] (2) No-load operation: At this time, the starting point of the shuttle vehicle is the storage point of the inbound operation, and the end point is the storage point of the next outbound task.

[0152] (3) Outbound operation: The shuttle vehicle starts from the goods storage point and ends at the buffer area on the layer where the task is located.

[0153] Based on the above situations, when the shuttle vehicle needs to perform path planning, it will use the output data of the scheduling module as the initial condition for path selection. When the system receives the path planning instruction, it will initialize the current state and position of each shuttle vehicle according to the order and vehicle status (inbound, outbound, no-load), and perform corresponding operations according to the tasks assigned by the scheduling module. The shuttle vehicle starts from the starting point and first detects whether there are other vehicles on the same layer. If a path overlap is found, avoidance will be carried out. When avoiding, first judge the priorities of the two vehicles, and the vehicle with the lower priority needs to avoid first. The vehicle with the lower priority tries to decelerate to avoid. If it fails to meet the requirements, it needs to stop and wait, and continue to plan the route. The specific process is as Figure 3 shown.

[0154] III. Solve the objective function.

[0155] To improve the storage efficiency, reasonable allocation of the storage locations for goods should be considered first. When the goods are not stored reasonably, it will not only affect the storage efficiency, but also may lead to concentrated storage of goods in the storage locations, thus causing traffic jams or collisions among multiple shuttle vehicles. Moreover, in an environment with high concurrency of multiple operation tasks, multiple shuttle vehicles and elevators need a scheduling optimization plan to improve the inbound and outbound efficiency and reduce the operation time to meet the customer's demand for picking up goods. To address the above problems, the present invention first reallocates the initial goods layout and the order storage locations based on the warehouse storage locations and the quantity of the initial goods. Secondly, according to the real-time information such as the speeds and positions of the shuttle vehicles and elevators, the position relationship between the vehicles and the storage locations is measured, and the working time for completing each task is analyzed. Then, with the minimum order completion time and the lowest energy consumption as the objectives, the genetic algorithm is used for solving, and the optimal scheduling plan is obtained. Finally, the inbound and outbound orders are re-sorted, and a task set is assigned to each shuttle vehicle to achieve the optimal multi-vehicle scheduling plan. To meet the above optimization requirements, the present invention uses an improved genetic algorithm for solving.

[0156] In the system operation, tasks are scheduled based on order information, the working status information of shuttle cars and elevators, and the storage status information of the system shelves. Among them, the order information includes in-out storage location information and task urgency information. System scheduling must first allocate tasks and then sort them according to the following rules.

[0157] (1) Task allocation rule: If there are multiple elevators in the system, allocate the task to the elevator with a shorter driving distance to the target storage location, and allocate a reasonable number of shuttle cars according to the task volume of the elevator.

[0158] (2) Task sorting rule: Based on the shuttle cars allocated to the elevator and the tasks randomly assigned to them, determine the order in which the elevator completes the tasks. The sorting must follow the following rules:

[0159] a) The elevator completes the first inbound operation of all shuttle cars in sequence before continuing with subsequent tasks, ensuring that all shuttle cars are in the system and have started operating.

[0160] b) The elevator gives priority to completing urgent order operations.

[0161] c) The elevator selects the shuttle car with the shortest distance for operation.

[0162] Based on rule a), the shuttle cars are transported into the system. If rule b) exists, then rule b) is given priority; otherwise, the sorting is carried out according to rule c).

[0163] In the present invention, after task allocation and preliminary sorting of the elevator and shuttle cars, the IGA is used to optimize the operation time. Taking the operation of a single elevator as an example, the genetic algorithm process is as Figure 4 shown, including:

[0164] (1) Initializing the shelf status: Initialize the storage status of the shelves in the shuttle car system. In the system, when the shuttle car picks up goods, the storage status on the shelf changes. The present invention mainly considers the path planning of shuttle cars for a batch of orders. Before planning, the shelf status is initialized, that is, whether there is goods stored in the storage location. If so, record the shelf status as 1; otherwise, record it as 0.

