JP6937054B1 - Three-dimensional automated warehouse in logistics - Google Patents

Abstract

PROBLEM TO BE SOLVED: To minimize a time in a warehousing process, and to easily perform an automatic process even in a situation where the combinatorial explosion occurs. SOLUTION: Order data reception A, delivery tray and delivery station determination B, collecting all luggage in a temporary storage space on a shelf facing one passage, and shipping to a picking station. In the flow of warehousing consisting of D, E to be packed by the worker, and shipping F, the explanatory variables related to the warehousing enter the input layer 11 of the artificial intelligence, and each is judged by the hidden layer 12 and output. The delivery tray and the delivery station are determined by deep learning that outputs the aggregated prediction time from the layer 13 and the prediction means by LightGBM based on the decision tree, and the reserved delivery tray is placed in one passage of the delivery station. Consolidate on the facing shelves. [Selection diagram] Fig. 1

Description

The present invention relates to a three-dimensional automated warehouse in physical distribution related to warehousing, among warehousing / warehousing / re-warehousing.

Recently, the mail-order business (e-commerce) using the Internet is flourishing, but the mechanism of the distribution center of such a business is that the best-selling products are dealt with frequently and the niche products that rarely receive orders (low). We have responded to the diverse demands of our customers by offering a large number of frequent long-tail products. For this reason, we would like to secure the site of the distribution center as large as possible, but the land price is high around large markets such as Tokyo and Osaka, and land acquisition is not possible. Therefore, the mechanism for shipping top-selling products is required to be performed efficiently in the narrowest possible space in order to secure a storage space for long-tail products, and as a material handling device that meets that demand, goods are temporarily stored at high speed. A mechanism was devised to use a high-capacity three-dimensional automated warehouse that can be delivered in order for loading at the time of shipment. The equipment configuration of this three-dimensional automated warehouse is provided in each stage to pick up goods in each stage using a storage rack consisting of multiple stages of shelves, a passage provided between the plurality of storage racks, and the passage. It is composed of a traveling trolley and a lifting device that controls the vertical movement of a multi-stage storage rack, and is operated in combination with a Goods to Person (GTP) station that quickly and appropriately loads the delivered goods. There are many things .

The basic operation when shipping the goods temporarily stored in such a three-dimensional automatic warehouse is to issue a warehouse management system (called Warehouse Management System WMS) to the goods stored on the shelves to issue a shipping order to the goods to be shipped. This is transmitted to the warehouse control system (called the Warehouse Control System WCS) that controls various matehan equipment such as the three-dimensional automatic warehouse, conveyor, and sorter in the distribution center, and the goods to be shipped stored in the three-dimensional automatic warehouse. (There may be a case where a plurality of target articles are stored in an at random place in the three-dimensional automatic warehouse), the article having a high throughput for unloading from the three-dimensional automatic warehouse is picked up, and the warehousing operation is started. First, a traveling trolley on the same stage as the target article shelf travels to the front of the target article shelf in order to pick up the target article, and the target article is picked up and traveled using the telescopic arm provided in the traveling trolley. Reprinted in the trolley. The traveling carriage travels to the temporary standby conveyor provided on the stage, and transfers the article to the temporary standby conveyor. The article transferred to the temporary standby conveyor is then picked up by an elevating device that can travel in the vertical direction and an arm provided in the elevating device, and elevates to the stage provided with the delivery transport line. Transfer to the temporary standby conveyor provided in. The transferred goods are transferred to the Goods to Person (GTP) station provided for arranging the goods and delivered to the delivery transfer line. The delivered goods are picked up by the worker at the GTP station and the delivery work is completed.

