An artificial intelligence-based dense library visualization modeling method and system

Abstract

The application provides an artificial intelligence-based dense warehouse visualization modeling method and system, which comprises the following steps: constructing a five-dimensional feature system, acquiring corresponding cargo feature data in real time according to the five-dimensional feature system, and generating a standardized cargo feature vector; adopting an improved deep clustering algorithm to dynamically cluster the cargo feature vector, automatically generating a twin model adapted to different cargo types, embedding logical parameters about cargo operation rules in the twin model, and forming a two-dimensional model library; collecting cargo constraints, space constraints and equipment constraints of the dense warehouse in real time, assigning weights based on the AHP (analytic hierarchy process) method, and then adopting a fusion algorithm to determine the optimal model matched with the cargo in the two-dimensional model library; constructing a three-dimensional twin model of the dense warehouse, collecting the three-dimensional space coordinates of the cargo in real time, and mapping the optimal model to the three-dimensional twin model in real time according to the three-dimensional space coordinates, and finally realizing dynamic visualization display of the real-time situation of the dense warehouse.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of visual modeling, and particularly relates to a dense warehouse visual modeling method and system based on artificial intelligence. BACKGROUND

[0002] In the field of warehouse management, dense warehouse is a warehouse mode with the core design goal of "maximizing the use of warehouse space and improving the storage capacity per unit area", and its core logic is to optimize the layout of shelves, storage equipment or operation process, reduce the space waste caused by channels, gaps and the like in traditional warehouses, and make the "space density" (i.e. the amount of goods that can be stored per unit area / volume) of the warehouse reach a higher level.

[0003] At present, the visual modeling of dense warehouse is only based on fixed goods types (such as "box type / pallet type") to build a static model library, which cannot adapt to the characteristics of variable goods sizes, large density differences and irregular package shapes. Further, when the corresponding model is called from the static model library, the model is usually called by a single goods type tag, which is prone to the problem of mismatch between the model and the actual scene. Furthermore, after calling the model, the existing technology adopts a visual method of static map + goods icon, which cannot intuitively show the spatial relationship between the goods and the dense warehouse structure and equipment, resulting in poor user experience. SUMMARY

[0004] Therefore, the embodiment of the present application provides a dense warehouse visual modeling method and system based on artificial intelligence, which aims to accurately dynamically visualize the real-time situation of the dense warehouse.

[0005] The first aspect of the embodiment of the present application provides a dense warehouse visual modeling method based on artificial intelligence, which comprises:

[0006] constructing a five-dimensional feature system, acquiring corresponding goods feature data in real time according to the five-dimensional feature system, and generating a standardized goods feature vector, wherein the five-dimensional feature system comprises physical features, packaging features, logistics features, safety features and dense warehouse adaptation features;

[0007] adopting an improved deep clustering algorithm to dynamically cluster the goods feature vector, automatically generating a twin model adapted to different goods types, and embedding logical parameters about goods operation rules in the twin model to form a two-dimensional model library, wherein the improved deep clustering algorithm is a DBSCAN algorithm based on dense warehouse structure constraints;

[0008] Real-time collection of goods constraints, space constraints and equipment constraints of the dense library, and based on AHP hierarchical analysis method to allocate weights, and then adopt fusion algorithm, determine the optimal model matched with the goods in the two-dimensional model library;

[0009] Construct a three-dimensional twin model of the dense library, real-time collection of three-dimensional spatial coordinates of the goods, and according to the three-dimensional spatial coordinates, real-time mapping of the optimal model to the three-dimensional twin model.

[0010] Further, the physical characteristics include size, weight and density, the packaging features include material, hardness and sealing, the logistics features include turnover frequency, storage time and loading and unloading requirements, the safety features include fragility, corrosiveness and explosion-proof level, and the dense library adaptation features include shelf load-bearing matching degree and access mechanical arm compatibility.

[0011] Further, in the step of adopting an improved deep clustering algorithm to dynamically cluster the goods feature vector, automatically generating a twin model adapted to different goods types, and embedding logical parameters about goods operation rules in the twin model to form a two-dimensional model library, the goods feature vector is dynamically clustered with the shelf structure parameters and logistics equipment parameters as constraints, wherein the shelf structure parameters include layer height, load-bearing and channel width, and the logistics equipment parameters include mechanical arm gripping range and AGV carrying capacity.

[0012] Further, in the step of adopting an improved deep clustering algorithm to dynamically cluster the goods feature vector, automatically generating a twin model adapted to different goods types, and embedding logical parameters about goods operation rules in the twin model to form a two-dimensional model library, at least the logical parameters about the storage location rules and access action rules of the goods are embedded in the twin model.

[0013] Further, in the step of real-time collection of goods constraints, space constraints and equipment constraints of the dense library, and based on AHP hierarchical analysis method to allocate weights, and then adopt fusion algorithm, determine the optimal model matched with the goods in the two-dimensional model library, the goods constraints include the real-time collected goods feature vector, the space constraints at least include the shelf load-bearing, layer height and channel width of the target storage area, and the equipment constraints at least include the mechanical arm gripping force and AGV carrying capacity responsible for accessing the corresponding goods.