[0165] (2) Population initialization and coding: According to the working characteristics of shuttle cars and elevators, design a chromosome with a three-layer structure. The first layer is the operation sequence coding of the shuttle car, with positive integers for inbound operations and negative numbers for outbound operations; the second layer is the operation sequence coding of the shuttle car, and the operation tasks are randomly assigned according to the number of shuttle cars; the third layer is the operation sequence of the elevator, sorted according to the tasks executed by the shuttle cars with layer-changing requirements, and calculated based on the current position of the elevator and the operation relationship between the first and second layers.

[0166] (3) Chromosome decoding: Assume the chromosome is as Figure 5As shown in the figure, this chromosome contains the operation information of 6 operation sequences, 3 shuttle cars, and 1 elevator. According to the data in the first column of the first two layers of the chromosome, it can be obtained that the inbound order No. 2 is carried by the shuttle car No. 1; according to the data in the first column of the third layer, it can be known that the shuttle car No. 1 borrows the elevator No. 1 to realize the floor change operation from the first layer to the second layer.

[0167] (4) Fitness calculation: With the goal of the shortest time for the system to complete tasks and the minimum energy consumption, this goal is used as the fitness. When calculating, it is necessary to consider the shelf state after the system completes a composite operation and then calculate the time for the next composite operation. The fitness function is: f = ω1×minT makespan +ω2×minE t .

[0168] (5) Genetic operators: Include selection, crossover, and mutation.

[0169] a) In the selection design, in order to be able to select individuals with higher fitness while maintaining diversity, the tournament selection method is adopted. First, a certain number of individuals are randomly selected from the population, the fitness of these individuals is compared, and the individual with the highest fitness is selected to enter the next generation, thereby improving the calculation efficiency and effectively maintaining the diversity of the population.

[0170] b) The embodiment of the present invention designs a two-point crossover method. Two crossover points are selected, and the chromosome segments between these two points are exchanged, which can provide more gene exchanges than single-point crossover and help better explore the solution space. Set the crossover probability to p c , when the random probability p i <p c for each group of chromosomes, crossover operation is performed.

[0171] c) In the mutation operation, an adaptive mutation method is designed, which can dynamically adjust the mutation rate at different stages of the genetic algorithm to improve the performance and efficiency of the algorithm. As the number of iterations increases, the mutation rate is reduced. This method can not only ensure a high exploration ability in the early stage but also gradually focus on the potential optimal solutions for optimization in the later stage. The adaptive mutation rate is calculated as shown in the following formula, where P m1 represents the initial mutation rate, P m2 represents the final mutation rate, t represents the current number of iterations, and G represents the total number of iterations:

[0172]

[0173] In summary, the present invention completes the scheduling optimization of inbound and outbound orders through the scheduling optimization model to improve the inbound and outbound efficiency, reduce the order completion time, and uses the optimization result as the initial data of the path planning module for path planning.

[0174] IV. Design of the RGV Path Planning Algorithm

[0175] In the actual warehousing operation environment, when multiple RGVs execute their respective tasks simultaneously, it is inevitable that the running trajectories of the vehicles overlap, resulting in the problem of multi-vehicle conflict and collision. Therefore, when there is an overlap in the paths, avoidance needs to be carried out according to the priority of the RGVs. Based on the above problem description, the present invention designs a path planning based on the A* algorithm in the path planning module according to the dynamic information of the RGV operation

[0176] The path planning rules include:

[0177] (1) Different priorities are set according to the operation conditions of the RGVs: outbound operation has a high priority, inbound operation has a medium priority, and empty-load operation has a low priority.

