WMS system warehouse dynamic management method and system based on Internet of Things

Abstract

The invention provides a WMS system storage dynamic management method and system based on the Internet of Things. The method comprises the steps of extracting historical logistics time sequence features from historical storage data after data cleaning based on a fusion attention time sequence model; the method comprises the following steps: carrying out multi-modal data acquisition and data cross verification on inventory materials based on an Internet of Things sensing layer to obtain real-time storage data, and extracting real-time logistics time sequence characteristics from the real-time storage data; carrying out logistics analysis on the real-time logistics time sequence characteristics based on the historical logistics time sequence characteristics to obtain analysis logistics data, and carrying out material dynamic allocation on the real-time storage data based on the analysis logistics data to obtain a material scheduling list; performing position category clustering on the warehouse digital twin model based on the real-time storage data to obtain a material distribution model; and carrying out position calibration and scheduling path optimization on the material scheduling list based on the material distribution model to obtain a scheduling path list, and carrying out dynamic management on inventory materials based on the scheduling path list.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of warehouse management, in particular to a WMS system warehouse dynamic management method and system based on Internet of Things. BACKGROUND

[0002] Warehouse Management System (WMS) as the core link of modern supply chain, its intelligent level is directly related to the operation efficiency and cost control of enterprises, the traditional warehouse management system mainly relies on information record, realizes the basic inventory management and order processing function, but there are obvious deficiencies in dynamic regulation and control, real-time decision-making, at present, many enterprises' warehouse management still faces systematic technical bottleneck. In the data acquisition link, the traditional way highly depends on manual input or simple barcode scanning, not only low efficiency, but also easy to cause inaccurate or delayed data due to human factors, it is difficult to realize real-time state tracking and verification of inventory materials, in the material allocation link, the traditional way often relies on manual experience for location planning, which is easy to cause material congestion during allocation, resulting in low operation efficiency and insufficient resource utilization of the existing warehouse management method. SUMMARY

[0003] (I) Technical problems solved In view of the defects of the prior art, the present application provides a WMS system warehouse dynamic management system based on Internet of Things, which has the advantages of accurate data perception, intelligent decision analysis and dynamic material scheduling optimization, and solves the problem of low warehouse management efficiency caused by low data entry accuracy and poor warehouse scheduling flexibility in warehouse management.

[0004] (II) Technical scheme In order to achieve the above purpose, the present application provides the following technical scheme: The present application provides a WMS system warehouse dynamic management method based on Internet of Things, comprising the following steps: The pre-acquired historical warehouse data is cleaned, and the historical logistics time sequence features are extracted from the cleaned historical warehouse data based on the fusion attention time sequence model; The multi-modal data of the inventory materials is collected and cross-verified based on the Internet of Things perception layer, the real-time warehouse data is obtained, and the real-time logistics time sequence features are extracted from the real-time warehouse data; The real-time logistics time sequence features are analyzed based on the historical logistics time sequence features, the analysis logistics data is obtained, and the real-time warehouse data is dynamically allocated based on the analysis logistics data, and the material scheduling list is obtained; construct a warehouse digital twin model based on the real-time warehouse data, and perform position and category clustering on the inventory in the warehouse digital twin model to obtain a material distribution model; based on the material distribution model, perform position calibration and scheduling path optimization on the material scheduling list to obtain a scheduling path list, and perform dynamic management on the inventory based on the scheduling path list.

[0005] According to one of the preferred embodiments of the present application, the historical logistics time sequence features are extracted from the historical warehouse data after data cleaning based on the fusion attention time sequence model, which includes: extract historical logistics data from the historical warehouse data after data cleaning, and perform feature embedding and feature normalization on the historical logistics data based on a preset time granularity to obtain a logistics material feature set; sort the logistics material feature set based on time sequence to obtain a historical material feature sequence; extract a material category feature set from the historical logistics data, and perform feature clustering on the material category feature set to obtain a material category feature cluster group; perform multi-scale time sequence convolution and residual connection on the historical material feature sequence to obtain a standard material feature sequence; perform time sequence attention coding on the standard material feature sequence based on the fusion attention time sequence model to obtain a material time sequence feature sequence; perform feature weighting on the material time sequence feature sequence based on the category attention weight of the material category feature cluster group to obtain an attention material time sequence feature sequence; extract historical logistics time sequence features from the attention material time sequence feature sequence based on the bidirectional long short-term memory network of the fusion attention time sequence model.

[0006] According to another preferred embodiment of the present application, the feature weighting on the material time sequence feature sequence based on the category attention weight of the material category feature cluster group to obtain an attention material time sequence feature sequence includes: map the material time sequence feature sequence and the material category feature cluster group to a unified vector space for dimension alignment to obtain a time sequence feature vector matrix and a category feature vector matrix; construct an attention query vector based on the time sequence feature vector matrix, and construct an attention key-value vector based on the category feature vector matrix; calculate the attention weight between the attention query vector and the attention key-value vector based on the scaled dot-product attention mechanism; Extract the feature variance contribution rate of each material category feature cluster in the material category feature cluster group, and aggregate the attention weight based on the feature variance contribution rate to obtain a category attention weight; Based on the category attention weight, the material time sequence feature sequence is subjected to feature weighted summation and feature regularization operation to obtain an attention material time sequence feature sequence.

[0007] According to another preferred embodiment of the present application, the logistics analysis based on the historical logistics time sequence feature on the real-time logistics time sequence feature obtains analysis logistics data, which includes: The historical logistics time sequence feature and the real-time logistics time sequence feature are subjected to time axis splicing and dimension expansion to obtain a standard logistics time sequence feature; The hidden state parameters of the bidirectional long short-term memory network in the fusion attention time sequence model are obtained, and the standard logistics time sequence feature is subjected to context-dependent decoding based on the hidden state parameters to obtain a context time sequence feature sequence; The context time sequence feature sequence is subjected to time sequence attention calculation to obtain a time sequence dependent weight; The context time sequence feature sequence is subjected to category attention calculation to obtain a category attention weight; The context time sequence feature sequence is subjected to double-weight weighted aggregation based on the time sequence dependent weight and the category attention weight to obtain an aggregated time sequence feature sequence; The aggregated time sequence feature sequence is subjected to time deconvolution analysis to obtain an analysis logistics feature sequence, and the analysis logistics feature sequence is subjected to feature de-normalization and numerical correction to obtain analysis logistics data.