[0014] Further, the step of adopting fusion algorithm to determine the optimal model matched with the goods in the two-dimensional model library includes:

[0015] According to the cosine similarity algorithm, calculate the similarity of the real-time collected goods feature vector and the goods feature vector of each twin model in the two-dimensional model library, and according to the similarity, determine the candidate model;

[0016] According to the checking results of the storage location rule and the access action rule of the candidate model and the corresponding space constraint and device constraint respectively, the optimal model is determined.

[0017] Further, the step of constructing a five-dimensional feature system, according to the five-dimensional feature system, real-time acquisition of corresponding cargo feature data, and generating a standardized cargo feature vector includes:

[0018] Introducing environmental features, and fusing the environmental features, the physical features, and the logistics features, constructing a feature embedding space through dynamic weighted contrast learning;

[0019] Based on the LSTM-GARCH hybrid model and the feature vectors in the feature embedding space, combining the cargo type change trend and the logistics link adjustment, predicting potential drift features, and according to the potential drift features, generating a standardized cargo feature vector;

[0020] According to the potential drift features, taking a healthy feature sample library as a benchmark, adjusting the feature distribution through adaptive maximum mean difference, so that the drifted features meet the requirements of generating a standardized cargo feature vector, wherein the healthy feature sample library contains feature samples obtained through artificial labeling.

[0021] The second aspect of the embodiment of the present application provides an artificial intelligence-based dense warehouse visualization modeling system for realizing the artificial intelligence-based dense warehouse visualization modeling method of the first aspect, and the system comprises:

[0022] A first construction module is configured to construct a five-dimensional feature system, according to the five-dimensional feature system, real-time acquisition of corresponding cargo feature data, and generating a standardized cargo feature vector, wherein the five-dimensional feature system includes physical features, packaging features, logistics features, security features, and dense warehouse adaptation features;

[0023] A clustering module is configured to use an improved deep clustering algorithm to dynamically cluster the cargo feature vectors, automatically generate twin models adapted to different cargo types, and embed logical parameters about cargo operation rules in the twin models to form a two-dimensional model library, wherein the improved deep clustering algorithm is a DBSCAN algorithm based on dense warehouse structure constraints.

[0024] A model determination module is configured to real-time collect cargo constraints, space constraints, and device constraints of a dense warehouse, assign weights based on an AHP hierarchical analysis method, and then use a fusion algorithm to determine an optimal model matching the cargo in the two-dimensional model library.

[0025] A second construction module is configured to construct a three-dimensional twin model of the dense warehouse, collect three-dimensional spatial coordinates of the goods in real time, and map the optimal model to the three-dimensional twin model in real time according to the three-dimensional spatial coordinates.

[0026] A third aspect of the embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the artificial intelligence-based dense warehouse visualization modeling method provided in the first aspect.

[0027] A fourth aspect of the embodiment of the present application provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the artificial intelligence-based dense warehouse visualization modeling method provided in the first aspect when executing the program.

[0028] The artificial intelligence-based dense warehouse visualization modeling method and system provided in the embodiment of the present application, by constructing a five-dimensional feature system, acquiring corresponding goods feature data in real time according to the five-dimensional feature system, and generating a standardized goods feature vector, using an improved deep clustering algorithm to dynamically cluster the goods feature vector, automatically generating a twin model adapted to different goods types, embedding logical parameters about goods operation rules in the twin model to form a two-dimensional model library, collecting goods constraints, spatial constraints and equipment constraints of the dense warehouse in real time, assigning weights based on the AHP hierarchical analysis method, and then using a fusion algorithm to determine the optimal model matched with the goods in the two-dimensional model library, constructing a three-dimensional twin model of the dense warehouse, collecting three-dimensional spatial coordinates of the goods in real time, and mapping the optimal model to the three-dimensional twin model in real time according to the three-dimensional spatial coordinates, specifically, by breaking through the limitations of the prior art "static modeling - manual matching - isolated display" through the core process of "model library construction - dynamic retrieval - visualization". BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 An implementation flowchart of the artificial intelligence-based dense warehouse visualization modeling method provided for the first embodiment of the present application;

[0030] Figure 2 A structural block diagram of the artificial intelligence-based dense warehouse visualization modeling system provided for the third embodiment of the present application;

[0031] Figure 3 A structural block diagram of the electronic device provided for the fourth embodiment of the present application. DETAILED DESCRIPTION

[0032] For the purpose of promoting an understanding of the application, the application will now be described in greater detail with reference to the figures. Several embodiments of the application are depicted in the drawings. However, the application can be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete. It will be apparent that the application is not limited to the embodiments set forth herein.

[0033] It should be noted that when an element is referred to as being "on" another element, it can be directly on the other element or intervening elements can also be present. In addition, it should be noted that when an element is referred to as being "connected", "coupled", or "linked" to another element, it can be directly connected, coupled, or linked to the other element or intervening elements can also be present. The terms "vertical", "horizontal", "left", "right", and the like as used herein are used for explanation purposes only and not intended to be limiting.

[0034] 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 application belongs. The terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0035] Embodiment One

[0036] According to the embodiments of the present application, a dense warehouse visualization modeling method based on artificial intelligence is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0037] In this embodiment one, a dense warehouse visualization modeling method based on artificial intelligence is provided, which can be used in electronic devices such as computers. Please refer to Figure 1 , Figure 1 The implementation flowchart of the dense warehouse visualization modeling method based on artificial intelligence provided by the embodiment one of the present application is shown, which specifically includes steps S01 to S04.

