Data-driven based cold storage automated storage and sorting system and method
By using a data-driven approach, a predictive model is trained using historical data from cold storage, and the A algorithm is used to optimize cold storage and sorting. This solves the problem of low efficiency in traditional methods and achieves efficient and accurate sorting and storage strategies, which are suitable for high-timeliness and high-dynamic scenarios in automated cold storage.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI BAOYE GRP CORP
- Filing Date
- 2026-03-09
- Publication Date
- 2026-06-09
AI Technical Summary
Traditional cold storage strategies and sorting methods fail to fully consider the dynamic characteristics of goods and environmental changes, resulting in low sorting efficiency, unreasonable route planning, and inability to meet the demand for small-batch, high-time-efficiency orders.
By collecting historical operation data from cold storage, performing feature engineering, training prediction models, generating sorting task execution sequences and movement paths, and combining real-time environmental data to optimize storage location and path planning, the A algorithm is used to avoid obstacles and collaborative equipment.
It achieves synergistic optimization of cold storage and sorting, improves sorting efficiency and accuracy, adapts to dynamic changes, avoids path conflicts, and enhances the utilization rate and economic benefits of cold storage.
Smart Images
Figure CN122175508A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of cold storage sorting technology, specifically to a data-driven automated cold storage and sorting system and method. Background Technology
[0002] Cold storage facilities are a core link in the storage and distribution of temperature-sensitive goods such as fresh agricultural products, frozen foods, and biopharmaceuticals. Their sorting efficiency directly impacts supply chain response speed and operating costs. With the rapid development of e-commerce and new retail models, cold storage orders are characterized by "small batches, multiple shipments, and high timeliness." Traditional sorting methods relying on manual experience or fixed rules are no longer sufficient to meet these demands. Methods such as first-in-first-out (FIFO) and category-based sorting face the following technical bottlenecks:
[0003] 1. Traditional cold storage strategies are mostly based on static rules, such as fixed partitioning by product category, without fully considering dynamic characteristics such as sorting frequency, storage duration, and order association.
[0004] 2. The execution order of sorting tasks directly affects order fulfillment efficiency. However, traditional methods often use static rules such as "first-come, first-served" or "sorted by order amount," without considering dynamic factors such as the urgency of the goods (e.g., the time requirements of fresh produce orders) and the aggregation of related orders (e.g., multiple orders from the same customer need to be sorted together). For example, a fresh produce order requires delivery within 30 minutes, but because it is not given a high priority, sorting is delayed; or the correlation between the goods and other orders is not considered, resulting in wasted duplicate paths.
[0005] 3. The environment inside cold storage is complex. Traditional path planning algorithms (such as Dijkstra's algorithm) are mostly based on static maps and do not update environmental data in real time, which leads to conflicts between the planned path and the actual environment. For example, the path may be blocked by obstacles, or the path congestion caused by the collaboration of multiple devices may not be considered, which increases sorting time. For example, when a sorting device moves according to a static path, if another device temporarily occupies the channel, the path needs to be replanned, which leads to a decrease in efficiency.
[0006] Among existing patented technologies, such as patent CN117132186A, "A Cold Storage and Sorting Method Based on Cold Storage Automation," an invention discloses a cold storage and sorting method based on cold storage automation, relating to the field of cold chain transportation and storage. This method includes the following steps: S1, obtaining parameters of cold storage transport vehicles and refrigerated products, and calculating refrigerated transport time and refrigerated product priority; S2, calculating initial refrigerated transport loss parameters based on refrigerated transport time and refrigerated product priority; S3, planning the original cold storage transfer route based on refrigerated transport time and refrigerated product priority, and generating an initial emergency adjustment plan and an initial equipment allocation plan based on the cold storage transfer route planning results. The invention incorporates refrigerated transport time and refrigerated product priority into the cold storage transportation... In the operational planning of warehousing systems, the goal is to reduce losses of refrigerated products during transportation and storage, and to effectively allocate resources and reduce unnecessary waiting and delays through detailed planning of priorities and transportation times. For example, patent CN112623588B, "An Automated Artificial Intelligence Cold Storage and Sorting Method for Cold Storage," relates to the field of cold chain transportation and storage technology and includes the following steps: S1, management of vehicle and product information in transit; S2, scanning and registration of products upon arrival at the warehouse; S3, traceability of the intelligent cold storage system; S4, product entry into the warehouse; S5, setting the cold storage safety mode after entry into the warehouse; S6, ethylene discharge during the storage of fruits and vegetables; S7, control of the refrigeration room system; S8, automation of initial processing for production entry and exit; and S9, diversity of cold storage equipment management systems.The system allows for switching between automated and manual operation within the cold storage facility. When the business is simple, automated equipment can be used; when the business is complex, a tiered system of manual and automated operations can be implemented, significantly improving the flexibility of equipment application scenarios and on-site business conditions, increasing operational efficiency, and reducing manual operating costs. For example, patent CN111842203A, "A Centralized Sorting, Classification, and Transfer System for Cold Storage," relates to the field of cold storage technology, specifically a centralized sorting, classification, and transfer system for cold storage. It includes a sorting and conveying unloading component and a tracking and classifying transfer component. The sorting and conveying unloading component includes a sorting conveyor belt, a conveyor belt mounting frame, a weighing mechanism, a detection mechanism, and a classification and unloading