Intelligent WMS system with intelligent scheduling function

Abstract

The invention, which relates to the warehouse management field, discloses an intelligent WMS system with an intelligent scheduling function, comprising a cargo sensing module, a data processing and analysis module, a scheduling decision module, a decision optimization module, and an execution control module. And the multi-agent scheduling module is used for defining each independent operation task in the warehouse as an agent, modeling the warehouse environment, and setting a reward function for each agent for guiding the agent to make an optimal decision. According to the intelligent WMS system with the intelligent scheduling function, the task completion time, the resource utilization rate, the cost saving and the service quality can be comprehensively considered through the set reward function, an intelligent agent can be guided to make an optimal decision, the accuracy and efficiency of future decision making can be improved through learning, and the intelligent WMS system has good application prospects. Therefore, the intelligent agent can gradually master the optimal scheduling strategy in different environments, and maximization of the overall warehouse operation efficiency is realized.

Description

Technical Field

[0001] The present invention relates to the field of warehouse management, and in particular to an intelligent WMS system with intelligent scheduling function. Background Art

[0002] The WMS system, or warehouse management system, is an information system used to manage the entire process of warehouse operations. From the actual application effect, the WMS system has many advantages: through automated task allocation and process optimization, it reduces the time and errors of manual operations. For example, the optimization of picking paths can allow pickers to avoid unnecessary steps, thereby increasing the picking speed and further increasing the turnover rate of goods in the entire warehouse; all warehouse operation data are recorded and updated by the system, avoiding possible errors in manual records, which makes inventory data more accurate and provides a reliable basis for the company's procurement, sales and other decisions; managers can use the system to understand the warehouse's operating status in real time, such as inventory levels, in and out of the warehouse dynamics, etc., and the system can standardize the warehouse operation process so that each operation has rules to follow, which is convenient for management and supervision.

[0003] In traditional warehouse scheduling, there is an over-reliance on existing empirical strategies, which makes it impossible to find better scheduling solutions, so that the scheduling solutions will always use the old and inefficient methods. At the same time, the existing models may ignore some physical details and behavioral characteristics of goods and equipment, resulting in deviations between simulation results and actual operations, making it impossible to accurately evaluate the actual impact of scheduling strategies on equipment and goods. In addition, traditional scheduling strategy recommendations may rely on manual experience or simple rules, and lack comprehensive analysis of historical data and complex scenarios. Based on this, staff cannot make optimal scheduling decisions.

[0004] Therefore, it is necessary to propose an intelligent WMS system with intelligent scheduling function to solve the above problems. Summary of the invention

[0005] The main purpose of the present invention is to provide an intelligent WMS system with intelligent scheduling function, which can effectively solve the problems in the background technology.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] An intelligent WMS system with intelligent scheduling function An intelligent WMS system with intelligent scheduling function includes a cargo sensing module, a data processing and analysis module, a scheduling decision module, a decision optimization module, and an execution control module. The data processing and analysis module predicts the warehouse status through data collection, processing and analysis, and formulates corresponding strategies according to the prediction results;

[0008] The scheduling decision module is electrically connected to the data processing and analysis module. The scheduling decision module includes a multi-agent scheduling module and an auxiliary scheduling decision module. The multi-agent scheduling module is used to define each independent operation task in the warehouse as an agent, model the warehouse environment, and set a multi-dimensional reward function for each agent to guide it to make the best decision;

[0009] The auxiliary scheduling decision module truly reflects the actual status of the warehouse based on the constructed digital twin model, and sets up multiple simulation scenarios to formulate multiple scheduling strategies, including different task allocation rules, equipment scheduling plans, and inventory layout adjustment strategies.

[0010] Preferably, the cargo sensing module is based on the integration of an RFID antenna array, an environmental sensing module and an intelligent visual recognition module. The RFID antenna array acquires basic cargo information based on an RFID tag arranged on the cargo packaging. The visual recognition module is used for detailed detection and image acquisition of the cargo's appearance, shape and packaging integrity. The environmental sensing module constructs a comprehensive environmental monitoring system based on the arranged sensors.

[0011] Preferably, the data processing and analysis module is electrically connected to the cargo sensing module, and the data processing and analysis module also includes an external data expansion and integration module, an abnormality monitoring module, and a predictive analysis module. The data expansion and integration module is used to establish a data interface with the e-commerce platform, the supplier system, and the traffic management department, and is used to obtain order information, customer evaluation, and sales trend forecast data from the e-commerce platform for planning logistics resources in advance; obtain raw material supply plans and delivery times from the supplier system for optimizing inventory management; and connect with the data of the traffic management department to obtain real-time road condition information, road construction information, and traffic control information to provide a basis for transportation route planning;

[0012] The anomaly monitoring module detects anomalies in logistics data based on convolutional neural networks and recurrent neural networks. Convolutional neural networks are used to process image data, including identifying damaged cargo packaging and damaged vehicle appearance. Recurrent neural networks are used to process time series order data and vehicle driving trajectory data, based on which abnormal fluctuations in order volume and deviations in vehicle driving routes are detected.

[0013] The anomaly detection steps include the following:

[0014] S301: Collect and pre-process the data obtained from the cargo sensing module, and mark normal data and abnormal data. The normal data and abnormal data are set based on the actual situation of vehicle driving data, cargo status data, inventory data, warehouse environment data, order data, and customer information data;

[0015] S302: Feature extraction: extract image features from the data through a convolutional neural network, extract time series data from the data through a recurrent neural network, and perform training, feature learning, and optimization on them;

[0016] S303: The pre-processed data obtained from the cargo sensing module is input into the convolutional neural network and the recurrent neural network according to its characteristics, and anomalies are detected based on this, and corresponding processing is performed according to the monitored abnormal data.