[0038] Step S01, a five-dimensional feature system is constructed, according to the five-dimensional feature system, the corresponding cargo feature data is acquired in real time, and the standardized cargo feature vector is generated, wherein the five-dimensional feature system includes physical features, packaging features, logistics features, security features and dense warehouse adaptation features.

[0039] In the embodiment of the present application, the traditional "size + weight" binary characteristics are broken through, and a five-dimensional feature system composed of physical features, packaging features, logistics features, security features and dense library adaptation features is constructed, wherein the physical features include size, weight and density, the packaging features include material, hardness and sealing performance, the logistics features include turnover frequency, storage time and loading and unloading requirements, the security features include fragility, corrosivity and explosion-proof level, and the dense library adaptation features include shelf load-bearing matching degree and access mechanical arm compatibility.

[0040] Specifically, the cargo feature data can be collected in real time by laser profile scanning + millimeter wave density detection + Internet of Things sensor array, and a standardized cargo feature vector is generated, for example, [1.2m x 0.8m x 0.5m, 50kg, 0.8g / cm 3 , paper hard box, 3 times per month, ordinary goods, shelf load-bearing level B].

[0041] In step S02, an improved deep clustering algorithm is used to dynamically cluster the cargo feature vector, automatically generate a twin model adapted to different cargo types, and embed logical parameters about cargo operation rules in the twin model to form a two-dimensional model library, wherein the improved deep clustering algorithm is a DBSCAN algorithm based on dense library structure constraints.

[0042] It should be noted that the cargo feature vector is dynamically clustered with the shelf structure parameters and logistics equipment parameters as constraint conditions, wherein the shelf structure parameters include layer height, load-bearing and channel width, and the logistics equipment parameters include mechanical arm grabbing range and AGV (Automated Guided Vehicle) carrying capacity. In addition, at least logical parameters about storage location rules and access action rules of the goods are embedded in the twin model. It can be understood that the storage location rules are, for example, high-density goods prefer low-layer shelves, and the access action rules are, for example, fragile goods are grabbed at low speed by the mechanical arm.

[0043] In the embodiment of the present application, all constraint parameters in the constraint conditions are converted into a numerical format comparable to the cargo feature vector. For example, if the "cargo height" in the cargo feature vector is h_goods, it needs to satisfy h_goods<=h_i (i is the target shelf layer), and if the "cargo weight" in the cargo feature vector is w_goods, it needs to satisfy w_goods<=w_i (shelf load-bearing) and w_goods<=w_max (mechanical arm grabbing), w_goods<=w_agv (AGV carrying capacity). Finally, a constraint threshold matrix is formed as a pre-selection condition for subsequent clustering (only the cargo feature vectors meeting the constraints can enter the clustering link).

[0044] Based on the multi-modal perception device (laser scanner, weight sensor, material identification camera, etc.) of the intelligent dense library, the characteristic data of the goods is collected, and the initial characteristic vector is constructed, and then the abnormal value is eliminated and the numerical standardization processing is carried out, the standardized characteristic vector is compared with the constraint threshold matrix one by one, and only the vector meeting all the constraints is reserved. Further, the traditional DBSCAN relies on two core parameters of neighborhood radius ε and minimum point number MinPts, which are improved and associated with the dense library constraint, and the parameter values are dynamically calculated. Specifically, for the neighborhood radius ε, i.e. the feature similarity threshold, it is determined based on the shelf load difference and the layer height difference. As can be understood, if the load difference of a certain shelf area is ≤50 kg and the layer height difference is ≤0.3 m, then ε is set to 0.2 (in the standardized feature space, it represents that the feature similarity ≥80%); if the regional hardware parameter difference is large (such as the load difference is 100 kg), then ε is reduced to 0.1 (the similarity is ≥90%), so as to ensure that the goods in the same cluster are adapted to the same type of hardware; for the minimum point number MinPts, i.e. the minimum sample number for forming a cluster, it is determined based on the number of shelf layers. As can be understood, if there are 20 layers in a certain area, then MinPts is set to 5 (to ensure that each cluster corresponds to at least 5 layer adaptation requirements), so as to avoid the model template redundancy caused by too few samples in the cluster.

[0045] Further, the traditional DBSCAN only determines whether to be classified into the same cluster by the feature distance, and the improved DBSCAN increases the hardware constraint verification when merging the clusters at each step. Specifically, the feature distance (such as the Euclidean distance) between the sample point p and the sample point q in the neighborhood is calculated, if the distance ≤ε, it is preliminarily determined as a potential same type; the features of p and q are verified whether they are adapted to the same group of hardware parameters, for example, the weight of p is 300 kg and is adapted to the A area 1 layer shelf (the load is 500 kg), the weight of q is 200 kg and is adapted to the A area 2 layer shelf (the load is 300 kg), although the feature distance ≤ε, but since the adapted shelves are different, it is determined as a non-same type and is not merged into the cluster; if p and q meet the feature distance requirement and are adapted to the same group of hardware parameters, they are merged into the same cluster, and gradually form the hardware adaptation type cluster.