mechanism. Each tracking and classifying transfer component includes a tracking trolley, an X-shaped lifting mechanism, a deflecting and tilting mechanism, a transfer box, and a box wall opening and closing mechanism. The top two sides of the conveyor belt mounting frame are vertically equipped with a first baffle and a second baffle. This invention provides a centralized sorting, classification, and transfer system for cold storage. The system can efficiently and automatically classify and transfer inbound products, reducing manual labor intensity, improving work efficiency, and reducing safety hazards for personnel. Products of different weights are pushed onto tracking carts by a classification and unloading mechanism, and then delivered to the corresponding positions in the cold storage, achieving efficient classification and storage. The above patents are all related to cold storage and sorting, and most of them are limited to automated cold storage and sorting to replace manual operations and static optimization of cold storage and sorting processes. They have not truly achieved optimization and zoning of cold storage storage and paths to achieve efficient utilization of cold storage. To this end, the applicant proposes a data-driven automated cold storage and sorting system and method based on actual needs, which optimizes cold storage efficiency and utilization. It can avoid temporary obstacles in real time, adjust the paths that need to be detoured due to temperature sensitivity, and avoid path conflicts through multi-device location coordination, achieving a comprehensive improvement in sorting efficiency, accuracy, and adaptability, and has significant technical and economic value. Summary of the Invention
[0007] To address the aforementioned technical problems, this invention proposes a data-driven automated cold storage and sorting system and method. The system collects historical operation data from the cold storage through a data acquisition module. This historical data includes historical storage location data, historical sorting task data, historical order data, cargo attribute data, and operation time data. A processing module performs feature engineering on the historical operation data to obtain multiple feature vectors. These feature vectors characterize the storage characteristics, sorting frequency characteristics, and order association characteristics of the cargo. Operational parameters and raw data are stored in a storage module. A training module uses these feature vectors to train a prediction model. This model predicts the recommended storage location for cargo and the execution priority of sorting tasks. The prediction model is trained with the goal of minimizing the deviation between the predicted and actual storage locations and maximizing the efficiency of sorting task execution. The prediction module receives real-time sorting tasks. The real-time sorting task request includes the identification information of the goods to be sorted. Based on the identification information, corresponding goods attribute data is extracted from historical operation data and input into the trained prediction model to obtain the storage location recommendation degree of the goods to be sorted and the execution priority of the sorting task. The sorting generation module determines the target storage location of the goods to be sorted based on the storage location recommendation degree and generates a sorting task execution sequence according to the execution priority of the sorting task. The path generation module generates the movement path of the sorting equipment based on the sorting task execution sequence and real-time environmental data in the cold storage through a path planning algorithm. The real-time environmental data includes the temperature distribution in the cold storage, the location of obstacles, and the current position of other sorting equipment. When the guidance module receives the output movement path and sorting task execution sequence, it guides the sorting equipment to perform the sorting operation, and finally completes the automated storage and sorting of the cold storage.
[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0009] A data-driven automated cold storage and sorting system is disclosed. The system comprises a data acquisition module, a processing module, a training module, a prediction module, a sorting generation module, a path generation module, and a guidance module. The data acquisition module collects historical operation data from the cold storage, including historical storage location data, historical sorting task data, historical order data, cargo attribute data, and operation time data. The processing module performs feature engineering on the historical operation data to obtain multiple feature vectors. These feature vectors characterize the storage characteristics, sorting frequency characteristics, and order association characteristics of the cargo. A storage module is included within the processing module. The training module uses the feature vectors to train a prediction model. This model predicts the recommended storage location for cargo and the execution priority of sorting tasks. The prediction model aims to minimize the deviation between the predicted and actual storage locations and maximize the efficiency of sorting task execution. The target is trained; the prediction module receives real-time sorting task requests, which include the identification information of the goods to be sorted. Based on the identification information of the goods to be sorted, it extracts the corresponding goods attribute data from historical operation data and inputs the goods attribute data into the trained prediction model to obtain the storage location recommendation degree of the goods to be sorted and the execution priority of the sorting task; the sorting generation module determines the target storage location of the goods to be sorted based on the storage location recommendation degree and generates a sorting task execution sequence according to the execution priority of the sorting task; the path generation module generates the movement path of the sorting equipment based on the sorting task execution sequence and the real-time environmental data in the cold storage through a path planning algorithm. The real-time environmental data includes the temperature distribution in the cold storage, the location of obstacles, and the current position of other sorting equipment; the guidance module guides the sorting equipment to perform sorting operations when it receives the output movement path and the sorting task execution sequence.
[0010] Furthermore, the steps of the data-driven automated cold storage and sorting method are as follows:
[0011] Step 1: The data acquisition module collects historical operation data of the cold storage. The historical operation data includes historical storage location data of goods, historical sorting task data, historical order data, goods attribute data, and operation time data.