In the warehousing work in this three-dimensional automated warehouse, when various goods are transported by truck and arrive at the distribution center, they are unboxed and then stored in the goods storage. After that, according to the delivery plan of the distribution center, the target goods are sequentially delivered from the goods storage according to the sunrise storage plan and stored in the three-dimensional automated warehouse for loading. As for the storage process, WMS issues a warehousing order for the goods to be warehousing, and most of them are warehousing transport lines, temporary standby conveyors, elevating devices, and temporary standby conveyors in order to store the goods in a specific place in the three-dimensional automated warehouse. , Transport the goods to the frontage of the target storage rack shelf while relaying the goods in the order of the traveling trolley. The goods arriving at the frontage are stored on the shelves by the arms of the traveling trolley, and the warehousing work is completed.

Next, in the three-dimensional automated warehouse, there is a mechanism for collecting goods in a specific aisle across the aisles and issuing the goods to improve the total efficiency of the warehousing work. The mechanism to cross the inter-aisle using a conveyor provided outside the side of the aisle, and a method of crossing by aggregating, using arms of the traveling carriage between each aisle, stored in the storage rack There are two types of mechanisms for moving between shelves, in which goods are picked up by the traveling trolley of the adjacent Isle (hereinafter referred to as the shuttle), and then picked up by the traveling trolley of the adjacent Isle.

As mentioned above, the mechanism of the 3D automated warehouse has become complicated these days, and there are both a method of using a conveyor for aggregation work outside the 3D automated warehouse and a method of moving between shelves. Which method is used and how much is the final throughput higher? What about the busy and off-seasons? And so on, various problems came up. In particular, when the number of aisles in a three-dimensional automated warehouse increases, the number of combinations of shipping routes becomes enormous, and even if humans try to solve this by programming, it has become extremely difficult to derive the optimum solution. Therefore, it is required to use artificial intelligence to solve such problems.

In this three-dimensional automated warehouse, the flow of warehousing is the reception of order data, the determination of the warehousing station provided at the end of the returnable box (delivery tray) containing the warehousing goods and the warehousing rack, and 1 for all luggage. It consists of collecting in a temporary storage space on a shelf facing each aisle, shipping to a GTP station (picking station), packing by an operator, and shipping.

nothing special

However, in the conventional determination of the delivery tray and the delivery station, the calculation process is performed based on the shortest moving distance of the delivery tray in the three-dimensional automated warehouse based on the original logic, and the number of decisions that can obtain the shortest time among them is several. It was artificially selected using individual explanatory variables. Dozens of explanatory variables are required to obtain the shortest time more accurately. When there are N necessary explanatory variables in the decision, when there are 3 conditional branches by each explanatory variable, 10 explanatory variables are about 60,000 (3 to the Nth power), and 20 are 3 The shortest time from the result of the arithmetic processing is that the calculation processing amount increases to the N (number of explanatory variables) power each time the number of explanatory variables increases, such as ,000,000,000,000. It was very difficult for humans to make the decision in terms of processing power.

In determining the issuing tray and the issuing station, there is a tray aggregation step of aggregating the issuing trays to the issuing station and a station issuing step of aggregating the aggregated trays to the picking station. It accounts for 84.6% of the total shipping process. Therefore, shortening the tray aggregation is indispensable for improving the delivery speed.

Therefore, the present invention has been devised in view of the above-mentioned conventional circumstances, and it is possible to shorten the time required for the warehousing process to the minimum, and even under the situation where the combinatorial explosion occurs. An object of the present invention is to provide a three-dimensional automated warehouse in physical distribution that can be easily and automatically processed.

In order to solve the above-mentioned problems, in the present invention, the order data is received, the delivery tray and the delivery station are determined, and all the luggage is stored in the temporary storage space of the shelf facing one passage. In the flow of shipping, which consists of collecting, shipping to the picking station, packing by the worker, and shipping, the explanatory variables related to shipping enter the input layer of artificial intelligence, and each is important in the hidden layer. The delivery tray and delivery station are currently determined by deep learning that determines various explanatory variables and outputs the optimum aggregation prediction time from the output layer (objective variable) and the prediction means by LightGBM based on the decision tree. It is characterized in that the delivery trays determined and allocated for each product of the order are collected on a shelf facing one passage of the delivery station.