[0178]

[0179] (2) Priority allocation: When operating on the same floor, the vehicle with a higher priority will be granted the right of way; if the priorities of two vehicles are the same, the order of passage will be determined according to the distance to the target position, and the vehicle with a shorter distance will have priority; when the RGV departs, it needs to detect other vehicles on the same floor in real time. If a path overlap is found, avoidance is required; the vehicle with a lower priority tries to decelerate to avoid conflicts with vehicles with a higher priority; if deceleration cannot meet the avoidance requirements, the vehicle with a lower priority needs to stop and wait, and continue to plan the route; all vehicles should share the current status information to ensure that other vehicles can obtain the avoidance information in time, thereby reducing the risk of traffic jams and collisions.

[0180] The A* algorithm evaluates the priority of each node by using a cost function. The cost function is usually expressed as: f(n) = g(n) + h(n); where f(n) represents the estimated total cost from the starting point to the target node; g(n) represents the actual cost from the starting point to the current node; h(n) represents the estimated cost from the current node to the target node (Manhattan distance).

[0181] To use the A* algorithm to solve the path planning and avoidance problems in a dynamic environment, it is necessary to extend and adjust the standard A* algorithm so that it can handle dynamic obstacles, priority scheduling, and multi-vehicle cooperative avoidance.

[0182] (1) Introduce a priority scheduling mechanism.

[0183] The essence of the A* algorithm is to select the next node to be expanded according to the total cost f(n) of the node. In a dynamic environment, the priority mechanism is used to affect the values of g(n) and h(n), making it easier for vehicles with a higher priority to find a preferred path.

[0184] a) Priority weight adjustment: When calculating the g value (the path cost from the starting point to the current node), a priority factor can be introduced to adjust the path cost of vehicles with different priorities. For example, for high-priority vehicles, the g value can be reduced to give them a greater advantage in path selection. The formula can be adjusted to: p i represents the priority factor, and g′ represents the path cost after priority adjustment.

[0185] b) Processing when priorities are equal: If two vehicles have the same priority, the order can be determined by the distance h to the target location (heuristic estimation), and the vehicle closer to the target location has priority.

[0186] (2) Dynamic avoidance mechanism.

[0187] In the A* algorithm, dynamic avoidance is achieved by introducing a real-time detection and path update mechanism. Each time a node is expanded, the path of the current vehicle is dynamically adjusted according to the paths or current positions of other vehicles. Each time a node is expanded, it is detected whether the current node may overlap with the paths of other vehicles. If it overlaps, the cost of the current node is increased, prompting the A* algorithm to choose other paths. The paths or expected positions of other vehicles are marked as "temporary obstacles", and the g values of these nodes are increased, so that A* is not inclined to choose these paths.

[0188] (3) Local path adjustment and replanning.

[0189] In a dynamic environment, path planning cannot be completed in one go. Even if the optimal path is calculated, some paths may become infeasible as other vehicles move, so local path adjustment and dynamic replanning are required.

[0190] a) Local path adjustment: Each time A* path planning is executed, the vehicle only executes part of the path (such as a few nodes within a period of time), and then replans the remaining path based on the latest position information of other vehicles. This can be achieved by calling A* to replan the path every once in a while.

[0191] b) Path replanning trigger: In some cases, the vehicle may enter a dead end or all paths are occupied by dynamic obstacles. At this time, the vehicle needs to pause and wait for the environment to change or actively trigger path replanning. This can be done by periodically re-evaluating the path or immediately replanning when a conflict is detected.

[0192] In a multi-vehicle system, the shuttle may encounter other vehicles when performing path planning, resulting in the risk of path overlap and collision. To address this problem, the design of the A* algorithm will include an avoidance mechanism to ensure that each vehicle can reasonably avoid other vehicles when selecting a path.

[0193] The flowchart of the A* algorithm of the present invention is as Figure 6 shown.

[0194] In the DQN algorithm, random exploration is the key to enriching the algorithm's decision-making. However, random exploration is a relatively inefficient strategy, especially when applied to the shuttle vehicle to find the optimal path. Random exploration sometimes fails to make effective decisions, which may cause the shuttle vehicle to move away from the target position. In the design of the present invention, an approximate optimal path can be found in the state space through the A* algorithm, and the state-action pairs of this path are used as the training data for DQN, helping DQN quickly learn a better strategy in the initial stage. That is, using the data of the A* algorithm to guide the decision-making process of the DQN algorithm, thereby making the path exploration more intelligent and improving the decision-making efficiency.

