Intelligent warehousing optimization management platform based on digital twinning and space-time prediction

Abstract

The invention discloses an intelligent warehouse optimization management platform based on digital twinning and space-time prediction, which relates to the technical field of intelligent warehouse management and comprises a digital twinning model construction module, a space-time prediction module, an inventory optimization management module and a user interaction module. The digital twinborn model construction module realizes digital mapping of a physical warehousing system by constructing a warehousing space three-dimensional model and associating warehousing sensing data, and the space-time prediction module constructs a prediction model to obtain inventory prediction information in a future time period, and performs inventory risk assessment and early warning according to the inventory prediction information; the inventory optimization management module generates an inventory strategy and an optimized storage position according to the prediction result, the storage cost and the cargo demand, and formulates a warehouse-in and warehouse-out task scheduling scheme; and the user interaction module visually presents the warehouse management data and receives a user interaction operation instruction to realize man-machine collaborative management, and the platform reduces the vacancy rate of the warehouse space and the interruption risk of the supply chain, and reduces the invalid carrying energy consumption.

Description

Technical Field

[0001] This application relates to the technical field of intelligent warehouse management, and specifically relates to an intelligent warehouse optimization management platform based on digital twin and spatio-temporal prediction. Background Art

[0002] In traditional warehouse management, inventory management often relies on manual experience and simple statistical methods, making it difficult to grasp the dynamic changes of warehouse inventory in real time and accurately. With the continuous expansion of warehouse scale and the increasing variety of goods, many problems have emerged in the traditional management method, such as frequent occurrence of inventory backlog or out-of-stock phenomena, low utilization rate of warehouse space, and low efficiency of goods in and out of the warehouse.

[0003] In traditional warehouse logistics management, with the continuous expansion of warehouse scale and the increasing variety of goods, the management method has exposed many limitations. First, the data collection and processing ability lags behind, relying too much on manual data entry, resulting in delayed information update and difficult to guarantee data accuracy, and unable to reflect the dynamic changes of the warehouse in time. Second, the decision-making support ability is weak, inventory management overly relies on manual experience and simple statistical methods, lacking scientific quantitative analysis, and difficult to adapt to the complex and changeable market demands, leading to frequent occurrence of inventory backlog or out-of-stock phenomena, increasing warehouse costs and business risks. Third, the existing warehouse management system has low integration, and the information island phenomenon in each business link is serious. There is a lack of effective coordination in links such as warehouse space planning, inventory allocation, and in and out of warehouse management, resulting in low utilization rate of warehouse space and difficult to improve the efficiency of goods in and out of the warehouse, seriously restricting the modernization development of warehouse logistics management. The digital twin technology, as an emerging technology, can realize real-time monitoring, simulation and optimization of physical entities by constructing virtual models of physical entities. The spatio-temporal prediction method can accurately predict the spatio-temporal change trend of warehouse inventory. However, there is currently no comprehensive platform that effectively combines digital twin technology and spatio-temporal prediction method and effectively applies them to the optimization management of intelligent warehouse inventory. Summary of the Invention

[0004] In view of this, this application provides an intelligent warehouse optimization management platform based on digital twin and spatio-temporal prediction, which solves the technical problems of lagging data collection and processing ability, weak decision-making support ability, difficult to adapt to the complex and changeable market demands in traditional warehouse logistics management, as well as low integration of the existing warehouse management system, serious information island phenomenon in each business link, lack of effective coordination in links such as warehouse space planning, inventory allocation, and in and out of warehouse management, resulting in low utilization rate of warehouse space and difficult to improve the efficiency of goods in and out of the warehouse.

[0005] To achieve the above object, the present invention provides the following technical solutions: The present invention mainly consists of an intelligent warehousing optimization management platform based on digital twin and spatio-temporal prediction. The platform includes a digital twin model construction module, a spatio-temporal prediction module, an inventory optimization management module, and a user interaction module. These four modules work closely together to specifically address the limitations of traditional warehousing logistics management and have significant advantages. The digital twin model construction module constructs a virtual warehouse model that is fully mapped to the physical warehouse, and synchronizes in real time the shelf layout, cargo location, environmental sensor data (temperature, humidity, light), AGV operation status, and inventory quantity of the physical warehouse. The spatio-temporal prediction module is used to construct a spatio-temporal prediction model, input the real-time collected warehousing inventory data into the spatio-temporal prediction model for prediction, obtain the inventory prediction information for a future period of time, and use the inventory prediction information for inventory risk assessment and early warning.

