Order prediction and picking out-of-warehouse method, device and storage medium based on digital twin

Abstract

The present invention relates to an order prediction, picking and dispatching method, device and storage medium based on digital twins, which is applied to the field of logistics warehousing technology, including: based on a digital twin model, by predicting future orders, while processing current orders, orders with a high probability of being generated in the next time period can be considered, and the goods contained therein can be dispatched to the conveyor line in advance, thereby improving the response speed of the picking system to orders, and by simulating the order processing sales volume and order response time of the picking system under different dispatching strategies, the dispatching strategy introducing order prediction is optimized to ensure that the strategy can improve the performance of the current picking system.

Description

Technical Field

[0001] The present invention relates to the field of logistics warehousing technology, and in particular to an order prediction picking and delivery method, device and storage medium based on digital twins. Background Art

[0002] The logistics warehousing system is an important part of the on-site logistics of manufacturing enterprises, usually composed of conveyor lines, shuttle vehicles, stackers, three-dimensional warehouses, AGVs, management systems, and dispatching systems. With the rapid development of the warehousing and logistics industry in recent years, its automation level has become increasingly higher. Suppliers of warehousing and logistics systems can provide customized solutions based on customer needs.

[0003] Picking, also known as the goods-to-person picking system, in the warehousing and logistics system means that people do not move, and the goods are automatically transported to people for selection by equipment; usually, the picking system includes two processes: warehousing and outbound.

[0004] Warehousing process: The goods are manually stacked in boxes (or pallets), and the product information (including variety, quantity, etc.) is bound to the boxes in the management system. The boxes are placed on the conveyor line, which transports the boxes to the warehouse end conveyor of the automated warehouse. The stacker forks the boxes and stores them in the automated warehouse.

[0005] Outbound process: According to the outbound order, the stacker takes out the box containing the required goods from the automated warehouse, transports it to the warehouse end conveyor, and then the conveyor system delivers it to the worker. The worker takes out the corresponding number of goods from the box according to the type of goods and the required quantity prompted by the system (screen, computer).

[0006] An e-commerce distribution center is a typical logistics warehousing system. Especially on days like “Double Eleven” when there are a lot of online shopping orders, how to improve picking efficiency is the focus of the e-commerce distribution center.

[0007] The current picking strategies are all based on orders that have already been generated. That is, after an order is generated, the corresponding goods must first be arranged to be shipped out based on its commodity composition. The drawback of this method is that it takes a long time for the goods to enter the conveyor line and then flow to the picking station, resulting in a long shipping sorting time for an order, which is not conducive to improving the competitiveness of merchants. Especially on days such as "Double Eleven" when the order volume may change suddenly, if the shipping and transportation strategy is only studied based on the sales orders at this moment, and the product shipment is not arranged in advance when the order volume is not large, it will cause the strategy to lag and bring about a waste of shipping capacity. Summary of the invention

[0008] In view of this, the purpose of the present invention is to provide an order prediction picking and delivery method, device and storage medium based on digital twins, so as to solve the problems in the prior art of delayed delivery of goods and long order processing time.

[0009] According to a first aspect of an embodiment of the present invention, a method for order prediction and picking out of warehouse based on digital twins is provided, the method comprising:

[0010] Build a digital twin model of the warehouse picking system based on the actual warehouse picking system's component equipment and process flow;

[0011] Acquire actual production data of an actual warehouse picking system over a period of time, input the actual production data into the digital twin model, and obtain the first average completion time of each picking order under the original picking strategy;

[0012] Obtaining the actual order data generated within a preset historical time period before the time to be predicted in the actual warehouse picking system, inputting the actual order data generated within the preset historical time period into a pre-trained LSTM prediction model, and the pre-trained LSTM prediction model outputs the predicted order data at the time to be predicted;

[0013] The predicted order data at the time to be predicted is added to the remaining order data, and the actual order data at each time to be predicted is recorded. If the actual order data newly generates order data that is not included in the predicted order data, the order data that is not included in the predicted order data is added to the remaining order data;

[0014] Inputting the actual order data after adding the predicted order data into the digital twin model to obtain the second average completion time of each picking order under the improved picking strategy;

[0015] Setting multiple prediction ratio coefficients, respectively obtaining multiple groups of first average completion times and second average completion times, and obtaining a system performance improvement index after adding the predicted order based on the multiple groups of first average completion times, second average completion times and multiple prediction ratio coefficients;

[0016] Selecting the prediction proportional coefficient with the highest system performance improvement index among the multiple prediction proportional coefficients as the final prediction proportional coefficient;

[0017] In an actual warehouse picking system, the order data predicted by the pre-trained LSTM model at each discrete moment is weighted by the final prediction ratio coefficient and then output as a final predicted order, which is then added to the current delivery plan.

