A smart storage rack and its usage method

Abstract

This invention relates to the field of shelving technology, providing an intelligent shelving unit and its usage method. The intelligent shelving unit includes a shelving body, a controller, and at least one feature acquisition module for acquiring feature information of goods to be stored. The shelving body has multiple storage slots. The controller stores historical access data for the storage slots. The feature acquisition module is connected to the controller and feeds back feature information to the controller. The controller is also used to acquire the popularity value of each storage slot based on the historical access data, and construct an access popularity map based on the popularity values ​​of multiple storage slots; acquire the overall popularity of the goods to be stored based on the historical access data and feature information; and allocate the optimal storage location based on the access popularity map and the overall popularity of the goods. This invention can optimize the goods access path, shorten operation time, improve space utilization, and reduce overall costs, thus improving the intelligence level of the shelving unit.

Description

Technical Field

[0001] This invention relates to the field of shelving technology, and more specifically, to an intelligent shelving unit and its usage method. Background Technology

[0002] Shelving units are widely used in various industries, such as shelves in shopping malls, bookshelves in libraries or bookstores, and storage racks in warehouses. With the development of intelligent and robotic technologies, more and more shelves / shelving units are being integrated with intelligent or robotic technologies to improve the efficiency of accessing and storing goods.

[0003] For example, Chinese invention patent application number CN201810188194.0 discloses an intelligent shelf and its control method. It involves installing image acquisition devices on the shelf body and / or shelves to collect image information of target objects. The control device can adjust the height of the intelligent shelf according to the height of shoppers based on the image information of the target objects. Another example is Chinese invention patent application number CN202111364038.3, which can automatically allocate space according to shelf size, efficiently utilize storage racks, automatically handle inbound and outbound operations, and record detailed shelf information, thus improving shelf space utilization. Yet another example is Chinese invention patent application number CN202410010597.1, which discloses a shelf recognition method, robot control method, terminal equipment, and computer storage medium. It employs a combination of coarse 3D point cloud recognition and fine 2D point cloud recognition to accurately identify the center position of the shelf.

[0004] Therefore, for smart storage racks, there are still many unsolved technical problems in their practical application and many unproven solutions. Summary of the Invention

[0005] Based on this, in order to improve the intelligence level of the storage rack, the present invention provides an intelligent storage rack and its usage method, the specific technical solution of which is as follows:

[0006] A smart storage rack includes a rack body, a controller, and at least one feature acquisition module for acquiring feature information of goods to be stored. The rack body has multiple storage slots. The controller stores historical access data of the storage slots. The feature acquisition module is connected to the controller and feeds back the feature information to the controller. The controller is also used for:

[0007] The popularity value of each storage location is obtained based on historical access data, and an access popularity map is constructed based on the popularity values ​​of multiple storage locations.

[0008] The overall popularity of the goods to be stored is obtained based on the historical access data and feature information.

[0009] And allocate the optimal storage location based on the access heat map and the overall popularity of the goods.

[0010] The intelligent shelving unit constructs a heat map and obtains the overall popularity of goods. Based on the heat map and the overall popularity of goods, it allocates the optimal storage location, which can optimize the storage and retrieval path of goods, shorten operation time, improve space utilization and reduce overall costs, thus improving the intelligence level of the shelving unit.

[0011] Preferably, the controller includes:

[0012] The distance acquisition module is used to acquire the three-dimensional Euclidean distance between the storage location and each access location in the historical access events;

[0013] The contribution acquisition module is used to acquire the contribution of each historical access event to the storage location based on the three-dimensional Euclidean distance.

[0014] The heatmap construction module is used to weight and superimpose the contribution values ​​to obtain the heat value, and construct the access heatmap based on the heat value.

[0015] Preferably, the controller further includes:

[0016] The access frequency acquisition module is used to acquire the overall access frequency and recent access frequency of the goods to be stored;

[0017] The related popularity acquisition module is used to acquire the popularity of related goods;

[0018] The comprehensive popularity acquisition module is used to acquire the comprehensive popularity of goods based on comprehensive access frequency, recent access frequency, and the popularity of related goods.

[0019] Preferably, the heatmap construction module is based on the formula Construct an access heatmap;

[0020] in, Represents the three-dimensional coordinates of the storage location. Popularity value This represents the total number of historical access events. Represents the three-dimensional coordinates of the storage location. With the The three-dimensional Euclidean distance of the next access location. This represents the diffusion coefficient of the Gaussian kernel. Indicates the distance offset. Represents the natural constant. Indicates the first The contribution of each historical access event to the storage location.

[0021] Preferably, the comprehensive popularity acquisition module is based on the formula Obtain the overall popularity of goods;

[0022] in, These represent the overall access frequency, recent access frequency, and related goods popularity, respectively. These represent the overall access frequency, recent access frequency, and related goods popularity weighting coefficient, respectively.

[0023] Preferably, the intelligent shelving unit further includes a storage and retrieval robotic arm connected to the controller via a signal. The robotic arm is used to respond to control commands and perform order sorting. The controller further includes:

[0024] The path planning module is used to plan the path according to the Q function after receiving the sorting order. Obtain the picking sequence and the planned path, and generate control instructions based on the picking sequence and the planned path;

[0025] in, Indicates the current system status. Indicates the system at time 10:00 The running status, Indicates the currently selected action. Indicates the time when the access robot arm is at time Specific actions, Indicates the discount factor. Indicates the system at time 10:00 Instant rewards Represents the expectation operator. Indicates from the current moment Starting future time step index, This represents the reward function, which includes time efficiency rewards, energy consumption penalties, and collision risk penalties.

[0026] A method of using a smart storage rack, comprising:

[0027] Obtain the characteristic information of the goods to be stored and the historical access data of the storage location;

[0028] The heat value of each storage location is obtained based on the historical access data, and an access heat map is constructed based on the heat values ​​of multiple storage locations.

[0029] The overall popularity of the goods to be stored is obtained based on the historical access data and feature information.

[0030] The optimal storage location is allocated based on the access heat map and the overall popularity of the goods.

[0031] Preferably, the specific method for constructing the access heatmap includes:

[0032] Obtain the three-dimensional Euclidean distance between the storage location and each access location in the historical access events. ;

[0033] The contribution of each historical access event to the storage location is obtained based on the three-dimensional Euclidean distance. ;

[0034] The contribution values ​​are weighted and summed to obtain the popularity value, and an access popularity map is constructed based on the popularity value;

[0035] in, Represents the three-dimensional coordinates of the storage location. Popularity value This represents the total number of historical access events. This represents the diffusion coefficient of the Gaussian kernel. Indicates the distance offset. Represents the natural constant.

[0036] Preferably, the specific method for obtaining the overall heat index of the goods to be stored includes:

[0037] Obtain the overall access frequency of the goods to be stored Recent Access Frequency and the popularity of related goods ;

[0038] The overall popularity of goods is obtained based on the overall access frequency, recent access frequency, and the popularity of related goods. .

[0039] in, These represent the weighting coefficients for the overall access frequency, recent access frequency, and related goods popularity, respectively.