[0046] It should be noted that after the clustering is completed, the effectiveness of the result is evaluated by two indexes, first, the proportion of the samples adapted to the same hardware parameters in each cluster is calculated, which needs to be ≥95% (if 30% of the samples in a certain cluster are adapted to the A area shelf and 70% are adapted to the B area shelf, it is determined as an invalid cluster), so as to evaluate the hardware adaptation rate; by counting the number of small clusters with the sample number <MinPts, if the proportion >10%, then the ε and MinPts parameters are adjusted, the clustering is re-performed, so as to evaluate the cluster redundancy; finally, the cluster set with "hardware adaptation effective and no redundancy" is output, and each cluster corresponds to a type of hardware adapted goods.

[0047] In the embodiments of the present application, a two-dimensional model library is constructed from geometric dimensions and logical dimensions. For each cluster, the physical feature mean of all cargo feature vectors in the cluster is extracted as the basic parameter of the 3D model. Based on the hardware parameters adapted to the cluster, a logical rule script is automatically generated, embedded in the 3D model, and supplemented with logical dimensions. Specifically, for storage location rules, the rule is generated according to the shelf layer adapted to the cargo weight and height (for example, "cluster C1 cargo weight ≤ 300 kg, height ≤ 1.5 m, then preferentially allocate A area 2-3 layer shelf"). For access action rules, the device action parameters are generated according to the cargo material hardness and fragility (for example, "cluster C2 cargo fragility = 1, then the mechanical arm grabbing speed ≤ 0.2 m / s, grabbing force ≤ 50 N; AGV transport speed ≤ 0.5 m / s"). For rule conflicts, if a cargo simultaneously meets multiple storage rules (for example, both adapted to A area 2 layer and B area 3 layer), the priority rule is embedded (for example, "turnover frequency > 5 times / month, then preferentially allocate A area 2 layer close to the delivery port"). Finally, each cluster corresponds to one logical rule package, which is bound to the 3D geometric model to form a twin model.

[0048] Step S03, real-time collection of cargo constraints, space constraints and device constraints in the dense warehouse, and distribution of weights based on AHP hierarchical analysis method, and then using a fusion algorithm to determine the optimal model matched with the cargo in the two-dimensional model library.

[0049] Among them, the cargo constraints, space constraints and device constraints are collected in real time by sensors arranged in the dense warehouse. The cargo constraints include real-time collected cargo feature vectors. The space constraints at least include the shelf load capacity, layer height and channel width of the target storage area. The device constraints at least include the mechanical arm grabbing force and AGV carrying capacity responsible for accessing the corresponding cargo. The weights can be adjusted according to business needs.

[0050] Specifically, according to the cosine similarity algorithm, the similarity between the real-time collected cargo feature vector and the cargo feature vector of each twin model in the two-dimensional model library is calculated, and the candidate model is determined according to the similarity.

[0051] According to the checking results of the storage location rule and the access action rule of the candidate model with the corresponding space constraint and device constraint respectively, the optimal model is determined. It can be understood that the rule matching algorithm is used for secondary screening of the candidate model, that is, checking whether the "storage location rule" of the candidate model matches the space constraint of the target area and whether the "access action rule" matches the device constraint, and eliminating the models that do not meet the constraints.

[0052] Step S04, constructing a three-dimensional twin model of the dense warehouse, real-time collecting three-dimensional space coordinates of the cargo, and real-time mapping the optimal model to the three-dimensional twin model according to the three-dimensional space coordinates.

[0053] Specifically, a full-scene three-dimensional twin model of a dense warehouse can be constructed based on a laser SLAM+BIM technology, containing all physical elements such as shelves, mechanical arms, AGVs, passages, entrances and exits, and being completely aligned with the space coordinate system of the actual dense warehouse (error ≤ 1 cm), thereby providing a high-precision scene carrier for visual display. Meanwhile, the three-dimensional space coordinates (X / Y / Z axes) of goods are collected in real time through a UWB ultra-wideband positioning technology, and finally the optimal model is mapped to the three-dimensional twin model in real time according to the three-dimensional space coordinates, and visual presentation is performed.

[0054] In summary, the above-mentioned embodiment of the application provides a dense warehouse visual modeling method based on artificial intelligence. The method constructs a five-dimensional feature system, acquires corresponding goods feature data in real time according to the five-dimensional feature system, and generates a standardized goods feature vector. An improved deep clustering algorithm is used to dynamically cluster the goods feature vector, automatically generate a twin model adapted to different goods types, embed logical parameters about goods operation rules in the twin model, form a two-dimensional model library, collect goods constraints, space constraints and equipment constraints of the dense warehouse in real time, assign weights based on an AHP hierarchical analysis method, and then use a fusion algorithm to determine the optimal model matched with the goods in the two-dimensional model library. A three-dimensional twin model of the dense warehouse is constructed, the three-dimensional space coordinates of the goods are collected in real time, and the optimal model is mapped to the three-dimensional twin model in real time according to the three-dimensional space coordinates. Specifically, by breaking through the limitations of the prior art such as “static modeling, manual matching and isolated display” through the core process of “model library construction, dynamic retrieval and visualization”, the limitations of the prior art such as “static modeling, manual matching and isolated display” are broken through.