[0017] Preferably, the predictive analysis module includes a deep feature mining and adaptive selection module, an ARIMA model and GARCH model hybrid module, a dynamic adjustment module, a deep neural network model construction module, a prediction enhancement module, and a model evaluation and optimization module. The deep feature mining and adaptive selection module performs feature mining based on the autoencoder technology in deep learning. The autoencoder is used to learn the deep feature representation of data, compress the original high-dimensional data into a low-dimensional feature space, and then decode and restore the data, optimize feature extraction by minimizing reconstruction errors, and perform adaptive optimization of feature selection based on a reinforcement learning algorithm. The reinforcement learning algorithm determines feature selection and weight adjustment based on performance feedback of the predictive analysis module, including accuracy and recall rate;

[0018] In the hybrid module of the ARIMA model and the GARCH model, the ARIMA model is an autoregressive integrated moving average model, and the GARCH model is a generalized autoregressive conditional heteroscedasticity model, which is used to integrate the ARIMA model with the GARCH model to address the common volatility and heteroscedasticity problems in logistics data, including the following steps:

[0019] S401: Use the ARIMA model to model the mean part of the time series to capture the linear trend and seasonality of the data, including predicting the long-term demand trend of goods and the seasonal changes in transportation flow;

[0020] S402: Input the residual sequence of the ARIMA model into the GARCH model. The GARCH model is used to model the variance of the residuals and effectively capture the volatility and heteroscedasticity characteristics of the data, including predicting the fluctuation range of cargo prices in a specific time period and the sudden change amplitude of transportation demand. The hybrid module of the ARIMA model and the GARCH model is used to provide more comprehensive time series forecasting results, including the forecast of the data mean, the confidence interval of the forecast and the risk assessment;

[0021] The dynamic adjustment module automatically adjusts based on the data fed back by the cargo sensing module and the hybrid module of the ARIMA model and the GARCH model. Based on the sliding window technology, it monitors the difference between the newly-inflowed data and the time series prediction results in real time. When the difference exceeds the set threshold, the system automatically triggers the model parameter optimization process, and uses the particle swarm optimization algorithm and the genetic algorithm to re-optimize the parameters of the ARIMA model and the GARCH model. In the optimization process, the objective function is to minimize the prediction error, while considering the complexity and stability of the model.

[0022] The deep neural network model building module is used to build a deep neural network model that can integrate multiple logistics data modalities. The logistics data includes structured data: order information, inventory data; semi-structured data: cargo description, logistics log; unstructured data: cargo images, transport vehicle monitoring videos. The convolutional neural network is used to process image and video data to extract the appearance features of the cargo and the operating status features of the transport vehicle; the recurrent neural network is used to process structured and semi-structured sequence data, including the time series of orders and the change history of inventory; the features of different modalities are integrated through the fully connected layer, and the prediction task is performed on the integrated feature space;

[0023] The prediction enhancement module is based on a generative adversarial network to solve the common small sample problem in logistics data, including goods and businesses with less historical data. The generative adversarial network consists of a generator and a discriminator. The generator is used to generate new simulated sample data based on limited real sample data. The simulated sample data is similar to the real sample in data distribution.

[0024] The model evaluation and optimization module performs model optimization based on the NSGA-II algorithm, while considering multiple conflicting evaluation indicators, including prediction accuracy, model complexity, and computing resource consumption. It is used to perform non-dominated sorting and crowding calculation on individuals in the model parameter combination population according to multiple evaluation indicators, and select the optimal model parameter combination.

[0025] Preferably, the scheduling decision module is electrically connected to the data processing and analysis module, and the scheduling decision module includes a multi-agent scheduling module and an auxiliary scheduling decision module. The multi-agent scheduling module includes an agent and environment modeling module, a reward mechanism design module, and a learning and decision module. The agent and environment modeling module is used to define each independent work task in the warehouse as an agent, including but not limited to order picking tasks, goods shelving tasks, and replenishment tasks. Each agent has its own status information, including task priority, current execution progress, required resource status, and association with other tasks. The agent and environment modeling module also includes warehouse resources and environmental status. The warehouse resources include but are not limited to various types of handling equipment, idle storage spaces, and human resources. The warehouse resources constitute the environmental elements in which the agent is located, and the environmental status is described by the real-time availability, capacity, location distribution, and environmental conditions of the warehouse resources.

[0026] Preferably, the reward mechanism design module sets a multi-dimensional reward function for each agent to guide it to make the best decision. The reward function comprehensively considers task completion time, resource utilization, cost savings and service quality. If the agent can complete the task in a short time and use resources efficiently, a higher positive reward will be given; if the task is delayed and causes resource waste, a negative reward will be given;

[0027] The learning and decision-making module learns based on the continuous interaction between the agent and the environment. At each decision-making moment, the agent selects an action from the action space based on its current state and the state of the environment using the proximal policy optimization algorithm. The action includes choosing which device to use to perform the task, determining the path and sequence of task execution, and deciding the specific location for cargo storage.

[0028] Preferably, the reward mechanism design module specifically includes:

[0029] Determine the reward factors and weights, including the task completion time factor: used to calculate the ratio of the time taken from the start to the completion of the task to the expected standard time. The formula is:

[0030]

[0031] Where R time is the task completion time reward factor, T actual is the actual completion time of the task, T standard is the expected standard time of the task. When the task is completed ahead of schedule, R time If the task is completed late, R time Less than 1, the task completion time factor weight is set to w time ;

[0032] Resource utilization factor: used to calculate the utilization of different resources and take them into consideration. The formula is:

[0033] R resource =∑ i w i U i ;

[0034] Where R resource is the comprehensive resource utilization rate, w i is the weight of different resource utilization, U i is the utilization rate of the i-th resource, and the weight of the resource utilization factor is set to w resource ;

[0035] Cost saving factor: used to consider the cost saved during task execution, including equipment operating costs and labor costs. The formula is:

[0036]

[0037] Where R cost is the cost saving incentive factor, C saved To save costs, C budest is the budget cost, and the weight of this cost saving factor is set to w cost ;

[0038] Service quality factor: used to measure the accuracy and completeness of task completion, the formula is:

[0039] R service =A accuracy *A completeness ;

[0040] Where R service is the service quality reward factor, A accuracy is the accuracy of the service, A completeness is the service completeness rate, and the weight of this service quality factor is set to w service ;

[0041] Based on the above content, the reward function is constructed as follows:

[0042] R=w time R time +w resource R resource +w cost R cost +w service R service ;

[0043] Where R is the reward function.