According to this configuration, by adopting artificial intelligence, the delivery tray and the delivery station are automatically determined according to the learning by deep learning or LightGBM, so that the delivery processing time can be minimized in the three-dimensional automated warehouse. ..

The predictive estimation means uses deep learning and LightGBM to learn a model by separating the movement between the trolley shelves and the movement on the off-shelf conveyor, and aggregates the objective variables from a plurality of explanatory variables via artificial intelligence. By outputting the estimated time, it has become possible to realize a warehouse equipped with a traveling trolley that runs on the aisle between the warehousing / delivery section and the storage rack, which has a short warehousing / delivery processing time.

According to this configuration, since the artificial intelligence learns in a pattern, it is possible to quickly perform the process of outputting the aggregated predicted time, which is the objective variable, from the N explanatory variables via the artificial intelligence.

According to the present invention, it is possible to provide a three-dimensional automated warehouse in physical distribution that can reduce the time required for delivery processing to a minimum and can easily perform automatic processing even in a situation where a combinatorial explosion occurs.

It is a perspective view which shows the flow of delivery in one form for carrying out this invention together with the flowchart. It is a schematic diagram which shows the aggregation time prediction model in one embodiment for carrying out this invention. It is a figure which shows the explanatory variable of the movement time prediction model between traveling carriage shelves in one embodiment for carrying out this invention. It is a figure which shows the explanatory variable of the off-shelf conveyor movement time prediction model in one embodiment for carrying out this invention. It is a figure which shows the correlation of the moving model between traveling carriage shelves, and the explanatory variable in one embodiment for carrying out this invention. It is a figure which shows the correlation of the off-shelf conveyor movement model objective variable and explanatory variable in one embodiment for carrying out this invention.

Hereinafter, the three-dimensional automated warehouse according to the embodiment of the present invention will be described in detail with reference to the drawings.

The logic of the three-dimensional automated warehouse according to this embodiment is roughly classified into warehousing, warehousing, and re-warehousing. Of these, the explanation will be limited to shipping. First, as shown in FIG. 1, in a three-dimensional automated warehouse having a warehousing unit and a traveling trolley traveling in a passage (isle) between storage racks, a transport mechanism that travels horizontally between adjacent storage racks is high. It is arranged in each stage in the horizontal direction. Then, when the delivery tray moves to the delivery section, the transport mechanism has a picking mechanism that allows the delivery tray to be taken in and out of both storage racks located in the left-right direction.

In the present embodiment, artificial intelligence for determining the delivery tray and the delivery station for collecting all the luggage in the temporary storage space of the delivery passage, particularly related to deep learning and LightGBM, is related to receiving order data A and artificial intelligence. Determining the delivery tray and delivery station B, collecting all the luggage in the temporary storage space of the shelf facing one passage C, shipping to the picking station D, the worker packing E, shipping F Will be done.

That is, when the traveling trolley that travels in the passage between the warehousing / delivery section and the storage rack is controlled, the transport mechanism that travels horizontally between the adjacent storage racks is controlled, and the reserved warehousing tray moves to the warehousing section. The control of the picking mechanism, which is provided in the transport mechanism and allows the delivery trays to be taken in and out of both storage racks located in the left-right direction, is all determined by artificial intelligence (particularly deep learning and LightGBM). The allocation tray is a tray determined for each item in the current order.

Conventionally, the shipping process is performed in the following two stages. A tray aggregation step of aggregating the goods issue tray unloading stations, there is a station issuing step of issuing the aggregated tray picking station. Focusing on the tray aggregation step among these, in the present embodiment, this is performed by using artificial intelligence (AI), particularly by deep learning or a predictive estimation means by LightGBM. Until now, the time required for tray aggregation accounted for 84.6% of the total shipping process. Therefore, it is considered indispensable to improve the delivery speed by shortening the tray aggregation.

