Order allocation method and system for preventing order picking and waiting of order picker
By configuring the number of random stores in the RF unified picking system and combining order processing time prediction, task priority sorting and dynamic load balancing strategies, the problems of pickers picking orders and other orders are solved, and fair and efficient allocation of picking tasks is achieved, and the work fluency and accuracy of the system is improved.
Patent Information
- Application Number
- CN202411972168.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-30
- Publication Date
- 2025-04-15
AI Technical Summary
In the existing RF unified picking system, pickers independently choose tasks, which leads to order picking and other orders, which affects operating efficiency and team collaboration.
By configuring the number of random stores, combining order processing time prediction, task priority sorting and dynamic load balancing strategies, fair and efficient allocation of picking tasks can be achieved. Specific steps include responding to picking requests, filtering valid orders, synthesising task allocation combinations, randomly selecting order allocations, and adjusting allocation strategies based on Q-learning.
It realizes fair and efficient allocation of picking tasks, improves the smoothness and accuracy of the entire picking system, and reduces the pickers' picking or waiting for orders due to uneven tasks.
Smart Images
Figure CN119831282A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and in particular to an order allocation method and system for avoiding pickers from picking orders and waiting for orders. Background Art
[0002] In modern retail operations, RF centralized picking systems have been widely used in diversified sales channels such as online platforms and offline stores. The system uses wireless radio frequency technology to help pickers obtain order information in real time and pick according to instructions, thereby improving warehouse operation efficiency and inventory management accuracy and reducing human errors. Although the existing RF centralized picking system has greatly improved the automation and accuracy of warehouse operations, it still has problems such as picking orders, waiting for orders, and imbalanced operation sequences in actual applications. These problems not only affect operating efficiency, but may also lead to an unfair working environment and reduce teamwork spirit. For example: ① In the RF centralized picking system, pickers can usually see detailed information about the order (such as the type of goods, quantity, and assigned picking area, etc.). Therefore, some pickers give priority to "easy to pick" orders based on their own judgment, such as orders with few items and large quantities, in order to improve their personal work performance. However, this kind of picker's autonomous task selection behavior may cause individual pickers to delay the processing of other orders in order to obtain orders with lower difficulty, resulting in reduced work efficiency.
[0003] ② Since pickers select tasks independently, some pickers may not be able to pick orders in the order assigned by the system, which disrupts the order acquisition sequence and affects the smooth and orderly progress of the operation process. Summary of the invention
[0004] The purpose of the invention is to solve the above technical problems and propose an order allocation method and system that avoids pickers from picking and waiting for orders. By configuring the number of random stores, combined with order processing time prediction, task priority sorting and dynamic load balancing strategy, fair and efficient allocation of picking tasks is achieved. The purpose of the invention can be achieved through the following technical solutions: The present invention provides an order allocation method that avoids pickers from picking orders and waiting for orders, comprising: Step S100: responding to the picking request of the picker; Step S200: According to the picking wave and picking zone, select valid orders that meet preset conditions from the orders to be allocated and allocate them; Step S300: according to the configured random store quantity and the task priority of the valid order, the valid order synthesis task is assigned and combined; Step S400: Randomly select an order from the task allocation combination and assign it to the picker for picking until all valid orders in the task allocation combination are assigned; Step S500: based on the completion of all valid orders in the current task allocation combination, the unallocated valid orders are regenerated into a new task allocation combination according to step S200 and allocated until all valid orders are allocated.
[0005] Further, in step S200, according to the picking wave and the picking zone, the valid orders meeting the preset conditions are selected from the orders to be allocated for order allocation, which includes: ; in, is a set of valid orders, For the i-th order, For waves, For waves, state.
[0006] Further, in step S300, according to the configured random store quantity and the task priority of the valid order, the valid order synthesis task allocation combination includes: Step S310: configuring the random number of stores R in each picking area, so as to randomly select valid orders of R stores from each picking area for distribution; Step S320: Calculate the predicted processing time of each valid order; Step S330: sorting the task priorities according to the predicted processing time, urgency and quantity of the valid orders selected based on the random number R of stores, and generating a task allocation combination.