[0018] Preferably,

[0019] The digital twin model of the warehouse picking system constructed based on the components and process flow of the actual warehouse picking system includes:

[0020] Obtain the equipment composition and process information of the warehouse picking system;

[0021] The equipment composition and process flow information include: commodity information, warehouse capacity information, conveyor line information, material transportation logic, picking station information, equipment information and historical operation data;

[0022] The commodity information includes: commodity type, quantity and location;

[0023] The warehouse capacity information includes: shelves and storage units;

[0024] The conveyor line information includes: conveyor line transportation speed and capacity;

[0025] The picking station information includes: the number of picking stations and work efficiency;

[0026] The digital twin model includes:

[0027] Geometric model: used to represent the appearance, size and structural relationship of the stereoscopic warehouse, conveyor line, picking station and goods;

[0028] Physical model: used to represent the force changes, mechanical coupling characteristics and temperature coupling characteristics of the equipment;

[0029] Behavior model: used to represent the motion state and action relationship of the equipment during operation;

[0030] Rule model: used to represent the constraints and range of motion of the device;

[0031] The geometric model is modeled in proportion and rendered using Autodesk 3ds Max software; the physical model is simulated and calculated using Ansys software; the behavior model and rule model are set in Unity3D environment;

[0032] The digital twin model of the warehouse picking system is driven by the historical operation data.

[0033] Preferably,

[0034] The order data predicted by the pre-trained LSTM model at each discrete moment is weighted by the final prediction ratio coefficient and then outputs the final predicted order, which includes:

[0035] Obtain the order data of each type of order quantity predicted by the pre-trained LSTM model at each discrete moment, weighted by the final prediction ratio coefficient, and compare the weighted order data with the preset order data critical value;

[0036] If the weighted order data is less than or equal to the preset order data critical value, the weighted order data will be used as the final predicted order;

[0037] If the weighted order data is greater than the preset order data critical value, the weighted order data is multiplied by the preset critical ratio coefficient until the order data is less than or equal to the preset order data critical value.

[0038] Preferably,

[0039] The preset order data critical values ​​include:

[0040] Generating goods to be shipped out of the warehouse in the virtual stereoscopic warehouse of the digital twin model of the warehousing and picking system;

[0041] Arrange the goods to be shipped out in sequence according to the material transportation logic. During the time period when the goods shipment tends to be stable, obtain the total number of orders processed by all picking stations in the digital twin simulation during this time period, and use this total number of orders as the critical value of the order quantity.

[0042] Preferably,

[0043] The step of selecting the prediction proportional coefficient with the highest system performance improvement index from among the multiple prediction proportional coefficients as the final prediction proportional coefficient comprises:

[0044] Obtain the predicted order data at the time to be predicted output by the pre-trained LSTM prediction model, multiply the predicted order data at each time to be predicted by multiple prediction ratio coefficients to obtain multiple weighted predicted order data;

[0045] Each weighted predicted order data is added to the actual order data to obtain multiple weighted actual order data;

[0046] Inputting the plurality of weighted actual order data into the digital twin model respectively, and obtaining the second average completion time of each picking order under different prediction proportional coefficients;

[0047] According to the second average completion time and the first average completion time of each picking order under different prediction ratio coefficients, the system performance improvement index under different prediction ratio coefficients is calculated respectively;

[0048] The system performance improvement indicators under different prediction proportional coefficients are sorted, and the prediction proportional coefficient corresponding to the largest system performance improvement indicator is selected as the final prediction proportional coefficient.

[0049] Preferably, it also includes:

[0050] In the digital twin model, with a daily cycle, the system performance improvement index under different prediction ratio coefficients is obtained according to the current actual order data and the predicted order data, and the optimal prediction ratio coefficient is selected and output to the actual warehouse picking system;

[0051] The actual warehousing and picking system updates the current final prediction ratio coefficient according to the optimal prediction ratio coefficient, and adjusts the proportion of the predicted order data added during the actual warehousing and picking system's outbound delivery process in the next day.