[0040] Preferably, the method of using the smart shelf further includes:

[0041] After receiving the sorting order, according to the Q function Obtain the picking sequence and the planned path, and generate control instructions based on the picking sequence and the planned path;

[0042] The robotic arm responds to control commands and performs order sorting;

[0043] in, Indicates the current system status. Indicates the system at time 10:00 The running status, Indicates the currently selected action. Indicates the time when the access robot arm is at time Specific actions, Indicates the discount factor. Indicates the system at time 10:00 Instant rewards Represents the expectation operator. Indicates from the current moment Starting future time step index, This represents the reward function, which includes time efficiency rewards, energy consumption penalties, and collision risk penalties. Attached Figure Description

[0044] The invention will be further understood from the following description taken in conjunction with the accompanying drawings. The components in the drawings are not necessarily drawn to scale, but rather the emphasis is on illustrating the principles of the embodiments. In different views, the same reference numerals designate corresponding parts.

[0045] Figure 1 This is a schematic diagram of the overall structure of the intelligent storage rack in one embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of the overall process of using the intelligent storage rack in one embodiment of the present invention;

[0047] Figure 3 This is a flowchart illustrating a specific method for constructing an access heatmap in one embodiment of the present invention;

[0048] Figure 4 This is a flowchart illustrating a specific method for obtaining the overall popularity of goods to be stored in one embodiment of the present invention.

[0049] Figure 5 This is one of the flowcharts illustrating the method of using the intelligent storage rack in another embodiment of the present invention;

[0050] Figure 6 This is a second flowchart illustrating the method of using the intelligent storage rack in another embodiment of the present invention;

[0051] Figure 7 This is a schematic diagram of the structure of an intelligent storage rack according to an embodiment of the present invention;

[0052] Figure 8 This is a schematic diagram of the functional modules of the controller in one embodiment of the present invention;

[0053] Figure 9 This is a second schematic diagram of the functional modules of the controller in one embodiment of the present invention.

[0054] Explanation of reference numerals in the attached diagram: 1. Shelf body; 2. Controller. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to its embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of the invention.

[0056] It should be noted that when an element is referred to as being "fixed to" another element, it can be directly attached to the other element or there may be an intervening element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or there may be an intervening element. The terms "vertical," "horizontal," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0058] In this invention, "first" and "second" do not represent a specific quantity or order, but are merely used to distinguish names.

[0059] Before describing the specific embodiments of the present invention, a brief introduction to the prior art will be given first.

[0060] The allocation of storage spaces on shelving units directly impacts the efficiency of goods entering and leaving the warehouse. Optimizing this allocation can often improve storage efficiency and reduce overall operating costs. Currently, the overall temperature of goods is often determined by the experience of technicians or managers, making it difficult to guarantee the accuracy of this assessment. Allocating storage spaces based on manually judged temperature not only reduces the intelligence level of the shelving system but also decreases the efficiency of goods storage and retrieval.

[0061] Existing technologies such as CN202211552812.8 ("An Intelligent Automated Warehouse Shelf Location Allocation System and Method"), CN201911138483.0 ("A Location Allocation Method for a Mobile Shelf Storage System for Cold Storage"), CN202211594440.5 ("An Optimization Method for Location Allocation in an Automated Warehouse with Non-Tall Shelves"), and CN202010568008.3 ("A Shelf Allocation Method, Device, Electronic Equipment, and Readable Storage Medium") all involve the allocation of shelf locations. While these shelf location allocation methods can improve the speed of goods entering and leaving the warehouse, they do not integrate the popularity of shelf locations with the overall popularity of the goods to be stored when considering the allocation of locations. Therefore, there is still room for further improvement in their level of intelligence and warehousing efficiency.

[0062] To improve the efficiency of goods storage and retrieval on the storage rack and its level of intelligence, one embodiment of the present invention provides an intelligent storage rack, such as... Figure 1 as well as Figure 7 As shown, the device includes a shelf body 1, a controller 2, and at least one feature acquisition module for acquiring feature information of goods to be stored. The shelf body has multiple storage slots. The controller is used to store historical access data of the storage slots. The feature acquisition module is connected to the controller and feeds back the feature information to the controller.

[0063] The shelving unit is placed in a warehouse or goods sorting workshop. The shelving unit can be a multi-layered shelf, with multiple storage locations on each layer for storing goods. The controller includes, but is not limited to, a server, a processing center, and a client. It is connected to the feature acquisition module via cable or wireless communication module to receive feature information of the goods to be stored acquired by the feature acquisition module, analyze the feature information, and determine the type of goods to be stored.

[0064] The feature acquisition module includes, but is not limited to, an RGB-D camera (e.g., depth resolution 1280×720), a weight sensor (e.g., range 0-50kg, accuracy ±0.1g), and an RFID scanning module (e.g., identification distance 1.5m). Correspondingly, the feature information includes, but is not limited to, image information, weight information, and tag information. Through the RGB-D camera, weight sensor, and RFID scanning module, a three-dimensional spatial-weight-tag joint sensing network can be formed, facilitating the controller to identify the type of goods to be stored based on feature information, and to obtain storage location access data and generate historical access data based on feature information.

[0065] The RGB-D camera, weight sensor, and RFID scanning module can be installed on the shelving unit or in appropriate locations (such as entrances and exits) in warehouses or sorting workshops, as needed.

[0066] Preferably, the feature acquisition module covers the static storage area, dynamic operation area, and interactive interface area of ​​the shelf, and the specific installation scheme is as follows:

[0067] 1. Top central area of ​​the shelf

[0068] 1. An RGB-D main camera is fixedly installed on the top crossbeam of the shelving unit, tilted downwards at a 15° angle. Multiple cameras are set up and evenly distributed along the length of the crossbeam. It is used to monitor the storage status of the shelf globally, generate a 1280×720 depth point cloud (accuracy ±2mm), and eliminate blind spots (such as obstruction at the edge of high-rise shelves) through multi-view stitching.

[0069] The RGB-D camera can be used to support real-time obstacle avoidance in robotic arm path planning, such as detecting sudden cargo deviations.

[0070] 2. The RFID scanning module is coaxially mounted with the RGB-D camera, with the antenna facing the entrance of the shelving unit. It is used to identify the unique ID of tagged goods within the range. By fusing with visual data, it can solve the problem of confusion between goods with similar appearances.

[0071] By setting an RGB-D main camera and RFID scanning module in the center of the shelf, the shelf can be monitored globally, solving the problem of "seeing but not being able to identify".

[0072] II. Shelving Layer Structure Area

[0073] 1. Embedded node weight sensors are installed at the bottom of the support columns of each shelf layer. They adopt a four-point weighing structure and are used to monitor the weight change of goods from 0-50kg in real time (accuracy below 0.1%) and detect abnormal storage and retrieval, such as partial retrieval without reset.

[0074] 2. Layered RGB-D edge cameras are installed 5cm below the front edge of each shelf, with a horizontal viewing angle covering the storage cell of that shelf. They are used to assist the RGB-D main camera in completing high-precision positioning (such as millimeter-level coordinate calibration of small parts) and to independently record the operation logs of each shelf.

[0075] By setting embedded node weight sensors and layered RGB-D edge cameras in the layered structure area of ​​the shelving, goods can be accurately positioned and their weight verified, preventing misplacement or missed retrieval.