[0007] Further, in step S320, calculating the predicted processing time of each valid order includes: The time prediction model is trained based on the collected order dataset: ,in, is the characteristic vector of order i, is the actual processing time of order i, is the target variable; the feature vector ; are input features, including product type, quantity, weight, difficulty, distance, and historical processing time, and N is the number of orders; The calculation formula in the time prediction model is: ; ; in, is the predicted processing time of order i, K is the number of trees, is the predicted value of the kth tree for order i; is the number of leaf nodes of the kth tree, is the weight of the j-th leaf node of the k-th tree, is the indicator function, used when the input feature The region that falls into the jth leaf node of the kth tree 1 when it is, otherwise 0; During the training process, the model parameters are optimized by minimizing the objective function, which is, ; ; ; in, is the loss function used to calculate the prediction processing time The actual processing time The gap, is a regularization term used to control the complexity of the time prediction model; , is the regularization hyperparameter.
[0008] Based on the trained time prediction model, the feature vector As input, the predicted processing time for valid orders is calculated.
[0009] Further, in step S330, task priorities are sorted according to the predicted processing time, urgency and quantity of the valid orders selected based on the random number of stores R, and a task allocation combination is generated, including: The calculation formula for task priority ranking is: ; in, is the task priority, Forecasted processing time for active orders, The urgency of the valid order, ranging from 0 to 1, is the order quantity of the valid order, The processing time for valid orders. , and is the weight parameter.
[0010] All valid orders are sorted according to their task priorities to generate a task allocation combination.
[0011] Further, in step S400, randomly selecting an order from the task allocation combination and assigning it to a picker for picking until all valid orders in the task allocation combination are assigned include: Define the state space based on the picker's workload, the task priority of valid orders, and the picker's processing efficiency ; Define the action space of selecting an order from the task assignment combination and assigning it to the picker ; Design reward function based on distribution results , used to evaluate the effect of each allocation; among them, , , and is the weight coefficient, is the task priority, is the picker's workload and the number of orders to be processed, To predict processing time, For processing efficiency; Based on Q-learning, the reward value Q assigned to the order is calculated, and the reward value is assigned to the corresponding picker based on the preset standard. The Q value update formula is: ; in, is the learning rate, is the discount factor, is the maximum Q value in the next state.
[0012] Based on the same inventive concept, the present invention provides an order distribution system that avoids pickers from picking orders and waiting for orders, using the order distribution method as described above, including: A response module, used to respond to the picking request of the picker; The screening module is used to screen out valid orders that meet preset conditions from the orders to be allocated according to the picking waves and picking zones for order allocation; A data processing module is used to allocate and combine the effective order synthesis tasks according to the configured random store quantity and the task priority of the effective orders; The allocation module is used to randomly select an order from the task allocation combination and assign it to the picker for picking until all valid orders in the task allocation combination are allocated; based on the completion of the allocation of all valid orders in the current task allocation combination, the unallocated valid orders are regenerated into a new task allocation combination for allocation until all valid orders are allocated.
[0013] Furthermore, the preset conditions for screening in the screening module are: ; in, is a set of valid orders, For the i-th order, For waves, For waves, state.
[0014] Furthermore, the data processing module includes: A random store data generation unit, used to configure the random number of stores R in each picking area, and to randomly select valid orders of R stores from each picking area for distribution; A predicted processing time calculation unit, used to calculate the predicted processing time of each valid order; The task priority sorting unit is used to sort the task priorities according to the predicted processing time, urgency and quantity of the valid orders selected based on the random number R of stores, and generate task allocation combinations.
[0015] Furthermore, the allocation module also includes, Dynamic load balancing unit for defining state space based on picker workload, task priority of valid orders, and picker processing efficiency ; Define the action space of selecting an order from the task allocation combination and assigning it to the picker ; Design reward function based on distribution results , used to evaluate the effect of each allocation; among them, , , and is the weight coefficient, is the task priority, is the picker's workload and the number of orders to be processed, To predict processing time, For processing efficiency; based on Q-learning, the reward value Q assigned to the order is calculated, and the reward value is assigned to the corresponding picker based on the preset standard. The Q value update formula is: ;in, is the learning rate, is the discount factor, is the maximum Q value in the next state.