[0052] Preferably,

[0053] The actual order data recorded at each time to be predicted also includes:

[0054] The predicted order data within a period of time and the actual order data within the period of time are obtained respectively. If the predicted order data within the period of time exceeds the actual order data within the period of time, the goods for the excess order data are returned to the warehouse.

[0055] According to a second aspect of an embodiment of the present invention, an order prediction picking and outbound device based on digital twin is provided, and the device includes:

[0056] Model building module: used to build a digital twin model of the warehouse picking system based on the components and process flow of the actual warehouse picking system;

[0057] Original strategy module: used to obtain actual production data of the actual warehouse picking system over a period of time, input the actual production data into the digital twin model, and obtain the first average completion time of each picking order under the original picking strategy;

[0058] Prediction module: used to obtain the actual order data generated in the preset historical time period before the time to be predicted of the actual warehouse picking system, input the actual order data generated in the preset historical time period into the pre-trained LSTM prediction model, and the pre-trained LSTM prediction model outputs the predicted order data at the time to be predicted;

[0059] Adding module: used to add the predicted order data at the time to be predicted to the remaining order data, record the actual order data at each time to be predicted, and if the actual order data newly generates order data not included in the predicted order data, add the order data not included in the predicted order data to the remaining order data;

[0060] Improvement strategy module: used to input the actual order data after adding the predicted order data into the digital twin model, and obtain the second average completion time of each picking order under the improved picking strategy;

[0061] Performance improvement calculation module: used to set multiple prediction ratio coefficients, obtain multiple groups of first average completion times and second average completion times respectively, and obtain the system performance improvement index after adding the predicted order based on the multiple groups of first average completion times, second average completion times and multiple prediction ratio coefficients;

[0062] Optimal proportional coefficient acquisition module: used to select the predicted proportional coefficient with the highest system performance improvement index from multiple predicted proportional coefficients as the final predicted proportional coefficient;

[0063] Outbound Optimization Module: It is used in the actual warehouse picking system to output the final predicted order after the order data predicted by the pre-trained LSTM model at each discrete moment is weighted by the final prediction ratio coefficient, and the final predicted order is added to the current outbound plan.

[0064] According to a third aspect of an embodiment of the present invention, there is provided a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a main controller, each step in the above method is implemented.

[0065] The technical solution provided by the embodiments of the present invention may have the following beneficial effects:

[0066] This application is based on a digital twin model. By predicting future orders, it can consider the orders with a high probability of being generated in the next time period while processing the current orders, and send the goods contained in them to the conveyor line in advance, thereby improving the response speed of the picking system to orders. By simulating the order processing sales volume and order response time of the picking system under different delivery strategies, the delivery strategy that introduces order prediction is optimized to ensure that the strategy can improve the performance of the current picking system.

[0067] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0069] Figure 1 It is a flowchart of an order prediction picking and delivery method based on digital twins according to an exemplary embodiment;

[0070] Figure 2is a system schematic diagram of an order prediction picking and dispatching device based on digital twins according to another exemplary embodiment;

[0071] In the attached figure: 1-model building module, 2-original strategy module, 3-prediction module, 4-addition module, 5-improvement strategy module, 6-performance improvement calculation module, 7-optimal proportional coefficient acquisition module, 8-warehouse optimization module. DETAILED DESCRIPTION

[0072] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0073] Embodiment 1

[0074] Figure 1 is a flow chart of an order prediction picking and delivery method based on digital twins according to an exemplary embodiment. Figure 1 As shown, the method includes:

[0075] S1, build a digital twin model of the warehouse picking system based on the actual warehouse picking system’s components and process flow;

[0076] S2, obtaining actual production data of an actual warehouse picking system over a period of time, inputting the actual production data into the digital twin model, and obtaining the first average completion time of each picking order under the original picking strategy;

[0077] S3, obtaining the order data actually generated in the preset historical time period before the time to be predicted of the actual warehouse picking system, inputting the order data actually generated in the preset historical time period into the pre-trained LSTM prediction model, and the pre-trained LSTM prediction model outputs the predicted order data at the time to be predicted;

[0078] S4, adding the predicted order data at the time to be predicted to the remaining order data, recording the actual order data at each time to be predicted, and if the actual order data newly generates order data that is not included in the predicted order data, adding the order data that is not included in the predicted order data to the remaining order data;

[0079] S5, inputting the actual order data after adding the predicted order data into the digital twin model to obtain the second average completion time of each picking order under the improved picking strategy;