[0006] The inventory optimization management module is connected to the spatio-temporal prediction module. The inventory optimization management module is used to generate an inventory strategy using an optimization algorithm based on the prediction results of the spatio-temporal prediction module, warehousing costs, and cargo demand, and is also used to perform a simulation analysis of the warehousing space based on the digital twin model. According to the characteristics of the weight, volume, and inbound / outbound frequency of the goods, a heuristic algorithm is used to optimize the storage location of the goods, and an inbound / outbound task scheduling plan is formulated based on the inventory strategy and the storage location of the goods.

[0007] The user interaction module is used to visually display the digital twin model, real-time data of the warehousing inventory, spatio-temporal prediction results, and inventory optimization management solution information, and receive the interactive operation instructions of the user.

[0008] Furthermore, the digital twin model construction module includes a data processing unit, a spatial three-dimensional model construction unit, and a model association unit. The data processing unit and the spatial three-dimensional model construction unit are connected to the model association unit. The data processing unit is used to obtain the cargo location, warehousing temperature and humidity, and cargo weight data uploaded by the warehousing Internet of Things device end, and use the Kalman filter algorithm to suppress the noise of the sensor data, and achieve multi-source data synchronization through timestamp alignment; the spatial three-dimensional model construction unit is used to perform parametric basic modeling of the warehousing structure using three-dimensional modeling software, and use the Unity engine to achieve kinematic modeling of the equipment to obtain a three-dimensional model of the warehousing space; the model association unit is used to associate the cargo location, warehousing temperature and humidity, cargo weight data, and the three-dimensional model of the warehousing space to establish a digital twin model, and update the digital twin model accordingly when the physical entity of the warehousing space changes.

[0009] The data acquisition layer deploys an IoT gateway based on the LoRaWAN protocol, and collects the following data through the RS485 / Modbus interface. The goods positioning data is collected by UWB ultra-wideband positioning tags with a positioning accuracy of ±5 cm. The environmental monitoring data is collected by temperature and humidity sensors and a light sensor group. The goods status data is collected by weighing sensors. When preprocessing the collected data, the Kalman filter algorithm is used to suppress the noise of the sensor data, and data encapsulation is achieved based on the MQTT protocol. Set QoS = 1 to ensure transmission reliability, and use Apache Kafka to build a message queue to achieve data stream processing. Basic modeling uses Blender 4.2 for parametric modeling of the warehouse structure, and dynamic modeling realizes the kinematic modeling of equipment (rigid body kinematic simulation) through the Unity engine. The dynamic objects in motion include AGVs / stackers. When the model association unit binds data, a unique mapping between physical devices and digital models is achieved based on UUID, and a spatial coordinate system conversion algorithm (homogeneous transformation matrix) is used to achieve position synchronization. The update of the model includes real-time update and incremental update. The real-time update establishes a two-way communication channel through WebSocket, and the incremental update is only used to transmit differential data packets, saving network resources. When detecting model changes, a hash tree can be used to compare the differences between model versions.

[0010] Furthermore, the spatio-temporal prediction module constructs a spatio-temporal prediction engine based on deep learning to achieve accurate prediction and risk warning of inventory dynamics. The spatio-temporal prediction module includes a data preprocessing unit, a model establishment unit, and a prediction and warning unit; The data preprocessing unit is used to collect the historical data of goods in and out from the warehouse management system. The historical data includes in and out time, quantity, and type information, and cleans the collected historical data to remove duplicate data, error data, and missing values, and then extracts time features and spatial features; The time features include year, month, day, hour, etc., and the spatial features include the goods storage area, shelf position, etc. When extracting time features, a time domain feature set including hour, week, and month cycle terms is constructed, and when extracting spatial features, GeoHash coding is used to achieve area division.

[0011] The model establishment unit is used to construct a spatio-temporal prediction model based on the spatio-temporal convolutional neural network, and divide the preprocessed historical data into a training set and a test set. Use the training set to train the spatio-temporal prediction model, and adjust the parameters of the model through the backpropagation algorithm to obtain a trained spatio-temporal prediction model.