[0016] Compared with the prior art, the present invention has at least one of the following technical effects: The present invention configures random store numbers according to picking zones, randomly selects orders from multiple stores for task combination, and prioritizes them based on factors such as processing time and order urgency, so as to fairly and efficiently distribute picking tasks and improve the working fluency and accuracy of the entire picking system.
[0017] (1) By introducing the XGBoost model to predict order processing time, we can accurately estimate the processing time of each order, identify orders that require more processing time in advance, and prioritize task allocation. Combined with random store quantity configuration and task priority sorting, we can intelligently optimize task allocation, avoid inefficiency or incorrect allocation caused by manual order selection in traditional methods, and improve overall picking efficiency and accuracy.
[0018] (2) The Q-learning algorithm in reinforcement learning is used to dynamically adjust the allocation strategy of picking tasks. Tasks are allocated according to factors such as the workload, processing efficiency, and historical performance of each picker, thereby achieving load balancing and avoiding the situation where some pickers are overloaded while other pickers are idle. Through this intelligent optimization, the phenomenon of pickers "picking orders" or "waiting for orders" caused by uneven tasks can be greatly reduced, thereby improving the work efficiency of the overall system.
[0019] (3) The random number of stores configuration method can avoid the concentration of picking tasks in specific stores or areas and make full use of the resources of the entire picking area. When allocating tasks each time, the system will randomly select a corresponding number of stores for task allocation based on the actual situation of the stores and orders, so that the picking tasks are evenly distributed, improving resource utilization and reducing unnecessary resource waste. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the following is a brief introduction to the drawings required for describing the embodiment: Figure 1 Flow chart of the steps of the order allocation method to avoid pickers picking and waiting for orders. DETAILED DESCRIPTION
[0021] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. It should be understood that the specific implementation methods described herein are only used to explain the present invention and are not used to limit the present invention.
[0022] Example 1 RF centralized picking system plays an important role in modern retail warehousing. However, due to the phenomenon of "picking orders" and "waiting for orders" caused by factors such as the difficulty of the task and the picking area, the pickers have low work efficiency, uneven workload, and affected the teamwork atmosphere. In order to improve this problem, a new method is needed to dynamically allocate tasks so that task allocation is both fair and efficient.
[0023] Therefore, the inventors proposed a multi-level task allocation mechanism based on reinforcement learning, XGBoost algorithm and intelligent load balancing, and optimized the dynamic allocation of the order by combining the random number of stores during RF centralized picking.
[0024] Ideas proposed for this invention proposal: First, in order to prevent pickers from picking orders, the preference of pickers to choose tasks independently is broken by randomly assigning a range of store numbers, thus ensuring fair distribution of tasks.
[0025] The second is to use XGBoost to predict the difficulty of picking tasks, the processing time of pickers, and the priority of tasks, so as to intelligently optimize task allocation. The specific implementation methods are as follows: The present invention provides an order allocation method that avoids pickers from picking orders and waiting for orders, comprising: Step S100: responding to the picking request of the picker; Step S200: According to the picking wave and picking zone, select valid orders that meet preset conditions from the orders to be allocated and allocate them; Step S300: according to the configured random store quantity and the task priority of the valid order, the valid order synthesis task is assigned and combined; Step S400: Randomly select an order from the task allocation combination and assign it to the picker for picking until all valid orders in the task allocation combination are assigned; Step S500: based on the completion of all valid orders in the current task allocation combination, the unallocated valid orders are regenerated into a new task allocation combination according to step S200 and allocated until all valid orders are allocated.
[0026] Further, in step S200, according to the picking wave and the picking zone, the valid orders meeting the preset conditions are selected from the orders to be allocated for order allocation, which includes: ; in, is a set of valid orders, For the i-th order, For waves, For waves, state.
[0027] Further, in step S300, according to the configured random store quantity and the task priority of the valid order, the valid order synthesis task allocation combination includes: Step S310: configuring the random number of stores R in each picking area, so as to randomly select valid orders of R stores from each picking area for distribution; Step S320: Calculate the predicted processing time of each valid order; Step S330: sorting the task priorities according to the predicted processing time, urgency and quantity of the valid orders selected based on the random number R of stores, and generating a task allocation combination.