[0080] S6, setting multiple prediction ratio coefficients, respectively obtaining multiple groups of first average completion times and second average completion times, and obtaining a system performance improvement index after adding the predicted order based on the multiple groups of first average completion times, second average completion times and multiple prediction ratio coefficients;

[0081] S7, selecting the prediction proportional coefficient with the highest system performance improvement index from among the multiple prediction proportional coefficients as the final prediction proportional coefficient;

[0082] S8, in the actual warehouse picking system, the order data predicted by the pre-trained LSTM model at each discrete moment is weighted by the final prediction ratio coefficient and then the final predicted order is output, and the final predicted order is added to the current delivery plan;

[0083] It can be understood that the basis for the implementation of this application includes two parts, one is the digital twin model, and the other is the LSTM prediction model;

[0084] Among them, the digital twin model is a digital simulation model based on the physical system. It reflects the status and behavior of the system in real time by modeling and simulating the actual system. It can be regarded as a mapping of the real site. By establishing a digital twin model, the operation of the warehousing and picking system can be simulated (simulating the next picking strategy-warehousing and picking system-box turnover effect), and the effects of different warehouse transportation strategies in the virtual environment can be evaluated.

[0085] The construction of the digital twin model includes:

[0086] Collect the composition and process information of the warehousing logistics system, including: collecting warehouse inventory information such as commodity type, quantity and location through the warehouse management system or logistics system, determining the capacity of the three-dimensional warehouse, including shelves, storage units, etc. through on-site engineering drawings; determining the transportation speed, capacity and other parameters of the conveyor line according to the type of on-site equipment, and determining the material transportation logic from the three-dimensional warehouse to the conveyor line according to the control program; calculating the number and work efficiency of manual picking stations according to the actual on-site conditions;

[0087] According to the layout of the picking system, 3D modeling software (such as 3Ds Max) is used to build 3D models of each equipment. The appearance, size, structure and assembly relationship of the equipment are characterized by original scale construction and texture rendering restoration. These geometric models will be used to represent the appearance and shape of the automated stereoscopic warehouse, warehouse-end conveyor line and manual picking station in the picking system; on the basis of building the geometric model, physical simulation software (such as Ansys and MATLAB) is used to improve the physical model of each equipment. These physical models are used to reflect the force changes and mechanical properties of the equipment, and can more realistically simulate the physical behavior of the equipment during operation; then the behavior model and rule model are set, and these geometric models are imported into the 3D physical engine software, and the behavior model and rule model of each equipment are set in the Unity3D environment. The behavior model is used to represent the motion state and action of the equipment during operation, and the rule model is used to represent the operation trajectory, constraints, motion range and expert knowledge of the equipment; at this point, the digital twin models of all equipment have been built;

[0088] After the model is established, the actual on-site data (inventory information, real-time orders that change over time) are input into the digital twin model, and the operation results of the digital twin model are recorded (including the outbound situation at each moment, the operation of the conveyor line and picking station, and the order situation of the completed picking). By comparing the actual on-site operation data with the digital twin system operation data, the digital twin model is optimized so that it can basically match the actual on-site operation situation;

[0089] The training of the LSTM prediction model includes:

[0090] Build an LSTM model for sales forecasting. LSTM is a recurrent neural network model with memory units and gating mechanisms that can capture long-term dependencies in sequences. The LSTM network consists of three parts:

[0091] The input layer receives sequence data. The input sample of each time step (for example, 10 minutes) is a tensor of size M×N, where M is the number of samples (i.e., the previous M time steps) and N is the number of features. Since the product only has one-dimensional feature of sales data, the number of N is the number of inventory products.

[0092] The LSTM layer includes multiple LSTM units, which are used to learn patterns and trends in sequences. The size and number of LSTM layers can be set according to actual conditions.

[0093] The output layer is set as a fully connected layer to output the sales situation in the next time step. The output sample size is 1×N, that is, the expected sales volume of all commodities in the next time step is output;

[0094] Use the training set to train the LSTM model. You can use an appropriate loss function (such as mean square error) and optimizer to train the model, and set appropriate hyperparameters such as batch size and number of iterations.

[0095] The trained model is used in the test set to obtain sales forecast results, which are compared with the actual sales situation in the test set. The root mean square error evaluation index is used to evaluate the prediction performance of the model. If the performance meets the standard, the model training is completed.

[0096] The trained artificial intelligence algorithm is stored in a computer-readable storage medium and deployed to an on-site industrial computer or computer. The model can receive a certain length of historical order data and then generate predicted order information for a period of time in the future. The order information includes the types of goods sold and the sales volume of different types of goods.