[0076] III. Entrance / Exit of the Area Where the Shelves Are Located

[0077] 1. A channel-type weight detection platform composed of weight sensors is embedded in the ground at the entrance and exit of the area where the shelving is located. The platform measures 80cm × 120cm and is used to verify the total weight of goods entering and leaving the warehouse (by comparing the data with the data of the embedded node weight sensors on the shelf layer; if the error is greater than the preset weight value, an alarm will be triggered).

[0078] 2. An RFID gate array composed of RFID scanning modules is used to set the entrance and exit as an arch structure. Three sets of directional antennas are deployed on each side of the arch for batch reading of all cargo tags in the cargo turnover box.

[0079] By setting up a channel-type weight detection platform and RFID gate array at the entrance and exit of the area where the shelving is located, dual verification of entry and exit can be achieved.

[0080] The controller is also used to obtain the heat value of each storage location based on historical access data, and construct an access heat map based on the heat values ​​of multiple storage locations; obtain the overall heat of the goods to be stored based on the historical access data and feature information; and allocate the optimal storage location based on the access heat map and the overall heat of the goods.

[0081] Preferably, such as Figure 8 As shown, the controller includes a distance acquisition module, a contribution acquisition module, and a heatmap framework module.

[0082] The distance acquisition module is used to obtain the three-dimensional Euclidean distance between the storage location and each access location in historical access events. Specifically, the three-dimensional Euclidean distance characterizes spatial proximity and represents the three-dimensional coordinates of the storage location. With the The three-dimensional Euclidean distance of the next access location can be denoted as: , Indicates the first The three-dimensional coordinates of the next access location.

[0083] The contribution acquisition module is used to obtain the contribution of each historical access event to the storage location based on the three-dimensional Euclidean distance. Specifically, the heat contribution of each historical access event to the surrounding space follows a normal distribution, and the contribution of the event to the surrounding space follows a normal distribution. The contribution of each historical access event to a storage location can be expressed as: It can be understood as the first The next historical access event at the storage location's three-dimensional coordinates. The degree of contribution.

[0084] The heatmap construction module is used to weight and sum the contribution values ​​to obtain the heat value, and then construct an access heatmap based on the heat value. Here, the heat value is determined according to the formula... It is used to obtain the three-dimensional coordinates of the storage location. The frequency of access or the intensity of demand.

[0085] Preferably, the heatmap construction module is based on the formula Construct an access heatmap. Among other things, Represents the three-dimensional coordinates of the storage location. Popularity value; This represents the total number of historical access events, reflecting the time span and sample size of the historical access data points; The diffusion coefficient of the Gaussian kernel is used to control the range of influence of a single access event on the surrounding area. This represents the distance offset, used to correct the model's sensitivity to access events (usually set to 0, indicating the event location as the center). Represents the natural constant.

[0086] Specifically, by acquiring the popularity value of each storage location, a heatmap is plotted using professional data analysis tools (such as Python's Matplotlib and Seaborn libraries) based on the popularity value of each storage location. Different colors in the heatmap represent different popularity levels, with darker colors indicating higher popularity. Based on the heatmap, a popularity threshold is set, and areas with popularity values ​​higher than this threshold are identified as high-frequency areas (typically areas where goods enter and leave the warehouse most frequently).

[0087] Diffusion coefficient of Gaussian kernel The larger the value, the wider the impact range of a single access event, but the lower the peak intensity; the diffusion coefficient of the Gaussian kernel... The smaller the value, the more concentrated the impact area, and the more significant the local peak. Regarding distance offset... ,like It can simulate the directional shift of access behavior (such as goods closer to the aisle being accessed more frequently).

[0088] The contribution of all historical access events can be superimposed to form a continuous heat map. The H-value of high-frequency access areas (such as shelf entrances) is significantly higher than that of low-frequency areas (such as corners). Based on this heat map, warehouse optimization can be guided to allocate optimal storage locations for goods, for example: 1. Prioritize high-demand, high-heat goods (… (Large value) placed 1. Shorten picking routes in peak areas; 2. Identify 3. Analyze local extreme points of the value to divide the storage area into "hot spots" and "cold spots" to optimize the shelf zoning design; 4. Dynamically adjust parameters and adjust the diffusion coefficient according to the cargo turnover rate. For example, seasonal products increase their diffusion coefficient during promotional periods. To expand the scope of inventory,

[0089] By weighting the contribution of historical access events using a Gaussian distribution function, the intensity of access demand at spatial locations can be dynamically quantified; by analyzing the popularity distribution during peak order periods, the location demand for popular products can be predicted, enabling automatic replenishment.

[0090] Preferably, such as Figure 9 As shown, the controller also includes an access frequency acquisition module, an association popularity acquisition module, and a comprehensive popularity acquisition module.

[0091] The access frequency acquisition module is used to acquire the overall access frequency and recent access frequency of the goods to be stored. Specifically, the overall access frequency... The recent access frequency .in, Indicates goods to be stored In the statistical period The total number of accesses within the specified timeframe can be counted using an RFID scanning module or RGB-D camera image recognition logs, with a statistical period of [missing information]. This can be understood as the total number of days the system runs or the time span of goods being put on the shelves. For example, if goods have been on the shelves for 30 days, then the statistical period is... . Indicates goods to be stored In recent The number of accesses recently It is a preset time window (e.g., the last 7 days). By calculating the recent access frequency, the problem of lagging popularity of long-tail goods can be solved.

[0092] The related popularity acquisition module is used to acquire the popularity of related goods. Specifically, the related goods popularity... .

[0093] in, Indicates related goods Quantity, Indicates goods and The number of co-accesses can be understood as the number of goods accessed within the statistical period. and The total number of times items are accessed simultaneously (e.g., a user accesses two items at the same time in one operation) can be statistically analyzed using user behavior logs (e.g., records from an RFID scanning module) to determine the co-occurrence frequency. This is used to quantify the strength of the association between two items; the higher the co-occurrence frequency, the stronger the association. For example, a high co-occurrence frequency of screwdrivers and screws reflects a functional association.

[0094] Indicates goods The total number of accesses can be understood as the number of goods. The total number of independent accesses within the statistical period can be determined by directly analyzing the goods in the log. Access records obtained This is used to normalize co-occurrence frequencies, avoiding excessive influence of high-frequency goods on correlation popularity. For example, if goods... Frequent access will weaken the chances of its occasional co-occurrence with multiple goods.

[0095] The time decay factor can be understood as a measure of the time decay effect of co-occurring events based on Newton's law of cooling. This indicates the attenuation coefficient (which can be set by technicians based on experience). This represents the time difference (in days) between the current time and the most recent co-occurrence. The larger the value, the faster the decay rate (e.g.) At that time, the co-occurrence weight 30 days ago dropped to ), The smaller the value (recent co-occurrence), the higher its weight. Time decay factor. It is used to emphasize recent correlations and can solve the problem of outdated historical co-occurrence data, such as the significant changes in the correlation of seasonal products over time.

[0096] This represents the semantic similarity weight, which can be understood as goods and The semantic association strength is calculated based on the second degree of the text description or knowledge graph. For example, based on the Word2Vec word embedding model, the cosine similarity is calculated by generating vectors from the description text of the goods. Alternatively, based on the knowledge graph, if the goods belong to the same category or have an upstream or downstream relationship (such as "coffee cup" and "coffee machine"), they are given higher weights. Its purpose is to supplement the lack of behavioral co-occurrence data and identify implicit associations. For example, even if the number of co-occurrences is small, "battery" and "remote control" may be associated due to semantic association.