[0012] Step 2: The processing module performs feature engineering based on historical operation data to obtain multiple feature vectors. These feature vectors are used to characterize the storage characteristics, sorting frequency characteristics, and order association characteristics of the goods. The operation parameter data and raw data are stored in the storage module.
[0013] Step 3: The training module uses feature vectors to train a prediction model. The prediction model is used to predict the recommended storage location of goods and the execution priority of sorting tasks. The prediction model is trained with the goal of minimizing the deviation between the predicted location and the actual storage location and maximizing the execution efficiency of sorting tasks.
[0014] Step 4: The prediction module receives real-time sorting task requests. The real-time sorting task requests contain the identification information of the goods to be sorted. Based on the identification information of the goods to be sorted, the module extracts the corresponding goods attribute data from the historical operation data and inputs the goods attribute data into the trained prediction model to obtain the recommended storage location of the goods to be sorted and the execution priority of the sorting task.
[0015] Step 5: The sorting generation module determines the target storage location of the goods to be sorted based on the storage location recommendation degree, and generates the sorting task execution sequence according to the execution priority of the sorting task;
[0016] Step 6: The path generation module generates the movement path of the sorting equipment based on the sorting task execution sequence and real-time environmental data in the cold storage through a path planning algorithm. The real-time environmental data includes the temperature distribution in the cold storage, the location of obstacles, and the current position of other sorting equipment.
[0017] Step 7: When the guidance module receives the output movement path and sorting task execution sequence, it guides the sorting equipment to perform sorting operations, ultimately completing the automated storage and sorting of the cold storage.
[0018] Furthermore, in the data-driven automated cold storage and sorting method, the data acquisition module collects historical cold storage data specifically as follows:
[0019] 1) Collect storage records, sorting records, and order records of goods in the cold storage according to the preset time window;
[0020] 2) Perform data cleaning on storage records, sorting records, and order records to remove outliers and missing values;
[0021] 3) The cleaned data is associated with cargo identifiers, timestamps, and operation types to obtain structured historical operation data.
[0022] Furthermore, in the data-driven automated cold storage and sorting method, the processing module performs feature engineering on historical operation data to obtain multiple feature vectors, specifically as follows:
[0023] 1) Perform one-hot encoding or label encoding on categorical features in historical operational data, and normalize or standardize numerical features;
[0024] 2) Extract time-series features from historical operational data. These features include storage duration, sorting interval, and order frequency. The formulas for calculating storage duration and sorting interval are:
[0025]
[0026] in: Indicates storage duration;
[0027] Indicates the time when goods are received into the warehouse;
[0028] Indicates the time the goods left the warehouse;
[0029]
[0030] in: Indicates the sorting interval time;
[0031] Indicates the current sorting task time;
[0032] Indicates the time of the last sorting task for the same type of goods;
[0033] 3) Based on the attribute data of goods, generate goods association features, which are used to characterize the correlation between goods; combine the processed categorical features, numerical features, time series features and goods association features to obtain multiple feature vectors.
[0034] Furthermore, in the data-driven automated cold storage and sorting method, the training module uses feature vectors to train the prediction model specifically as follows:
[0035] 1) Divide the feature vectors into training set, validation set, and test set;
[0036] 2) Construct the initial prediction model. The initial prediction model is a multi-task learning model based on deep learning, including a shared layer, a storage location prediction branch, and a sorting priority prediction branch.
[0037] 3) Train the initial prediction model using the training set, and adjust the model parameters using the backpropagation algorithm to minimize the cross-entropy loss of the prediction branch at the storage location. and the mean square error loss of sorting priority prediction branches The total loss function L_total is:
[0038]
[0039] Where: α is the cross-entropy loss;
[0040] β is the weighting coefficient for the mean squared error loss, used to balance the importance of the two tasks;
[0041] 4) Validate the trained model using the validation set, and adjust the model hyperparameters based on the validation results until the model performance meets the preset conditions;
[0042] 5) Test the adjusted model using the test set to obtain the final prediction model.
[0043] Furthermore, in the data-driven automated cold storage and sorting method, the prediction module extracts corresponding cargo attribute data from historical operation data based on the identification information of the goods to be sorted, and inputs the cargo attribute data into the trained prediction model to obtain the recommended storage location of the goods to be sorted and the execution priority of the sorting task. Specifically:
[0044] 1) Based on the identification information of the goods to be sorted, obtain the attribute data of the goods to be sorted from the cold storage's goods information database. The attribute data includes the category, size, weight, shelf life, and storage temperature requirements of the goods.
[0045] 2) Convert the attribute data into feature vectors that match the input format of the prediction model;
[0046] 3) Input the feature vector into the trained prediction model to obtain the storage location recommendation degree and the sorting task execution priority. The storage location recommendation degree is the recommendation probability of the goods to be sorted in each available storage location, and the sorting task execution priority is the numerical value of the execution order of the goods to be sorted relative to other goods to be sorted.