In addition, as the logic for tray aggregation, in determining the delivery station based on a predetermined program, first, the tray path and prediction for the movement between the trolley shelves facing one aisle and the movement of the conveyor outside the shelves. Calculate the aggregation time. This predicts the aggregation time from the trays that satisfy the aggregation conditions to the shelves whose processing amount is less than the threshold value, and is calculated for all trays and all shelves. The trays considered here are all trays that can be aggregated.

The WCS program for issuing goods is basically a rule-based program in which human beings create a program while considering a large number of conditions. It is executed in the order of determination, and in the aggregation decision of the delivery tray, the tray is determined for each product of the current order in the following priority order. First, in the single tray allocation, a search is performed for trays having inventory equal to or greater than the required number of products, and the products are aggregated. If there is no tray that meets the required number of products in the delivery trays that are subject to multiple allocations, the trays that have one or more corresponding products are searched until the number of allocations meets the required number, and the trays are aggregated. In the out-of-stock delivery tray, if the required number of products is not satisfied even if multiple trays are allocated, the out-of-stock tray with the corresponding product is searched and aggregated.

In determining the delivery tray for each shelf facing one passage, the delivery tray with the shortest path and predicted aggregation time is required for all shelves and all product IDs (face to one passage). Shelf, product ID, tray), delivery tray by moving between trolley shelves and delivery tray by moving off-shelf conveyor, the throughput is maximized based on the determined rule base.

Here, the issue AI determines the aggregation aisle and the issue tray so that the aggregation time is the shortest. Here, the aggregated aisle is an aisle on which a traveling trolley travels when the delivery trays are aggregated on a shelf facing one aisle (isle). Since this aggregation time is not directly known, it is predicted from past data using a model, and various information is considered for model prediction. The rule base determines the aggregate aisle and the shipping tray so that the travel distance is the shortest. In aggregate aisle selection, the number of issue trays that can be allocated is large, the amount of processing (number of items waiting to be issued) is small, and the youth of the index is considered in order. ..

It is not enough to reduce the travel distance in order to shorten the aggregation time. Therefore, in the present embodiment, the prediction estimation means by the following aggregation time prediction model is handled. This emphasizes high prediction accuracy and ease of analysis, and uses deep learning, which is a model similar to the cranial nerves, or LightGBM for the delivery AI. The model is learned separately for the movement between the trolley shelves and the movement of the conveyor outside the shelves. That is, as shown in FIG. 2, the explanatory variables 1, the explanatory variables 2, the explanatory variables 3, ... The explanatory variables N enter the input layer 11 of the artificial intelligence, and N first hidden layers 12a and second, respectively. The hidden layer 12b determines important explanatory variables. Then, the aggregate prediction time, which is an objective function, is output (objective variable) from the output layer 13.

In prediction estimating means creates a pair of unloading trays candidate is provision between the current Yakua yl candidate, predicts the aggregation time by AI model for each pair. At this time, the traveling trolley inter-shelf movement model and the off-shelf conveyor movement model are used properly, the aggregation time is predicted, and the shortest pair of aggregation aisle and delivery tray is determined.

Next, an example of use / operation of the form configured as described above will be described.