[0044] Preferably, the auxiliary scheduling decision module includes a digital twin warehouse construction module, a simulation scenario and strategy setting module, and a simulation evaluation and optimization module. The digital twin warehouse construction module creates a digital twin model that completely corresponds to the real warehouse based on 3D modeling technology and Internet of Things data. The digital twin model is used to restore the physical layout of the warehouse, including the location, size, and number of layers of shelves, the running tracks and docking points of handling equipment, and the storage location, shape, and volume of goods. The digital twin model is also embedded with the logistics process, equipment operation logic, and personnel operation specifications and business rules in the warehouse to truly reflect the actual operation status of the warehouse;

[0045] The simulation scenario and strategy setting module is used to set up multiple simulation scenarios in the digital twin model according to actual business needs and possible situations, including different order volumes: daily order volume, promotional activity order volume, peak season order volume; different cargo distribution: evenly distributed, concentrated in a specific area; equipment failure: forklift failure, conveyor belt failure; emergency order insertion; formulate multiple scheduling strategies for each scenario, including different task allocation rules, equipment scheduling plans, and inventory layout adjustment strategies;

[0046] The simulation evaluation and optimization module is used to simulate the operation of different scheduling strategies in each simulation scenario, collect and analyze relevant data, including the time of goods entering and leaving the warehouse, equipment utilization, inventory turnover rate, and task completion rate. By comparing the performance of different strategies in the same scenario, the advantages and disadvantages of each strategy are evaluated, and the strategies with better performance are selected for further optimization. The optimization process can adjust the parameters in the strategy, including the equipment allocation ratio and task priority weight, and then perform simulation evaluation again until the optimal scheduling strategy combination is found.

[0047] Preferably, the decision optimization module includes a digital twin model update module and a simulation scenario generation and optimization module. The digital twin model update module is used to improve the synchronization with the physical warehouse, optimize the data fusion mechanism, and is used to fuse with the computer vision system, thereby obtaining real-time image data of the status and location of the goods and real-time motion capture data of the employee's operation, and dynamically adjust the model update frequency according to the dynamic degree of warehouse operation and the importance of the data;

[0048] In the construction of the digital twin model, more physical details and behavioral characteristics are added, including considering the impact of the physical characteristics of the goods on the equipment handling operation, and simulating the energy consumption changes and mechanical wear of the equipment under different loads and operating speeds;

[0049] The simulation scenario generation and optimization module is used to add more complex and rare scenarios, including but not limited to the impact of natural disasters on warehouses and supply chain disruption scenarios. A combinatorial optimization method is used to perform combined simulations on different scenarios to study scheduling strategies when multiple adverse factors occur simultaneously.

[0050] Preferably, the execution control module is used to realize high-speed, low-latency data transmission of scheduling instructions through a device control interface of communication technology and a personnel interaction terminal.

[0051] Compared with the prior art, the present invention provides an intelligent WMS system with intelligent scheduling function, which has the following beneficial effects:

[0052] 1. The intelligent WMS system with intelligent scheduling function can be connected with external platforms through the data expansion and integration module, so as to expand the system. It can obtain order information, customer evaluation and sales trend forecast data from the e-commerce platform, based on which it can be used to plan logistics resources in advance; obtain raw material supply plans and delivery times from the supplier system, based on which inventory management can be optimized; connect with the data of the traffic management department to obtain real-time traffic information, road construction information and traffic control information, based on which it can provide a basis for transportation route planning, and in subsequent scheduling, it can provide higher reliability.

[0053] 2. The intelligent WMS system with intelligent scheduling function can also detect anomalies in the warehouse's logistics data, detect abnormal situations, and deal with them accordingly, including the allocation of logistics resources, preventing theft and transportation accidents of goods, and accurately identifying subtle differences and abnormal situations of goods in the warehouse. It can predict the fluctuation range of goods prices and the sudden change in transportation demand within a specific time period through the hybrid module of ARIMA model and GARCH model, and can predict the average quantity of transportation demand within a certain period of time and the upper and lower limits of demand fluctuations. Logistics companies can reasonably arrange transportation resources and inventory levels based on this information, and formulate strategies to deal with the risks of demand fluctuations, such as signing flexible supply contracts with suppliers and preparing emergency transportation plans.

[0054] 3. The intelligent WMS system with intelligent scheduling function can guide the intelligent agent to make the best decision when managing the warehouse by comprehensively considering the task completion time, resource utilization, cost savings and service quality through the set reward function. It can improve the accuracy and efficiency of future decisions through learning. After a lot of interactive learning, the intelligent agent can gradually master the optimal scheduling strategy in different environments and maximize the overall warehouse operation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a system block diagram of the present invention. DETAILED DESCRIPTION

[0056] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0057] Embodiment 1:

[0058] like Figure 1 As shown, an intelligent WMS system with intelligent scheduling function, an intelligent WMS system with intelligent scheduling function, the cargo sensing module is based on the integration of RFID antenna array, environmental sensing module and intelligent visual recognition module, the RFID antenna array acquires basic cargo information based on the RFID tag arranged on the cargo packaging; the visual recognition module is used for detailed detection and image acquisition of the appearance, shape and packaging integrity of the cargo, and the environmental sensing module builds a comprehensive environmental monitoring system based on the arranged sensors.

[0059] Embodiment 2:

[0060] like Figure 1 As shown, an intelligent WMS system with intelligent scheduling function, the data processing and analysis module is electrically connected to the cargo sensing module, the data processing and analysis module also includes an external data expansion and integration module, an abnormality monitoring module, and a predictive analysis module. The data expansion and integration module is used to establish a data interface with the e-commerce platform, the supplier system, and the traffic management department, and is used to obtain order information, customer evaluation, and sales trend forecast data from the e-commerce platform for planning logistics resources in advance; obtain raw material supply plans and delivery times from the supplier system for optimizing inventory management; and connect with the traffic management department's data to obtain real-time road conditions, road construction information, and traffic control information to provide a basis for transportation route planning;

[0061] The anomaly monitoring module detects anomalies in logistics data based on convolutional neural networks and recurrent neural networks. Convolutional neural networks are used to process image data, including identifying damaged cargo packaging and damaged vehicle exteriors. Recurrent neural networks are used to process time series order data and vehicle driving trajectory data, based on which abnormal fluctuations in order volume and deviations from vehicle driving routes are detected. By learning from historical order data, the recurrent neural network model can identify abnormal situations in which the order volume in a certain area suddenly increases by more than 50% within a specific time period. This indicates that there are promotional activities and sudden demands in the area, and an early warning is automatically triggered at this time, suggesting an increase in logistics resource allocation. For the driving trajectory of transport vehicles, the recurrent neural network model is used to monitor in real time when the vehicle deviates from the scheduled route by more than 5 kilometers, issue an alarm in time, and notify relevant personnel to verify, to prevent theft and transportation accidents of goods.