[0008] According to another preferred embodiment of the present application, the multi-modal data acquisition and data cross-verification of the inventory material based on the Internet of Things perception layer obtains real-time warehouse data, which includes: The radio frequency identification component based on the Internet of Things perception layer performs radio frequency identification and time stamp labeling on the inventory material in and out of the warehouse and the inventory material on the shelf to obtain real-time radio frequency data; The bar code identification component based on the Internet of Things perception layer performs bar code identification and time stamp labeling on the inventory material in and out of the warehouse and the inventory material on the shelf to obtain real-time bar code data; The quality detection component based on the Internet of Things perception layer performs quality identification and time stamp labeling on the inventory material in and out of the warehouse and the inventory material on the shelf to obtain real-time quality data; The camera component based on the Internet of Things perception layer photographs and time stamp labels the inventory material in and out of the warehouse and the inventory material on the shelf to obtain real-time photo data; data alignment and data correlation are performed on the real-time radio frequency data, the real-time barcode data, the real-time quality data and the real-time photo data, to obtain real-time multi-modal data; data cross verification is performed on the real-time multi-modal data based on the unique identifier of the inventory material, to obtain real-time warehouse data.

[0009] According to another preferred embodiment of the present application, the real-time warehouse data is dynamically allocated based on the analysis logistics data, to obtain a material scheduling list, which comprises: real-time logistics data is extracted from the real-time warehouse data, and time axis alignment and data fusion are performed on the real-time logistics data and the analysis logistics data, to obtain standard logistics data; The standard logistics data is split into logistics data sequences based on a preset time granularity; real-time inventory data is extracted from the real-time warehouse data, and time alignment is performed on the real-time inventory data and the logistics data sequences, to obtain inventory data sequences; Inventory deviation sequences are obtained by performing inventory deviation calculation on the inventory data sequences based on the logistics data sequences; The scheduling priority of each inventory material in the real-time warehouse data is calculated, and the real-time warehouse data is dynamically allocated according to the inventory deviation sequences and the scheduling priority, to obtain a material scheduling list.

[0010] According to another preferred embodiment of the present application, the warehouse digital twin model is constructed based on the real-time warehouse data, which comprises: The pre-acquired warehouse point cloud data is subjected to Gaussian filtering and point cloud smoothing, to obtain smoothed point cloud data; The smoothed point cloud data is subjected to three-dimensional reconstruction and texture mapping, to obtain a primary warehouse model, and entity recognition is performed on the primary warehouse model, to obtain a secondary warehouse model; Physical properties are added to each entity in the secondary warehouse model, to obtain a warehouse physical model; The unique identifier, size data and quality data of each inventory material are extracted from the real-time warehouse data, and a material model set is constructed according to the size data and the quality data; Each material model in the material model set is assigned to the corresponding entity of the warehouse physical model based on the unique identifier, to obtain a warehouse digital twin model.

[0011] According to another preferred embodiment of the present application, the inventory materials in the warehouse digital twin model are subjected to location and category clustering, to obtain a material distribution model, which comprises: Multi-dimensional category feature extraction is performed on each inventory in the warehouse digital twin model to obtain a warehouse inventory category feature set; Multi-dimensional spatial feature extraction is performed on each inventory in the warehouse digital twin model to obtain a warehouse inventory location feature set; Feature standardization and feature fusion are performed on the warehouse inventory category feature set and the warehouse inventory location feature set to obtain an inventory location category feature set; Feature clustering is performed on the inventory location category feature set based on a clustering method optimized based on a contour coefficient to obtain a location category feature cluster group; Spatial correlation and category calibration are performed on each location category feature cluster in the location category feature cluster group to obtain a cluster group spatial region group and a region inventory category group; The warehouse digital twin model is divided into regions based on the cluster group spatial region group, and the warehouse digital twin model after the region division is annotated and mapped based on the region inventory category group to obtain an inventory distribution model.

[0012] According to another preferred embodiment of the present application, the location calibration and dispatch path optimization of the inventory dispatch list based on the inventory distribution model obtain a dispatch path list, which includes: The to-be-dispatched inventory and the corresponding dispatch direction are filtered from the inventory dispatch list in time sequence one by one; The initial location information of the to-be-dispatched inventory and the to-be-dispatched inventory category are obtained, and the location calibration of the inventory distribution model is performed based on the dispatch direction, the initial location information, and the to-be-dispatched inventory category to obtain destination location information; An initial path is generated on the inventory distribution model based on the initial location information and the destination location information, and all initial paths corresponding to the inventory dispatch list are collected into an initial path list; The initial path list is subjected to dispatch duration calculation to obtain an initial path duration set, and the initial path list is subjected to dispatch timestamp annotation based on the initial path duration set to obtain an annotated dispatch path list; The annotated dispatch path list is split according to a preset time period to obtain a dispatch path group sequence, and each dispatch path group in the dispatch path group sequence is subjected to path conflict detection to obtain a path conflict probability sequence; The initial path list is subjected to path iteration optimization based on the path conflict probability sequence to obtain a dispatch path list.

[0013] In order to achieve at least one of the above-mentioned purposes, the application further provides a warehouse dynamic management system based on a WMS system of Internet of Things, which comprises a feature extraction module, a real-time acquisition module, a material allocation module, a digital twin module and a dynamic scheduling module, wherein: The feature extraction module is used for data cleaning on the pre-acquired historical warehouse data, and extracting historical logistics time sequence features from the data cleaned historical warehouse data based on a fusion attention time sequence model. The real-time acquisition module is used for multi-modal data acquisition and data cross verification on the inventory materials based on the Internet of Things perception layer, obtaining real-time warehouse data, and extracting real-time logistics time sequence features from the real-time warehouse data. The material allocation module is used for logistics analysis on the real-time logistics time sequence features based on the historical logistics time sequence features, obtaining analysis logistics data, and dynamically allocating materials based on the analysis logistics data, obtaining a material scheduling list. The digital twin module is used for constructing a warehouse digital twin model based on the real-time warehouse data, and clustering the inventory materials in the warehouse digital twin model by location and category, obtaining a material distribution model. The dynamic scheduling module is used for location calibration and scheduling path optimization on the material scheduling list based on the material distribution model, obtaining a scheduling path list, and dynamically managing the inventory materials based on the scheduling path list.

[0014] The application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the above-mentioned warehouse dynamic management method based on a WMS system of Internet of Things.