[0055] Embodiment two

[0056] Embodiment two of the application also provides a dense warehouse visual modeling method based on artificial intelligence. The difference from embodiment one of the application is that it solves the feature drift in a dynamic scene, providing high-quality feature input for subsequent clustering. It should be noted that features are the basis for mapping and clustering, and feature drift (such as shift of goods weight feature distribution and addition of packaging material feature dimension) is prone to occur in a dynamic scene (such as changes in goods types, environmental temperature and humidity, and frequency of intelligent dense warehouse). Therefore, the step of constructing a five-dimensional feature system, acquiring corresponding goods feature data in real time according to the five-dimensional feature system, and generating a standardized goods feature vector includes:

[0057] Step one, introducing environmental features, and fusing the environmental features, the physical features and the logistics features to construct a feature embedding space through dynamic weighted comparison learning.

[0058] The environmental features include the temperature and humidity of the storage area (real-time collected by temperature and humidity sensors deployed at different positions in the warehouse, with a sampling frequency of 1 time per minute), vibration frequency (vibration sensors are installed at the transportation channel and beside the shelves to record the vibration of the goods during transportation and storage, in hertz), after data cleaning and standardization processing of the environmental features, physical features and logistics features, for the stable features, the initial weight of the stable features is determined through historical data statistical analysis combined with the experience of domain experts, for the variable features, the weight adjustment coefficient is determined according to the size of the fluctuation coefficient, and the initial weight of the variable features is determined. It can be understood that the priority of the variable features such as temporary labeling will fluctuate with the changes of business demand and logistics scene, and the dynamic weight needs to be calculated according to the fluctuation coefficient of the last 3 logistics cycles.

[0059] Further, a part of the feature data after data cleaning and standardization processing is randomly selected as anchor samples, and then positive samples and negative samples are selected for each anchor sample. The positive sample is a sample with high similarity to the anchor sample (the cosine similarity of the feature vector is calculated, and the sample with similarity greater than 0.8 is selected) which belongs to the same type of goods; the negative sample is a sample with low similarity to the anchor sample (the sample with similarity less than 0.3 is selected) which belongs to different types of goods; the feature data of the anchor sample, the positive sample and the negative sample are input into the neural network model, and the feature embedding vector of each sample is calculated through the model. According to the initial weight determined before, the contribution of different features in the embedding process is weighted to construct a dynamic weighted contrast loss function. The goal of this loss function is to make the feature embedding vector distance between the anchor sample and the positive sample as small as possible, and the feature embedding vector distance between the anchor sample and the negative sample as large as possible. Through the back propagation algorithm, the parameters of the neural network model are continuously adjusted to minimize the dynamic weighted contrast loss function until the model converges, and the construction of the initial feature embedding space is completed.

[0060] Step two, based on the LSTM-GARCH hybrid model and the feature vectors in the feature embedding space, combined with the change trend of the type of goods and the adjustment of the logistics link, the potential drift features are predicted, and the standardized goods feature vectors are generated according to the potential drift features.

[0061] In this way, the feature drift can be predicted before it occurs, providing sufficient time for subsequent correction measures. It should be noted that the collection of cargo type trend data can be achieved by the warehouse management system to count the daily storage quantity, delivery quantity and inventory quantity of different types of goods, and calculate the change of the proportion of each type of goods in the total goods. For example, before the e-commerce promotion, the change of the storage quantity proportion of small fast-moving consumer goods (such as snacks, daily necessities, etc.) is counted to determine whether it shows an upward trend. At the same time, combined with market demand prediction data (such as e-commerce platform sales prediction report), the possible change trend of the goods type in the future period of time is obtained. The logistics link adjustment data can focus on the transportation route planning of the logistics enterprise, the change of the transportation mode (such as changing from road transportation to railway transportation), the increase and decrease of the transfer station and other logistics link adjustment information. Through data docking with logistics partners, the specific content and implementation time of logistics link adjustment are obtained in real time, such as the specific route of the newly added short-distance transfer link, the location and operation time of the transfer station, etc.

[0062] In some other embodiments of the present application, other associated factor data that may affect feature drift, such as seasonal change data and policy and regulation adjustment data, can also be included. The relevant information is obtained through weather forecast data from meteorological departments, policy documents published by government departments, etc., and is converted into quantifiable feature data (such as converting seasonal change into quarterly coding, spring as 1, summer as 2, autumn as 3, winter as 4).

[0063] It should be noted that first, a healthy feature sample library is constructed, and samples that can accurately reflect the normal feature state of the goods are selected from the historical feature data as initial healthy samples. The selection criteria include: the goods corresponding to the sample have not appeared any abnormal situation (such as damage, loss, delay, etc.) in the storage and transportation process; the feature data of the sample is complete and accurate, and there is no abnormal value after the preprocessing in step one; the sample can represent the typical features of different types and different batches of goods. Through artificial annotation screening, or artificial annotation combined with model screening (such as using clustering algorithm to classify samples with similar features into a class, and selecting center samples from each class), the sample set of the initial healthy feature sample library is determined, and in addition, to ensure that the healthy feature sample library can reflect the normal state of the current goods features, the sample library is updated regularly.