[0195] The algorithm of A* guiding DQN decision-making provides an initial path planning suggestion for the intelligent agent through the A* algorithm in the initial stage of training. The A* guiding DQN algorithm is designed as follows:

[0196] (1) Action selection: At each step of training, the intelligent agent can select the action planned by the A* algorithm according to the current state, rather than relying entirely on random exploration. This way can improve the exploration efficiency of the intelligent agent in the early stage.

[0197] (2) Experience replay: The paths generated by the A* algorithm are used as the experience data of DQN and stored in the experience replay pool. These paths can provide high-quality samples for the training of DQN and help the intelligent agent learn an effective strategy faster.

[0198] (3) Dynamic adjustment: During the training process, as the intelligent agent's learning gradually deepens, the dependence on the A* algorithm is gradually reduced, and instead, more reliance is placed on the strategy learned by DQN itself.

[0199] The present invention designs an exploration rate ε in the algorithm to determine the probability of using the neural network for decision-making, and 1 - ε represents the probability of using the A* algorithm. For example, when the value of ε is 0.8, there is an 80% probability of making a decision using the neural network. The purpose of training is to make the neural network converge and make correct decisions based on the input observations. Therefore, the probability of using the A* algorithm is relatively high at the beginning of training. After training to a certain extent, it is necessary to completely use the neural network for decision-making to ensure the final convergence of the neural network. Figure 7 Shows the change process of the exploration rate. The abscissa represents the training progress, and the ordinate represents the value of the exploration rate. The exploration rate is fixed at 0.8 in the first third of the training progress, and then the exploration rate gradually increases to 1 to ensure the convergence of the neural network.

[0200] The structure of the DQN algorithm guided by A* path planning is as Figure 8As shown. The algorithm contains two neural networks: the current value network for calculating state values and the target value network for evaluating state actions. The parameters of the target value network are derived from the current value network. After a certain number of training iterations, the parameters of the current value network are copied to the target value network. The parameters of the current value network are updated using the loss function. The following equation represents the Q-value update formula, where α represents the learning rate, r is the current reward, γ is the discount factor, and S t+1 is the next state after the execution of action a, represents the Q-value for selecting the best action under S t+1 :

[0201]

[0202] A state represents all the information of the environment at a certain moment and is the basis for the agent to make decisions. In this design, four matrices are created based on the grid map established for the warehouse to represent the current state of the system: the valid position matrix, the prohibited position matrix, the current position matrix, and the target position matrix. The specific formulas for the matrices are as follows:

[0203] M1 = {m ij | m = 1 when the cell is a valid position, and 0 otherwise}; ij

[0204] M2 = {m ij | m = 1 when the cell is an area where AGV entry is prohibited, and 0 otherwise}; ij

[0205] M3 = {m ij | m = 1 for the cell where the AGV is located, and 0 otherwise}; ij

[0206] M4 = {m ij | m = 1 for the cell at the target position, and 0 otherwise}. ij

[0207] The reward function in reinforcement learning is used to measure how good an agent's action is in a specific state. Designing an effective reward function is a key step in achieving learning. In this case, the reward is based on the action results given by the shuttle vehicle executing DQN to evaluate the value of an action. When the shuttle vehicle executes an action, running out of the specified area, entering the shelf area, and not reaching the specified final position will result in a reward value of -1; when the shuttle vehicle can reach the target position, it will receive a reward value of 1; when the shuttle vehicle is in operation and does not reach the final destination, it will receive a reward value of 0. The above reward function is shown in the following equation, where (x', y') is the new position of the shuttle vehicle after executing action a.