[0015] (Three) beneficial effects Compared with the prior art, the application provides a warehouse dynamic management method and system based on a WMS system of Internet of Things, which has the following beneficial effects: The warehouse dynamic management method based on a WMS system of Internet of Things can realize efficient fusion of multi-dimensional features through the attention mechanism of category and time sequence interaction, guide the dynamic attention of time sequence features through category features, strengthen the response ability of key time slices to specific categories, and at the same time can suppress redundant features and highlight difference information, thereby improving the discriminability and time sequence sensitivity of the fusion attention time sequence model, and realizing more accurate logistics feature or event identification.

[0016] The WMS system warehouse dynamic management method based on the Internet of Things realizes digital-driven warehouse management and intelligent decision-making through time sequence splicing, attention mechanism and dynamic calculation, can predict the future logistics change situation in advance, thereby scheduling and optimizing the inventory materials, significantly improving the inventory turnover rate and reducing the operation cost, and improving the efficiency of warehouse management.

[0017] The WMS system warehouse dynamic management method based on the Internet of Things can intelligently match the shelves based on the clustering results through position calibration and scheduling path optimization, improve the scientific nature of the warehouse location allocation, reduce invalid scheduling movement, and through time sequence path scheduling and path conflict judgment, can quantify the channel occupancy rate to avoid congestion risk, improve the robustness and adaptability of the scheduling path, and further improve the efficiency of warehouse management. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A WMS system warehouse dynamic management method based on the Internet of Things is shown. DETAILED DESCRIPTION

[0019] The following description is used to disclose the present application so that those skilled in the art can implement the present application. The preferred embodiments in the following description are only used as examples, and other obvious modifications can be thought by those skilled in the art. The basic principles of the present application defined in the following description can be applied to other embodiments, modifications, improvements, equivalents and other technical solutions without departing from the spirit and scope of the present application.

[0020] It can be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of one element can be one, and in another embodiment, the number of the element can be multiple, and the term "one" cannot be understood as a limitation on the number.

[0021] Embodiment one: Please combine Figure 1 The present application discloses a WMS system warehouse dynamic management method based on the Internet of Things, the method comprises the following steps: The pre-acquired historical warehouse data is data cleaned, and the historical logistics time sequence features are extracted from the data cleaned historical warehouse data based on the fusion attention time sequence model.

[0022] The historical warehouse data is related data of warehouse-in and warehouse-out of inventory materials recorded by an Internet perception layer in a past time period, including warehouse-in time, warehouse-out time, inventory location, category, quality, bar code, Radio Frequency Identification (RFID) tag, and warehouse-in photo of each inventory material, and the fusion attention time sequence model includes a time sequence attention encoding layer and a bidirectional long short-term memory network.

[0023] In detail, in order to realize systematic management of the inventory, an inventory behavior model needs to be extracted from the historical warehouse data, so as to obtain a more accurate inventory management scheme. The pre-acquired historical warehouse data is subjected to data cleaning, including format unification and logical consistency verification of the pre-acquired historical warehouse data, and data deduplication, outlier screening, and missing value filling of the historical warehouse data after the logical consistency verification.

[0024] Specifically, the historical logistics time sequence features are extracted from the historical warehouse data after data cleaning based on the fusion attention time sequence model, including: The historical logistics data is extracted from the historical warehouse data after data cleaning, and the historical logistics data is subjected to feature embedding and feature normalization based on a preset time granularity, to obtain a logistics material feature set; The logistics material feature set is sorted based on time sequence, to obtain a historical material feature sequence; A material category feature set is extracted from the historical logistics data, and the material category feature set is subjected to feature clustering, to obtain a material category feature cluster group; The historical material feature sequence is subjected to multi-scale time sequence convolution and residual connection, to obtain a standard material feature sequence; The standard material feature sequence is subjected to time sequence attention encoding based on the fusion attention time sequence model, to obtain a material time sequence feature sequence; The material time sequence feature sequence is subjected to feature weighting based on a category attention weight of the material category feature cluster group, to obtain an attention material time sequence feature sequence; The historical logistics time sequence features are extracted from the attention material time sequence feature sequence based on a bidirectional long short-term memory network of the fusion attention time sequence model.

[0025] The historical logistics data is data corresponding to the warehouse and the warehouse-out of the warehouse materials in the historical logistics data, the time granularity can be one hour, half a day or one day, the feature embedding refers to vectorizing the storage data in each time granularity in the historical logistics data, including vectorizing each warehouse material and each data corresponding to the warehouse material, and the feature normalization is used to unify each feature of the historical logistics data after the feature embedding.

[0026] In the embodiment of the application, the extracting the material category feature set from the historical logistics data refers to extracting the category, value, turnover rate and residence time of each warehouse material in the historical logistics data, and taking the category, value, turnover rate and residence time as category features for feature embedding to obtain the material category feature set.

[0027] The feature clustering can be performed by using a k-nearest neighbor clustering algorithm or a density clustering algorithm, and each material category feature cluster corresponds to a variety of warehouse dynamic mode categories, such as a high turnover material category and a low warehouse fluctuation material category. The multi-scale time sequence convolution is performed by using convolution kernels of different time scales to convolve the historical material feature sequence, the time scale can be one day, one week or one month, the residual connection refers to residual connection of the convolution results of each time scale, thereby avoiding the problem of feature detail loss caused by multi-layer convolution and improving the ability to capture local time sequence patterns, and the time sequence attention encoding refers to calculating the attention weight distribution between the standard material features in each time step in the standard material feature sequence by using an attention mechanism, thereby highlighting the standard material features corresponding to the time steps that play a key role in the warehouse change trend.

[0028] In detail, the category attention weight based on the material category feature cluster group is used for feature weighting on the material time sequence feature sequence to obtain an attention material time sequence feature sequence, including: mapping the material time sequence feature sequence and the material category feature cluster group to a unified vector space for dimension alignment to obtain a time sequence feature vector matrix and a category feature vector matrix; constructing an attention query vector based on the time sequence feature vector matrix and constructing an attention key-value vector based on the category feature vector matrix; calculating the attention weight between the attention query vector and the attention key-value vector based on the scaled dot-product attention mechanism; extracting the feature variance contribution rate of each material category feature cluster in the material category feature cluster group, and weighting and aggregating the attention weight based on the feature variance contribution rate to obtain a category attention weight; The category attention weight is used for feature weighted summation and feature regularization operation on the material time sequence feature sequence, to obtain an attention material time sequence feature sequence.