[0097] Specifically, if semantic similarity weights are calculated based on the cosine similarity of word vectors, then , Representing goods and Word vectors can be generated using models such as Word2Vec and BERT. This represents the vector magnitude. If semantic similarity weights are calculated based on path-weighted similarity from a knowledge graph, then... , Representing goods in a knowledge graph and The number of shortest paths, Representing paths The weights of the edges in the relation and the path length (number of nodes).

[0098] Indicates goods The independent popularity of related goods. The higher the popularity of the related goods themselves, the greater the impact on... The greater the contribution. For example, if related goods Even if a product appears only once in a short period of time, its popularity will still be boosted. Value. For goods The independent heat can be determined by the formula. .in, Indicates goods In the statistical period Total number of accesses within.

[0099] The goods Independent popularity .in, Representing goods The search popularity, discussion popularity, and dissemination popularity of [the topic]. Representing goods The weighting coefficients for search popularity, discussion popularity, and dissemination popularity can be set by technical personnel. Preferably, the weighting coefficients are dynamically adjusted by predicting user behavior trends using an LSTM neural network. That is, based on time series prediction, weight coefficients are dynamically allocated.

[0100] Search popularity can be calculated based on search volume from platforms such as Baidu Index and Google Trends. Preferably, a regional weighting coefficient (which can be set by technical or managerial personnel based on experience) is introduced, based on the acquisition of goods... The search volume of keywords on different platforms in different regions is combined with the regional platform search volume and regional weighting coefficient to calculate a weighted value, in order to distinguish the differences in search popularity in different regions.

[0101] Discussion popularity can be calculated based on the number of social media comments and forum posts, combined with user influence weighting (e.g., comments from Weibo influencers have higher weighting). By weighting user influence, interference from low-quality discussions and artificially inflated numbers can be avoided. Dissemination popularity can be calculated based on the number of reposts and shares, combined with the depth of the dissemination path (e.g., three-level dissemination on WeChat Moments has higher weighting). By introducing a dissemination depth factor, shallow dissemination and deep dissemination can be distinguished.

[0102] This indicates the time-based enhancement factor for search popularity. Indicates the time augmentation factor. This indicates a time variable, showing an exponential increase in recent search volume, used to distinguish between short-term surges and long-term stable popularity. This indicates the decay factor of the S-shaped curve of discussion popularity ( (Peak time of popularity), indicating that the popularity of discussion first rises and then declines over time (such as the cooling-off period of a hot topic), used to fit the actual decline curve of discussion popularity. This represents the kurtosis parameter of the sigmoid function, used to control the transition speed of weights from low to high (e.g., ...). Smooth transition, (Quick switching) Indicates the logarithmic growth factor of the spread popularity. This indicates that the reach of content increases slowly over time (such as long-tail content), and is used to prevent excessive reach from distorting the popularity score. This represents the logarithmic growth scaling factor, used to adjust the magnitude of potential value growth over time; This represents the interaction quality factor (range: 0-1). Interaction types include collection, like, comment, and share. It can be obtained by weighting the interaction type, for example, collection = 0.4, like = 0.3, comment = 0.2, share = 0.1. Of course, the values ​​of different interaction types can also be defined according to the actual situation, which is used to distinguish the value of interaction behavior in a fine-grained way. This represents the user profile factor (range: 0-1), which can be understood as the user value stratification (e.g., VIP users have higher interaction weight), and is used to combine user profiles to improve the accuracy of popularity calculation. These represent the weights of interaction and user profile, determined through A / B testing (e.g., in e-commerce scenarios). It is divided into settings of 0.6 and 0.4, used to dynamically balance the impact of user behavior and user profile; This represents the outlier filtering coefficient (range: 0-1), which is used to filter robot behavior (such as a large number of repeated interactions in a short period of time) based on the chi-square test to eliminate interference from abnormal data.

[0103] The goods The independent heat formula has the following advantages:

[0104] 1. Multi-dimensional dynamic weight allocation. By predicting user behavior trends using LSTM, the weight of search, discussion, and dissemination popularity is automatically adjusted. For example, search weight is increased during the new product launch period, and dissemination weight is increased during the long-tail content period.

[0105] 2. Enhanced time sensitivity. Search popularity is captured using an exponential growth factor to identify short-term surges; discussion popularity is simulated using an S-shaped curve to represent its lifecycle; and dissemination popularity is smoothed using a logarithmic function to mitigate long-tail effects.

[0106] 3. Refined user behavior modeling. Activity quality factors differentiate the value of behaviors such as saving and liking, while user profile factors enhance the contribution of high-value users. For example, in the maternal and infant product category, the interaction weight of mothers is higher.

[0107] 4. Noise immunity. Outlier filtering coefficient. Chi-square tests are used to exclude bots from inflating user numbers, ensuring the authenticity of popularity metrics.

[0108] The comprehensive popularity acquisition module is used to acquire the comprehensive popularity of goods based on comprehensive access frequency, recent access frequency, and the popularity of related goods. Specifically, the comprehensive popularity acquisition module uses the formula... Obtain the overall popularity of goods; among them, These represent the overall access frequency, recent access frequency, and related goods popularity, respectively. These represent the weighting coefficients for the overall access frequency, recent access frequency, and related goods popularity, respectively.

[0109] The popularity of related goods is enhanced by integrating co-occurrence frequency, time decay, and semantic similarity, thus amplifying the influence of strongly related goods and preventing the impact of highly popular related goods from being evenly distributed. This is combined with behavioral data. Time decay and semantic information This improves the accuracy of related popularity metrics. Based on the popularity of related goods, the final comprehensive popularity of goods is more realistic and more accurate.

[0110] The specific method for allocating the optimal storage location based on the access heat map and the overall popularity of goods includes: pre-constructing a mapping relationship between the overall popularity of goods and access heat, and allocating corresponding storage locations for the goods to be stored according to the constructed mapping relationship.

[0111] Specifically, storage locations can be divided into zones based on the storage and retrieval heat map. Storage locations in different zones correspond to different overall cargo heat, and corresponding storage locations are allocated based on the overall cargo heat of the goods to be stored.

[0112] By improving the accuracy of overall cargo popularity and storage location popularity, and integrating storage location popularity with the overall popularity of the goods to be stored when considering cargo allocation, it is beneficial to optimize the allocation of storage locations for goods by smart shelving, thereby optimizing the cargo storage and retrieval path.

[0113] In other words, the intelligent storage rack can optimize the storage and retrieval path, shorten operation time, improve space utilization and reduce overall costs by constructing a heat map and obtaining the overall popularity of goods, and allocate the optimal storage location based on the heat map and the overall popularity of goods, which is conducive to improving the intelligence level of the storage rack.

[0114] As a preferred technical solution, the outlier filtering coefficient Its core objective is to identify and eliminate abnormal behaviors such as bot-driven traffic manipulation through statistical testing, specifically including the following steps:

[0115] The first step is data collection. Since bots typically exhibit high-frequency, repetitive, or abnormal interaction patterns within a short period, data collection can be used to identify related goods. and Further analysis will be conducted on the interactive behavior data (such as likes, comments, sharing timestamps, user IDs, IP addresses, device fingerprints, etc.) to determine whether the interactive data is abnormal.