[0097] After completing the construction of the digital twin model and the training of the LSTM model, the simulation runs of the original picking strategy and the improved picking strategy can be performed as follows:

[0098] Picking efficiency calculation under the original scheduling strategy: Collect the actual production data of the actual picking system to calculate the picking efficiency under the original strategy. The specific method is as follows: t First, select N time periods of length t, which together constitute a time span of N t Time period, record from N t From the moment to 2N t The length of this moment is N t All orders generated in reality during the time period [O1, O2…O m ] are generated at the time [t1, t2…t m ], and simultaneously record the various moments when these orders are picked in the real picking system [t 1d , t 2d …t md ], by calculating The average completion time of each picking order in this period when the original scheduling strategy is adopted can be obtained (i.e., the first average completion time mentioned above);

[0099] Picking efficiency calculation under the improved scheduling strategy with order prediction: The picking system digital twin model is used to simulate the operation results of the picking system after adding order prediction. The specific method is as follows: tAt the beginning, at the next N moments separated by t from each other, the order data actually generated in the past N continuous time periods of length t are input into the trained LSTM prediction model, so as to generate the expected number of all types of orders at that moment through the LSTM prediction model (for example, at the moment (N+D)t, the past N moments are [Dt, (D+1)t, (D+2)t...(N+D-1)t], and the data input to the LSTM model are the order data generated in the time periods Dt~(D+1)t, (D+1)t~(D+2)t...(N+D-1)t~(N+D)t, respectively. The LSTM model predicts the expected number of orders for each commodity in the next time period (N+D)t~(N+D+1) through the order quantity of each commodity in these N time periods), and at the same time, these expected orders are added to the remaining orders at that moment (the remaining orders are the orders that the digital twin picking system has not currently arranged to ship out of the warehouse); then, in this span of N t In the time period, all orders generated in this period [O1, O2…O m ] are generated at the time [t1, t2…t m ]; As the digital twin model continues to run, the picking completion time of these orders [t′ 1d , t′ 2d …t′ md ], then, by calculating The average completion time of each picking order in the period (i.e., the second average completion time mentioned above) can be obtained when the improved picking strategy with order prediction is adopted;

[0100] It is worth emphasizing that if the actual order data generates other orders that are not included in the predicted order data within the predicted time period, the unpredicted actual orders will also be added to the remaining orders to avoid order omissions. In addition to the situation where the orders are not predicted, there is also a situation where the predicted orders exceed the actual orders. It is necessary to arrange for the goods exceeding the orders to be returned to the warehouse. Generally speaking, the efficiency impact of the return of orders due to inaccurate predictions on the overall system can be analyzed and evaluated by the digital twin system (through the comparison of improved strategy performance, if the system efficiency is improved after adding orders, it means that even if some orders are returned to the warehouse due to inaccurate predictions, the efficiency of the overall picking system is still improved);

[0101] Performance comparison between the original picking strategy and the improved picking strategy:

[0102] In Q consecutive N tDuring the time period, according to the first average duration calculation method and the second average duration calculation method mentioned above, the multiple groups of original strategy operation results and the multiple groups of improved strategy operation results are input into the bin picking strategy evaluation model. The model evaluates the performance of the order prediction scheduling strategy through the following formula, that is, calculates the performance improvement index P:

[0103]

[0104] P represents the percentage of picking efficiency improvement of the scheduling strategy with order prediction compared to the original scheduling strategy. A higher percentage indicates a higher efficiency improvement.

[0105] In order to further improve the performance improvement index, this application also introduces the prediction ratio coefficient K. In simple terms, after the order prediction is completed by the pre-trained LSTM model, all predicted orders will not be added to the remaining orders. Instead, 10%, 20%...90% of the predicted order data will be selected and added to the remaining orders at that moment. The improvement in the picking efficiency of the picking system after adding different predicted order ratios [P 10％ , P 20％ …P 90％ ], select the predicted order addition ratio with the highest improvement, and determine it as the final prediction ratio coefficient K (for example, K = 0.7 means that the picking efficiency of adding 70% predicted order group under the twin system is the highest);

[0106] In the actual operation process, the picking system arranges some orders to be shipped out in advance by predicting the number of future orders, thereby improving the operation efficiency of the picking system. However, the market situation is constantly changing. The K value obtained by simulating historical data and digital twin models will continue to decrease in reference value as the number and distribution of orders continue to change. Therefore, a strategy update method combining newly generated order data is adopted. As the picking system runs, a better predicted order addition ratio coefficient K is continuously found through order data and digital twin models.