[0028] Further, in step S320, calculating the predicted processing time of each valid order includes: The time prediction model is trained based on the collected order dataset: ,in, is the feature vector of order i, is the actual processing time of order i, is the target variable; the feature vector ; are input features, including product type, quantity, weight, difficulty, distance, and historical processing time, and N is the number of orders; The calculation formula in the time prediction model is: ; ; in, is the predicted processing time of order i, K is the number of trees, is the predicted value of the kth tree for order i; is the number of leaf nodes of the kth tree, is the weight of the j-th leaf node of the k-th tree, is the indicator function, used when the input feature The region that falls into the jth leaf node of the kth tree 1 when it is, otherwise 0; During the training process, the model parameters are optimized by minimizing the objective function, which is, ; ; ; in, is the loss function used to calculate the prediction processing time The actual processing time The gap, is a regularization term used to control the complexity of the time prediction model; , is the regularization hyperparameter.
[0029] Based on the trained time prediction model, the feature vector As input, the predicted processing time for valid orders is calculated.
[0030] Further, in step S330, task priorities are sorted according to the predicted processing time, urgency and quantity of the valid orders selected based on the random number of stores R, and a task allocation combination is generated, including: The calculation formula for task priority ranking is: ; in, is the task priority, Forecasted processing time for active orders, The urgency of the valid order, ranging from 0 to 1, is the order quantity of the valid order, The processing time for valid orders. , and is the weight parameter.
[0031] All valid orders are sorted according to their task priorities to generate a task allocation combination.
[0032] Further, in step S400, randomly selecting an order from the task allocation combination and assigning it to a picker for picking until all valid orders in the task allocation combination are assigned include: Define the state space based on the picker's workload, the task priority of valid orders, and the picker's processing efficiency ; Define the action space of selecting an order from the task assignment combination and assigning it to the picker ; Design reward function based on distribution results , used to evaluate the effect of each allocation; among them, , , and is the weight coefficient, is the task priority, is the picker's workload and the number of orders to be processed, To predict processing time, For processing efficiency; Based on Q-learning, the reward value Q assigned to the order is calculated, and the reward value is assigned to the corresponding picker based on the preset standard. The Q value update formula is: ; in, is the learning rate, is the discount factor, is the maximum Q value in the next state.
[0033] Example 2 Based on the same inventive concept, the present invention provides an order distribution system that avoids pickers from picking orders and waiting for orders, using the order distribution method as described above, including: A response module, used to respond to the picking request of the picker; The screening module is used to screen out valid orders that meet preset conditions from the orders to be allocated according to the picking waves and picking zones for order allocation; A data processing module is used to allocate and combine the effective order synthesis tasks according to the configured random store quantity and the task priority of the effective orders; The allocation module is used to randomly select an order from the task allocation combination and assign it to the picker for picking until all valid orders in the task allocation combination are allocated; based on the completion of the allocation of all valid orders in the current task allocation combination, the unallocated valid orders are regenerated into a new task allocation combination for allocation until all valid orders are allocated.
[0034] Furthermore, the preset conditions for screening in the screening module are: ; in, is a set of valid orders, For the i-th order, For waves, For waves, state.
[0035] Furthermore, the data processing module includes: A random store data generation unit, used to configure the random number of stores R in each picking area, and to randomly select valid orders of R stores from each picking area for distribution; A predicted processing time calculation unit, used to calculate the predicted processing time of each valid order; The task priority sorting unit is used to sort the task priorities according to the predicted processing time, urgency and quantity of the valid orders selected based on the random number R of stores, and generate task allocation combinations.
[0036] Furthermore, the allocation module also includes, Dynamic load balancing unit for defining state space based on picker workload, task priority of valid orders, and picker processing efficiency ; Define the action space of selecting an order from the task allocation combination and assigning it to the picker ; Design reward function based on distribution results , used to evaluate the effect of each allocation; among them, , , and is the weight coefficient, is the task priority, is the picker's workload and the number of orders to be processed, To predict processing time, For processing efficiency; based on Q-learning, the reward value Q assigned to the order is calculated, and the reward value is assigned to the corresponding picker based on the preset standard. The Q value update formula is: ;in, is the learning rate, is the discount factor, is the maximum Q value in the next state.
[0037] Example 3 In order to enable technical personnel in this field to have a deeper understanding of the present technical solution, the following will combine with the third specific embodiment to elaborate on the order allocation method based on the random number of stores, task priority and dynamic load balancing, and explain its role and effect in practical applications.