[0062] The anomaly detection steps include the following:

[0063] S301: Collect and pre-process the data obtained from the cargo sensing module, and mark normal data and abnormal data. The normal data and abnormal data are set based on the actual situation of vehicle driving data, cargo status data, inventory data, warehouse environment data, order data, and customer information data;

[0064] S302: Feature extraction: extract image features from the data through a convolutional neural network, extract time series data from the data through a recurrent neural network, and perform training, feature learning, and optimization on them;

[0065] S303: The pre-processed data obtained from the cargo sensing module is input into the convolutional neural network and the recurrent neural network according to its characteristics, and anomalies are detected based on this, and corresponding processing is performed according to the monitored abnormal data.

[0066] The predictive analysis module includes a deep feature mining and adaptive selection module, an ARIMA model and GARCH model hybrid module, a dynamic adjustment module, a deep neural network model building module, a prediction enhancement module, and a model evaluation and optimization module. The deep feature mining and adaptive selection module performs feature mining based on the autoencoder technology in deep learning. The autoencoder is used to learn the deep feature representation of the data, compress the original high-dimensional data into a low-dimensional feature space, and then decode and restore the data. It optimizes feature extraction by minimizing the reconstruction error. When processing cargo image data, the autoencoder automatically learns the shape, color, and texture features of the cargo. Compared with traditional manually designed features, these features can more accurately identify subtle differences in cargo. and abnormal situations, and adaptively optimize feature selection based on the reinforcement learning algorithm. The reinforcement learning algorithm determines the selection and weight adjustment of features based on the performance feedback of the predictive analysis module, including accuracy and recall. In cargo demand forecasting, if a feature contributes less to the current predictive analysis module, the reinforcement learning algorithm can try to reduce its weight, temporarily remove it from the feature set, and then observe the changes in model performance. When the model performance is improved, continue to optimize the feature selection strategy; when the performance decreases, restore the previous feature settings and try other adjustment methods. Through this adaptive feature selection mechanism, it can ensure that the predictive analysis module always uses the most valuable feature combination to improve the accuracy and efficiency of the prediction;

[0067] In the hybrid module of ARIMA model and GARCH model, ARIMA model is an autoregressive integrated moving average model, and GARCH model is a generalized autoregressive conditional heteroskedasticity model. It is used to integrate ARIMA model and GARCH model to solve the common volatility and heteroskedasticity problems in logistics data, including the following steps:

[0068] S401: Use the ARIMA model to model the mean part of the time series to capture the linear trend and seasonality of the data, including predicting the long-term demand trend of goods and the seasonal changes in transportation flow;

[0069] S402: Input the residual sequence of the ARIMA model into the GARCH model. The GARCH model is used to model the variance of the residual and effectively capture the volatility and heteroscedasticity characteristics of the data, including predicting the fluctuation range of the price of goods in a specific time period and the sudden change amplitude of the transportation demand. The hybrid module of the ARIMA model and the GARCH model is used to provide more comprehensive time series prediction results, including the prediction of the data mean, the confidence interval of the prediction and the risk assessment. For example, when predicting the transportation demand of a certain type of high-value electronic products, the hybrid module of the ARIMA model and the GARCH model can predict the average quantity of transportation demand in the next month and the upper and lower limits of demand fluctuations. Logistics companies can reasonably arrange transportation resources and inventory levels based on this information, and formulate strategies to deal with the risk of demand fluctuations, such as signing flexible supply contracts with suppliers and preparing emergency transportation plans.

[0070] The dynamic adjustment module automatically adjusts based on the data fed back by the cargo sensing module and the hybrid module of the ARIMA model and the GARCH model. Based on the sliding window technology, it monitors the difference between the newly-incoming data and the time series prediction results in real time. When the difference exceeds the set threshold, the system automatically triggers the model parameter optimization process, and uses the particle swarm optimization algorithm and the genetic algorithm to re-optimize the parameters of the ARIMA model and the GARCH model. During the optimization process, the objective function is to minimize the prediction error, while considering the complexity and stability of the model. By continuously and dynamically adjusting the model parameters according to the new data, the time series model can quickly adapt to the changing laws of logistics data, improve the timeliness and accuracy of the prediction, and when an emergency occurs in the logistics market, the model can adjust the parameters in time, accurately predict the changes in logistics demand and transportation time of different regions and different types of goods during the emergency, and provide strong support for the emergency decision-making of logistics companies.

[0071] The deep neural network model building module is used to build a deep neural network model that can integrate multiple logistics data modalities. Logistics data includes structured data: order information, inventory data; semi-structured data: cargo description, logistics log; unstructured data: cargo image, transport vehicle monitoring video. Convolutional neural network is used to process image and video data to extract the appearance features of cargo and the running status features of transport vehicles; recurrent neural network is used to process structured and semi-structured sequence data, including the time series of orders and the change history of inventory; the features of different modalities are fused through the fully connected layer, and prediction tasks are performed on the fused feature space. When predicting the quality risk of goods, the model can simultaneously analyze the image data of the goods, including checking whether the packaging is damaged, whether there are abnormal changes in the appearance of the goods, video data during transportation, including monitoring the driving stability of the transport vehicle, whether there are severe bumps and collisions, as well as order information and logistics logs, including understanding the storage conditions and transportation time of the goods. Through the comprehensive analysis of multi-modal data, the quality risk of goods can be predicted more comprehensively and accurately, and preventive measures can be taken in advance to reduce logistics losses.