[0107] The above method is used to calculate the efficiency of the original strategy on a daily basis. Then, different predicted order addition ratios are used.

[0108]

[0109] Compare the optimal predicted order addition ratio K' under the current market sales situation, and then the digital twin model transmits the simulated K' to the running real picking system through the data interface. The real picking system updates the original K value to K', that is, adjusts the predicted order addition ratio in the picking system's outbound process within the next 24 hours, thereby making full use of the asynchronous operation and simulation functions of the digital twin model, dynamically adjusting the predicted order addition ratio according to real-time market changes, and optimizing the picking system efficiency;

[0110] In addition to the proportional coefficient K, the present application further introduces the critical value of the order quantity. It can be known that since the present application adds the predicted order at each moment, the corresponding order outbound pressure will increase. Although there is a predicted order proportional coefficient, the number of predicted orders added will not be too large, but it is still possible to exceed the outbound capacity of the system. Although the above discussion on the efficiency impact of the order return to the warehouse due to inaccurate prediction on the overall system can be analyzed and evaluated by the digital twin system (through the comparison of improved strategy performance, if the system efficiency is improved after adding the order, it means that even if some orders are inaccurately predicted and lead to return to the warehouse, the efficiency of the overall picking system is still improved), but in the case of a large number of orders, the conveying pressure of the warehouse-side conveyor line is close to the critical situation. At this time, the return of goods caused by inaccurate prediction will have a relatively large impact on the overall efficiency of the picking system (too many outbound orders will cause congestion in the production line). Therefore, the operation of the picking system when the number of orders is too large is simulated by virtual operation of the digital twin, and the critical value of the number of orders that will cause congestion of the warehouse-side conveyor line and greatly reduce the efficiency is obtained according to the simulation operation results. Among them, the critical value of the order quantity is obtained as follows:

[0111] With the help of the picking system digital twin virtual scene, generate enough goods to be shipped out in the three-dimensional warehouse of the virtual scene, and then arrange the goods to be shipped out in sequence. During the time period when the shipment tends to be stable, calculate the total number of orders M processed by all picking stations in the digital twin simulation scene within the time period t; then M is the upper limit of the number of orders that the picking system can process in continuous operation within the time period t, which is regarded as the critical value of the order quantity;

[0112] After obtaining the critical value of the order quantity, when calculating the predicted order quantity for each time period, first weight the predicted order quantity by the prediction proportion coefficient K, and compare the weighted predicted order quantity with the critical value of the order quantity M. If the order quantity is greater than M, then continue to multiply the weighted order quantity by the proportion coefficient so that it does not exceed the critical value of the order M. If the weighted order quantity is less than or equal to M, then add the weighted predicted order to the current remaining orders;

[0113] After verifying the optimal ratio K of predicted orders to be added to the picking system through the digital twin model, the current LSTM model is deployed on the industrial computer or server on site to make rolling predictions of the expected orders at the next moment (including the expected number of all different types of orders) at each discrete moment;

[0114] A communication interface for the LSTM model is added to the task dispatching information system of the original picking system on site. At each discrete moment, the quantity of each type of order predicted by the LSTM model is weighted by the proportional coefficient K, and then the weighted predicted order quantity is compared with the critical value of the overall order of the system to obtain the final predicted order. Subsequently, communication with the original picking information system is completed through the communication interface, and these predicted orders are added to the current delivery plan of the automated warehouse, helping the actual picking system to reduce the response time of an order from placing an order to completing picking and packaging by delivering the goods in advance, thereby improving the picking efficiency of the bin picking system.

[0115] Embodiment 2

[0116] Figure 2 is a system schematic diagram of an order prediction picking and outbound device based on digital twin according to another exemplary embodiment, including:

[0117] Model building module 1: used to build a digital twin model of the warehouse picking system based on the components and process flow of the actual warehouse picking system;

[0118] Original strategy module 2: used to obtain actual production data of the actual warehouse picking system within a period of time, input the actual production data into the digital twin model, and obtain the first average completion time of each picking order under the original picking strategy;

[0119] Prediction module 3: used to obtain the actual order data generated in the preset historical time period before the prediction time of the actual warehouse picking system, input the actual order data generated in the preset historical time period into the pre-trained LSTM prediction model, and the pre-trained LSTM prediction model outputs the predicted order data at the prediction time;