[0038] The present invention provides an order allocation method that avoids pickers from picking and waiting for orders, aiming to achieve fairness and efficiency in task allocation for pickers through dynamic task allocation and a reasonable load balancing mechanism. The specific implementation steps are as follows: Step S100: responding to the picking request of the picker; Specifically, the system receives the picking request from the picker and triggers the picking task allocation mechanism according to the content of the request. When the picker issues a picking request, the system prepares the order data to be allocated according to the current picking wave, picking partition and workload. This step ensures that the picker's request can be responded to in a timely manner and initiates the subsequent task allocation process.
[0039] Step S200: According to the picking wave and picking zone, select valid orders that meet preset conditions from the orders to be allocated and allocate them; Specifically, in this step, valid orders that meet the conditions in the pending orders are screened out according to the current picking wave and picking partition. First, valid orders that meet the current conditions are screened out according to the picking wave, picking partition, and order status of each order, such as pending, assigned, etc. This step ensures that pickers only take tasks from orders that are consistent with their work scope and are currently assignable, thereby avoiding unnecessary waste and incorrect allocation of orders. For example, the definition formula of the valid order set is set as, ; in, is a set of valid orders, For the i-th order, For waves, For waves, state.
[0040] Step S300: according to the configured random store quantity and the task priority of the valid order, the valid order synthesis task is assigned and combined; According to the configuration of each picking area, determine the number of random stores R in each area. This number will affect the number of valid orders selected from the area. The random store selection mechanism can prevent pickers from picking certain specific orders, thereby avoiding skewed distribution of tasks. Each picking area may have a different number of random stores configured, which is dynamically adjusted according to actual needs. The specific configuration rules can be optimized based on factors such as the order volume of the area and the difficulty of order processing.
[0041] Secondly, XGBoost is used to predict the processing time of each order to ensure that more complex or urgent orders are processed first. At the same time, tasks are prioritized based on the predicted processing time, urgency, and number of orders. The specific calculation process is as follows: (1) Training the time prediction model based on the collected order data set: including: ①Build a training data set: The dataset includes the feature vector for each order i. and actual processing time, where the feature vector includes product type, quantity, weight, difficulty, distance, and historical processing time, etc. For example, the order data set is ,in, is the characteristic vector of order i, is the actual processing time of order i, is the target variable; the feature vector ; are input features, including product type, quantity, weight, difficulty, distance, and historical processing time, and N is the number of orders; According to these N order data, the feature vector As input, the actual processing time is used as output to train a time prediction model In the time prediction model, a decision tree is used to predict the processing time of each order. Assume that we use K trees for prediction, and the output of the Kth tree is , each tree makes predictions through leaf nodes, and the calculation formula of the time prediction model is: ; ; in, is the predicted processing time of order i, K is the number of trees, is the predicted value of the kth tree for order i; is the number of leaf nodes of the kth tree, is the weight of the jth leaf node of the kth tree, is the indicator function, used when the input feature The region that falls into the jth leaf node of the kth tree 1 when it is, otherwise 0; During the training process, in order to optimize the model parameters, that is, the split point of each tree, the weight of the leaf node, etc., the error between the predicted value and the actual processing time is minimized through the objective function. The objective function is, The loss function is ,The difference between the processing time predicted by the model and the actual processing time is calculated based on the mean square error; , is a regularization term used to control the complexity of the time prediction model and prevent overfitting. , is the regularization hyperparameter.
[0042] is the final objective function, Based on the trained time prediction model, the feature vector As input, the predicted processing time for valid orders is calculated.
[0043] (2) The calculation formula for task priority ranking is: ; in, is the task priority, Forecasted processing time for active orders, The urgency of the valid order, ranging from 0 to 1, is the order quantity of the valid order, The processing time for valid orders. , and is the weight parameter. All valid orders are sorted according to their task priorities to generate a task allocation combination.