[0208]

[0209] ​​​​An action is an operation that an agent can perform in a specific state. The shuttle vehicle needs to perform four action operations when executing tasks: going up (0), going down (1), going left (2), and going right (3). Since the neural network can only output results in digital format, each action is assigned a corresponding number. When running the system, whenever the shuttle vehicle enters a grid, it sends the observation result to the system. DQN returns the action to be executed to the shuttle vehicle and then determines whether the action is feasible. When the action is not feasible, the action will terminate. The action is represented by the vector a, where a ∈ {0, 1, 2, 3} corresponding to the movement operations in four directions. When the shuttle vehicle executes the action a, the new position (x', y') can be calculated by the following formula:

[0210]

[0211] After designing the algorithm, start training the shuttle vehicle for path planning. The training process of the shuttle vehicle is as Figure 9 shown. When the shuttle vehicle starts to execute the first task, it starts to learn. The shuttle vehicle transmits the current state to the system, and DQN feeds back an action. When the reward obtained by executing the action is -1, the shuttle vehicle will send a failure signal and then start a new round of training; otherwise, the shuttle vehicle will make corresponding reactions according to the action instructions and continue to transmit the observations of the current state and reward of this action. When the shuttle vehicle completes all tasks under the guidance of DQN, the shuttle vehicle will send a "task completed" signal, the current training phase ends, and a new training phase begins. This cycle will continue until the termination condition is met. In the DQN algorithm, each time a decision is made, the input observation, the output action, and the obtained reward are combined into an experience and stored in the experience pool. Once the number of experiences reaches the set value, the neural network will start to update. The neural network is updated once after each decision. First, extract some experiences from the experience pool, calculate their loss values, and then use this loss value to adjust the parameters of the neural network in combination with the learning rate.

[0212] Based on the above-mentioned operation optimization system of the intensive warehousing system proposed by the present invention, it is applied to the intensive warehousing system and consists of a storage location optimization unit, a scheduling optimization unit, and a shuttle vehicle path planning unit; wherein, the storage location optimization unit is used to establish a storage location optimization target model with the shelf center of gravity, inbound and outbound efficiency, and storage density as the targets, and allocate orders to the storage locations in the intensive warehousing system based on the goods optimization target model; the scheduling optimization unit is used to establish a shuttle vehicle - elevator scheduling model with the minimum order completion time and minimum energy consumption as the optimization targets, and solve the model based on the improved genetic algorithm to obtain the optimal scheduling plan; the shuttle vehicle path planning unit is used to establish a path planning model with the shortest path and vehicle avoidance as the targets, and solve the model based on the A* algorithm to guide the decision-making of the DQN algorithm to obtain the optimal operation path of the shuttle vehicle.

[0213] The intensive warehousing operation optimization system proposed by the present invention aims at storage efficiency, warehouse center of gravity, and storage location density. It optimizes the layout of warehouse storage locations according to the initial quantity of goods in the warehouse, and reasonably allocates the access points in the order. While ensuring the safety and high-efficiency storage of the warehouse, it effectively avoids collisions during multi-vehicle collaborative operations. With the goal of minimizing order completion time and energy consumption, it reasonably schedules the task allocation between the shuttle vehicles and elevators to ensure the efficient utilization of resources and the balanced distribution of operation loads, thereby reducing the phenomenon of excessive busyness or idleness of a single vehicle. With the goal of the shortest path and vehicle avoidance, it uses intelligent algorithms to plan the optimal driving route for each vehicle, avoiding long paths, reducing the risk of collisions and congestion between vehicles. In path planning, it takes into account the dynamic avoidance strategy of multi-vehicle collaborative operations to ensure that vehicles can efficiently complete tasks in a dense operation environment.

[0214] In the simulation test stage, this design allocated different numbers of shuttle vehicles and elevators for different scenarios and conducted a systematic performance evaluation. The test results show that this optimized design has achieved significant improvements in terms of running time, energy consumption, and avoidance of multi-vehicle collaborative operations. Specifically, the optimized system can not only complete the inbound and outbound tasks faster, reducing the empty running and waiting time of vehicles, but also significantly reduce the overall energy consumption of the system. It effectively avoids traffic jams and vehicle collisions during multi-vehicle parallel operations, improving the overall stability and efficiency of the system.