[0029] The dimension alignment can be performed by using a shared embedding layer or a linear mapping layer, and the dimension-aligned time sequence feature vector matrix and category feature vector matrix are consistent in dimension, thereby facilitating subsequent feature weighting calculation; the constructing an attention query vector refers to multiplying the time sequence feature vector matrix by a preset query vector, the constructing an attention key-value vector refers to multiplying the category feature vector matrix by a preset key vector and value vector, the scaling dot product attention mechanism refers to dividing the product of the attention query vector and the transpose of the attention key vector by the dimension of the key vector before multiplying by the attention value vector, to realize scaling of the dimension, and the extracting the feature variance contribution rate of each material category feature cluster in the material category feature cluster group refers to performing principal component analysis on each material category feature cluster to obtain the total variance within each material category feature cluster, calculating the global total variance based on the total variance within each cluster, and finally calculating the corresponding feature variance contribution rate according to the ratio between the global total variance and the total variance within the cluster. Feature regularization can be performed by using Dropout and Layer Normalization technologies.

[0030] In detail, the category and time sequence interactive attention mechanism can realize efficient fusion of multi-dimensional features, dynamically focus on the time sequence features guided by the category features, and strengthen the response ability of key time slices to specific categories; at the same time, it can suppress redundant features and highlight difference information, thereby improving the discriminability and time sequence sensitivity of the fusion attention time sequence model, and realizing more accurate logistics feature or event recognition.

[0031] Based on the Internet of Things perception layer, multi-modal data of the inventory materials is collected and cross-verified, to obtain real-time warehouse data, and real-time logistics time sequence features are extracted from the real-time warehouse data.

[0032] The inventory material refers to the smallest storage unit in warehouse management, and can be express package, the real-time warehouse data is the Internet of Things sensing layer, which is used for sensing real-time inventory material changes through Internet of Things technology, and the existing material data usually only uses a bar code identification component to extract material data, and in actual use, a single information input method may have data abnormalities or misinputting conditions, so that multi-modal data collection and cross verification are required to improve the accuracy of data, the Internet of Things sensing layer includes a radio frequency identification component, a bar code identification component, a quality detection component and a camera component, the radio frequency identification component, the camera component, the bar code identification component and the quality detection component are arranged at the entrance and exit of the warehouse and the material shelf, the radio frequency identification component and the bar code identification component at the entrance and exit of the warehouse are used to identify the unique identifier of the inventory material entering and leaving the warehouse, and the camera component is used to take a photo of the inventory material, so that the inventory material information can be input in combination with the input time, and the category and quality attribute information of the inventory material can be matched according to the unique identifier, the radio frequency identification component, the bar code identification component, the camera component and the quality detection component arranged on the shelf can be used to verify the corresponding inventory material information again, and the number of the shelf can be used to collect the storage position of the inventory material.

[0033] In the embodiment of the application, the multi-modal data collection and data cross verification of the inventory material based on the Internet of Things sensing layer are carried out to obtain real-time warehouse data, including: The radio frequency identification component based on the Internet of Things sensing layer performs radio frequency identification and timestamp labeling on the inventory material entering and leaving the warehouse and the inventory material on the shelf to obtain real-time radio frequency data; The bar code identification component based on the Internet of Things sensing layer performs bar code identification and timestamp labeling on the inventory material entering and leaving the warehouse and the inventory material on the shelf to obtain real-time bar code data; The quality detection component based on the Internet of Things sensing layer performs quality identification and timestamp labeling on the inventory material entering and leaving the warehouse and the inventory material on the shelf to obtain real-time quality data; The camera component based on the Internet of Things sensing layer performs shooting and timestamp labeling on the inventory material entering and leaving the warehouse and the inventory material on the shelf to obtain real-time photo data; The real-time radio frequency data, the real-time bar code data, the real-time quality data and the real-time photo data are aligned and associated to obtain real-time multi-modal data; The real-time multi-modal data is cross-verified based on the unique identifier of the inventory material to obtain real-time warehouse data.

[0034] The data alignment refers to time window alignment according to time stamp data marked in the real-time radio frequency data, the real-time barcode data, the real-time quality data and the real-time photo data, that is, taking radio frequency data, barcode data, quality data and photos recorded at the same position in the same time window as the data of the same inventory material, the data correlation refers to data correlation according to unique identifiers obtained through radio frequency identification and barcode identification, and attribute information of the inventory material corresponding to the unique identifier, and the data cross-checking refers to verifying the unique identifier by comparing the unique identifiers obtained through radio frequency identification and barcode identification, realizing visual verification of the material by comparing the appearance of the warehouse-in photo, the inventory photo and the warehouse-out photo with the inventory material photo in the database, and verifying the quality by comparing the change of the quality data, so as to identify and filter out abnormal data, and ensure the accuracy of real-time warehouse data.

[0035] In detail, the extracting real-time logistics time sequence features from the real-time warehouse data refers to extracting real-time logistics data from the real-time warehouse data, and labeling the inventory state of the real-time logistics data, for example, labeling the state of "warehousing", "out of stock" and the like, so that the real-time logistics data has the same data format as the historical logistics data, and extracting real-time logistics time sequence features from the real-time logistics data based on a fusion attention time sequence model, the method of extracting real-time logistics time sequence features is consistent with the method of extracting historical logistics time sequence features, so as to facilitate the subsequent data input of logistics analysis.

[0036] Based on the historical logistics time sequence features, the real-time logistics time sequence features are analyzed to obtain analysis logistics data, and based on the analysis logistics data, the real-time warehouse data is dynamically allocated to obtain a material scheduling list.

[0037] In the embodiment of the application, the existing logistics analysis method often performs logistics analysis based on experience, that is, analyzes and predicts the number of in and out of stock materials in a future time period according to historical material management experience, so as to adjust the scheduling strategy of the inventory material, in actual use, the logistics analysis method based on management experience has a large error, and it is difficult to accurately analyze different types of materials, so that the efficiency of inventory management is low.

[0038] In detail, the logistics analysis based on the historical logistics time sequence features and the real-time logistics time sequence features to obtain analysis logistics data comprises: Time axis splicing and dimension expansion are performed on the historical logistics time sequence features and the real-time logistics time sequence features to obtain standard logistics time sequence features; obtaining the hidden state parameters of the bidirectional long short-term memory network in the fusion attention time sequence model, and performing context-dependent decoding on the standard logistics time sequence features based on the hidden state parameters to obtain a context time sequence feature sequence; performing time sequence attention calculation on the context time sequence feature sequence to obtain time sequence dependent weights; performing category attention calculation on the context time sequence feature sequence to obtain category attention weights; performing double-weight weighted aggregation on the context time sequence feature sequence based on the time sequence dependent weights and the category attention weights to obtain an aggregated time sequence feature sequence; performing time deconvolution analysis on the aggregated time sequence feature sequence to obtain an analysis logistics feature sequence, and performing feature de-normalization and numerical correction on the analysis logistics feature sequence to obtain analysis logistics data.