[0047] Furthermore, in the data-driven automated cold storage and sorting method, the sorting generation module generates the sorting task execution sequence as follows:
[0048] 1) Select the storage location with the highest recommendation from the storage location recommendations as the target storage location for the goods to be sorted;
[0049] 2) Sort multiple goods to be sorted in descending order of sorting task execution priority to generate a sorting task execution sequence.
[0050] Furthermore, in the data-driven automated cold storage and sorting method, the path generation module generates the movement path of the sorting equipment through a path planning algorithm as follows:
[0051] 1) Acquire real-time environmental data inside the cold storage. The real-time environmental data is collected in real time by sensors installed inside the cold storage.
[0052] 2) Determine the sequence of storage locations that the sorting equipment needs to access based on the sorting task execution sequence;
[0053] 3) Based on the storage location sequence and the real-time environmental data, Algorithm A is used for path planning to generate the movement path of the sorting equipment. The movement path satisfies the constraints of shortest path and / or minimum time consumption. The cost function of Algorithm A is... Defined as:
[0054]
[0055] Where: n represents the node in the current path;
[0056] This represents the actual cost from the starting point to node n;
[0057] This represents the estimated cost from node n to the destination.
[0058] Furthermore, in the data-driven automated cold storage and sorting method, the guidance module guides the sorting equipment to perform sorting operations as follows:
[0059] 1) Receive actual execution data fed back by the sorting equipment. The actual execution data includes the actual movement path, actual sorting time, and actual storage location.
[0060] 2) Compare the actual execution data with the movement path, sorting task execution sequence, and target storage location to calculate the execution deviation, which includes the path deviation rate and the sorting time deviation rate; wherein:
[0061] Path deviation rate The calculation formula is:
[0062]
[0063] in: Indicates the actual length of the movement path;
[0064] Indicates the length of the planned movement path;
[0065] Sorting time deviation rate The general calculation formula is:
[0066]
[0067] in: Indicates the actual sorting time;
[0068] Indicates the planned sorting time;
[0069] 3) Adjust and optimize the prediction model and / or path planning algorithm based on the execution deviation.
[0070] Furthermore, in the data-driven automated cold storage and sorting method, the path generation module uses the A algorithm for path planning, specifically generating the movement path of the sorting equipment as follows:
[0071] 1) Based on real-time environmental data, construct an environmental map of the cold storage, which includes obstacle information and passable area information;
[0072] 2) Use each storage location in the storage location sequence as the target point for path planning;
[0073] 3) Starting from the current position of the sorting equipment, and taking each storage location as the target point in turn, the A algorithm is used to plan the path and generate the path segment from the current position to the next target point.
[0074] 4) Connect all path segments to obtain the movement path of the sorting equipment.
[0075] The benefits of this application are:
[0076] 1. A data-driven automated cold storage and sorting method trains a prediction model to simultaneously output the recommended storage location of goods and the execution priority of sorting tasks, thereby achieving coordinated optimization of storage strategy and sorting sequence.
[0077] 2. A data-driven automated cold storage and sorting method solves the problem of traditional methods that handle storage and sorting independently. The method optimizes both through a multi-task learning model, which can comprehensively consider dynamic factors such as order timeliness and cargo correlation, and introduces real-time environmental data in the cold storage during the path planning stage.
[0078] 3. The data-driven automated cold storage and sorting method combines the A algorithm to dynamically generate movement paths. Compared with traditional path planning based on static maps, such as Dijkstra's algorithm, this method can avoid temporary obstacles in real time, adjust the detour path required for temperature-sensitive goods, and avoid path conflicts through multi-device location coordination.
[0079] 4. The data-driven automated cold storage and sorting method has achieved a comprehensive improvement in sorting efficiency, accuracy and adaptability. It is especially suitable for the high-timeliness and high-dynamic operation scenarios of automated cold storage, and has significant technical and economic value. Attached Figure Description
[0080] Figure 1 This is a schematic diagram of the method flow of the present invention;
[0081] Figure 2 This is a schematic diagram illustrating the process of collecting historical operation data of cold storage according to the present invention;
[0082] Figure 3This is a schematic diagram of the feature vector generation process of the present invention;
[0083] Figure 4 This is a schematic diagram illustrating the process of training a prediction model using feature vectors according to the present invention.