[0064] In the embodiments of the present application, the cargo type change trend data and the logistics link adjustment data are fused with the feature data of the corresponding period in the health feature sample library to construct an input feature vector of the LSTM-GARCH hybrid model. The dimension of the input feature vector is determined according to the number of associated factors, for example, if the input feature vector includes n associated factors such as cargo type proportion change and logistics link adjustment times, then the dimension of the input feature vector is n. At the same time, the features that have drifted in the historical feature data are taken as label data of the model (1 represents drift, and 0 represents no drift).

[0065] Further, the prepared input feature vector and label data are divided into a training set (70% of the total data), a validation set (20% of the total data), and a test set (10% of the total data). The training set is used to train the LSTM-GARCH hybrid model, wherein the LSTM part is used to capture the time series characteristics of the associated factor data and predict the change trend of the cargo feature; the GARCH part is used to process the heteroscedasticity in the feature data and improve the prediction accuracy of the model for feature fluctuations. During the training process, the hyperparameters of the model (such as the number of hidden layer nodes of LSTM, learning rate, order of GARCH model, etc.) are constantly adjusted through the validation set, and AIC (Akaike information criterion) and BIC (Bayesian information criterion) are used as evaluation indexes for model selection, and the model with the smallest AIC and BIC values is selected as the optimal model.

[0066] Further, the associated factor data at the current time is input into the trained LSTM-GARCH hybrid model, and the model outputs the probability of each feature drifting in the future preset time. A drift probability threshold is set, and if the predicted drift probability of a certain feature is greater than the drift probability threshold, then the feature is determined to be a potential drift feature, and feature drift warning information is generated, including the name of the potential drift feature, the predicted drift time, the drift probability, etc.

[0067] Step three, according to the potential drift feature, taking the health feature sample library as a reference, adjusting the feature distribution through adaptive maximum mean difference, so that the drifted feature meets the requirements of generating a standardized cargo feature vector, wherein the health feature sample library includes feature samples obtained through manual annotation.

[0068] It should be noted that for the features confirmed to have drifted, the adaptive maximum mean difference (AMMD) of the real-time data distribution of the feature and the feature distribution in the health feature sample library is calculated. AMMD can adaptively adjust the parameters of the kernel function and better capture the differences between different distributions. The calculation formula is based on the expected difference of the kernel function, and the norm of the difference between the means of the two distributions is calculated in the reproducing kernel Hilbert space (RKHS).

[0069] According to the calculated AMMD value, the parameters of the feature normalization dynamic calibration are determined. The greater the AMMD value, the greater the difference between the real-time feature distribution and the healthy feature distribution, and the greater the adjustment range. Finally, the calibrated coefficient is used to perform normalization dynamic calibration on the real-time data of the drifted feature, and through calibration, the feature data distribution after drifting is as close as possible to the feature distribution in the healthy feature sample library.

[0070] In some other embodiments of the application, the KL divergence and the JS divergence of the calibrated feature data and the corresponding features in the healthy feature sample library are calculated again, and it is judged whether the calibrated feature distribution meets the requirements. If it meets the requirements, it means that the calibration effect is good, and the feature drift problem is solved. If it does not meet the requirements, the adaptive maximum mean difference adjustment is continued, the calibration coefficient is adjusted again, and the calibration is performed again until the feature distribution meets the requirements. The calibrated feature data is added to the healthy feature sample library as a new sample, and the LSTM-GARCH hybrid model is retrained and optimized using the updated sample library, the parameters of the LSTM-GARCH hybrid model are updated, and the prediction accuracy of the LSTM-GARCH hybrid model for future feature drift is improved. At the same time, according to the type and reason of the feature drift, the weights of the related factors input into the LSTM-GARCH hybrid model are adjusted, so that the LSTM-GARCH hybrid model can better capture the key factors affecting the feature drift.

[0071] Embodiment three

[0072] Please refer to Figure 2 , Figure 2 is a structural block diagram of an artificial intelligence-based dense library visualization modeling system provided by Embodiment Three of the present application. The artificial intelligence-based dense library visualization modeling system 200 is used to implement the above-mentioned embodiments and preferred embodiments, and will not be described again. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, hardware, or a combination of software and hardware implementations are also possible and contemplated.

[0073] Specifically, the artificial intelligence-based dense library visualization modeling system 200 includes a first construction module 21, a clustering module 22, a model determination module 23, and a second construction module 24, wherein:

[0074] The first construction module 21 is configured to construct a five-dimensional feature system, acquire corresponding cargo feature data in real time according to the five-dimensional feature system, and generate a standardized cargo feature vector, wherein the five-dimensional feature system comprises physical features, packaging features, logistics features, security features and dense warehouse adaptation features, the physical features comprise size, weight and density, the packaging features comprise material, hardness and sealing performance, the logistics features comprise turnover frequency, storage duration and loading and unloading requirements, the security features comprise fragility, corrosiveness and explosion-proof level, and the dense warehouse adaptation features comprise rack load-bearing matching degree and access mechanical arm compatibility;

[0075] The clustering module 22 is configured to dynamically cluster the cargo feature vector by using an improved deep clustering algorithm, automatically generate a twin model adapted to different cargo types, and embed logical parameters about cargo operation rules in the twin model to form a two-dimensional model library.