[0116] The second step is feature construction. Extract anomaly detection features, including but not limited to the frequency of interaction per unit time (such as the same user liking multiple times within 1 minute), the deviation of device / account activity time from user profile (such as high-frequency interaction late at night), and the contradiction between geographical location and device information (such as multiple devices interacting simultaneously from the same IP address).

[0117] The third step is to implement the chi-square test. First, the hypotheses are set: since the chi-square test detects anomalies by comparing the observed frequency with the expected frequency, normal interactive behavior follows a Poisson or normal distribution, while abnormal behavior deviates from this distribution. Next, binning is performed; the continuous data is discretized to suit the chi-square test, grouping interactive behavior by time or user dimension (e.g., number of interactions per hour). Finally, the chi-square value is calculated. A significance test is then performed. If the chi-square value is greater than the critical value, an anomaly is considered to exist. These represent the observed frequency and the expected frequency, respectively.

[0118] The fourth step is to dynamically adjust the outlier filtering coefficient based on the proportion of abnormalities. If the detected percentage of abnormal interactions is ≤5%, the outlier filtering coefficient will be adjusted. If abnormal interactions account for 5%-15%, then the outlier filtering coefficient should be adjusted accordingly. (For example, if the percentage of abnormal interactions is 10%, then the outlier filtering coefficient is...) If the percentage of abnormal interactions is greater than 15%, then the outlier filtering coefficient will be adjusted. This involves forcibly filtering high-risk data. Of course, the outlier filtering coefficients corresponding to different proportions of abnormal interactions can be dynamically adjusted according to the actual situation and needs.

[0119] Based on the outlier filtering coefficient, the cargo can be dynamically adjusted. The credibility weight of independent popularity has the following effects on the independent popularity formula: 1. Improved noise resistance by eliminating bot-driven volume manipulation (such as a large number of repeated likes in a short period of time) and avoiding false interactions that artificially inflate popularity values; 2. Enhanced authenticity of popularity values ​​by introducing an outlier filtering coefficient. The independent popularity formula is closer to real user behavior. For example, in the mother and baby product category, if the interaction rate of VIP users (high-value users) is diluted by abnormal behavior, This will reduce the overall weight, ensuring that the behavior of high-value users dominates the popularity calculation; 3. Dynamic adaptive optimization, The value changes with the proportion of anomalies, giving the independent popularity formula an adaptive capability. For example, during major e-commerce promotions, the frequency of normal interactions may temporarily increase. In this case, the chi-square test can be combined with historical peak values ​​to adjust the expected frequency and avoid misjudgment; 4. Multi-dimensional collaborative filtering. Complementing the interaction quality factor and user profile factor, the interaction quality factor is used to differentiate behavioral value (e.g., collection > like), while the user profile factor is used to weight user value (e.g., VIP has higher weight). This eliminates systematic anomalies from a statistical perspective, and the three factors work together to improve the quality of goods. The accuracy of independent heat.

[0120] The outlier filtering coefficient dynamically identifies and suppresses robot-generated data through chi-square tests, significantly improving the anti-interference capability and authenticity of the independent popularity value formula. Its synergistic effect with interaction quality factors and user profile factors makes the goods... The independent popularity value more accurately reflects real user behavior and is suitable for high-noise scenarios such as e-commerce recommendations.

[0121] To improve the intelligence of goods storage and retrieval in the intelligent shelving system, as a preferred technical solution, the intelligent shelving system further includes a storage and retrieval robotic arm connected to a controller. This robotic arm responds to control commands and performs order sorting. The robotic arm can be mounted on an AGV (Automated Guided Vehicle) trolley, which is connected to the controller and responds to control commands, moving along a planned path. Once it reaches the designated position, the robotic arm picks up the goods according to the picking sequence.

[0122] Specifically, the controller also includes a path planning module.

[0123] The path planning module is used to determine the path based on the Q function after receiving the sorting order. Obtain the picking sequence and the planned path, and generate control instructions based on the picking sequence and the planned path.

[0124] in, Indicates the current system status. Indicates the system at time 10:00 The running status, Indicates the currently selected action. Indicates the time when the access robot arm is at time Specific actions, Indicates the discount factor. Indicates the system at time 10:00 Instant rewards Represents the expectation operator. Indicates from the current moment The starting future time step index, whose value range is: Corresponding reward sequence , The reward function includes time efficiency rewards, energy consumption penalties, and collision risk penalties. Indicates the future number The decay weight of the step reward. In the Q function, by accumulating the decay rewards of all time steps, it can comprehensively evaluate the action. In state The long-term benefits.

[0125] Specifically, the system at any time The operational status includes warehouse layout (such as shelf positions, aisle width, etc.), real-time location and remaining battery power of the storage / retrieval robotic arm / AGV, priority of orders to be sorted, location of target shelf storage positions, and other dynamic factors (such as paths of other storage / retrieval robotic arms, obstacle distribution). The storage / retrieval robotic arm is constantly... Specific actions include movement direction (forward, turning, obstacle avoidance path planning) and cargo grabbing / releasing operations. Discount factor This is used to balance the importance of current rewards and future rewards, when the discount factor... When the value is 0, it means that only the immediate reward is considered, and all subsequent rewards are ignored. (When the discount factor...) When prioritizing long-term cumulative benefits (such as global path optimization), the Q function assigns almost the same weight to future rewards as to current rewards.

[0126] Time efficiency refers to the number of orders completed or the efficiency of sorting operations per unit of time. Time efficiency rewards include positive and negative rewards. Positive rewards can be understood as an increase in reward value (e.g., +1 or +0.5) for each completed order sorting or a shortened sorting cycle. Negative rewards can be understood as a deduction in reward when a sorting task is not completed within the time limit (e.g., -0.5 or -1). Time efficiency rewards are used to reduce ineffective movement time and improve order throughput by optimizing paths.

[0127] The energy consumption includes the energy consumed by the AGV's movement and the grasping actions of the robotic arm. Energy consumption penalties include negative penalties, which can be understood as being based on the distance traveled (e.g., per meter, with a penalty value defined as -1 or -2) and acceleration frequency (e.g., each rapid acceleration, with a penalty value defined as -0.5 or -1). Energy consumption penalties are used to reduce repetitive paths and lower overall energy consumption through a layered strategy.

[0128] Collision risk represents the probability of the access robot colliding with other equipment, people, or obstacles. Collision risk penalties include negative penalties (such as real-time detection of surrounding dynamic obstacles, calculating penalty values ​​based on distance and speed; when the risk level is high, the negative penalty value is defined as -1.5 or -2) and safety rewards (positive rewards are gradually increased when there are consecutive collision-free operations, such as +0.3 or 0.6 every 10 minutes).

[0129] Obtaining the picking sequence and planning the path based on the Q function specifically includes: maximizing the expected cumulative reward during the picking process, i.e., finding the optimal strategy. Make By adjusting the weight coefficients of each indicator in the reward function, the strategy bias under different business needs can be realized. In high-priority order scenarios, the weight of time efficiency reward is increased, in energy-saving mode, the weight of energy consumption penalty is increased, and in the face of complex environment, the weight of collision risk penalty is strengthened.