[0012] The architecture of constructing a spatio-temporal prediction model based on the spatio-temporal convolutional neural network includes a spatial convolutional layer that uses a 3D convolutional kernel (3×3×3) to capture spatial correlation, a temporal convolutional layer that uses causal convolution to maintain temporal causality, and an attention module.

[0013] The prediction and early warning unit is used to obtain the real-time collected warehousing inventory data and input it into the trained spatio-temporal prediction model to obtain the inventory prediction results for a future period of time. It compares the predicted inventory level with the preset replenishment point threshold and safety inventory level. When the predicted inventory level is lower than the replenishment point, it automatically outputs a replenishment warning signal. When the predicted inventory level is higher than the safety inventory level, it automatically outputs an inventory backlog warning signal. It is also used to integrate a multi-task learning model to synchronously output the inventory shortage probability, hot spots of high-turnover goods, and warning of warehousing equipment failure risks.

[0014] The multi-task learning model includes a shared encoder and independent decoders. The shortage probability branch uses an LSTM network (128 hidden units), and the risk warning branch adopts a hierarchical warning method.

[0015] Furthermore, the inventory optimization management module includes an inventory policy generation unit. The policy generation unit is used to generate an inventory policy according to the prediction results of the spatio-temporal prediction module, warehousing costs, and goods demand, and determine the safety inventory level, replenishment point, and replenishment quantity using a genetic algorithm.

[0016] Warehousing costs include goods procurement costs, storage costs, and shortage costs. Goods demand includes historical sales data and market trends.

[0017] Even further, the inventory optimization management module also includes a goods storage optimization unit. The goods storage optimization unit is used to perform simulation analysis on the warehousing space using a digital twin model, and optimize the storage location of goods using a simulated annealing algorithm according to the weight, volume, and inbound / outbound frequency characteristics of the goods.

[0018] Optimizing the storage location of goods using a simulated annealing algorithm includes: determining the state space: the bin coordinates (x, y, z) and goods attributes (weight / volume / SKU); neighborhood generation: the exchange strategy uses randomly exchanging the goods in two bins, the insertion strategy uses the method of inserting goods into a new bin, and the cooling strategy uses exponential annealing (initial temperature T0 = 1000, attenuation coefficient α = 0.95).

[0019] Even further, the inventory optimization management module also includes an inbound / outbound task scheduling unit. The inbound / outbound task scheduling unit is used to formulate an inbound / outbound task scheduling plan according to the inventory policy and the storage location of goods, and perform simulation on the inbound / outbound process using a digital twin model, considering factors such as the running speed of handling equipment, path length, and goods handling sequence, to optimize the handling path of goods and the scheduling of handling equipment.

[0020] The path planning algorithm combines the global path and local obstacle avoidance. The global path planning uses the A* algorithm combined with the improved Dijkstra, and the local obstacle avoidance uses the dynamic window method to achieve real-time obstacle avoidance. The device scheduling adopts a distributed task allocation method based on the auction algorithm and uses a time window negotiation mechanism to resolve conflicts.

[0021] Furthermore, the user interaction module includes a visual display unit and an interaction editing unit; The visual display unit is used to visually display the digital twin model, real-time data of warehouse inventory, spatio-temporal prediction results, and inventory optimization management solution information, and receive the user's interaction operation instructions; The interaction editing unit is used to obtain the inventory strategy parameters input by the user, adjust the storage location of goods, and the parameters for issuing inbound and outbound tasks, and convert them into logical execution instructions, verify and process the instructions to ensure the legality and accuracy of the operations, feedback the operation results to the user, and update and display them on the visual interface. The operation verification includes permission control and operation auditing. The permission control is carried out through the RBAC model (role-permission-operation), and the operation auditing part records the operation logs. The ANTLR4 is used to implement the instruction syntax check. The deployment architecture of this intelligent platform includes the edge layer: deploying the digital twin engine (NVIDIA Jetson AGX Orin), and the cloud: the spatio-temporal prediction model service (AWS SageMaker).