[0039] The time axis splicing refers to splicing the historical logistics time sequence features and the real-time logistics time sequence features in time axis order, the dimension expansion refers to dimension alignment and unification of the historical logistics time sequence features and the real-time logistics time sequence features after time axis splicing, the hidden state parameters are hidden states and cell state parameters of the bidirectional long short-term memory network when encoding the historical logistics time sequence features, the context-dependent decoding of the standard logistics time sequence features based on the hidden state parameters refers to the initial state of an initial state of a time sequence decoder based on the hidden state parameters, the time sequence decoder can be a pre-trained long short-term memory (LSTM) model, the time sequence attention calculation, the category attention calculation, and the double-weight weighted aggregation are performed in each time step of the context-dependent decoding, the methods of the time sequence attention calculation and the category attention calculation are the same as those of the time sequence attention encoding and the category attention calculation in the above steps, which will not be described here, the time deconvolution analysis refers to using a multi-layer time deconvolution network opposite to the multi-scale time sequence convolution and the residual connection to upsample and expand features of the aggregated time sequence feature sequence, the method of the feature de-normalization is opposite to that of the feature normalization, the numerical correction refers to necessary numerical correction based on business rules, for example, ensuring that the number of outbound is less than the total amount of inventory, and the analysis logistics data is related data of inventory materials in and out in a future time period.

[0040] In detail, the real-time warehouse data is dynamically allocated based on the analysis logistics data to obtain a material scheduling list, including: extracting real-time logistics data from the real-time warehouse data, and performing time axis alignment and data fusion on the real-time logistics data and the analysis logistics data to obtain standard logistics data; splitting the standard flow data into flow data sequences based on a preset time granularity; extracting real-time inventory data from the real-time warehouse data, and performing time alignment on the real-time inventory data and the flow data sequences to obtain inventory data sequences; performing inventory deviation calculation on the inventory data sequences based on the flow data sequences to obtain inventory deviation sequences; calculating scheduling priorities of each inventory material in the real-time warehouse data, and performing dynamic deployment on the real-time warehouse data according to the inventory deviation sequences and the scheduling priorities to obtain a material scheduling list.

[0041] In detail, the real-time flow data is the in-out warehouse data in the real-time warehouse data, the real-time inventory data is the inventory material related data in the real-time warehouse data in the inventory state, the inventory deviation calculation refers to determining inventory gap data or inventory excess data according to the total amount of warehouse inventory, combining the in-out warehouse data amount and the inventory quantity of each time granularity, and collecting the inventory gap data or the inventory excess data as the inventory deviation to form the inventory deviation sequence, the scheduling priority can be calculated by weighted fusion according to the inventory time, the warehouse cost and the average turnover rate of each inventory material, the dynamic deployment refers to taking out the inventory material with inventory excess in each inventory deviation sequence or putting in the inventory material with inventory gap in order of the scheduling priority, to generate a scheduling scheme list for each inventory material that needs to be scheduled, and the material scheduling list contains the scheduling time period and the scheduling direction of each inventory material, wherein the scheduling direction includes in-out warehouse, for example, the inventory material "0-2025-3-67849" needs to be taken out in the time period from 10:30 to 11:00 on April 12, 2025.

[0042] Among them, through time sequence splicing, attention mechanism and dynamic calculation, digital driven warehouse management and intelligent decision are realized, which can predict the future logistics change in advance, so as to optimize the scheduling of inventory materials, thereby significantly improving the inventory turnover rate and reducing the operating cost, and improving the efficiency of warehouse management.

[0043] constructing a warehouse digital twin model based on the real-time warehouse data, and performing position and category clustering on the inventory materials in the warehouse digital twin model to obtain a material distribution model.

[0044] In the prior art, in the embodiments of the present application, the inventory is usually stored based on unique identifiers, and the unique identifiers are usually identifiers generated according to the entry timestamp, which is difficult to express the category association between the inventories, and may cause repeated search in multiple storage blocks when each inventory is taken out, thereby affecting the warehouse management efficiency.

[0045] The warehouse digital twin model is a digital twin model of the warehouse for warehouse management. The digital twin model is an equivalent information model of a physical entity in a digital space. The warehouse digital twin model can intuitively reflect the physical information of the warehouse and the inventories in the warehouse.

[0046] In detail, the warehouse digital twin model is constructed based on the real-time warehouse data, including: The pre-acquired warehouse point cloud data is subjected to Gaussian filtering and point cloud smoothing to obtain smoothed point cloud data; The smoothed point cloud data is subjected to three-dimensional reconstruction and texture mapping to obtain a primary warehouse model, and entity recognition is performed on the primary warehouse model to obtain a secondary warehouse model; Physical properties are added to each entity in the secondary warehouse model to obtain a warehouse physical model; The unique identifier, size data and quality data of each inventory are extracted from the real-time warehouse data, and a set of inventory models is constructed based on the size data and the quality data; Each inventory model in the set of inventory models is assigned to the corresponding entity of the warehouse physical model based on the unique identifier to obtain the warehouse digital twin model.

[0047] The warehouse point cloud data is obtained by point cloud scanning of the warehouse during idle time, the entity recognition is semantic recognition and labeling of spatial structures such as main roads and partition roads in the primary warehouse model and entity devices such as shelves, and can be performed by using a geometric segmentation or deep learning model method; the physical properties refer to the length, width, height and maximum weight that can be carried of each entity, the unique identifier is obtained according to real-time radio frequency data and real-time barcode data, the size data is obtained by comparison and analysis according to real-time photo data, and the quality data is extracted from real-time quality data; the construction of the material model set according to the size data and the quality data refers to establishing a primary material model according to the size data of each inventory material, and adding quality data and size data to each primary material model based on quality data to obtain a material model; and the distribution of each material model in the material model set to the corresponding entity in the warehouse physical model based on the unique identifier refers to corresponding distribution of each unique identifier and the identifier corresponding to the shelf entity in the warehouse physical model, and distribution of the corresponding material model to the corresponding shelf entity.