[0120] Adding module 4: used to add the predicted order data at the time to be predicted to the remaining order data, record the actual order data at each time to be predicted, and if the actual order data newly generates order data not included in the predicted order data, add the order data not included in the predicted order data to the remaining order data;

[0121] Improvement strategy module 5: used to input the actual order data after adding the predicted order data into the digital twin model, and obtain the second average completion time of each picking order under the improved picking strategy;

[0122] Performance improvement calculation module 6: used to set multiple prediction ratio coefficients, respectively obtain multiple groups of first average completion time and second average completion time, and obtain the system performance improvement index after adding the predicted order based on the multiple groups of first average completion time, second average completion time and multiple prediction ratio coefficients;

[0123] Optimal proportional coefficient acquisition module 7: used to select the predicted proportional coefficient with the highest system performance improvement index from among multiple predicted proportional coefficients as the final predicted proportional coefficient;

[0124] Outbound Optimization Module 8: It is used in the actual warehouse picking system to output the final predicted order after weighting the order data predicted by the pre-trained LSTM model at each discrete moment through the final prediction ratio coefficient, and add the final predicted order to the current outbound plan.

[0125] Embodiment three:

[0126] This embodiment provides a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a main controller, each step in the above method is implemented;

[0127] It is understandable that the storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0128] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0129] It should be noted that, in the description of the present invention, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of the present invention, unless otherwise specified, the meaning of "plurality" refers to at least two.

[0130] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.

[0131] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0132] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0133] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0134] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0135] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0136] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. The order prediction picking and delivery method based on digital twins is characterized by: The method comprises: Build a digital twin model of the warehouse picking system based on the actual warehouse picking system's component equipment and process flow; Acquire actual production data of an actual warehouse picking system over a period of time, input the actual production data into the digital twin model, and obtain the first average completion time of each picking order under the original picking strategy; Obtaining the actual order data generated within a preset historical time period before the time to be predicted in the actual warehouse picking system, inputting the actual order data generated within the preset historical time period into a pre-trained LSTM prediction model, and the pre-trained LSTM prediction model outputs the predicted order data at the time to be predicted; The predicted order data at the time to be predicted is added to the remaining order data, and the actual order data at each time to be predicted is recorded. If the actual order data newly generates order data that is not included in the predicted order data, the order data that is not included in the predicted order data is added to the remaining order data; Inputting the actual order data after adding the predicted order data into the digital twin model to obtain the second average completion time of each picking order under the improved picking strategy; Setting multiple prediction ratio coefficients, respectively obtaining multiple groups of first average completion times and second average completion times, and obtaining a system performance improvement index after adding the predicted order based on the multiple groups of first average completion times, second average completion times and multiple prediction ratio coefficients; Selecting the prediction proportional coefficient with the highest system performance improvement index among the multiple prediction proportional coefficients as the final prediction proportional coefficient; In an actual warehouse picking system, the order data predicted by the pre-trained LSTM model at each discrete moment is weighted by the final prediction ratio coefficient and then output as a final predicted order, which is then added to the current delivery plan.

2. The method according to claim 1, characterized in that The digital twin model of the warehouse picking system constructed based on the components and process flow of the actual warehouse picking system includes: Obtain the equipment composition and process information of the warehouse picking system; The equipment composition and process flow information include: commodity information, warehouse capacity information, conveyor line information, material transportation logic, picking station information, equipment information and historical operation data; The commodity information includes: commodity type, quantity and location; The warehouse capacity information includes: shelves and storage units; The conveyor line information includes: conveyor line transportation speed and capacity; The picking station information includes: the number of picking stations and work efficiency; The digital twin model includes: Geometric model: used to represent the appearance, size and structural relationship of the stereoscopic warehouse, conveyor line, picking station and goods; Physical model: used to represent the force changes, mechanical coupling characteristics and temperature coupling characteristics of the equipment; Behavior model: used to represent the motion state and action relationship of the equipment during operation; Rule model: used to represent the constraints and range of motion of the device; The geometric model is modeled in proportion and rendered using Autodesk 3ds Max software; the physical model is simulated and calculated using Ansys software; the behavior model and rule model are set in Unity3D environment; The digital twin model of the warehouse picking system is driven by the historical operation data.