[0215] Through these optimizations, this design provides strong technical support for the efficient operation of the shuttle vehicle system in a complex warehousing environment, ensuring the smooth progress of inbound and outbound operations and improving the overall operational efficiency of the warehousing system.

[0216] It should be noted that in the specific implementation process, the above control part can be implemented by a processor in hardware form executing computer-executable instructions in software form stored in the memory, which will not be elaborated here. And the programs corresponding to the actions executed by the above control circuit can all be stored in the computer-readable storage medium of the system in software form for the processor to call and execute the operations corresponding to each module above.

[0217] The computer-readable storage medium in the above text can include volatile memory, such as random access memory; it can also include non-volatile memory, such as read-only memory, flash memory, hard disk, or solid-state drive; it can also include a combination of the above types of memory.

[0218] The processor mentioned above may also be a general term for multiple processing elements. For example, the processor may be a central processing unit, or may be other general-purpose processors, digital signal processors, application-specific integrated circuits, field-programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc., and may also be a dedicated processor.

[0219] It should be noted that the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by those of ordinary skill in the art within the scope of the essence of the present invention shall also fall within the protection scope of the present invention.

Claims

1. An intensive warehousing operation optimization system, characterized in that, Including: A location optimization unit, which is used to establish a location optimization objective model with the shelf center of gravity, inbound and outbound efficiency, and storage density as the goals, and allocate orders to the locations in the intensive warehousing system based on the goods optimization objective model; A scheduling optimization unit, which is used to establish a shuttle-lift scheduling model with the minimum order completion time and minimum energy consumption as the optimization goals, and solve the model based on the improved genetic algorithm to obtain the optimal scheduling plan; A shuttle path planning unit, which is used to establish a path planning model with the shortest path and vehicle avoidance as the goals, and solve the model by guiding the decision-making of the DQN algorithm based on the A* algorithm to obtain the optimal operation path of the shuttle; 2. The intensive warehousing operation optimization system according to claim 1, characterized in that The location optimization objective model is: Among them, is the target function of the shelf center of gravity; m xyz represents the weight of the goods placed in the location (x rows, y columns, z layers), z represents the number of layers where the goods are stored, and L0 represents the height of each layer of the shelf; is the objective function of the storage density, where R i represents the turnover rate of the i-th cargo, and R m is the average value of the turnover rates of all cargos; is the objective function of the inbound and outbound efficiency, where x i , y i , z i represent the coordinates of the i-th cargo, and v x , v y , v z represent the traveling speeds of the shuttle vehicle in the x, y, and z directions during the handling operation, and R i represents the turnover rate of the i-th cargo.

3. The optimized system for intensive warehousing operations according to claim 1, wherein The shuttle-lift scheduling model includes: Minimum time model: f = ω1 × minT makespan + ω2 × minE t ; where ω1 and ω2 are the weighting coefficients of the minimum time and the minimum energy consumption; among them, is the horizontal movement time of the shuttle car from the starting position to the elevator or the target position; is the waiting time of the shuttle car for the elevator; is the vertical movement time of the elevator; is the horizontal movement time of the shuttle car from the elevator to the target position; Minimum energy consumption model: E t = E St + E Lt ; where, E St = ∑E S is the total energy consumption of the shuttle vehicle, and E Lt = ∑E L is the total energy consumption of the elevator.