[0076] The model determination module 23 is configured to collect cargo constraints, space constraints and equipment constraints of the dense warehouse in real time, assign weights based on an AHP (Analytic Hierarchy Process) method, and then determine an optimal model matched with the cargo in the two-dimensional model library by using a fusion algorithm, wherein the cargo constraints comprise the cargo feature vector collected in real time, the space constraints comprise at least rack load-bearing, layer height and channel width of a target storage area, and the equipment constraints comprise at least mechanical arm gripping force and AGV (Automated Guided Vehicle) carrying capacity.

[0077] The second construction module 24 is configured to construct a three-dimensional twin model of the dense warehouse, collect three-dimensional space coordinates of the cargo in real time, and map the optimal model to the three-dimensional twin model in real time according to the three-dimensional space coordinates.

[0078] Further, in some optional embodiments of the present application, the model determination module 23 comprises:

[0079] The calculation unit is configured to calculate the similarity between the cargo feature vector collected in real time and the cargo feature vector of each twin model in the two-dimensional model library according to a cosine similarity algorithm, and determine a candidate model according to the similarity.

[0080] A determination unit is configured to determine the optimal model according to the checking results of the storage location rule and the access action rule of the candidate model and the corresponding space constraint and device constraint, respectively.

[0081] Further, in some optional embodiments of the present application, the first construction module 21 comprises:

[0082] A construction unit is configured to introduce environmental features and fuse the environmental features, the physical features and the logistics features to construct a feature embedding space through dynamic weighted contrast learning.

[0083] A prediction unit is configured to predict potential drift features based on an LSTM-GARCH hybrid model and a feature vector in the feature embedding space, in combination with a cargo type change trend and a logistics link adjustment, and generate a standardized cargo feature vector according to the potential drift features.

[0084] An adjustment unit is configured to adjust a feature distribution through adaptive maximum mean difference based on the potential drift features and a healthy feature sample library, so that the drifted features meet the requirement of generating the standardized cargo feature vector, wherein the healthy feature sample library contains feature samples obtained through manual labeling.

[0085] Embodiment Four

[0086] Another aspect of the present application also provides an electronic device, please refer to Figure 3 , which is an electronic device in the fourth embodiment of the present application, comprising a memory 20, a processor 10 and a computer program 30 stored in the memory and executable on the processor, wherein the processor 10 implements the artificial intelligence-based dense library visualization modeling method as described above when executing the computer program 30.

[0087] In some embodiments, the processor 10 can be a central processing unit (CPU), a controller, a microcontroller, a microprocessor or other data processing chip, configured to run program codes or process data stored in the memory 20, such as executing access restriction programs.

[0088] The memory 20 includes at least one type of readable storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., SD or DX memory, etc.), a magnetic memory, a magnetic disk, an optical disk, etc. The memory 20 can be an internal storage unit of the electronic device in some embodiments, such as a hard disk of the electronic device. The memory 20 can also be an external storage device of the electronic device in other embodiments, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device. Further, the memory 20 can include both the internal storage unit and the external storage device of the electronic device. The memory 20 can be used not only to store application software and various data of the electronic device, but also to temporarily store data that has been output or will be output.

[0089] It should be noted that, Figure 3 The illustrated structure does not limit the electronic device, and in other embodiments, the electronic device can include fewer or more components than illustrated, or combine certain components, or arrange different components.

[0090] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program. The program is executed by a processor to implement the artificial intelligence-based dense library visualization modeling method as described above.

[0091] Those skilled in the art can understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a list of executable instructions for implementing logical functions, which can be specifically implemented in any computer readable medium for use by or in conjunction with an instruction execution system, device or apparatus, such as a computer-based system, a system including a processor or other system that can fetch and execute instructions from the instruction execution system, device or apparatus. For the present specification, the "computer readable medium" can be any device that can contain, store, communicate, propagate or transport programs for use by or in conjunction with the instruction execution system, device or apparatus, or in conjunction with these instruction execution systems, devices or apparatus.

[0092] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that is then suitable for use by the computer. Program code embodied on a computer-readable medium can be transmitted using any apparatus adapted to transmit such a program code, including a modem, or other fetal communication apparatus.

[0093] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the embodiments described above, various steps or methods can be implemented, for example, by software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following techniques, which are well known in the art, can be used to implement the application: a hybrid of the techniques mentioned above; a combination of one or more of the techniques mentioned above; or one or more other techniques suitable for use in the art.

[0094] In the description of the specification, the description of the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the application. In the specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in one or more embodiments or examples.

[0095] The above embodiments only express several embodiments of the application, which are described in detail and in detail, but cannot be understood as limiting the scope of the patent of the application. It should be noted that for those skilled in the art, without departing from the concept of the application, a number of modifications and improvements can be made, which are within the scope of protection of the application. Therefore, the scope of protection of the patent of the application should be subject to the appended claims.

Claims

1. An artificial intelligence based dense library visualization modeling method, characterized in that, The method comprises: constructing a five-dimensional feature system, acquiring corresponding cargo feature data in real time according to the five-dimensional feature system, and generating a standardized cargo feature vector, wherein the five-dimensional feature system comprises physical features, packaging features, logistics features, security features, and dense warehouse adaptation features; using an improved deep clustering algorithm to dynamically cluster the cargo feature vector, automatically generating twin models adapted to different cargo types, and embedding logical parameters related to cargo operation rules in the twin models to form a two-dimensional model library, wherein the improved deep clustering algorithm is a DBSCAN algorithm based on dense warehouse structure constraints; real-time collection of cargo constraints, space constraints, and equipment constraints of the dense warehouse, and weight allocation based on AHP hierarchical analysis, followed by fusion algorithm to determine the optimal model matching the cargo in the two-dimensional model library; constructing a three-dimensional twin model of the dense warehouse, real-time collection of three-dimensional spatial coordinates of the cargo, and real-time mapping of the optimal model to the three-dimensional twin model according to the three-dimensional spatial coordinates.