[0130] In this embodiment, the picking sequence and planned path are obtained based on the Q function, and the sorting efficiency (time), energy consumption, and collision risk are balanced through the reward function, thereby reducing the collision risk. The rack body, feature access module, controller, and access robot are combined, and the optimal storage location is allocated based on the access heat map and the overall popularity of the goods. This not only optimizes the allocation of storage space for goods by the intelligent rack, thereby optimizing the goods access path, but also further improves space utilization and reduces overall costs, further enhancing the intelligence level of the rack and the warehousing efficiency.

[0131] An embodiment of the present invention also provides a method for using an intelligent storage rack, such as... Figure 2 As shown, it includes:

[0132] S1, obtain the characteristic information of the goods to be stored and the historical access data of the storage location.

[0133] The feature information includes, but is not limited to, image information, weight information, and tag information of the goods to be stored. The type of goods to be stored is identified based on this feature information. Storage location access data can be obtained first based on the feature information and processed and stored by a controller such as a client, processing center, or server to generate the historical access data. Specifically, each historical access event corresponds to one piece of historical access data.

[0134] S2, obtain the heat value of each storage location based on the historical access data, and construct an access heat map based on the heat values ​​of multiple storage locations.

[0135] Preferably, in step S2, as Figure 3 As shown, the specific methods for constructing an access heatmap include:

[0136] S21, Obtain the three-dimensional Euclidean distance between the storage location and each access location in the historical access events. .

[0137] S22, Obtain the contribution of each historical access event to the storage location based on the three-dimensional Euclidean distance. .

[0138] S23, the contribution values ​​are weighted and superimposed to obtain the heat value, and an access heat map is constructed based on the heat value.

[0139] in, Represents the three-dimensional coordinates of the storage location. Popularity value This represents the total number of historical access events. This represents the diffusion coefficient of the Gaussian kernel. Indicates the distance offset. Represents the natural constant.

[0140] Specifically, the three-dimensional Euclidean distance characterizes spatial proximity, representing the three-dimensional coordinates of the storage location. With the The three-dimensional Euclidean distance of the next access location can be denoted as: , Indicates the first The three-dimensional coordinates of the next access location.

[0141] The heatmap is generated by acquiring the heat value of each storage location and using professional data analysis tools (such as Python's Matplotlib and Seaborn libraries) to create a heatmap based on that value. Different colors represent different heat levels in the heatmap, with darker colors indicating higher heat. Based on the heatmap, a heat threshold is set, and areas with heat values ​​exceeding this threshold are identified as high-frequency areas (typically areas where goods enter and leave the warehouse most frequently).

[0142] S3, obtain the overall popularity of the goods to be stored based on the historical access data and feature information.

[0143] Preferably, the goods to be stored Overall popularity of goods ;in, Indicates goods to be stored In the statistical period Total number of accesses within, This can be understood as goods to be stored. In the statistical period The internal access density, or heat value, Indicates related goods In the corresponding statistical period Total number of accesses within, This can be understood as related goods. In the statistical period The internal access density, or heat value, Indicates related goods Quantity, This indicates the goods to be stored. Related goods The average related popularity, These represent the goods to be stored. The weighting coefficients for the average popularity of related factors can be set by technical personnel.

[0144] Specifically, with the goods to be stored Other goods that are related to the access or retrieval of goods are called related goods. It can be based on user behavior data mining (such as co-occurrence access, semantic association, etc.).

[0145] S4. Allocate the optimal storage location based on the access heat map and the overall popularity of the goods. Specifically, the storage locations can be divided into regions based on the access heat map, with different regions corresponding to different overall popularity of the goods. The corresponding storage location is then allocated based on the overall popularity of the goods to be stored.

[0146] In summary, the method of using the intelligent storage rack, by constructing a heat map and obtaining the overall popularity of goods, and allocating the optimal storage location based on the heat map and the overall popularity of goods, can optimize the storage and retrieval path of goods, shorten operation time, improve space utilization and reduce overall costs, which is conducive to improving the intelligence level of the storage rack.

[0147] Goods awaiting storage Overall popularity of goods In the formula, with the goods to be stored Related goods The average association popularity has two shortcomings. First, the influence of highly associated goods is spread evenly, and taking the average value can easily weaken the popularity value of strongly associated goods. Second, it does not distinguish between direct co-occurrence and weak semantic association, and ignores the differences in association strength.

[0148] In order to assess the overall popularity of goods Formula optimization is a preferred technical solution, such as... Figure 4 As shown, in step S3, the specific method for obtaining the overall popularity of the goods to be stored includes:

[0149] S31, Obtain the overall storage and retrieval frequency of the goods to be stored. Recent access frequency and the popularity of related goods ;

[0150] S32, Obtain the overall popularity of goods based on the overall access frequency, recent access frequency, and the popularity of related goods. .

[0151] in, The weighting coefficients representing the overall access frequency, recent access frequency, and related goods popularity can all be set by technical personnel.

[0152] The comprehensive storage frequency The recent access frequency .in, Indicates goods to be stored In the statistical period The total number of accesses within the specified timeframe can be counted using an RFID scanning module or RGB-D camera image recognition logs, with a statistical period of [missing information]. This can be understood as the total number of days the system runs or the time span of goods being put on the shelves. For example, if goods have been on the shelves for 30 days, then the statistical period is... . Indicates goods to be stored In recent The number of accesses recently It is a preset time window (e.g., the last 7 days). By calculating the recent access frequency, the problem of lagging popularity of long-tail goods can be solved.

[0153] Related goods popularity .

[0154] in, Indicates related goods Quantity, Indicates goods and The number of times the data is accessed in the current state. Indicates goods Total number of accesses, Indicates the time decay factor. This indicates the attenuation coefficient (which can be set by technicians based on experience). This represents the time difference (in days) between the current time and the most recent co-occurrence. Represents semantic similarity weights. Indicates goods The independent popularity of related goods. The higher the popularity of the related goods themselves, the greater the impact on... The greater the contribution. For example, if related goods Even if a product appears only once in a short period of time, its popularity will still be boosted. Value. For goods The independent heat can be determined by the formula. .in, Indicates goods In the statistical period Total number of accesses within.

[0155] The improved overall popularity ranking of goods can achieve the following: by using a triple weighting of co-occurrence frequency, time decay, and semantic similarity, it strengthens the influence of strongly associated goods and avoids the averaging effect; and by combining behavioral data... Time decay and semantic information This improves the accuracy of related popularity.

[0156] Based on the popularity of related goods, the final comprehensive popularity of goods is more in line with reality and more accurate.

[0157] In other words, by improving the accuracy of the overall popularity of goods and the popularity of storage locations, and by integrating the popularity of storage locations with the overall popularity of the goods to be stored when considering the allocation of goods, it is beneficial to optimize the allocation of storage locations for goods by smart shelves, thereby optimizing the storage and retrieval path of goods.