[0072] The prediction enhancement module is based on the generative adversarial network to solve the common small sample problem in logistics data, including goods and businesses with less historical data. The generative adversarial network consists of a generator and a discriminator. The generator is used to generate new simulated sample data based on limited real sample data. The simulated sample data is similar to the real sample in data distribution. For the fault prediction of a new type of intelligent logistics equipment, since the equipment has just been put into use, the historical fault data is very limited. The generative adversarial network generates more simulated fault data based on the existing small amount of fault data and a large amount of normal operation data. The generated simulated samples are used together with the real samples to train the predictive analysis module, which can effectively expand the amount of training data and improve the generalization ability and accuracy of the predictive analysis module in the case of small samples. By continuously optimizing the training process of GAN, the simulated samples generated by it are more real and diverse, further improving the performance of the predictive analysis module;

[0073] The model evaluation and optimization module performs model optimization based on the NSGA-II algorithm, while considering multiple conflicting evaluation indicators, including prediction accuracy, model complexity, and computing resource consumption. It is used to perform non-dominated sorting and congestion calculation on individuals in the model parameter combination population according to multiple evaluation indicators, and select the optimal model parameter combination. When optimizing the cargo demand predictive analysis module, it is necessary to pursue high prediction accuracy while controlling the complexity of the model. The NSGA-II algorithm can find a balance between these two conflicting goals and generate a series of different model parameter combinations. Each combination has relative advantages in prediction accuracy and model complexity. Logistics companies can select the most suitable model parameter configuration from the optimal solutions of these model parameter combinations according to their actual needs and resource constraints, so as to maximize model performance and optimize resource utilization.

[0074] Embodiment three:

[0075] like Figure 1 As shown, an intelligent WMS system with intelligent scheduling function, the scheduling decision module includes a multi-agent scheduling module and an auxiliary scheduling decision module. The multi-agent scheduling module is used to define each independent operation task in the warehouse as an agent, and model the warehouse environment, and set a multi-dimensional reward function for each agent to guide it to make the best decision;

[0076] The auxiliary scheduling decision module truly reflects the actual status of the warehouse based on the constructed digital twin model, and sets multiple simulation scenarios to formulate multiple scheduling strategies, including different task allocation rules, equipment scheduling plans, and inventory layout adjustment strategies;

[0077] The scheduling decision module includes a multi-agent scheduling module and an auxiliary scheduling decision module. The multi-agent scheduling module includes an agent and environment modeling module, a reward mechanism design module, and a learning and decision module. The agent and environment modeling module is used to define each independent operation task in the warehouse as an agent, including but not limited to order picking tasks, goods shelving tasks, and replenishment tasks. Each agent has its own status information, including the priority of the task, the current execution progress, the required resources, and the relationship with other tasks. The agent and environment modeling module also includes warehouse resources and environmental status. Warehouse resources include but are not limited to various types of handling equipment, idle storage locations, and human resources. Warehouse resources constitute the environmental elements in which the agent is located. The environmental status is described by the real-time availability, capacity, location distribution, and environmental conditions of warehouse resources. For example, in a warehouse, the status of an order picking agent is that the order contains 5 kinds of goods, 2 have been picked, the priority is high, forklifts are required for transportation, and it is suitable to be stored on the shelves in the normal temperature shipping area; while the environmental status is that there are currently 3 forklifts available, the shelves in the normal temperature shipping area have 10 idle locations, and the temperature in the warehouse is 20 degrees Celsius.

[0078] The reward mechanism design module sets a multi-dimensional reward function for each agent to guide it to make the best decision. The reward function comprehensively considers the task completion time, resource utilization, cost savings and service quality. If the agent can complete the task in a short time and use resources efficiently, it will be given a higher positive reward; if the task is delayed and causes resource waste, it will be given a negative reward. Taking the goods shelving agent as an example, if it can accurately place the goods in the appropriate location, which not only meets the requirements of the goods storage conditions but also improves the shelf space utilization rate, and completes the shelving operation within the specified time, the system will give a certain reward score. This score will increase with the improvement of resource utilization efficiency and the shortening of time. On the contrary, if the shelving location is wrong and exceeds the specified time, the corresponding score will be deducted;

[0079] The learning and decision-making module learns based on the continuous interaction between the agent and the environment. At each decision moment, the agent selects an action from the action space using the proximal policy optimization algorithm based on its current state and the state of the environment. The action includes choosing which device to use to perform the task, determining the path and sequence of task execution, and deciding the specific location of the goods storage. For example, when a replenishment agent finds that a certain type of goods is out of stock, it will select the most suitable forklift and plan the optimal replenishment path based on the current forklift position, idle status, distance to the target storage area, and goods storage layout information, and move the goods from the temporary storage area to the designated shelf location. After executing the action, the environmental state changes, and the agent obtains corresponding rewards based on the new environmental state and task completion status, and updates its own policy network to improve the accuracy and efficiency of future decisions. After a lot of interactive learning, the agent can gradually master the optimal scheduling strategy in different environments and maximize the overall warehouse operation efficiency.

[0080] The reward mechanism design module specifically includes:

[0081] Determine the reward factors and weights, including the task completion time factor: used to calculate the ratio of the time taken from the start to the completion of the task to the expected standard time. The formula is:

[0082]

[0083] Where R time is the task completion time reward factor, T actual is the actual completion time of the task, T standard is the expected standard time of the task. When the task is completed ahead of schedule, R time If the task is completed late, R time Less than 1, the task completion time factor weight is set to w time ;

[0084] Resource utilization factor: used to calculate the utilization of different resources and take them into consideration. The formula is:

[0085] R resource =∑ i w i U i ;

[0086] Where R resource is the comprehensive resource utilization rate, w i is the weight of different resource utilization, U i is the utilization rate of the i-th resource, and the weight of the resource utilization factor is set to w resource ;

[0087] Cost saving factor: used to consider the cost saved during task execution, including equipment operating costs and labor costs. The formula is:

[0088]

[0089] Where R cost is the cost saving incentive factor, C saved To save costs, C budest is the budget cost, and the weight of this cost saving factor is set to w cost ;

[0090] Service quality factor: used to measure the accuracy and completeness of task completion, the formula is:

[0091] R service =A accuracy *A completeness ;

[0092] Where R service is the service quality reward factor, A accuracy is the accuracy of the service, A completeness is the service completeness rate, and the weight of this service quality factor is set to w service ;

[0093] Based on the above content, the reward function is constructed as follows:

[0094] R=w time R time +w resource R resource +w cost R cost +w service R service ;

[0095] Where R is the reward function.