FIG. 3 shows the explanatory variables of the traveling time prediction model between the traveling carriage shelves, and there are explanatory variable names and corresponding English names. That is, the prediction estimation means enters the input layer 11 of the artificial intelligence into the following explanatory variables, determines important explanatory variables in the first hidden layer 12a and the second hidden layer 12b, and then aggregates the predicted time from the output layer 13. Is output (objective variable), making it possible to realize a three-dimensional automated warehouse with a short delivery processing time.
(1) Source passage [from_aisle]
(2) Destination passage [to_aisle]
(3) Distance between passages [aisle_distance]
(4) Elevating device moving distance of the moving source passage [from_floor_distance]
(5) Elevating device moving distance of the destination passage [to_floor_distance]
(6) Number of warehousing tasks for trolleys in the aisle (moving source, mediation, moving destination) [recept_task_num_k (k is 0 to 8, 0 is the moving source, 8 is the moving destination)]
(7) Number of tasks for moving between shelves of traveling carts in the aisle (moving source, mediation, moving destination) [inter_aisure_task_num_k (k is 0 to 8, 0 is the moving source, 8 is the moving destination)]
(8) Number of tasks for leaving the trolley in the aisle (moving source, mediation, moving destination) [retrieval_task_num_k (k is 0 to 8, 0 is the moving source, 8 is the moving destination)]
(9) Distance between the initial position of the traveling carriage and the tray [initial_shottle_dostance]
(10) Number of trays that need to be moved to the rearrangement frontage in order to take out the tray [is_num]
(11) Number of trays existing in the delivery buffer [retrieval_buffer_occupancy]
(12) Flag for busy hours (0: no, 1: yes) [busy_time]
(13) Day of the week [weekday]
(14) Day [day]
(15) Number of waits in the moving source passage [from_aisle_waiting_num]
(16) Number of waits in the destination passage [to_aisle_waiting_num]
(17) Number of trays waiting to be delivered (number of trays placed in the temporary storage space to be brought to ST) [waiting_picking_num]
(18) Number of warehousing instructions in the moving source passage [from_aisle_store_comand_num]
(19) Number of delivery instructions in the moving source passage [from_aisle_retrieval_comand_num]
(20) Total number of warehousing requests for the entire 3D automated warehouse within 10 minutes [total_store_request]
(21) Total number of requests for issue of the entire 3D automated warehouse within 10 minutes [total_retrieval_request]
(22) Number of warehousing trays allocated in the source aisle [from_aisle_store_request]
(23) Number of delivery trays allocated in the source aisle [from_aisle_retrieval_request]
(24) Number of warehousing trays allocated in the destination aisle [to_aisle_store_request]
(25) Number of delivery trays allocated in the destination aisle [to_aisle_retriever_request]

FIG. 4 shows the explanatory variables of the off-shelf conveyor movement time prediction model, and there are explanatory variable names and corresponding English names. That is, the prediction estimation means outputs the aggregated prediction time from the output layer 13 after the following explanatory variables enter the input layer 11 of the artificial intelligence and are judged by the first hidden layer 12a and the second hidden layer 12b (objective variable). ) And make it possible to realize a three-dimensional automated warehouse with a short delivery processing time.
(1) Source passage [from_aisle]
(2) Destination passage [to_aisle]
(3) Distance between passages [aisle_distance]
(4) Number of unprocessed warehousing tasks held by all passing trolleys at the start of inter-shelf movement [recept_task_num]
(5) Number of unprocessed inter-shelf movement tasks held by all passing carriages at the start of inter-shelf movement [inter_aisle_task_num]
(6) Number of unprocessed warehousing tasks held by all passing trolleys at the start of inter-shelf movement [retrieval_task_num]
(7) Distance from the initial position of the traveling bogie (previous task completion position) to the Rack position (absolute value of integer difference) [intial_shuttle_distance]
(8) Distance to the rack (closer) for moving between shelves (absolute value of the difference in the number of racks) [near_rac_distance]
(9) Distance to the rack (farther) for moving between shelves (absolute value of the difference in the number of racks) [far_rac_distance]
(10) Number of trays that need to be moved to the alternative space in order to take out the target tray (= number of obstructing luggage. 0, 1, 2) [is_num]
(11) Number of trays existing in the delivery buffer [retrieval_buffer_occupancy]
(12) Day of the week [weekday]
(13) Day [day]
(14) Number of waits in the moving source passage [from_aisle_waiting_num]
(15) Number of waits in the destination passage [to_aisle_waiting_num]
(16) Number of trays waiting to be delivered (number of trays placed in the temporary storage space to be brought to ST) [waiting_picking_num]
(17) Number of warehousing trays allocated in the source aisle [from_aisle_store_comand_num]
(18) Number of delivery trays allocated in the source aisle [from_aisle_retrieval_comand_num]
(19) Number of warehousing trays allocated in the destination aisle [to_aisle_store_request]
(20) Number of delivery trays allocated in the destination aisle [to_aisle_retriever_request]