[0158] As a preferred technical solution, such as Figure 5 As shown, the method of using the smart storage rack also includes:

[0159] S5, after receiving the sorting order, according to the Q function Obtain the picking sequence and the planned path, and generate control instructions based on the picking sequence and the planned path;

[0160] S6, the robotic arm responds to control commands and performs order sorting;

[0161] in, Indicates the current system status. Indicates the system at time 10:00 The running status, Indicates the currently selected action. Indicates the time when the access robot arm is at time Specific actions, Indicates the discount factor. Indicates the system at time 10:00 Instant rewards Represents the expectation operator. Indicates from the current moment Starting future time step index, This represents the reward function, which includes time efficiency rewards, energy consumption penalties, and collision risk penalties.

[0162] Obtaining the picking sequence and planning the path based on the Q function specifically includes: maximizing the expected cumulative reward during the picking process, i.e., finding the optimal strategy. Make By adjusting the weight coefficients of each indicator in the reward function, the strategy bias under different business needs can be realized. In high-priority order scenarios, the weight of time efficiency reward is increased, in energy-saving mode, the weight of energy consumption penalty is increased, and in the face of complex environment, the weight of collision risk penalty is strengthened.

[0163] In this embodiment, the picking sequence and planned path are obtained based on the Q function, and the sorting efficiency (time), energy consumption, and collision risk are balanced through the reward function, thereby reducing the collision risk. The rack body, feature access module, controller, and access robot are combined, and the optimal storage location is allocated based on the access heat map and the overall popularity of the goods. This not only optimizes the allocation of storage space for goods by the intelligent rack, thereby optimizing the goods access path, but also further improves space utilization and reduces overall costs, further enhancing the intelligence level of the rack and the warehousing efficiency.

[0164] As a preferred technical solution, such as Figure 6 As shown, the method of using the smart storage rack also includes the following steps:

[0165] S7, Obtain the basic path cost item; wherein, the basic path cost item includes the cost from the starting point to the current node. Actual moving cost and the current node Estimated cost to the destination .

[0166] S8, Construct a spatiotemporal conflict prediction term; wherein, the spatiotemporal conflict prediction term is composed of a conflict penalty factor. Prediction time window length and spatiotemporal conflict indication function The spatiotemporal conflict indicator function is represented as the function when a node... exist Output 1 if there is always a risk of collision, otherwise output 0.

[0167] S9: Construct a cost function based on the basic path cost item and the spatiotemporal conflict prediction item, and obtain the planned path based on the cost function.

[0168] Specifically, the cost function Actual movement costs include, but are not limited to, distance, time, and energy consumption, and can be calculated by weighting the path length, movement time, and movement energy consumption.

[0169] Preferably, the estimated cost , Indicates the coordinates of the endpoint. This indicates the current node coordinates, i.e., the real-time position of the robotic arm's end effector, obtained through a joint encoder or vision positioning system. This represents the obstacle density influence coefficient, used to adjust the strength of the impact of obstacle density on path estimation. Its value typically ranges from 0.2 to 0.5. For low-risk scenarios (such as open passages), it can be... Set to 0.2 to make a slight correction to the path (increasing the path estimate by 20%). For high-risk scenarios (such as temporary storage areas), the value can be... Set to 0.5 to significantly correct the path (path estimation improved by 50%). The obstacle density influence coefficient can be dynamically adjusted according to different scenarios.

[0170] Obstacle density can be expressed by the formula , Indicates the preset radius; effective obstacles can be understood as those at the node. Surrounding radius Within a circular area, dynamic / static objects whose height is within the motion envelope of the robotic arm (e.g., 0.5~1.5m).

[0171] The traditional A* algorithm's cost prediction function only considers geometric distance, while the formula... Here, through the correction item Dynamic weighting of distance estimates: in areas with sparse obstacles It can maintain the original distance estimate and prioritize straight paths; however, in areas with dense obstacles... This can significantly improve the path estimation in the area, guiding the algorithm to detour. Therefore, based on the improved predicted cost function, it can not only reduce the probability of collisions and avoid high-risk areas in advance, but also reduce the expansion of invalid nodes. By dynamically adjusting the path estimation through quantifying environmental risks, it enables the robotic arm to achieve both efficiency and safety in complex scenarios.

[0172] The conflict penalty factor can be adjusted appropriately according to the actual scenario. For example, in a security-sensitive environment (such as human-computer collaboration), the value of the conflict penalty factor can be appropriately increased, such as by setting it to [value missing]. The value is set between 2.0 and 5.0. In efficiency-priority environments (such as pure machine environments), the value can be appropriately lowered, for example, by setting it to... Set it between 0.5 and 1.5.

[0173] Prediction time window length The unit is seconds, which determines the depth of the algorithm's prediction of future risks, such as short periods ( (seconds), suitable for high-speed sorting scenarios (robotic arm speed ≥ 2m / s), such as long cycle ( (seconds), then it is used for large-scale AGV cluster scheduling.

[0174] For spatiotemporal conflict indication function The collision conditions are: ;in, These represent the current state of the robotic arm. The predicted position at any given time, the predicted positions of other robotic arms / obstacles, and the safety radius. This indicates the start time of the path planning decision. For the spatiotemporal conflict indicator function collision condition, after each path planning iteration, Updated to the latest time window Scroll forward, It represents three-dimensional spatial distance and is used to characterize the actual distance between two objects in space.

[0175] The collision condition can be understood as: the existence of a certain time. Add from the current time to the current time Within the window, if the distance between the predicted position of the current robotic arm and the predicted positions of other robotic arms or obstacles is less than the safe radius, the output is 1; otherwise, the output is 0.

[0176] The specific methods for obtaining the planned path based on the cost function include: algorithm initialization and parameter setting (such as environment modeling, configuration of input features / output results / update frequency of the LSTM prediction module, setting parameters such as conflict penalty factor and prediction time window length in the cost function), conflict probability assessment, node expansion and cost calculation, priority queue management, and incremental path update. Conflict probability assessment includes predicting the trajectory of all adjacent robotic arms for the next 30 steps using LSTM, generating a spatiotemporal occupancy heatmap, and determining collision risk; node expansion and cost calculation includes calculating actual movement cost, predicted cost, and conflict penalty superposition; priority queue management includes using a binary heap to store nodes to be expanded, arranging them in ascending order by cost function, and evaluating the head node of the queue every 100ms. If the value of the cost function increases beyond a preset threshold due to environmental changes, path backtracking is triggered; incremental path update includes retaining the first 50% of the planned path and restarting the A* search from intermediate nodes when a new obstacle is detected or the prediction error is >5cm.

[0177] Here, the LSTM prediction results are transformed into a probabilistic occupancy heatmap to replace the traditional binary collision detection, which can improve the tolerance for uncertainty.

[0178] In summary, the path planning method based on the cost function predicts the trajectories of other robotic arms / AGVs within the next 3 seconds using an LSTM network, generates a spatiotemporal conflict heatmap, and adjusts the path cost function by incorporating a conflict penalty factor. This allows for the pre-generation of detour paths, avoiding sudden stops or path replanning, and further improves the intelligence level of the intelligent shelving usage method. Constructing the cost function based on the basic path cost term and the spatiotemporal conflict prediction term, and comprehensively considering multi-dimensional optimization objectives including distance, energy consumption, efficiency, and safety, enables better path optimization and resource conservation.