[0022] From the above technical solutions, the advantages of the present invention are as follows: At the data collection and processing level, by deploying the Internet of Things gateway and combining multiple devices such as UWB ultra-wideband positioning and temperature and humidity sensors, real-time collection is achieved, effectively solving the problems of data update delay and low accuracy in traditional systems, and providing reliable data support for the dynamic management of the warehouse. And through the deep integration of the spatio-temporal prediction module and the inventory optimization management module, not only can the change trend of inventory be accurately predicted, but also multi-dimensional warning information such as out-of-stock probability and equipment failure risk can be output, making up for the defect of lack of quantitative analysis in traditional manual experience decision-making. The inventory optimization management module effectively avoids the risks of inventory backlog or out-of-stock by formulating inventory strategies, optimizing storage locations, and scheduling inbound and outbound tasks, and improves the utilization efficiency of warehouse resources. The digital twin model construction module realizes the digital mapping of the physical warehouse system by establishing a three-dimensional model of the warehouse space and associating real-time sensing data, breaking the information silos between business links. Each module operates in coordination based on the digital twin model, promoting the intelligentization of warehouse logistics management. In addition, the visual presentation and interaction editing functions of the user interaction module realize efficient human-machine collaborative management. Description of the Drawings

[0023] The accompanying drawings, which form a part of this application, are used to provide a further understanding of this application. The schematic embodiments and descriptions thereof of this application are used to explain this application and do not constitute an improper limitation of this application.

[0024] Figure 1 It is a schematic structural diagram of the composition of the intelligent warehousing optimization management platform based on digital twin and spatio-temporal prediction of this application.

[0025] Figure 2 It is a system block diagram of this embodiment.

[0026] Figure 3 It is a schematic flow diagram of model update of this embodiment.

[0027] Figure 4 It is a schematic diagram of the steps of the warehousing inventory optimization method of this embodiment. Detailed implementation manners

[0028] To make the objectives, technical solutions and advantages of this application clearer and more understandable, the following further details this application in combination with the implementation manners and the accompanying drawings. Herein, the schematic implementation manners and descriptions thereof of this application are used to explain this application, but do not serve as a limitation of this application.

[0029] Refer to Figures 1 to 4 , this embodiment provides an intelligent warehousing optimization management platform based on digital twin and spatio-temporal prediction. By constructing a digital twin model of the warehousing system and combining spatio-temporal prediction methods, it can monitor, analyze and optimize the warehousing inventory in real time, avoid information silos, improve the utilization rate of warehousing space, reduce inventory costs and the risk of supply chain interruption, improve the efficiency of goods inbound and outbound, and reduce the energy consumption of ineffective handling at the same time. As Figure 1As shown in the figure, the platform includes: a digital twin model construction module, a spatio-temporal prediction module, an inventory optimization management module, and a user interaction module. The digital twin model construction module provides a real-time and accurate data basis for warehouse management by constructing a virtual model that fully maps the physical warehouse and synchronizing the shelf layout, goods location, environmental data, and AGV operation status information in real time. The spatio-temporal prediction module is used to construct a spatio-temporal prediction model, input the real-time collected warehouse inventory data into the spatio-temporal prediction model for prediction, obtain the inventory prediction information for a future period of time, and use the inventory prediction information for inventory risk assessment and early warning, realizing a scientific quantitative analysis of the inventory change trend and effectively avoiding inventory backlog or out-of-stock phenomena. The inventory optimization management module is connected to the spatio-temporal prediction module. The inventory optimization management module is used to generate inventory strategies using optimization algorithms based on the prediction results of the spatio-temporal prediction module, warehouse costs, and goods demands, and is also used to perform simulation analysis on the warehouse space based on the digital twin model. According to the characteristics of the weight, volume, and inbound and outbound frequency of the goods, heuristic algorithms are used to optimize the storage locations of the goods, and an inbound and outbound task scheduling plan is formulated based on the inventory strategy and the storage locations of the goods, breaking the information silos of each business link in traditional warehousing and improving the utilization rate of warehouse space and the efficiency of goods inbound and outbound. The user interaction module is used to visually display the digital twin model, real-time data of warehouse inventory, spatio-temporal prediction results, and inventory optimization management solution information, and receive the interactive operation instructions of users, realizing efficient human-machine collaborative management.