[0096] The auxiliary scheduling decision module includes a digital twin warehouse construction module, a simulation scenario and strategy setting module, and a simulation evaluation and optimization module. The digital twin warehouse construction module is based on 3D modeling technology and IoT data to create a digital twin model that completely corresponds to the real warehouse. The digital twin model is used to restore the physical layout of the warehouse, including the location, size, and number of layers of the shelves, the running tracks and docking points of the handling equipment, and the storage location, shape, and volume of the goods. The digital twin model is also embedded with the logistics process, equipment operation logic, personnel operation specifications, and business rules in the warehouse to truly reflect the actual operating status of the warehouse. In the digital twin model, each shelf location has a corresponding number and storage capacity information, and the movement speed, turning radius, and load capacity of the forklift and AGV are also accurately set. The actual location and status of the goods in the warehouse can be updated based on real-time IoT data;

[0097] The simulation scenario and strategy setting module is used to set up multiple simulation scenarios in the digital twin model according to actual business needs and possible situations, including different order volumes: daily order volume, promotional order volume, peak season order volume; different cargo distribution: evenly distributed, concentrated in specific areas; equipment failure: forklift failure, conveyor belt failure; emergency order insertion; formulate multiple scheduling strategies for each scenario, including different task allocation rules, equipment scheduling plans, and inventory layout adjustment strategies. In the simulation scenario where the promotional order volume increases dramatically, one strategy is to prioritize the allocation of automated equipment to handle large-volume orders and manually handle small-volume high-value orders; another strategy is to assign tasks to equipment and personnel in the corresponding area based on the storage area of ​​the ordered goods;

[0098] The simulation evaluation and optimization module is used to simulate the operation of different scheduling strategies under each simulation scenario, collect and analyze relevant data, including the time of goods entering and leaving the warehouse, equipment utilization, inventory turnover rate, and task completion rate. By comparing the performance of different strategies in the same scenario, the advantages and disadvantages of each strategy are evaluated, and the strategies with better performance are selected for further optimization. The optimization process can adjust the parameters in the strategy, including the equipment allocation ratio and task priority weight, and then perform simulation evaluation again until the optimal scheduling strategy combination is found. For example, when evaluating the equipment utilization index, it is found that under a certain strategy, the forklift is idle in certain time periods, while the AGV is overloaded. In this case, the task allocation rules can be adjusted to appropriately increase the task volume of the forklift and reduce the burden on the AGV. Then, the adjusted strategy effect can be verified by re-simulating, and finally the optimal scheduling strategy under the simulation scenario is determined and applied to actual warehouse operations.

[0099] Embodiment 4:

[0100] like Figure 1As shown in the figure, an intelligent WMS system with intelligent scheduling function, the decision optimization module includes a digital twin model update module and a simulation scene generation and optimization module. The digital twin model update module is used to improve the synchronization with the physical warehouse and optimize the data fusion mechanism. It is used to merge with the computer vision system, based on which the real-time image data of the status and location of the goods and the real-time motion capture data of the employee's operation are obtained. At the same time, according to the dynamic degree of warehouse operation and the importance of data, the model update frequency is dynamically adjusted. For example, for areas with frequent goods in and out of the warehouse, the model is updated every 10 minutes; for areas with relatively stable storage, the model is updated every hour;

[0101] In the construction of digital twin models, more physical details and behavioral characteristics are added, including considering the impact of the physical characteristics of the goods on the equipment handling operation, simulating the energy consumption changes and mechanical wear of the equipment under different loads and operating speeds. With a more accurate model, the impact of the scheduling strategy on the equipment life and energy consumption can be more realistically evaluated in the simulation, thereby optimizing the scheduling decision;

[0102] The simulation scenario generation and optimization module is used to add more complex and rare scenarios, including but not limited to the impact of natural disasters on warehouses and supply chain disruption scenarios. It uses a combinatorial optimization method to perform combined simulations on different scenarios and study scheduling strategies when multiple adverse factors occur at the same time. For example, it simulates warehouse fires and disruptions to key transportation routes at the same time to observe how the scheduling system adjusts inventory allocation, equipment scheduling, and order processing strategies.

[0103] The execution control module is used to achieve high-speed, low-latency data transmission of scheduling instructions through the equipment control interface of communication technology and the personnel interaction terminal. For automated equipment such as automatic sorting machines and stackers, a dual-network redundant control architecture that integrates industrial Ethernet and 5G networks is adopted to ensure the stability and reliability of equipment operation. Once the main network fails, the backup network can switch seamlessly to ensure the continuous normal operation of the equipment and avoid production stagnation due to network interruptions.

[0104] It should be noted that the present invention is an intelligent WMS system with intelligent scheduling function. When in use, it is based on the integration of RFID antenna array, environmental perception module and intelligent visual recognition module to collect warehouse status data, process and analyze the collected data, monitor the data for abnormalities and predict the status, and carry out corresponding planning and processing based on the expanded e-commerce platform, supplier system and traffic management department data;

[0105] The residual sequence of the ARIMA model is input into the GARCH model to predict the average quantity of transportation demand over a period of time and the upper and lower limits of demand fluctuations. Based on this information, logistics companies can rationally arrange the transportation resources and inventory levels of warehouses and formulate strategies to deal with the risk of demand fluctuations.

[0106] When quality risks are predicted for goods, the image data of the goods is analyzed at the same time, including checking whether the packaging is damaged, whether there are any abnormal changes in the appearance of the goods, and video data during transportation, including monitoring the driving stability of the transport vehicle, whether there are severe bumps and collisions, as well as order information and logistics logs, including understanding the storage conditions and transportation time of the goods, based on which warehouse managers can take preventive measures in advance;

[0107] Each independent work task in the warehouse is defined as an intelligent agent, and multiple simulation scenarios are set up to formulate multiple scheduling strategies, including different task allocation rules, equipment scheduling plans, and inventory layout adjustment strategies. A reward mechanism is designed to select the optimal decision for the scheduling of the intelligent agent, so that the scheduling decision of the intelligent agent can be used in the optimal choice.