[0084] Figure 5 This is a schematic diagram illustrating the process of generating a sorting task execution sequence for this invention. Detailed Implementation
[0085] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:
[0086] like Figure 1-5 The diagram illustrates a data-driven automated cold storage and sorting system and method. The data-driven automated cold storage and sorting method includes a data acquisition module, a processing module, a training module, a prediction module, a sorting generation module, a path generation module, and a guidance module. The data acquisition module collects historical operation data from the cold storage, including historical storage location data, historical sorting task data, historical order data, cargo attribute data, and operation time data. The processing module performs feature engineering on the historical operation data to obtain multiple feature vectors. These feature vectors represent the storage characteristics, sorting frequency characteristics, and order association characteristics of the cargo. A storage module is included within the processing module. The training module uses the feature vectors to train a prediction model. The prediction model predicts the recommended storage location of the cargo and the execution priority of the sorting task. The prediction model minimizes the deviation between the predicted location and the actual storage location and maximizes the priority of the sorting task. The system is trained with efficiency as the objective. The prediction module receives real-time sorting task requests, which include the identification information of the goods to be sorted. Based on this information, it extracts corresponding goods attribute data from historical operation data and inputs this data into the trained prediction model to obtain the recommended storage location for the goods and the execution priority of the sorting task. The sorting generation module determines the target storage location for the goods based on the recommended storage location and generates a sorting task execution sequence based on the execution priority. The path generation module generates the movement path of the sorting equipment using a path planning algorithm based on the sorting task execution sequence and real-time environmental data within the cold storage. Real-time environmental data includes the temperature distribution within the cold storage, obstacle locations, and the current positions of other sorting equipment. The guidance module, upon receiving the output movement path and sorting task execution sequence, guides the sorting equipment to perform the sorting operation.
[0087] like Figure 1 As shown, the steps of the data-driven automated cold storage and sorting method are as follows:
[0088] Step 1: The data acquisition module collects historical operation data of the cold storage. The historical operation data includes historical storage location data of goods, historical sorting task data, historical order data, goods attribute data, and operation time data.
[0089] Step 2: The processing module performs feature engineering based on historical operation data to obtain multiple feature vectors. These feature vectors are used to characterize the storage characteristics, sorting frequency characteristics, and order association characteristics of the goods. The operation parameter data and raw data are stored in the storage module.
[0090] Step 3: The training module uses feature vectors to train a prediction model. The prediction model is used to predict the recommended storage location of goods and the execution priority of sorting tasks. The prediction model is trained with the goal of minimizing the deviation between the predicted location and the actual storage location and maximizing the execution efficiency of sorting tasks.
[0091] Step 4: The prediction module receives real-time sorting task requests. The real-time sorting task requests contain the identification information of the goods to be sorted. Based on the identification information of the goods to be sorted, the module extracts the corresponding goods attribute data from the historical operation data and inputs the goods attribute data into the trained prediction model to obtain the recommended storage location of the goods to be sorted and the execution priority of the sorting task.
[0092] Step 5: The sorting generation module determines the target storage location of the goods to be sorted based on the storage location recommendation degree, and generates the sorting task execution sequence according to the execution priority of the sorting task;
[0093] Step 6: The path generation module generates the movement path of the sorting equipment based on the sorting task execution sequence and real-time environmental data in the cold storage through a path planning algorithm. The real-time environmental data includes the temperature distribution in the cold storage, the location of obstacles, and the current position of other sorting equipment.
[0094] Step 7: When the guidance module receives the output movement path and sorting task execution sequence, it guides the sorting equipment to perform sorting operations, ultimately completing the automated storage and sorting of the cold storage.
[0095] like Figure 2 As shown, the data acquisition module in the data-driven automated cold storage and sorting method specifically collects historical cold storage data as follows:
[0096] 1) Collect storage records, sorting records, and order records of goods in the cold storage according to the preset time window;
[0097] 2) Perform data cleaning on storage records, sorting records, and order records to remove outliers and missing values;
[0098] 3) The cleaned data is associated with cargo identifiers, timestamps, and operation types to obtain structured historical operation data.
[0099] like Figure 3 As shown, in the data-driven automated cold storage and sorting method, the processing module performs feature engineering on historical operation data to obtain multiple feature vectors. The specific process is as follows:
[0100] 1) Perform one-hot encoding or label encoding on categorical features in historical operational data, and normalize or standardize numerical features;
[0101] 2) Extract time-series features from historical operational data. These features include storage duration, sorting interval, and order frequency. The formulas for calculating storage duration and sorting interval are:
[0102]
[0103] in: Indicates storage duration;
[0104] Indicates the time when goods are received into the warehouse;
[0105] Indicates the time the goods left the warehouse;
[0106]
[0107] in: Indicates the sorting interval time;
[0108] Indicates the current sorting task time;
[0109] Indicates the time of the last sorting task for the same type of goods;
[0110] 3) Based on the attribute data of goods, generate goods association features, which are used to characterize the correlation between goods; combine the processed categorical features, numerical features, time series features and goods association features to obtain multiple feature vectors.
[0111] like Figure 4 As shown, in the data-driven automated cold storage and sorting method, the training module uses feature vectors to train the prediction model as follows:
[0112] 1) Divide the feature vectors into training set, validation set, and test set;
[0113] 2) Construct the initial prediction model. The initial prediction model is a multi-task learning model based on deep learning, including a shared layer, a storage location prediction branch, and a sorting priority prediction branch.
[0114] 3) Train the initial prediction model shown using the training set, and adjust the model parameters using the backpropagation algorithm to minimize the cross-entropy loss of the prediction branch at the storage location. and the mean square error loss of sorting priority prediction branches The total loss function L_total is:
[0115]
[0116] Where: α is the cross-entropy loss;
[0117] β is the weighting coefficient for the mean squared error loss, used to balance the importance of the two tasks;
[0118] 4) Validate the trained model using the validation set, and adjust the model hyperparameters based on the validation results until the model performance meets the preset conditions;
[0119] 5) Test the adjusted model using the test set to obtain the final prediction model.