[0044] Step S400: Randomly select an order from the task allocation combination and assign it to the picker for picking until all valid orders in the task allocation combination are assigned; In step S400, randomly selecting an order from the task allocation combination and assigning it to a picker for picking until all valid orders in the task allocation combination are assigned, including: Define the state space based on the picker's workload, the task priority of valid orders, and the picker's processing efficiency , ; Define the action space of selecting an order from the task assignment combination and assigning it to the picker That is, all possible operations of selecting an order from the task allocation combination and assigning it to a picker. Each selection action corresponds to randomly selecting an order from the task allocation combination for allocation.
[0045] In order to optimize when selecting the most suitable picker for order assignment, a reward function is designed , which is used to evaluate the effect of each order allocation action. The calculation of the reward function needs to consider multiple factors, such as ①The higher the task priority, the greater the impact of the order on the picker's load, so it should be assigned first.
[0046] ② The picker’s workload. The picker’s current workload affects his or her processing speed. Assigning more orders to a picker with a higher workload may result in lower processing efficiency.
[0047] ③ The predicted processing time of the order. Orders with longer processing time require more work, so the decision should be made based on a comprehensive consideration of the processing efficiency assigned to the current picker.
[0048] ④ The picker’s processing efficiency. The picker’s historical processing efficiency should affect task allocation. Pickers with lower processing efficiency should not be over-assigned high-priority complex orders.
[0049] Therefore, the reward function is designed as ;in, , , and is the weight coefficient, is the task priority, is the picker's workload and the number of orders to be processed, To predict processing time, For processing efficiency; Based on Q-learning, the reward value Q assigned to the order is calculated, and the reward value is assigned to the corresponding picker based on the preset standard. The Q value update formula is: ; in, is the learning rate, is the discount factor, is the maximum Q value in the next state.
[0050] Specifically, is in state Next select action The Q value, It is an immediate reward after assigning a task. is a discount factor used to consider the impact of future rewards, is the learning rate, which controls the speed of Q value update. In each task assignment, the system selects the best order for assignment based on the current Q value. Specifically, the system selects the order with the largest Q value to assign to the picker, thereby improving the overall efficiency of the picking task.
[0051] For example, suppose at a certain moment, the current state When the picker has a high workload, a high task priority, a large number of orders to be assigned, and a low picker processing efficiency, Q-learning will calculate the Q value and select the most suitable order for assignment.
[0052] Step S500: based on the completion of all valid orders in the current task allocation combination, the unallocated valid orders are regenerated into a new task allocation combination according to step S200 and allocated until all valid orders are allocated.
[0053] Although the present invention has been disclosed as above in the form of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art may make possible changes and modifications to the technical solution of the present invention by using the methods and technical contents disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope of the technical solution of the present invention.
Claims
1. An order distribution method that avoids pickers from picking orders and waiting for orders, characterized in that: include: Step S100: responding to the picking request of the picker; Step S200: According to the picking wave and picking zone, select valid orders that meet preset conditions from the orders to be allocated and allocate them; Step S300: according to the configured random store quantity and the task priority of the valid order, the valid order synthesis task is assigned and combined; Step S400: randomly selecting an order from the task allocation combination and assigning it to the picker for picking, until all the valid orders in the task allocation combination are assigned; Step S500: based on the completion of the allocation of all the valid orders in the current task allocation combination, the unallocated valid orders are regenerated into a new task allocation combination according to step S200 and allocated until all the valid orders are allocated.
2. The order allocation method according to claim 1, characterized in that: In step S200, according to the picking wave and picking zone, the valid orders meeting the preset conditions are selected from the orders to be allocated for order allocation, which includes: ； in, is a set of valid orders, For the i-th order, For waves, For waves, state.
3. The order allocation method according to claim 1, characterized in that: In step S300, according to the configured random store quantity and the task priority of the valid order, the valid order synthesis task allocation combination includes: Step S310: configuring the random number R of stores in each picking area, so as to randomly select the valid orders of R stores from each picking area for distribution; Step S320: Calculate the predicted processing time of each of the valid orders; Step S330: Prioritize the tasks according to the predicted processing time, urgency and quantity of the valid orders selected based on the random store number R, and generate the task allocation combination.