[0108] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. An intelligent WMS system with intelligent dispatching function, including a cargo sensing module, a data processing and analysis module, a dispatching decision module, a decision optimization module, and an execution control module, characterized in that: The data processing and analysis module predicts the warehouse status through data collection, processing and analysis, and formulates corresponding strategies according to the prediction results; The scheduling decision module includes a multi-agent scheduling module and an auxiliary scheduling decision module. The multi-agent scheduling module is used to define each independent operation task in the warehouse as an agent, model the warehouse environment, and set a multi-dimensional reward function for each agent to guide it to make the best decision; The auxiliary scheduling decision module truly reflects the actual status of the warehouse based on the constructed digital twin model, and sets up multiple simulation scenarios to formulate multiple scheduling strategies, including different task allocation rules, equipment scheduling plans, and inventory layout adjustment strategies.

2. The intelligent WMS system with intelligent scheduling function according to claim 1 is characterized in that: The cargo sensing module is based on the integration of RFID antenna array, environmental sensing module and intelligent visual recognition module. The RFID antenna array acquires basic cargo information based on the RFID tag arranged on the cargo packaging; the visual recognition module is used for detailed detection and image acquisition of the cargo's appearance, shape and packaging integrity; the environmental sensing module builds a comprehensive environmental monitoring system based on the arranged sensors.

3. The intelligent WMS system with intelligent scheduling function according to claim 1 is characterized in that: The data processing and analysis module is electrically connected to the cargo sensing module. The data processing and analysis module also includes an external data expansion and integration module, an abnormality monitoring module, and a predictive analysis module. The data expansion and integration module is used to establish a data interface with the e-commerce platform, the supplier system, and the traffic management department, and is used to obtain order information, customer evaluation, and sales trend forecast data from the e-commerce platform, so as to plan logistics resources in advance; Obtain raw material supply plans and delivery times from supplier systems to optimize inventory management; connect with traffic management departments to obtain real-time traffic information, road construction information, and traffic control information to provide a basis for transportation route planning; The anomaly monitoring module detects anomalies in logistics data based on convolutional neural networks and recurrent neural networks. Convolutional neural networks are used to process image data, including identifying damaged cargo packaging and damaged vehicle appearance. Recurrent neural networks are used to process time series order data and vehicle driving trajectory data, based on which abnormal fluctuations in order volume and deviations in vehicle driving routes are detected. The anomaly detection steps include the following: S301: Collect and pre-process the data obtained from the cargo sensing module, and mark normal data and abnormal data. The normal data and abnormal data are set based on the actual situation of vehicle driving data, cargo status data, inventory data, warehouse environment data, order data, and customer information data; S302: Feature extraction: extract image features from the data through a convolutional neural network, extract time series data from the data through a recurrent neural network, and perform training, feature learning, and optimization on them; S303: The pre-processed data obtained from the cargo sensing module is input into the convolutional neural network and the recurrent neural network according to its characteristics, and anomalies are detected based on this, and corresponding processing is performed according to the monitored abnormal data.

4. The intelligent WMS system with intelligent scheduling function according to claim 3 is characterized in that: The predictive analysis module includes a deep feature mining and adaptive selection module, an ARIMA model and GARCH model hybrid module, a dynamic adjustment module, a deep neural network model construction module, a prediction enhancement module, and a model evaluation and optimization module. The deep feature mining and adaptive selection module performs feature mining based on the autoencoder technology in deep learning. The autoencoder is used to learn the deep feature representation of data, compress the original high-dimensional data into a low-dimensional feature space, and then decode and restore the data. The feature extraction is optimized by minimizing the reconstruction error. At the same time, the feature selection is adaptively optimized based on the reinforcement learning algorithm. The reinforcement learning algorithm determines the selection of features and weight adjustment based on the performance feedback of the predictive analysis module, including the accuracy and recall rate; In the hybrid module of the ARIMA model and the GARCH model, the ARIMA model is an autoregressive integrated moving average model, and the GARCH model is a generalized autoregressive conditional heteroscedasticity model, which is used to integrate the ARIMA model with the GARCH model to address the common volatility and heteroscedasticity problems in logistics data, including the following steps: S401: Use the ARIMA model to model the mean part of the time series to capture the linear trend and seasonality of the data, including predicting the long-term demand trend of goods and the seasonal changes in transportation flow; S402: Input the residual sequence of the ARIMA model into the GARCH model. The GARCH model is used to model the variance of the residuals and effectively capture the volatility and heteroscedasticity characteristics of the data, including predicting the fluctuation range of cargo prices in a specific time period and the sudden change amplitude of transportation demand. The hybrid module of the ARIMA model and the GARCH model is used to provide more comprehensive time series forecasting results, including the forecast of the data mean, the confidence interval of the forecast and the risk assessment; The dynamic adjustment module automatically adjusts based on the data fed back by the cargo sensing module and the hybrid module of the ARIMA model and the GARCH model. Based on the sliding window technology, it monitors the difference between the newly-inflowed data and the time series prediction results in real time. When the difference exceeds the set threshold, the system automatically triggers the model parameter optimization process, and uses the particle swarm optimization algorithm and the genetic algorithm to re-optimize the parameters of the ARIMA model and the GARCH model. In the optimization process, the objective function is to minimize the prediction error, while considering the complexity and stability of the model. The deep neural network model building module is used to build a deep neural network model that can integrate multiple logistics data modalities. The logistics data includes structured data: order information, inventory data; semi-structured data: cargo description, logistics log; unstructured data: cargo images, transport vehicle monitoring videos. The convolutional neural network is used to process image and video data to extract the appearance features of the cargo and the operating status features of the transport vehicle; the recurrent neural network is used to process structured and semi-structured sequence data, including the time series of orders and the change history of inventory; the features of different modalities are integrated through the fully connected layer, and the prediction task is performed on the integrated feature space; The prediction enhancement module is based on a generative adversarial network to solve the common small sample problem in logistics data, including goods and businesses with less historical data. The generative adversarial network consists of a generator and a discriminator. The generator is used to generate new simulated sample data based on limited real sample data. The simulated sample data is similar to the real sample in data distribution. The model evaluation and optimization module performs model optimization based on the NSGA-II algorithm, while considering multiple conflicting evaluation indicators, including prediction accuracy, model complexity, and computing resource consumption. It is used to perform non-dominated sorting and crowding calculation on individuals in the model parameter combination population according to multiple evaluation indicators, and select the optimal model parameter combination.