[0030] Specifically, the digital twin model construction module includes a data processing unit, a spatial three-dimensional model construction unit, and a model association unit. The data processing unit and the spatial three-dimensional model construction unit are connected to the model association unit. The data processing unit is used to obtain the goods location, warehouse temperature and humidity, and goods weight data uploaded by the warehouse Internet of Things device end, suppress the noise of the sensor data using the Kalman filter algorithm, and achieve multi-source data synchronization through timestamp alignment; the spatial three-dimensional model construction unit is used to perform parametric basic modeling of the warehouse structure using three-dimensional modeling software and realize the kinematic modeling of the equipment through the Unity engine to obtain a three-dimensional model of the warehouse space; the model association unit is used to associate the goods location, warehouse temperature and humidity, goods weight data, and the three-dimensional model of the warehouse space to establish a digital twin model, and update the digital twin model accordingly when the physical entity of the warehouse space changes.

[0031] In this embodiment, an Internet of Things gateway based on the LoRaWAN protocol is deployed in the data acquisition layer. Relevant sensing data is collected through the RS485 / Modbus interface. The goods positioning data is collected through UWB ultra-wideband positioning tags, and three-dimensional positioning can be achieved through the existing TDOA algorithm with a positioning accuracy of ±5 cm. The environmental monitoring data of the storage space is collected through temperature and humidity sensors and a group of light sensors. The goods status data is collected using a cantilever beam weighing sensor. When preprocessing the collected data, the Kalman filter algorithm is used to suppress the noise of the sensor data, and data encapsulation is achieved based on the MQTT protocol. QoS = 1 is set to ensure transmission reliability. The MQTT protocol adopts a three-level topic structure of <warehouse ID> / <area> / <sensor type>. QoS = 1 ensures at least one delivery, and the message retain flag (RetainFlag) is enabled. And Apache Kafka is used to build a message queue to achieve data streaming processing. Basic modeling is carried out using Blender 4.2 for parametric modeling of the storage structure, and the shelf dimensions (length × width × height), aisle width, etc. are parametrically defined. Dynamic modeling is achieved through the Unity engine for kinematic modeling of equipment (rigid body kinematic simulation). The dynamic objects in motion include AGVs / stackers. When the model association unit binds data, a unique mapping between physical devices and digital models is achieved based on UUID, and the MySQL database stores the mapping table of device UUIDs and virtual model IDs. And a spatial coordinate system transformation algorithm (homogeneous transformation matrix) is adopted to achieve position synchronization. The update of the model includes real-time update and incremental update. The real-time update is through the two-way communication channel established by WebSocket, and the incremental update is only used to transmit differential data packets, saving network resources. When detecting model changes, a hash tree can be used to compare the differences between model versions to achieve block-level model change detection. The process of model update is as Figure 3 shown in the version control mechanism.

[0032] In this embodiment, the spatio-temporal prediction module constructs a spatio-temporal prediction engine based on deep learning to achieve accurate prediction and risk warning of inventory dynamics. The spatio-temporal prediction module includes a data preprocessing unit, a model establishment unit, and a prediction and warning unit; the data preprocessing unit is used to collect historical data on the inbound and outbound of goods from the warehouse management system. The historical data includes inbound and outbound time, quantity, and type information, and the collected historical data is cleaned to remove duplicate data, error data, and missing values, and then time features and spatial features are extracted; the time features include year, month, day, hour, etc., and the spatial features include the goods storage area, shelf location, etc. When extracting time features, a Fourier series periodic term is constructed to obtain a time-domain feature set including hour, week, and month periodic terms. When extracting spatial features, GeoHash coding is used to achieve area division and map it to the warehouse partition (such as A1 - Shelf Layer 1 - Column B).

[0033] The model building unit is used to construct a spatio-temporal prediction model based on a spatio-temporal convolutional neural network, divide the preprocessed historical data into a training set and a test set, train the spatio-temporal prediction model using the training set, and adjust the parameters of the model through the backpropagation algorithm to obtain a trained spatio-temporal prediction model. The architecture of the spatio-temporal prediction model constructed based on the spatio-temporal convolutional neural network includes a spatial convolutional layer, a temporal convolutional layer, and an attention module. Among them, the 3D convolutional kernel size of the spatial convolutional layer is 3×3×3, the stride is (1,1,1), zero-padding ensures that the output size remains unchanged, and the activation function is LeakyReLU (α = 0.01); the causal convolution kernel size of the temporal convolutional layer is 5, and the dilation factor d = 1, 2, 4 (the receptive field covers 15 time steps), and the attention mechanism uses multi-head self-attention, with the Key-Value dimension = 64.