4. The order allocation method according to claim 3, characterized in that: In step S320, calculating the predicted processing time of each of the valid orders includes: The time prediction model is trained based on the collected order data set: wherein the order data set is ,in, is the characteristic vector of order i, is the actual processing time of the order i, which is the target variable; the feature vector ; are input features, including product type, quantity, weight, difficulty, distance, and historical processing time, and N is the number of orders; The calculation formula in the time prediction model is: ； ； in, is the predicted processing time of the order i, K is the number of trees, is the predicted value of the kth tree for the order i; is the number of leaf nodes of the kth tree, is the weight of the j-th leaf node of the k-th tree, is an indicator function used when the input feature The region that falls into the jth leaf node of the kth tree 1 when it is, otherwise 0; During the training process, the model parameters are optimized by minimizing the objective function, which is, ; ； ； in, is the loss function used to calculate the prediction processing time The actual processing time The gap, is a regularization term used to control the complexity of the time prediction model; , is the regularization hyperparameter; Based on the trained time prediction model, the feature vector As input, the predicted processing time for the active orders is calculated.
5. The order allocation method according to claim 4, characterized in that: In step S330, the predicted processing time, urgency and quantity of the valid orders selected based on the random store number R are sorted by task priority to generate the task allocation combination, including: The calculation formula for the task priority ranking is: ； in, is the task priority, the predicted processing time of the active order, The urgency of the valid order, ranging from 0 to 1, is the order quantity of the valid order, is the processing time for the valid order, , and is the weight parameter; All the valid orders are sorted according to the task priorities of the valid orders to generate a task allocation combination.
6. The order distribution method according to claim 5, characterized in that: In step S400, randomly selecting an order from the task allocation combination and assigning it to the picker for picking until all the valid orders in the task allocation combination are assigned, includes: A state space is defined based on the workload of the picker, the task priority of the valid order, and the processing efficiency of the picker. ; Define an action space for selecting one of the order from the task assignment combination and assigning it to the picker ; Design reward function based on distribution results , used to evaluate the effect of each allocation; among them, , , and is the weight coefficient, is the task priority, is the workload of the picker and the number of orders to be processed, is the prediction processing time, is the processing efficiency; The reward value Q assigned to the order by the picker is calculated based on Q-learning, and is assigned to the corresponding picker based on the reward value reaching the preset standard. The Q value update formula is: ； in, is the learning rate, is the discount factor, is the maximum Q value in the next state.
7. An order distribution system that avoids pickers from picking orders and waiting for orders, using the order distribution method according to any one of claims 1 to 6, characterized in that: include, A response module, used to respond to the picking request of the picker; The screening module is used to screen out valid orders that meet preset conditions from the orders to be allocated according to the picking waves and picking zones for order allocation; A data processing module, used for allocating and combining the effective order synthesis tasks according to the configured random store quantity and the task priority of the effective order; The allocation module is used to randomly select an order from the task allocation combination and allocate it to the picker for picking until all the valid orders in the task allocation combination are allocated; based on the completion of the allocation of all the valid orders in the current task allocation combination, the unallocated valid orders are regenerated into a new task allocation combination for allocation until all the valid orders are allocated.
8. The order distribution system according to claim 7, characterized in that: The preset conditions screened in the screening module are: in, is a set of valid orders, For the i-th order, For waves, For waves, state.
9. The order distribution system according to claim 7, characterized in that: The data processing module comprises: A random store data generating unit, used to configure the random store quantity R of each picking area, and to randomly select the valid orders of R stores from each picking area for distribution; A predicted processing time calculation unit, used to calculate the predicted processing time of each of the valid orders; A task priority sorting unit is used to sort the task priorities according to the predicted processing time, urgency and quantity of the valid orders selected based on the random store number R, and generate the task allocation combination.
10. The order distribution system according to claim 7, characterized in that: The allocation module also includes: A dynamic load balancing unit, for defining a state space based on the workload of the picker, the task priority of the valid order, and the processing efficiency of the picker ; Define the action space of selecting one of the orders from the task assignment combination and assigning it to the picker ; Design reward function based on distribution results , used to evaluate the effect of each allocation; among them, , , and is the weight coefficient, is the task priority, is the workload of the picker and the number of orders to be processed, is the prediction processing time, is the processing efficiency; based on Q-learning, the reward value Q assigned to the order by the picker is calculated, and the reward value is assigned to the corresponding picker based on the preset standard. The Q value update formula is: ;in, is the learning rate, is the discount factor, is the maximum Q value in the next state.