5. The intelligent WMS system with intelligent scheduling function according to claim 1 is characterized in that: The scheduling decision module is electrically connected to the data processing and analysis module. The scheduling decision module includes a multi-agent scheduling module and an auxiliary scheduling decision module. The multi-agent scheduling module includes an agent and environment modeling module, a reward mechanism design module, and a learning and decision module. The agent and environment modeling module is used to define each independent operation task in the warehouse as an agent, including but not limited to order picking tasks, goods shelving tasks, and replenishment tasks. Each agent has its own status information, including task priority, current execution progress, required resource status, and association with other tasks. The agent and environment modeling module also includes warehouse resources and environmental status. The warehouse resources include but are not limited to various types of handling equipment, idle storage locations, and human resources. The warehouse resources constitute the environmental elements in which the agent is located. The environmental status is described by the real-time availability, capacity, location distribution, and environmental conditions of the warehouse resources.

6. The intelligent WMS system with intelligent scheduling function according to claim 5 is characterized in that: The reward mechanism design module sets a multi-dimensional reward function for each agent to guide it to make the best decision. The reward function comprehensively considers task completion time, resource utilization, cost savings and service quality. If the agent can complete the task in a short time and use resources efficiently, a higher positive reward will be given; if the task is delayed and causes resource waste, a negative reward will be given; The learning and decision-making module learns based on the continuous interaction between the agent and the environment. At each decision-making moment, the agent selects an action from the action space based on its current state and the state of the environment using the proximal policy optimization algorithm. The action includes choosing which device to use to perform the task, determining the path and sequence of task execution, and deciding the specific location for cargo storage.

7. The intelligent WMS system with intelligent scheduling function according to claim 6 is characterized in that: The reward mechanism design module specifically includes: Determine the reward factors and weights, including the task completion time factor: used to calculate the ratio of the time taken from the start to the completion of the task to the expected standard time. The formula is: Where R time is the task completion time reward factor, T actual is the actual completion time of the task, T standard is the expected standard time of the task. When the task is completed ahead of schedule, R time If the task is completed late, R time Less than 1, the task completion time factor weight is set to w time ; Resource utilization factor: used to calculate the utilization of different resources and take them into consideration. The formula is: R resource =∑ i w i U i ; Where R resource is the comprehensive resource utilization rate, w i is the weight of different resource utilization, U i is the utilization rate of the i-th resource, and the weight of the resource utilization factor is set to w resource ; Cost saving factor: used to consider the cost saved during task execution, including equipment operating costs and labor costs. The formula is: Where R cost is the cost saving incentive factor, C saved To save costs, C budest is the budget cost, and the weight of this cost saving factor is set to w cost ; Service quality factor: used to measure the accuracy and completeness of task completion, the formula is: R service =A accuracy *A completeness ; Where R service is the service quality reward factor, A accuracy is the accuracy of the service, A completeness is the service completeness rate, and the weight of this service quality factor is set to w service ; Based on the above content, the reward function is constructed as follows: R=w time R time +w resource R resource +w cost R cost +w service R service ; Where R is the reward function.

8. The intelligent WMS system with intelligent scheduling function according to claim 5 is characterized in that: The auxiliary scheduling decision module includes a digital twin warehouse construction module, a simulation scenario and strategy setting module, and a simulation evaluation and optimization module. The digital twin warehouse construction module creates a digital twin model that completely corresponds to the real warehouse based on 3D modeling technology and Internet of Things data. The digital twin model is used to restore the physical layout of the warehouse, including the location, size, and number of layers of the shelves, the running tracks and docking points of the handling equipment, and the storage location, shape, and volume of the goods. The digital twin model is also embedded with the logistics process, equipment operation logic, personnel operation specifications, and business rules in the warehouse to truly reflect the actual operation status of the warehouse; The simulation scenario and strategy setting module is used to set up multiple simulation scenarios in the digital twin model according to actual business needs and possible situations, including different order volumes: daily order volume, promotional activity order volume, peak season order volume; different cargo distribution: evenly distributed, concentrated in a specific area; equipment failure: forklift failure, conveyor belt failure; emergency order insertion; formulate multiple scheduling strategies for each scenario, including different task allocation rules, equipment scheduling plans, and inventory layout adjustment strategies; The simulation evaluation and optimization module is used to simulate the operation of different scheduling strategies in each simulation scenario, collect and analyze relevant data, including the time of goods entering and leaving the warehouse, equipment utilization, inventory turnover rate, and task completion rate. By comparing the performance of different strategies in the same scenario, the advantages and disadvantages of each strategy are evaluated, and the strategies with better performance are selected for further optimization. The optimization process can adjust the parameters in the strategy, including the equipment allocation ratio and task priority weight, and then perform simulation evaluation again until the optimal scheduling strategy combination is found.

9. The intelligent WMS system with intelligent scheduling function according to claim 1 is characterized in that: The decision optimization module includes a digital twin model update module and a simulation scenario generation and optimization module. The digital twin model update module is used to improve synchronization with the physical warehouse and optimize the data fusion mechanism, and is used to fuse with the computer vision system to obtain real-time image data of the status and location of goods and real-time motion capture data of employee operations. At the same time, the model update frequency is dynamically adjusted according to the dynamic degree of warehouse operation and the importance of data; In the construction of the digital twin model, more physical details and behavioral characteristics are added, including considering the impact of the physical characteristics of the goods on the equipment handling operation, and simulating the energy consumption changes and mechanical wear of the equipment under different loads and operating speeds; The simulation scenario generation and optimization module is used to add more complex and rare scenarios, including but not limited to the impact of natural disasters on warehouses and supply chain disruption scenarios. A combinatorial optimization method is used to perform combined simulations on different scenarios to study scheduling strategies when multiple adverse factors occur simultaneously.

10. The intelligent WMS system with intelligent scheduling function according to claim 1, characterized in that: The execution control module is used to realize high-speed and low-latency data transmission of scheduling instructions through the device control interface of the communication technology and the personnel interaction terminal.

Images