[0048] In the embodiment of the application, the position category clustering of the inventory materials in the warehouse digital twin model to obtain a material distribution model comprises: Multi-dimensional category feature extraction is performed on each inventory material in the warehouse digital twin model to obtain a warehouse material category feature set; Multi-dimensional spatial feature extraction is performed on each inventory material in the warehouse digital twin model to obtain a warehouse material position feature set; Feature standardization and feature fusion are performed on the warehouse material category feature set and the warehouse material position feature set to obtain a material position category feature set; Feature clustering is performed on the material position category feature set based on a contour coefficient optimization clustering method to obtain a position category feature cluster group; Spatial correlation and category calibration are performed on each position category feature cluster in the position category feature cluster group to obtain a cluster group spatial region group and a region material category group; The warehouse digital twin model is divided into regions based on the cluster group spatial region group, and the warehouse digital twin model after region division is annotated and mapped based on the region material category group to obtain a material distribution model.

[0049] The multi-dimensional category feature extraction refers to extracting dimensions such as the category, value, turnover rate and residence time of each inventory material, and collecting into warehouse material category features, the multi-dimensional space feature extraction refers to extracting dimensions such as the three-dimensional coordinates of the shelf and the distance from the exit of each inventory material, and collecting into warehouse material location features, the clustering method based on contour coefficient optimization for feature clustering refers to clustering based on the K-neighbor clustering method, and filtering out the optimal clustering method for feature clustering based on the contour coefficient of various clustering results, the space correlation refers to mapping the inventory material group corresponding to each location category feature cluster in space, and taking the shelf space area with a high inventory material proportion as the corresponding cluster group space area, and the category calibration refers to taking the multi-dimensional category attribute corresponding to the feature of the clustering center of each location category feature cluster as the corresponding area material category.

[0050] In detail, by constructing a warehouse digital twin model, visual warehouse management can be realized, physical properties can be given to the warehouse model, the accuracy of subsequent scheduling path optimization can be improved, by performing location category clustering, the existing inventory materials in the warehouse can be regionally divided according to the category attribute and the spatial position, thereby forming an intuitive business and space linkage distribution, facilitating subsequent storage location optimization and scheduling path design, and improving the efficiency of warehouse management.

[0051] Based on the material distribution model, the location calibration and scheduling path optimization of the material scheduling list are performed to obtain a scheduling path list, and the inventory materials are dynamically managed based on the scheduling path list.

[0052] Among them, the existing warehouse management method only dynamically adjusts the in-out quantity of inventory materials, and does not refine the actual scheduling path planning for each inventory material, and the inventory materials often appear channel congestion during actual scheduling, causing the efficiency of warehouse management to be reduced.

[0053] Specifically, the location calibration and scheduling path optimization of the material scheduling list based on the material distribution model to obtain a scheduling path list, comprising: According to the time sequence, the to-be-scheduled materials and the corresponding scheduling direction are filtered out from the material scheduling list one by one; The initial location information of the to-be-scheduled material and the to-be-scheduled material category are obtained, and the material distribution model is calibrated based on the scheduling direction, the initial location information and the to-be-scheduled material category to obtain destination location information; Based on the initial location information and the destination location information, an initial path is generated on the material distribution model, and all initial paths corresponding to the material scheduling list are collected into an initial path list; perform scheduling duration calculation on the initial path list to obtain an initial path duration set, and perform scheduling timestamp marking on the initial path list based on the initial path duration set to obtain a marked scheduling path list; split the marked scheduling path list according to a preset time period to obtain a scheduling path group sequence, and perform path conflict detection on each scheduling path group in the scheduling path group sequence to obtain a path conflict probability sequence; perform path iteration optimization on the initial path list based on the path conflict probability sequence to obtain a scheduling path list.

[0054] In detail, the to-be-scheduled material refers to the inventory material in the material scheduling list that needs to be scheduled, the scheduling direction includes outbound and inbound, and the initial position information refers to the initial position of the to-be-scheduled material. For example, when the inventory material is inbound, the warehouse entrance is the initial position information, and when the inventory material is in the inventory, the corresponding shelf position is the corresponding initial position information.

[0055] Specifically, the position calibration of the material distribution model based on the scheduling direction, the initial position information, and the to-be-scheduled material category to obtain the destination position information refers to determining that the to-be-scheduled material is an outbound path or an inbound path according to the scheduling direction. When it is an outbound path, the warehouse exit is taken as the destination position information. When it is an inbound path, a shelf with appropriate size and capable of bearing the weight of the to-be-scheduled material is selected from the corresponding cluster space region of the material distribution model as the destination position information according to the to-be-scheduled material category.

[0056] In detail, the initial path can be generated on the material distribution model based on the initial position information and the destination position information by using the Dijkstra algorithm. The scheduling duration calculation refers to calculating the initial path duration according to the moving speed of the warehouse robot or the average moving speed of the warehouse worker and the path length of each initial path in the initial path list. The scheduling timestamp marking refers to calculating the scheduling end timestamp according to the corresponding initial path duration on the basis of the scheduling start timestamp corresponding to each initial path list, and marking the scheduling start timestamp and the scheduling end timestamp on the corresponding initial path.

[0057] Wherein, each scheduling path group in the scheduling path group sequence is a path that needs to be scheduled in the same time period, the path conflict detection refers to judging whether each scheduling path in the scheduling path group has an intersection path, calculating a channel occupancy rate according to the channel width of the corresponding intersection path in the material distribution model and the size information of the corresponding inventory material, and calculating a path conflict probability according to the channel occupancy rate, the path iteration optimization of the initial path list based on the path conflict probability sequence refers to taking the scheduling path group with a path conflict probability greater than a preset probability threshold in the path conflict probability sequence as a conflict scheduling path group, and optimizing the path of the conflict scheduling path group based on an ant colony algorithm, and then re-performing path conflict detection until there is no path conflict probability greater than the preset probability threshold in the path conflict probability sequence, the corresponding initial path list is taken as a scheduling path list.

[0058] In detail, the dynamic management of the inventory material based on the scheduling path list refers to realizing the warehouse-in and warehouse-out operation of the inventory material according to the scheduling path list by time and by path, and updating the scheduling path list in real time, realizing dynamic circulation warehouse management, through position calibration and scheduling path optimization, the goods shelf can be intelligently matched based on the clustering result, the scientificity of the warehouse location allocation is improved, invalid scheduling movement is reduced, and through time sequence path scheduling and path conflict judgment, the channel occupancy rate can be quantified to avoid congestion risk, improve the robustness and adaptability of the scheduling path, and further improve the efficiency of warehouse management.