[0120] like Figure 5 As shown, in the data-driven automated cold storage and sorting method, the prediction module extracts corresponding cargo attribute data from historical operation data based on the identification information of the goods to be sorted, and inputs the cargo attribute data into the trained prediction model to obtain the recommended storage location of the goods to be sorted and the execution priority of the sorting task. Specifically:
[0121] 1) Based on the identification information of the goods to be sorted, obtain the attribute data of the goods to be sorted from the cold storage's goods information database. The attribute data includes the category, size, weight, shelf life, and storage temperature requirements of the goods.
[0122] 2) Convert the attribute data into feature vectors that match the input format of the prediction model;
[0123] 3) Input the feature vector into the trained prediction model to obtain the storage location recommendation degree and the sorting task execution priority. The storage location recommendation degree is the recommendation probability of the goods to be sorted in each available storage location, and the sorting task execution priority is the numerical value of the execution order of the goods to be sorted relative to other goods to be sorted.
[0124] The sorting generation module in the data-driven automated cold storage and sorting method shown in the figure generates the sorting task execution sequence as follows:
[0125] 1) Select the storage location with the highest recommendation from the storage location recommendations as the target storage location for the goods to be sorted.
[0126] 2) Sort multiple goods to be sorted in descending order of sorting task execution priority to generate a sorting task execution sequence.
[0127] The path generation module in the data-driven automated cold storage and sorting method shown generates the movement path of the sorting equipment through a path planning algorithm, specifically as follows:
[0128] 1) Acquire real-time environmental data inside the cold storage. The real-time environmental data is collected in real time by sensors installed inside the cold storage.
[0129] 2) Determine the sequence of storage locations that the sorting equipment needs to access based on the sorting task execution sequence;
[0130] 3) Based on the storage location sequence and the real-time environmental data shown, Algorithm A is used for path planning to generate the movement path of the sorting equipment. The movement path satisfies the constraints of shortest path and / or minimum time consumption. The cost function of Algorithm A is... Defined as:
[0131]
[0132] Where: n represents the node in the current path;
[0133] This represents the actual cost from the starting point to node n;
[0134] This represents the estimated cost from node n to the destination.
[0135] The data-driven automated cold storage and sorting method shown employs the A algorithm for path planning in its path generation module, generating the specific movement path for the sorting equipment as follows:
[0136] 1) Based on real-time environmental data, construct an environmental map of the cold storage, which includes obstacle information and passable area information;
[0137] 2) Use each storage location in the storage location sequence as the target point for path planning;
[0138] 3) Starting from the current position of the sorting equipment, and taking each storage location as the target point in turn, the A algorithm is used to plan the path and generate the path segment from the current position to the next target point.
[0139] 4) Connect all path segments to obtain the movement path of the sorting equipment.
[0140] The guidance module in the data-driven automated cold storage and sorting method shown guides the sorting equipment to perform sorting operations as follows:
[0141] 1) Receive actual execution data fed back by the sorting equipment. The actual execution data includes the actual movement path, actual sorting time, and actual storage location.
[0142] 2) Compare the actual execution data with the shown movement path, sorting task execution sequence, and target storage location to calculate the execution deviation, which includes the path deviation rate and the sorting time deviation rate; where:
[0143] Path deviation rate The calculation formula is:
[0144]
[0145] in: Indicates the actual length of the movement path;
[0146] Indicates the length of the planned movement path;
[0147] Sorting time deviation rate The general calculation formula is:
[0148]
[0149] in: Indicates the actual sorting time;
[0150] Indicates the planned sorting time;
[0151] 3) Adjust and optimize the prediction model and / or path planning algorithm based on the execution deviation.
[0152] The data-driven automated cold storage and sorting method described herein extracts multi-dimensional feature vectors, including storage characteristics, sorting frequency characteristics, and order association characteristics, by collecting historical operational data and performing feature engineering. Compared to traditional static rule-based storage strategies, this method dynamically identifies key categories such as high-frequency sorting goods and easily delayed goods by quantifying time-series features, and recommends optimal storage locations based on goods attributes. By training a prediction model, it simultaneously outputs the recommended storage location degree and sorting task execution priority, achieving collaborative optimization of storage strategy and sorting order. Traditional methods often handle storage and sorting independently, while this method jointly optimizes both through a multi-task learning model, comprehensively considering dynamic factors such as order timeliness (e.g., the urgency of fresh food orders) and goods association (e.g., multiple orders from the same customer need to be combined for sorting). For example, goods in urgent orders can be assigned high priority and simultaneously recommended to easily accessible storage locations, avoiding priority invalidation due to remote storage locations. In the path planning stage, real-time environmental data within the cold storage is introduced, combined with the cost function f(n) of the A algorithm. =g(n)+h(n) dynamically generates movement paths. Compared with traditional path planning based on static maps, such as Dijkstra's algorithm, this method can avoid temporary obstacles (such as stacked goods) in real time, adjust the detour paths required for temperature-sensitive goods (such as avoiding high-temperature areas), and avoid path conflicts through multi-device location collaboration. This method can avoid temporary obstacles in real time, adjust the detour paths required for temperature-sensitive goods, and avoid path conflicts through multi-device location collaboration, thereby achieving a comprehensive improvement in sorting efficiency, accuracy, and adaptability, and has significant technical and economic value.