4. The intensive warehousing operation optimization system according to claim 1, wherein Obtaining the optimal scheduling plan by solving the model based on the improved genetic algorithm includes: Initializing the shelf state: Initialize the storage state of the shelf. If there is goods on the shelf, record the shelf state as 1, otherwise record it as 0; Population initialization and encoding: Design a chromosome with a three-layer structure according to the working characteristics of the shuttle and the lift; the first layer is the operation sequence encoding of the shuttle, with positive integers for inbound operations and negative numbers for outbound operations; the second layer is the operation sequence encoding of the shuttle, and the operation tasks are randomly assigned according to the number of shuttles; the third layer is the operation sequence of the lift, sorted according to the tasks executed by the shuttles with layer-changing requirements, and calculated based on the current position of the lift and the operation relationship between the first layer and the second layer; Chromosome decoding: Construct a chromosome with the operation information and operation sequence of the shuttle and the lift; Fitness calculation: With the goal of the shortest system task completion time and minimum energy consumption, use this goal as the fitness; Genetic operators: including selection, crossover, and mutation; in the selection design, the tournament selection method is adopted. First, a certain number of individuals are randomly selected from the population, their fitness values are compared, and the individual with the highest fitness is selected to enter the next generation; in the crossover design, the two-point crossover method is adopted. Two crossover points are selected, and the chromosome segments between these two points are exchanged. The crossover probability is set to p c , when the random probability p of each group of chromosomes i <p c crossover operation is performed; in the mutation operation, the adaptive mutation method is adopted, and the mutation rate is dynamically adjusted at different stages of the genetic algorithm. As the number of iterations increases, the mutation rate is decreased; the adaptive mutation rate is calculated using is implemented, where P m1 represents the initial mutation rate, P m2 represents the final mutation rate, t represents the current number of iterations, and G represents the total number of iterations.

5. The intensive warehousing operation optimization system according to claim 1, wherein Introducing a priority scheduling mechanism in the A* algorithm, including: Priority weight adjustment: When calculating the g value, a priority factor is introduced to adjust the path cost of vehicles with different priorities; g represents the path cost from the starting point to the current node, and p i represents the priority factor, and g′ represents the path cost after priority adjustment; Processing when priorities are the same: If the priorities of two vehicles are the same, determine the order by the distance h to the target position, and the vehicle closer to the target position passes first.

6. The optimized intensive warehousing operation system according to claim 1, wherein, Introducing a dynamic avoidance mechanism in the A* algorithm, including: Introducing a real-time detection and path update mechanism. When each node is expanded, detect whether the current node may overlap with the paths of other vehicles; if there is an overlap, increase the cost of the current node, prompting the A* algorithm to select other paths, mark the paths or expected positions of other vehicles as "temporary obstacles", and increase the g value of these nodes, so that the A* algorithm does not tend to select these paths.

7. The intensive warehousing operation optimization system according to claim 1, wherein Introducing local path adjustment and replanning in the A* algorithm, including: Local path adjustment: When each A* path planning is executed, the vehicle only executes part of the path, and then replans the remaining path according to the latest position information of other vehicles; Path replanning trigger: Pause the vehicle when it enters a dead end or all paths are occupied by dynamic obstacles, and wait for the environment to change or actively trigger path replanning.

8. The intensive warehousing operation optimization system according to claim 1, characterized in that The A* guiding DQN algorithm includes: Action selection: At each step of training, the agent selects the action planned by the A* algorithm according to the current state, rather than relying entirely on random exploration; Experience replay: The paths generated by the A* algorithm are used as the experience data of the DQN and stored in the experience replay pool; these paths can provide high-quality samples for the training of the DQN and help the agent learn effective strategies faster; Dynamic adjustment: During the training process, as the agent's learning gradually deepens, the dependence on the A* algorithm is gradually reduced, and instead, more reliance is placed on the strategy learned by DQN itself.

9. The intensive warehousing operation optimization system according to claim 8, wherein, In the DQN algorithm, an exploration rate ε is designed to determine the probability of using the neural network for decision-making, and 1 - ε represents the probability of using the A* algorithm.

10. The intensive warehousing operation system according to claim 8, characterized in that, The DQN algorithm guided by A* path planning contains two neural networks: the current value network for calculating state values and the target value network for evaluating state actions; the parameters of the target value network are derived from the current value network, and after a certain number of training sessions, the parameters of the current value network are copied to the target value network.

Images