2. The artificial intelligence based dense library visualization modeling method of claim 1, wherein, The physical features include size, weight, and density, the packaging features include material, hardness, and sealing, the logistics features include turnover frequency, storage duration, and loading and unloading requirements, the security features include fragility, corrosiveness, and explosion-proof level, and the dense warehouse adaptation features include shelf load-bearing matching degree and access mechanical arm compatibility.

3. The artificial intelligence based dense library visualization modeling method of claim 2, wherein, In the step of using an improved deep clustering algorithm to dynamically cluster the cargo feature vector, automatically generating twin models adapted to different cargo types, and embedding logical parameters related to cargo operation rules in the twin models to form a two-dimensional model library, the cargo feature vector is dynamically clustered with shelf structure parameters and logistics equipment parameters as constraints, wherein the shelf structure parameters include layer height, load-bearing, and aisle width, and the logistics equipment parameters include mechanical arm gripping range and AGV carrying capacity.

4. The artificial intelligence-based dense library visualization modeling method of claim 3, wherein, In the step of using an improved deep clustering algorithm to dynamically cluster the cargo feature vector, automatically generating twin models adapted to different cargo types, and embedding logical parameters related to cargo operation rules in the twin models to form a two-dimensional model library, at least logical parameters related to storage location rules and access action rules of the cargo are embedded in the twin models.

5. The artificial intelligence based dense library visualization modeling method of claim 4, wherein, In the step of real-time collection of cargo constraints, space constraints, and equipment constraints of the dense warehouse, and weight allocation based on AHP hierarchical analysis, followed by fusion algorithm to determine the optimal model matching the cargo in the two-dimensional model library, the cargo constraints include real-time collected cargo feature vectors, the space constraints at least include shelf load-bearing, layer height, and aisle width of the target storage area, and the equipment constraints at least include mechanical arm gripping force and AGV carrying capacity responsible for accessing the corresponding cargo.

6. The artificial intelligence based dense library visualization modeling method of claim 5, wherein, The step of using a fusion algorithm to determine the optimal model matching the cargo in the two-dimensional model library comprises: calculating the similarity of the real-time collected cargo feature vector and the cargo feature vector of each twin model in the two-dimensional model library according to the cosine similarity algorithm, and determining a candidate model according to the similarity. According to the checking results of the storage location rule and the access action rule of the candidate model and the corresponding space constraint and device constraint respectively, the optimal model is determined.

7. The artificial intelligence based dense library visualization modeling method of claim 1, wherein, The step of constructing the five-dimensional feature system, acquiring corresponding cargo feature data in real time according to the five-dimensional feature system, and generating a standardized cargo feature vector includes: An environmental feature is introduced, and the environmental feature, the physical feature, and the logistics feature are fused to construct a feature embedding space through dynamic weighted contrast learning; Based on the LSTM-GARCH hybrid model and the feature vectors in the feature embedding space, in combination with the cargo type change trend and the logistics link adjustment, potential drift features are predicted, and a standardized cargo feature vector is generated according to the potential drift features; According to the potential drift features, a health feature sample library is taken as a benchmark, and the feature distribution is adjusted through adaptive maximum mean difference to make the drifted features meet the requirements of generating a standardized cargo feature vector, wherein the health feature sample library contains feature samples obtained through artificial labeling.

8. An artificial intelligence based dense library visualization modeling system, characterized in that, The system is used to implement the artificial intelligence-based dense warehouse visualization modeling method according to any one of claims 1-7. A first construction module is configured to construct a five-dimensional feature system, acquire corresponding cargo feature data in real time according to the five-dimensional feature system, and generate a standardized cargo feature vector, wherein the five-dimensional feature system includes physical features, packaging features, logistics features, security features, and dense warehouse adaptation features. A clustering module is configured to dynamically cluster the cargo feature vectors using an improved deep clustering algorithm, automatically generate twin models adapted to different cargo types, embed logical parameters about cargo operation rules in the twin models, and form a two-dimensional model library. A model determination module is configured to collect cargo constraints, space constraints, and device constraints of a dense warehouse in real time, assign weights based on an AHP (Analytic Hierarchy Process) method, and then determine an optimal model matching the cargo in the two-dimensional model library using a fusion algorithm. A second construction module is configured to construct a three-dimensional twin model of the dense warehouse, collect three-dimensional space coordinates of the cargo in real time, and map the optimal model to the three-dimensional twin model in real time according to the three-dimensional space coordinates.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by a processor to implement the artificial intelligence-based dense warehouse visualization modeling method according to any one of claims 1-7.

10. An electronic device, comprising: The computer program is stored in the memory and executable on the processor, and the processor implements the artificial intelligence-based dense warehouse visualization modeling method according to any one of claims 1-7 when executing the program.

Images