[0153] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any modifications or equivalent changes made based on the technical essence of the present invention shall still fall within the scope of protection claimed by the present invention.
Claims
1. A data-driven automated cold storage and sorting system, characterized in that: The data-driven automated cold storage and sorting method includes a data acquisition module, a processing module, a training module, a prediction module, a sorting generation module, a path generation module, and a guidance module. The data acquisition module collects historical operation data from the cold storage, including historical storage location data, historical sorting task data, historical order data, cargo attribute data, and operation time data. The processing module performs feature engineering on the historical operation data to obtain multiple feature vectors. These feature vectors characterize the storage characteristics, sorting frequency characteristics, and order association characteristics of the cargo. A storage module is included within the processing module. The training module uses the feature vectors to train a prediction model. This model predicts the recommended storage location for cargo and the execution priority of sorting tasks. The prediction model is trained with the goal of minimizing the deviation between the predicted and actual storage locations and maximizing the efficiency of sorting task execution. The module receives real-time sorting task requests, which include the identification information of the goods to be sorted. Based on the identification information, it extracts corresponding goods attribute data from historical operation data and inputs the goods attribute data into the trained prediction model to obtain the recommended storage location of the goods to be sorted and the execution priority of the sorting task. The sorting generation module determines the target storage location of the goods to be sorted based on the recommended storage location and generates a sorting task execution sequence based on the execution priority of the sorting task. The path generation module generates the movement path of the sorting equipment based on the sorting task execution sequence and real-time environmental data in the cold storage through a path planning algorithm. The real-time environmental data includes the temperature distribution in the cold storage, the location of obstacles, and the current location of other sorting equipment. When the guidance module receives the output movement path and the sorting task execution sequence, it guides the sorting equipment to perform the sorting operation.
2. The method of a data-driven automated cold storage and sorting system according to claim 1, characterized in that: The steps of the data-driven automated cold storage and sorting method are as follows: Step 1: The data acquisition module collects historical operation data of the cold storage. The historical operation data includes historical storage location data of goods, historical sorting task data, historical order data, goods attribute data, and operation time data. Step 2: The processing module performs feature engineering based on historical operation data to obtain multiple feature vectors. These feature vectors are used to characterize the storage characteristics, sorting frequency characteristics, and order association characteristics of the goods. The operation parameter data and raw data are stored in the storage module. Step 3: The training module uses feature vectors to train a prediction model. The prediction model is used to predict the recommended storage location of goods and the execution priority of sorting tasks. The prediction model is trained with the goal of minimizing the deviation between the predicted location and the actual storage location and maximizing the execution efficiency of sorting tasks. Step 4: The prediction module receives real-time sorting task requests. The real-time sorting task requests contain the identification information of the goods to be sorted. Based on the identification information of the goods to be sorted, the module extracts the corresponding goods attribute data from the historical operation data and inputs the goods attribute data into the trained prediction model to obtain the recommended storage location of the goods to be sorted and the execution priority of the sorting task. Step 5: The sorting generation module determines the target storage location of the goods to be sorted based on the storage location recommendation degree, and generates the sorting task execution sequence according to the execution priority of the sorting task; Step 6: The path generation module generates the movement path of the sorting equipment based on the sorting task execution sequence and real-time environmental data in the cold storage through a path planning algorithm. The real-time environmental data includes the temperature distribution in the cold storage, the location of obstacles, and the current position of other sorting equipment. Step 7: When the guidance module receives the output movement path and sorting task execution sequence, it guides the sorting equipment to perform sorting operations, ultimately completing the automated storage and sorting of the cold storage.
3. The method of a data-driven automated cold storage and sorting system according to claim 1, characterized in that: The data-driven automated cold storage and sorting method specifically involves the acquisition module collecting historical cold storage data as follows: 1) Collect storage records, sorting records, and order records of goods in the cold storage according to the preset time window; 2) Perform data cleaning on storage records, sorting records, and order records to remove outliers and missing values; 3) The cleaned data is associated with cargo identifiers, timestamps, and operation types to obtain structured historical operation data.
4. The method of a data-driven automated cold storage and sorting system according to claim 1, characterized in that: The data-driven automated cold storage and sorting method described above involves a processing module performing feature engineering on historical operation data to obtain multiple feature vectors. The specific process is as follows: 1) Perform one-hot encoding or label encoding on categorical features in historical operational data, and normalize or standardize numerical features; 2) Extract time-series features from historical operational data. These features include storage duration, sorting interval, and order frequency. The formulas for calculating storage duration and sorting interval are: ； in: Indicates storage duration; Indicates the time when goods are received into the warehouse; Indicates the time the goods left the warehouse; ； in: Indicates the sorting interval time; Indicates the current sorting task time; Indicates the time of the last sorting task for the same type of goods; 3) Based on the attribute data of goods, generate goods association features, which are used to characterize the correlation between goods; combine the processed categorical features, numerical features, time series features and goods association features to obtain multiple feature vectors.