[0034] The prediction and early warning unit is used to obtain the warehousing inventory data collected in real time and input it into the trained spatio-temporal prediction model to obtain the inventory prediction results for a future period of time. Compare the predicted inventory level with the preset replenishment point threshold and safety inventory level. When the predicted inventory level is lower than the replenishment point, automatically output a replenishment warning signal. When the predicted inventory level is higher than the safety inventory level, automatically output an inventory backlog warning signal. It is also used to integrate a multi-task learning model to synchronously output the inventory out-of-stock probability, hot spots of high-turnover goods, and warehousing equipment failure risk warnings. The multi-task learning model includes a shared encoder and independent decoders. The out-of-stock probability branch uses an LSTM network, and the risk warning branch uses a hierarchical warning method. Among them, the shared encoder includes 4 layers of residual convolutional blocks, and the output channels of each layer are 64, 128, 256, and 512 respectively. The out-of-stock probability branch uses an LSTM decoder with 128 hidden units, activated by the tanh function, and outputs a Sigmoid value. The risk warning branch uses a 3-level classification warning, including normal, attention, and danger, and outputs a Softmax probability distribution.

[0035] In this embodiment, the inventory optimization management module includes an inventory policy generation unit and a goods storage optimization unit. The policy generation unit is used to generate an inventory policy using a genetic algorithm according to the prediction results of the spatio-temporal prediction module, warehousing costs, and goods demands, and determine the safety inventory level, replenishment point, and replenishment quantity. Among them, the warehousing costs include goods procurement costs, storage costs, and out-of-stock costs. The goods demands include historical sales data and market trends. The goods storage optimization unit is used to simulate and analyze the warehousing space using a digital twin model, and optimize the storage location of goods using a simulated annealing algorithm according to the weight, volume, and inbound and outbound frequency characteristics of the goods.

[0036] The steps to generate an inventory policy using a genetic algorithm include: (1) Parameter initialization, safety inventory level: range [0, S] pieces, replenishment point: range [0, R] pieces, replenishment quantity: range [0, Q] pieces; (2) Input parameters, obtain demand prediction data D(t) from the spatio-temporal prediction module, and input cost parameters (purchase cost, unit holding cost rate, shortage penalty coefficient) and algorithm parameters (population size N = 100, maximum number of generations T = 200, crossover probability p_c = 0.85, mutation probability p_m = 0.01); (3) Perform chromosome encoding to generate an initial population; (4) Calculate the fitness function, perform selection, crossover, and mutation operations, and finally select the individual with the highest fitness in the final population, and perform decoding to output the inventory policy, safety inventory level, replenishment point, replenishment quantity, and expected total cost. Optimizing the storage location of goods using a simulated annealing algorithm includes: defining the state space, mapping the storage location coordinates (x, y, z) to the digital twin model grid, and using a triple (weight, volume, turnover_rate) for the goods attributes. Classify the inbound and outbound frequencies: Class A (>10 times / day), Class B (1 - 10 times / day), Class C (<1 time / day); Perform neighborhood operations: For the exchange strategy, randomly exchange the goods in two non-empty storage locations, with the volume of the goods ≤ the remaining volume of the target storage location. For the insertion strategy, randomly select goods and insert them into an empty storage location, with the weight of the goods ≤ the upper limit of the shelf load; The cooling strategy uses exponential annealing (initial temperature T0 = 1000, cooling function: T_{k+1}=α·T_k, attenuation coefficient α = 0.95, termination condition is that the improvement of the optimal solution in 10 consecutive generations <0.1% or T < 0.01).

[0037] More specifically, the inventory optimization management module further includes an inbound and outbound task scheduling unit. The inbound and outbound task scheduling unit is used to formulate an inbound and outbound task scheduling plan based on the inventory policy and the storage location of the goods, and use the digital twin model to simulate the inbound and outbound process, considering factors such as the running speed, path length, and goods handling sequence of the handling equipment, and optimize the handling path of the goods and the scheduling of the handling equipment.