[0179] The technical features of the embodiments described can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0180] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A smart storage rack, comprising a rack body, a controller, and at least one feature acquisition module for acquiring feature information of goods to be stored, wherein the rack body has multiple storage slots, the controller stores historical access data of the storage slots, and the feature acquisition module is connected to the controller and feeds back feature information to the controller, characterized in that, The controller is also used for: The popularity value of each storage location is obtained based on historical access data, and an access popularity map is constructed based on the popularity values ​​of multiple storage locations. The overall popularity of the goods to be stored is obtained based on the historical access data and feature information. And allocate the optimal storage location based on the access heat map and the overall popularity of the goods; The controller includes: The access frequency acquisition module is used to acquire the overall access frequency and recent access frequency of the goods to be stored; The related popularity acquisition module is used to acquire the popularity of related goods; The overall popularity acquisition module is used to acquire the overall popularity of goods based on the overall access frequency, recent access frequency, and the popularity of related goods. Related goods popularity ;in, Indicates related goods Quantity, Indicates goods and The number of times the data is accessed in the current state. Indicates goods Total number of accesses, Indicates the time decay factor. Indicates the attenuation coefficient. This represents the time difference between the current time and the most recent co-occurrence. Indicates goods and The strength of semantic association, Indicates goods Independent popularity; , Representing goods and Word vectors, Represents the magnitude of the vector. , Representing goods The search popularity, discussion popularity, and dissemination popularity of [the topic]. Representing goods The weighting coefficients for search popularity, discussion popularity, and dissemination popularity. This indicates the time-based enhancement factor for search popularity. Indicates the time augmentation factor. Represents a time variable. This indicates the decay factor of the S-shaped curve of discussion popularity. The peak time of popularity. This represents the kurtosis parameter of the S-shaped function. This represents the logarithmic growth factor of the spread's popularity. This represents the logarithmic scaling factor. Indicates the interaction quality factor. Indicates user profile factors, These represent the interaction and profile weights, respectively. This represents the outlier filtering coefficient.

2. The intelligent storage rack as described in claim 1, characterized in that, The controller includes: The distance acquisition module is used to acquire the three-dimensional Euclidean distance between the storage location and each access location in the historical access events; The contribution acquisition module is used to acquire the contribution of each historical access event to the storage location based on the three-dimensional Euclidean distance. The heatmap construction module is used to weight and superimpose the contribution values ​​to obtain the heat value, and construct the access heatmap based on the heat value.

3. The intelligent storage rack as described in claim 2, characterized in that, The heatmap construction module is based on the formula Construct an access heatmap; in, Represents the three-dimensional coordinates of the storage location. Popularity value This represents the total number of historical access events. Represents the three-dimensional coordinates of the storage location. With the The three-dimensional Euclidean distance of the next access location. This represents the diffusion coefficient of the Gaussian kernel. Indicates the distance offset. Represents the natural constant. Indicates the first The contribution of each historical access event to the storage location.

4. The intelligent storage rack as described in claim 3, characterized in that, The comprehensive heat acquisition module is based on the formula Obtain the overall popularity of goods; in, These represent the overall access frequency, recent access frequency, and related goods popularity, respectively. These represent the overall access frequency, recent access frequency, and related goods popularity weighting coefficient, respectively.

5. The intelligent storage rack as described in claim 4, characterized in that, The intelligent shelving unit also includes a storage and retrieval robotic arm connected to the controller via a signal. The robotic arm is used to respond to control commands and perform order sorting. The controller also includes: The path planning module is used to plan the path according to the Q function after receiving the sorting order. Obtain the picking sequence and the planned path, and generate control instructions based on the picking sequence and the planned path; in, Indicates the current system status. Indicates the system at time... The running status, Indicates the currently selected action. Indicates the time when the access robot arm is at time Specific actions, Indicates the discount factor. Indicates the system at time... Instant rewards Represents the expectation operator. Indicates from the current moment Starting future time step index, This represents the reward function, which includes time efficiency rewards, energy consumption penalties, and collision risk penalties.

6. A method of using an intelligent storage rack, characterized in that, The method of using the smart storage rack includes: Obtain the characteristic information of the goods to be stored and the historical access data of the storage location; The heat value of each storage location is obtained based on the historical access data, and an access heat map is constructed based on the heat values ​​of multiple storage locations. The overall popularity of the goods to be stored is obtained based on the historical access data and feature information. Based on the access heat map and the overall popularity of the goods, the optimal storage location is allocated; The methods for obtaining the overall popularity of goods include: obtaining the overall access frequency, recent access frequency and related goods popularity of the goods to be stored; and obtaining the overall popularity of goods based on the overall access frequency, recent access frequency and related goods popularity. Related goods popularity ;in, Indicates related goods Quantity, Indicates goods and The number of times the data is accessed in the current state. Indicates goods Total number of accesses, Indicates the time decay factor. Indicates the attenuation coefficient. This represents the time difference between the current time and the most recent co-occurrence. Indicates goods and The strength of semantic association, Indicates goods Independent popularity; , Representing goods and Word vectors Represents the magnitude of the vector. , Representing goods The search popularity, discussion popularity, and dissemination popularity of [the topic]. Representing goods The weighting coefficients for search popularity, discussion popularity, and dissemination popularity. This indicates the time-based enhancement factor for search popularity. Indicates the time augmentation factor. Represents a time variable. This indicates the decay factor of the S-shaped curve of discussion popularity. The peak time of popularity. This represents the kurtosis parameter of the S-shaped function. This represents the logarithmic growth factor of the spread's popularity. This represents the logarithmic scaling factor. Indicates the interaction quality factor. Indicates user profile factors, These represent the interaction and profile weights, respectively. This represents the outlier filtering coefficient.

7. The method of using an intelligent storage rack as described in claim 6, characterized in that, Specific methods for constructing access heatmaps include: Obtain the three-dimensional Euclidean distance between the storage location and each access location in the historical access events. ; The contribution of each historical access event to the storage location is obtained based on the three-dimensional Euclidean distance. ; The contribution values ​​are weighted and summed to obtain the popularity value, and an access popularity map is constructed based on the popularity value; in, Represents the three-dimensional coordinates of the storage location. Popularity value This represents the total number of historical access events. This represents the diffusion coefficient of the Gaussian kernel. Indicates the distance offset. Represents the natural constant.

8. The method of using an intelligent storage rack as described in claim 7, characterized in that, Specific methods for obtaining the overall popularity of goods to be stored include: Obtain the overall access frequency of the goods to be stored Recent access frequency and related goods popularity ; The overall popularity of goods is obtained based on the overall access frequency, recent access frequency, and the popularity of related goods. ; in, These represent the weighting coefficients for the overall access frequency, recent access frequency, and related goods popularity, respectively.

9. A method of using an intelligent storage rack as described in claim 8, characterized in that, The method of using the smart storage rack also includes: After receiving the sorting order, according to the Q function Obtain the picking sequence and the planned path, and generate control instructions based on the picking sequence and the planned path; The robotic arm responds to control commands and performs order sorting; in, Indicates the current system status. Indicates the system at time... The running status, Indicates the currently selected action. Indicates the time when the access robot arm is at time Specific actions, Indicates the discount factor. Indicates the system at time... Instant rewards Represents the expectation operator. Indicates from the current moment Starting future time step index, This represents the reward function, which includes time efficiency rewards, energy consumption penalties, and collision risk penalties.

Images