5. The method of a data-driven automated cold storage and sorting system according to claim 1, characterized in that: The data-driven automated cold storage and sorting method uses feature vectors to train a prediction model, specifically as follows: 1) Divide the feature vectors into training set, validation set, and test set; 2) Construct the initial prediction model. The initial prediction model is a multi-task learning model based on deep learning, including a shared layer, a storage location prediction branch, and a sorting priority prediction branch. 3) Train the initial prediction model using the training set, and adjust the model parameters using the backpropagation algorithm to minimize the cross-entropy loss of the prediction branch at the storage location. and the mean square error loss of sorting priority prediction branches The total loss function L_total is: ； Where: α is the cross-entropy loss; β is the weighting coefficient for the mean squared error loss, used to balance the importance of the two tasks; 4) Validate the trained model using the validation set, and adjust the model hyperparameters based on the validation results until the model performance meets the preset conditions; 5) Test the adjusted model using the test set to obtain the final prediction model.
6. The method of a data-driven automated cold storage and sorting system according to claim 1, characterized in that: In the data-driven automated cold storage and sorting method, the prediction module extracts corresponding cargo attribute data from historical operation data based on the identification information of the goods to be sorted, and inputs the cargo attribute data into the trained prediction model to obtain the recommended storage location of the goods to be sorted and the execution priority of the sorting task. Specifically: 1) Based on the identification information of the goods to be sorted, obtain the attribute data of the goods to be sorted from the cold storage's goods information database. The attribute data includes the category, size, weight, shelf life, and storage temperature requirements of the goods. 2) Convert the attribute data into feature vectors that match the input format of the prediction model; 3) Input the feature vector into the trained prediction model to obtain the storage location recommendation degree and the sorting task execution priority. The storage location recommendation degree is the recommendation probability of the goods to be sorted in each available storage location, and the sorting task execution priority is the numerical value of the execution order of the goods to be sorted relative to other goods to be sorted.
7. The method of a data-driven automated cold storage and sorting system according to claim 1, characterized in that: The sorting generation module in the data-driven automated cold storage and sorting method generates the sorting task execution sequence as follows: 1) Select the storage location with the highest recommendation from the storage location recommendations as the target storage location for the goods to be sorted; 2) Sort multiple goods to be sorted in descending order of sorting task execution priority to generate a sorting task execution sequence.
8. The method of a data-driven automated cold storage and sorting system according to claim 1, characterized in that: In the data-driven automated cold storage and sorting method, the path generation module generates the movement path of the sorting equipment through a path planning algorithm, specifically as follows: 1) Acquire real-time environmental data inside the cold storage. The real-time environmental data is collected in real time by sensors installed inside the cold storage. 2) Determine the sequence of storage locations that the sorting equipment needs to access based on the sorting task execution sequence; 3) Based on the storage location sequence and the real-time environmental data, Algorithm A is used for path planning to generate the movement path of the sorting equipment. The movement path satisfies the constraints of shortest path and / or minimum time consumption. The cost function of Algorithm A is... Defined as: ； Where: n represents the node in the current path; This represents the actual cost from the starting point to node n; This represents the estimated cost from node n to the destination.
9. The method of a data-driven automated cold storage and sorting system according to claim 1, characterized in that: In the data-driven automated cold storage and sorting method, the guidance module guides the sorting equipment to perform sorting operations as follows: 1) Receive actual execution data fed back by the sorting equipment. The actual execution data includes the actual movement path, actual sorting time, and actual storage location. 2) Compare the actual execution data with the movement path, sorting task execution sequence, and target storage location to calculate the execution deviation, which includes the path deviation rate and the sorting time deviation rate; wherein: Path deviation rate The calculation formula is: ； in: Indicates the actual length of the movement path; Indicates the length of the planned movement path; Sorting time deviation rate The general calculation formula is: ； in: Indicates the actual sorting time; Indicates the planned sorting time; 3) Adjust and optimize the prediction model and / or path planning algorithm based on the execution deviation.
10. The method of a data-driven automated cold storage and sorting system according to claim 8, characterized in that: The data-driven automated cold storage and sorting method employs the A algorithm for path planning in its path generation module, generating the specific movement path for the sorting equipment as follows: 1) Based on real-time environmental data, construct an environmental map of the cold storage, which includes obstacle information and passable area information; 2) Use each storage location in the storage location sequence as the target point for path planning; 3) Starting from the current position of the sorting equipment, and taking each storage location as the target point in turn, the A algorithm is used to plan the path and generate the path segment from the current position to the next target point. 4) Connect all path segments to obtain the movement path of the sorting equipment.