[0038] In this embodiment, the path planning algorithm combines global path planning with local obstacle avoidance. The global path planning uses the A* algorithm combined with the improved Dijkstra algorithm, and the local obstacle avoidance uses the dynamic window method to achieve real-time obstacle avoidance. The equipment scheduling uses a distributed task allocation method based on the auction algorithm, and uses a time window negotiation mechanism to resolve conflicts. The auction algorithm process includes: Task release: The central scheduler broadcasts a task list, and the task list contains task location / time limit / priority; AGV bidding: Calculate the cost C_i = k_1·distance + k_2·time window conflict degree; Winner determination: The AGV with the minimum cost obtains the task, updates its time window table, and uses the time window negotiation mechanism to resolve conflicts.

[0039] The user interaction module of this application includes a visualization display unit and an interaction editing unit. The visualization display unit is used to visually display the digital twin model, real-time data of warehouse inventory, spatio-temporal prediction results, and inventory optimization management solution information, and receive the user's interaction operation instructions. The interaction editing unit is used to obtain the inventory strategy parameters input by the user, adjust the storage locations of goods, and the parameters for issuing inbound and outbound tasks, and convert them into logical execution instructions, verify and process the instructions to ensure the legality and accuracy of the operations, feedback the operation results to the user, and update and display them on the visualization interface.

[0040] User operation verification includes permission control and operation auditing. Permission control is carried out through the RBAC model (Role-Permission-Operation), and the roles include administrator, dispatcher, and monitor; the operation auditing part records operation logs, and the log format is [timestamp][operator][object ID][operation type][SHA-256(parameters)]. The instruction processing process is: user input → syntax parsing (using ANTLR4 to implement instruction syntax checking) → semantic verification → permission checking → execution engine → result feedback. The deployment architecture of this intelligent platform includes an edge layer and a cloud layer. The digital twin engine (NVIDIA Jetson AGX Orin) is deployed in the edge layer, and services such as spatio-temporal prediction models (AWS SageMaker) are run in the cloud layer.

[0041] A method for optimizing warehouse inventory includes the steps of: synchronizing the full-element state of the physical warehouse to the digital twin; extracting spatio-temporal sequence data in the past 30 days to train the ST-GNN prediction model; rolling and updating the heat map of the demand in the storage area for the next 6 hours every 15 minutes; triggering the AGV dynamic rebalancing task when detecting local inventory imbalance in the storage location; generating an inventory health report every 24 hours, marking slow-moving goods and high-risk out-of-stock categories.

[0042] The above are only the preferred embodiments of this application and are not used to limit this application. For those skilled in the art, various changes and modifications can be made to the embodiments of this application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included within the protection scope of this application.

Claims

1. An intelligent warehousing optimization management platform based on digital twin and spatio-temporal prediction, characterized in that, Including: A digital twin model construction module, a spatio-temporal prediction module, an inventory optimization management module, and a user interaction module; The digital twin model construction module is used to construct a three-dimensional model of the storage space, and associate the collected storage sensing data with the three-dimensional model of the storage space to establish a digital twin model; The spatio-temporal prediction module is used to construct a spatio-temporal prediction model, input the real-time collected storage inventory data into the spatio-temporal prediction model for prediction, obtain inventory prediction information for a future period of time, and use the inventory prediction information for inventory risk assessment and early warning; The inventory optimization management module is connected to the spatio-temporal prediction module. The inventory optimization management module is used to generate an inventory strategy using an optimization algorithm according to the prediction result of the spatio-temporal prediction module, storage cost, and goods demand, and is also used to optimize the storage location of the goods, and formulate an inbound and outbound task scheduling plan according to the inventory strategy and the storage location of the goods; The user interaction module is used to visually display the digital twin model, real-time data of the storage inventory, spatio-temporal prediction results, and inventory optimization management plan information, and receive user interaction operation instructions.