[0059] Embodiment two: The application discloses a WMS system warehouse dynamic management system based on an Internet of Things, and the system comprises a feature extraction module, a real-time acquisition module, a material allocation module, a digital twin module and a dynamic scheduling module, wherein: The feature extraction module is used for data cleaning on the pre-acquired historical warehouse data, and extracting historical logistics time sequence features from the data cleaned historical warehouse data based on a fusion attention time sequence model. The real-time acquisition module is used for multi-modal data acquisition and data cross verification on the inventory material based on an Internet of Things perception layer, obtaining real-time warehouse data, and extracting real-time logistics time sequence features from the real-time warehouse data. The material allocation module is used for logistics analysis on the real-time logistics time sequence features based on the historical logistics time sequence features, obtaining analysis logistics data, and dynamically allocating materials on the real-time warehouse data based on the analysis logistics data, obtaining a material scheduling list. The digital twin module is used for constructing a warehouse digital twin model based on the real-time warehouse data, and clustering the inventory material in the warehouse digital twin model by position and category, obtaining a material distribution model. The dynamic scheduling module is used to perform location marking and scheduling path optimization on the material scheduling list based on the material distribution model, obtain a scheduling path list, and perform dynamic management of the inventory materials based on the scheduling path list.

[0060] The processes described above with reference to the flowcharts in the embodiments disclosed in this invention can be implemented as computer software programs. The embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wire segments, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless segments, wire segments, optical fibers, RF, etc., or any suitable combination thereof.

[0061] The computer program product of the present application can be a computer program product comprising a computer-readable medium bearing computer program code embodied therein for use with a computer. The computer program code can be code defining and / or implementing the present application. The computer program code can be written in any suitable computer readable programming language. The computer program code can be stored in a computer- readable storage medium, such as, but not limited to, any type of disk including an optical disk, a CD-ROM, a CD-R, a CD-RW, a DVD, a flash memory, a ROM, a RAM, a magnetic disk or hard drive, or any other suitable type of medium including a medium that holds the software for a particular or specialized computing purpose, or any suitable combination of media. The computer program product can be a computer program product distributed to end users, whether as a stand-alone program, as part of a physical system, or as a software download. The computer program product can be distributed on a physical medium, such as, but not limited to, a floppy disk, a CD-ROM, a CD-R, a CD-RW, a DVD, a flash memory, a ROM, a RAM, a magnetic disk or hard drive, or any other suitable type of medium, or any suitable combination of media. The computer program product can be distributed from a program distribution center, either as a tangible medium or via electronic delivery, such as from a Web site via the Internet, or from one computer to another via electronic transfer, such as by e-mail. The computer program product can be distributed in an encrypted manner, such as via encryption or via password protection.

[0062] Those skilled in the art will understand that the application described above and illustrated in the accompanying drawings is presented by way of example only and is not limiting as to the present application. The intent is to cover all modifications and alternatives of the present application falling within the scope of the application.