The correlation between the explanatory variable and the objective variable in the aggregation time prediction model will be described. First, in the correlation between the traveling model inter-shelf movement model objective variable and the explanatory variable, Correlation indicates a correlation coefficient and takes a value of -1 to +1. The Connection rate indicates how much the variable contributed to the prediction of the objective variable, and takes a value of 0 to 1, and the larger this value is, the more it contributes. In the figure, Correlation has the largest path with an aisle distance [aisle_distance] of "0.69", and Correlation rate has the largest number of picking [waiting_picking_num] waiting for "1".

Next, in the correlation between the objective variable and the explanatory variable of the off-shelf conveyor movement model, Correlation shows the correlation coefficient in the same manner and takes a value of -1 to +1. The Connection rate indicates how much the variable contributed to the prediction of the objective variable, and takes a value of 0 to 1, and the larger this value is, the more it contributes. In the figure, Correlation has the largest path of the distance between aisles [aisle_distance] of "0.82", and Correlation rate has the largest number of picking [waiting_picking_num] waiting for "1".

The performance evaluation index of the aggregation time prediction model and the accuracy of the prediction time of the aggregation time prediction model will be described. First, as a verification environment for AI model performance, a 5-fold examination and verification method is used in an aggregated time model by moving between trolley shelves, and the data is divided into training data and test data for verification. The average prediction time of this is 140.81 seconds, the average error is 36.98 seconds, and the average error / average prediction time = 22.5%.

In the aggregation time model by moving the off-shelf conveyor, the 5-fold cross-validation method is used, and the data is divided into training data and test data for verification. The average prediction time of this is 219.25 seconds, the average error is 27.57 seconds, and the average error / average prediction time = 11.1%.

As described above, in the present embodiment, it is possible to provide a three-dimensional automated warehouse that can minimize the time required for shipping processing and can easily automatically process even if a combinatorial explosion occurs. can.

A ... Receiving order data B ... Determining the artificially intelligent delivery tray and delivery station C ... Collect all luggage in the temporary storage space on the shelf facing one aisle D ... Ship to the picking station E ... Workers pack F ... Shipment 11 ... Input layer 12 ... Hidden layer 13 ... Output layer

Claims (2)

When the order data is received, the predictive estimation means determines which delivery tray related to the order data is to be aggregated in which aggregation aisle among the delivery trays containing the goods to be delivered, and all the delivery trays related to the order data . goods Issue tray said determined for luggage is collected on the determined aggregation aisle, the unloading tray collected are shipped to picking station, said goods issue object article from the unloading tray that is shipped to the picking station is taken are packaged, the packaged-the goods issue object article is shipped, a three-dimensional automatic warehouse,
The predictive estimation means has a traveling carriage inter-shelf movement model and an off-shelf conveyor moving model, and in the traveling carriage inter-shelf moving model and the off-shelf conveyor moving model, an aggregateable aisle candidate and an allocation tray candidate can be allocated. As explanatory variables related to warehousing, the source passage, the destination passage, the distance between passages, the number of tasks to move between the shelves of the traveling trolley of the passage, the number of trays existing in the warehousing buffer, the day, the day, and the movement source. Artificial intelligence has at least an explanatory variable that has at least the number of waiting passages, the number of waiting passages, the number of waiting trays for leaving, the number of warehousing trays allocated in the destination passage, and the number of shipping trays allocated in the destination passage. enters the input layer is determined by each of the explanatory variables hidden layer, an output layer or al current deep learning output approximately prediction time, or a decision tree to LightGB M that is based Accordingly, said aggregate estimated time solid automatic warehouse in distribution to unloading tray and determine the aggregate aisle Teisu characterized Rukoto according to the pair having the shortest. Solid automatic warehouse in further distribution of請Motomeko 1, further comprising an inlet-outlet unit.

Images