3. The method according to claim 2, characterized in that The order data predicted by the pre-trained LSTM model at each discrete moment is weighted by the final prediction ratio coefficient and then outputs the final predicted order, which includes: Obtain the order data of each type of order quantity predicted by the pre-trained LSTM model at each discrete moment, weighted by the final prediction ratio coefficient, and compare the weighted order data with the preset order data critical value; If the weighted order data is less than or equal to the preset order data critical value, the weighted order data will be used as the final predicted order; If the weighted order data is greater than the preset order data critical value, the weighted order data is multiplied by the preset critical ratio coefficient until the order data is less than or equal to the preset order data critical value.

4. The method according to claim 3, characterized in that The preset order data critical values ​​include: Generating goods to be shipped out of the warehouse in the virtual stereoscopic warehouse of the digital twin model of the warehousing and picking system; Arrange the goods to be shipped out in sequence according to the material transportation logic. During the time period when the goods shipment tends to be stable, obtain the total number of orders processed by all picking stations in the digital twin simulation during this time period, and use this total number of orders as the critical value of the order quantity.

5. The method according to claim 4, characterized in that The step of selecting the prediction proportional coefficient with the highest system performance improvement index from among the multiple prediction proportional coefficients as the final prediction proportional coefficient comprises: Obtain the predicted order data at the time to be predicted output by the pre-trained LSTM prediction model, multiply the predicted order data at each time to be predicted by multiple prediction ratio coefficients to obtain multiple weighted predicted order data; Each weighted predicted order data is added to the actual order data to obtain multiple weighted actual order data; Inputting the plurality of weighted actual order data into the digital twin model respectively, and obtaining the second average completion time of each picking order under different prediction proportional coefficients; According to the second average completion time and the first average completion time of each picking order under different prediction ratio coefficients, the system performance improvement index under different prediction ratio coefficients is calculated respectively; The system performance improvement indicators under different prediction proportional coefficients are sorted, and the prediction proportional coefficient corresponding to the largest system performance improvement indicator is selected as the final prediction proportional coefficient.

6. The method according to claim 5, characterized in that Also includes: In the digital twin model, with a daily cycle, the system performance improvement index under different prediction ratio coefficients is obtained according to the current actual order data and the predicted order data, and the optimal prediction ratio coefficient is selected and output to the actual warehouse picking system; The actual warehousing and picking system updates the current final prediction ratio coefficient according to the optimal prediction ratio coefficient, and adjusts the proportion of the predicted order data added during the actual warehousing and picking system's outbound delivery process in the next day.

7. The method according to claim 1, characterized in that The actual order data recorded at each time to be predicted also includes: The predicted order data within a period of time and the actual order data within the period of time are obtained respectively. If the predicted order data within the period of time exceeds the actual order data within the period of time, the goods for the excess order data are returned to the warehouse.

8. The order prediction picking and delivery device based on digital twin is characterized by: The device comprises: Model building module: used to build a digital twin model of the warehouse picking system based on the components and process flow of the actual warehouse picking system; Original strategy module: used to obtain actual production data of the actual warehouse picking system over a period of time, input the actual production data into the digital twin model, and obtain the first average completion time of each picking order under the original picking strategy; Prediction module: used to obtain the actual order data generated in the preset historical time period before the time to be predicted of the actual warehouse picking system, input the actual order data generated in the preset historical time period into the pre-trained LSTM prediction model, and the pre-trained LSTM prediction model outputs the predicted order data at the time to be predicted; Adding module: used to add the predicted order data at the time to be predicted to the remaining order data, record the actual order data at each time to be predicted, and if the actual order data newly generates order data not included in the predicted order data, add the order data not included in the predicted order data to the remaining order data; Improvement strategy module: used to input the actual order data after adding the predicted order data into the digital twin model, and obtain the second average completion time of each picking order under the improved picking strategy; Performance improvement calculation module: used to set multiple prediction ratio coefficients, obtain multiple groups of first average completion times and second average completion times respectively, and obtain the system performance improvement index after adding the predicted order based on the multiple groups of first average completion times, second average completion times and multiple prediction ratio coefficients; Optimal proportional coefficient acquisition module: used to select the predicted proportional coefficient with the highest system performance improvement index from multiple predicted proportional coefficients as the final predicted proportional coefficient; Outbound Optimization Module: It is used in the actual warehouse picking system to output the final predicted order after the order data predicted by the pre-trained LSTM model at each discrete moment is weighted by the final prediction ratio coefficient, and the final predicted order is added to the current outbound plan.

9. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the main controller, it implements each step of the digital twin-based order prediction picking and warehousing method as described in any one of claims 1 to 7.

Images