Claims

1. A method for dynamic warehouse management in a WMS system based on the Internet of Things, characterized in that, The method comprises: data cleaning on pre-acquired historical warehouse data, and extracting historical logistics time sequence features from the data-cleaned historical warehouse data based on a fusion attention time sequence model; multi-modal data collection and data cross-verification of inventory materials based on an Internet of Things sensing layer to obtain real-time warehouse data, and extracting real-time logistics time sequence features from the real-time warehouse data; logistics analysis on the real-time logistics time sequence features based on the historical logistics time sequence features to obtain analysis logistics data, and dynamic allocation of inventory materials based on the analysis logistics data to obtain a material scheduling list; construction of a warehouse digital twin model based on the real-time warehouse data, and position and category clustering of inventory materials in the warehouse digital twin model to obtain a material distribution model; position calibration and scheduling path optimization of the material scheduling list based on the material distribution model to obtain a scheduling path list, and dynamic management of the inventory materials based on the scheduling path list. 2.The warehouse dynamic management method based on the WMS system of the Internet of Things according to claim 1, wherein, The method comprises: extracting historical logistics data from the data-cleaned historical warehouse data, and performing feature embedding and feature normalization on the historical logistics data based on a preset time granularity to obtain a logistics material feature set; sorting the logistics material feature set based on time sequence to obtain a historical material feature sequence; extracting a material category feature set from the historical logistics data, and performing feature clustering on the material category feature set to obtain a material category feature cluster group; performing multi-scale time sequence convolution and residual connection on the historical material feature sequence to obtain a standard material feature sequence; performing time sequence attention coding on the standard material feature sequence based on a fusion attention time sequence model to obtain a material time sequence feature sequence; performing feature weighting on the material time sequence feature sequence based on category attention weights of the material category feature cluster group to obtain an attention material time sequence feature sequence; extracting historical logistics time sequence features from the attention material time sequence feature sequence based on a bidirectional long short-term memory network of the fusion attention time sequence model. 3.The warehouse dynamic management method based on the WMS system of the Internet of Things according to claim 2, characterized in that, The method comprises: mapping the material time sequence feature sequence and the material category feature cluster group to a unified vector space for dimension alignment to obtain a time sequence feature vector matrix and a category feature vector matrix; constructing an attention query vector based on the time sequence feature vector matrix, and constructing an attention key-value vector based on the category feature vector matrix; calculating attention weights between the attention query vector and the attention key-value vector based on a scaled dot-product attention mechanism; extracting feature variance contribution rates of each material category feature cluster in the material category feature cluster group, and performing weighted aggregation on the attention weights based on the feature variance contribution rates to obtain category attention weights; perform feature regularization operation on the material time sequence feature sequence based on the category attention weight to obtain an attention material time sequence feature sequence. 4.The warehouse dynamic management method based on the WMS system of the Internet of Things according to claim 2, characterized in that, The logistics analysis based on the historical logistics time sequence feature and the real-time logistics time sequence feature to obtain analysis logistics data, comprising: The time axis splicing and dimension expansion are performed on the historical logistics time sequence feature and the real-time logistics time sequence feature to obtain standard logistics time sequence feature. The hidden state parameters of the bidirectional long short-term memory network in the fusion attention time sequence model are obtained, and context time sequence feature sequence is obtained by decoding the standard logistics time sequence feature based on the hidden state parameters. The time sequence attention calculation is performed on the context time sequence feature sequence to obtain time sequence dependent weight. The category attention calculation is performed on the context time sequence feature sequence to obtain category attention weight. The double-weight weighted aggregation is performed on the context time sequence feature sequence based on the time sequence dependent weight and the category attention weight to obtain aggregated time sequence feature sequence. The analysis logistics feature sequence is obtained by performing time deconvolution analysis on the aggregated time sequence feature sequence, and the analysis logistics data is obtained by performing feature de-normalization and numerical correction on the analysis logistics feature sequence. 5.The warehouse dynamic management method based on the WMS system of the Internet of Things according to claim 1, wherein, The multi-modal data acquisition and data cross verification are performed on the inventory material based on the Internet of Things perception layer to obtain real-time warehouse data, comprising: The radio frequency identification and timestamp labeling are performed on the inventory material in and out of the warehouse and the inventory material on the shelf based on the radio frequency identification component of the Internet of Things perception layer to obtain real-time radio frequency data; The bar code identification and timestamp labeling are performed on the inventory material in and out of the warehouse and the inventory material on the shelf based on the bar code identification component of the Internet of Things perception layer to obtain real-time bar code data; The quality identification and timestamp labeling are performed on the inventory material in and out of the warehouse and the inventory material on the shelf based on the quality detection component of the Internet of Things perception layer to obtain real-time quality data; The shooting and timestamp labeling are performed on the inventory material in and out of the warehouse and the inventory material on the shelf based on the camera component of the Internet of Things perception layer to obtain real-time photo data; The data alignment and data association are performed on the real-time radio frequency data, the real-time bar code data, the real-time quality data and the real-time photo data to obtain real-time multi-modal data; The data cross verification is performed on the real-time multi-modal data based on the unique identifier of the inventory material to obtain real-time warehouse data. 6.The warehouse dynamic management method based on the WMS system of the Internet of Things according to claim 1, wherein, The material dynamic allocation is performed on the real-time warehouse data based on the analysis logistics data to obtain a material scheduling list, comprising: The real-time logistics data is extracted from the real-time warehouse data, and the time axis alignment and data fusion are performed on the real-time logistics data and the analysis logistics data to obtain standard logistics data; The standard logistics data is split into logistics data sequence based on a preset time granularity; The real-time inventory data is extracted from the real-time warehouse data, and the time alignment is performed on the real-time inventory data and the logistics data sequence to obtain inventory data sequence; perform inventory bias calculation on the inventory data sequence based on the logistics data sequence to obtain an inventory bias sequence; calculate a scheduling priority of each inventory material in the real-time warehouse data, and perform dynamic allocation on the real-time warehouse data based on the inventory bias sequence and the scheduling priority to obtain a material scheduling list. 7.The warehouse dynamic management method based on the WMS system of the Internet of Things according to claim 1, characterized in that, constructing a warehouse digital twin model based on the real-time warehouse data, comprising: performing Gaussian filtering and point cloud smoothing on the pre-acquired warehouse point cloud data to obtain smoothed point cloud data; performing three-dimensional reconstruction and texture mapping on the smoothed point cloud data to obtain a primary warehouse model, and performing entity recognition on the primary warehouse model to obtain a secondary warehouse model; adding physical properties to each entity in the secondary warehouse model to obtain a warehouse physical model; extracting unique identifiers, size data and quality data of each inventory material from the real-time warehouse data, and constructing a material model set based on the size data and the quality data; assigning each material model in the material model set to the corresponding entity of the warehouse physical model based on the unique identifier to obtain a warehouse digital twin model. 8.The warehouse dynamic management method based on the WMS system of the Internet of Things according to claim 1, wherein, performing position and category clustering on the inventory materials in the warehouse digital twin model to obtain a material distribution model, comprising: performing multi-dimensional category feature extraction on each inventory material in the warehouse digital twin model to obtain a warehouse material category feature set; performing multi-dimensional spatial feature extraction on each inventory material in the warehouse digital twin model to obtain a warehouse material location feature set; performing feature standardization and feature fusion on the warehouse material category feature set and the warehouse material location feature set to obtain a material location and category feature set; performing feature clustering on the material location and category feature set based on a clustering method optimized by contour coefficients to obtain a position and category feature cluster group; performing spatial correlation and category calibration on each position and category feature cluster in the position and category feature cluster group to obtain a cluster group spatial region group and a region material category group; performing region division on the warehouse digital twin model based on the cluster group spatial region group, and performing category labeling and mapping correlation on the warehouse digital twin model after region division based on the region material category group to obtain a material distribution model. 9.The warehouse dynamic management method based on the WMS system of the Internet of Things according to claim 1, wherein, performing position calibration and scheduling path optimization on the material scheduling list based on the material distribution model to obtain a scheduling path list, comprising: filtering out a to-be-scheduled material and a corresponding scheduling direction from the material scheduling list one by one in chronological order; obtaining initial position information of the to-be-scheduled material and a to-be-scheduled material category, and performing position calibration on the material distribution model based on the scheduling direction, the initial position information and the to-be-scheduled material category to obtain destination position information; generating an initial path on the material distribution model based on the initial position information and the destination position information, and collecting all initial paths corresponding to the material scheduling list into an initial path list; The initial path list is subjected to scheduling duration calculation to obtain an initial path duration set, and the initial path list is subjected to scheduling timestamp marking based on the initial path duration set to obtain a marked scheduling path list; The marked scheduling path list is split according to a preset time period to obtain a scheduling path group sequence, and each scheduling path group in the scheduling path group sequence is subjected to path conflict detection to obtain a path conflict probability sequence; The initial path list is subjected to path iteration optimization based on the path conflict probability sequence to obtain a scheduling path list.

10. An IoT-based WMS system warehouse dynamic management system, characterized in that, The system comprises a feature extraction module, a real-time acquisition module, a material allocation module, a digital twin module, and a dynamic scheduling module, wherein: The feature extraction module is configured to clean historical warehouse data obtained in advance, and extract historical logistics time sequence features from the cleaned historical warehouse data based on a fusion attention time sequence model; The real-time acquisition module is configured to acquire multi-modal data of the inventory materials and cross-check the data based on an Internet of Things sensing layer, to obtain real-time warehouse data, and extract real-time logistics time sequence features from the real-time warehouse data; The material allocation module is configured to analyze logistics based on the historical logistics time sequence features and the real-time logistics time sequence features, to obtain analysis logistics data, and dynamically allocate materials based on the analysis logistics data and the real-time warehouse data, to obtain a material scheduling list; The digital twin module is configured to construct a warehouse digital twin model based on the real-time warehouse data, and cluster the inventory materials in the warehouse digital twin model according to positions and categories, to obtain a material distribution model; The dynamic scheduling module is configured to calibrate positions and optimize scheduling paths based on the material distribution model and the material scheduling list, to obtain a scheduling path list, and dynamically manage the inventory materials based on the scheduling path list.

Images