2. The intelligent warehousing optimization management platform based on digital twin and spatio-temporal prediction according to claim 1, characterized in that, The digital twin model construction module includes a data processing unit, a spatial three-dimensional model construction unit, and a model association unit. The data processing unit and the spatial three-dimensional model construction unit are connected to the model association unit; The data processing unit is used to obtain the goods location, storage temperature and humidity, and goods weight data uploaded by the storage Internet of Things device end, suppress the noise of the sensor data using the Kalman filtering algorithm, and realize multi-source data synchronization through timestamp alignment; The spatial three-dimensional model construction unit is used to perform parametric basic modeling of the storage structure using three-dimensional modeling software according to the actual size and structure parameters of the storage space, and realize equipment kinematic modeling through the Unity engine to obtain a three-dimensional model of the storage space; The model association unit is used to associate the goods location, storage temperature and humidity, goods weight data, and the three-dimensional model of the storage space, realize the unique mapping of the physical device and the digital model based on UUID, adopt the spatial coordinate system conversion algorithm to realize position synchronization, obtain the digital twin model, and when the physical entity of the storage space changes, perform corresponding real-time updates and incremental updates on the digital twin model.

3. The intelligent warehousing optimization management platform based on digital twin and spatio-temporal prediction according to claim 1, wherein The spatio-temporal prediction module includes a data preprocessing unit, a model establishment unit, and a prediction and early warning unit; The data preprocessing unit is used to collect historical data of goods inbound and outbound from the warehouse management system. The historical data includes inbound and outbound time, quantity, and type information, clean the collected historical data, remove duplicate data, error data, and missing values, and then extract time features and spatial features. The time features include a time domain feature set, and the spatial features include regional data divided by GeoHash coding; The model building unit is used to construct a spatio-temporal prediction model based on a spatio-temporal convolutional neural network. The spatio-temporal prediction model includes a spatial convolutional layer, a temporal convolutional layer, and an attention mechanism module. The preprocessed historical data is divided into a training set and a test set. The spatio-temporal prediction model is trained using the training set, and the parameters of the model are adjusted through the backpropagation algorithm to obtain a trained spatio-temporal prediction model; The prediction and warning unit is used to obtain the warehousing inventory data collected in real time and input it into the trained spatio-temporal prediction model to obtain the inventory prediction results for a future period of time. The predicted inventory level is compared with a preset replenishment point threshold and a safety inventory level. When the predicted inventory level is lower than the replenishment point, a replenishment warning signal is automatically output. When the predicted inventory level is higher than the safety inventory level, an inventory overstock warning signal is automatically output. It is also used to integrate a multi-task learning model to synchronously output the inventory shortage probability, hot spots of high-turnover goods, and warehousing equipment failure risk warnings.

4. The intelligent warehousing optimization management platform based on digital twin and spatio-temporal prediction according to claim 1, characterized in that, The inventory optimization management module includes an inventory strategy generation unit. The strategy generation unit is used to generate an inventory strategy using a genetic algorithm based on the prediction results of the spatio-temporal prediction module, warehousing costs, and goods demands, and determine the safety inventory level, replenishment point, and replenishment quantity.

5. The intelligent warehousing optimization management platform based on digital twin and spatio-temporal prediction according to claim 4, wherein The inventory optimization management module further includes a goods storage optimization unit. The goods storage optimization unit is used to simulate and analyze the warehousing space using a digital twin model, and optimize the storage locations of goods using a simulated annealing algorithm according to the weight, volume, and inbound / outbound frequency characteristics of the goods.

6. The intelligent warehousing optimization management platform based on digital twin and spatio-temporal prediction according to claim 5, characterized in that The inventory optimization management module further includes an inbound / outbound task scheduling unit. The inbound / outbound task scheduling unit is used to formulate an inbound / outbound task scheduling plan based on the inventory strategy and the storage locations of the goods, and simulate and analyze the inbound / outbound process using a digital twin model. Considering factors such as the running speed of handling equipment, path length, and goods handling sequence, it optimizes the handling paths of goods and the scheduling of handling equipment.

7. The intelligent warehousing optimization management platform based on digital twin and spatio-temporal prediction according to claim 1, characterized in that, The user interaction module includes a visualization display unit and an interaction editing unit; The visualization display unit is used to visually display the digital twin model, real-time data of warehousing inventory, spatio-temporal prediction results, and inventory optimization management plan information, and receive interactive operation instructions from users; The interaction editing unit is used to obtain the inventory strategy parameters input by the user, adjust the parameters for the storage locations of goods, and issue inbound / outbound tasks, and convert them into logical execution instructions. It verifies and processes the instructions to ensure the legality and accuracy of the operations, feeds back the operation results to the user, and updates and displays them on the visualization interface.

Images