RFID warehouse logistics management method and system

Abstract

The invention discloses an RFID warehouse logistics management method and system, and relates to the technical field of warehouse logistics management.The method comprises the steps that a historical commodity information database is constructed, GAT is used for conducting feature extraction on commodity data, HDBSCAN is used for clustering commodities, the commodities in each cluster are trained through a kernel ridge regression model, and the commodities in each cluster are obtained; and when a new commodity enters the warehouse, information is obtained through an RFID tag, the trained GAT model is used for extracting the characteristics of the new commodity, a cluster Cmin corresponding to the minimum distance between the new commodity and the characteristic center of each cluster is found, Qmin is substituted, the optimal storage position of the new commodity is obtained, and the new commodity is stored in the warehouse. And generating all alternative paths from the storage position to the bin gate by using a backtracking algorithm, and optimizing the alternative paths by using a Pareto algorithm. According to the invention, the storage position of the new commodity can be quickly positioned, and the optimal path to the warehouse door during warehouse-out is planned in advance.

Description

Technical Field

[0001] The present invention relates to the technical field of warehousing logistics management, and particularly to an RFID warehousing logistics management method and system. Background Art

[0002] In today's complex and changeable warehousing logistics environment, traditional warehousing management methods face many challenges. With the sharp increase in the types and quantities of commodities, it is difficult for manual management to accurately determine the best storage locations of commodities and efficient retrieval paths, resulting in problems such as low utilization rate of warehouse space, low logistics efficiency, and rising costs. The emergence of RFID technology has brought new opportunities to warehousing management. Although RFID can quickly obtain the identification information of commodities, it is difficult to rely solely on RFID technology itself to achieve more efficient warehousing decisions, such as reasonably arranging storage locations and planning retrieval paths. Effective data analysis methods still need to be combined to further explore its value. Summary of the Invention

[0003] Aiming at the deficiencies of the prior art, the present invention provides an RFID warehousing logistics management method and system, which solves the problems that when new commodities enter the warehouse, their storage locations cannot be quickly and accurately found, and the best path to the warehouse door cannot be found based on the current positions of the commodities.

[0004] To achieve the above objectives, the present invention is realized through the following technical solutions: An RFID warehousing logistics management method includes:

[0005] S1. Construct a commodity feature database and an optimal storage location database according to the historical information of the warehouse;

[0006] S2. Use a graph attention network (GAT) to extract features from the data in the commodity feature database;

[0007] S3. Based on the features extracted by GAT, use the HDBSCAN density clustering algorithm to cluster the commodities to obtain r clusters;

[0008] S4. Use a kernel ridge regression model to train the commodities in each cluster Ct (t ∈ [1, r]), and use a particle swarm optimization algorithm to optimize its regularization parameter and kernel parameter to obtain the optimal storage location model QTt of Ct;

[0009] S5. When a new commodity enters the warehouse, scan the RFID tag of the new commodity through an RFID reader to obtain all its information, then use the trained GAT model to extract its features, calculate the distances between it and the feature centers of each cluster, find the cluster Cmin with the smallest distance, and substitute the features into the optimal storage location model Qmin corresponding to Cmin to obtain the optimal storage location of the new commodity;

[0010] S6. Based on the optimal storage location of the new product, use the backtracking algorithm to generate all alternative paths from the warehouse entrance to this storage location, and use the Pareto algorithm to perform multi-objective optimization on the alternative paths to obtain the optimal alternative path.

[0011] As a further solution of the present invention, the specific steps of using GAT to extract features from product data are as follows:

[0012] Perform standardization processing on the product data;

[0013] Construct an association graph G=(V, E) between each product;

[0014] Determine the hyperparameters of the GAT layer, including the number of attention heads K and the output feature dimension dout;

[0015] Initialize the weight matrix and attention coefficient vector of each attention head;

[0016] Use the Latin Hypercube Sampling (LHS) method to sample the neighbor nodes of each node;

[0017] Calculate the attention coefficient eij of each node vi and its neighbor node vj∈Ni after LHS sampling sampled (Ni sampled is the set of sampled neighbor nodes) under each attention head k k :

[0018] where LeakyReLU is the activation function, and its formula is LeakyReLU(a)=max(0.01a, a);

[0019] Obtain the aggregated feature of node vi under each attention head k where, is the normalized attention weight;

[0020] Concatenate the features zi under each attention head k k to obtain the final output feature of node vi as zi = concat(zi 1 , zi 2 ,..., zi K ), where concat() is the concatenation function.

[0021] As a further solution of the present invention, the specific process of LHS sampling includes:

[0022] Number the nodes in the neighbor node set Ni in a random order;

[0023] Divide each dimension (which can be understood as a feature dimension of each neighbor node) into m equally probable intervals;

[0024] For each divided interval, use a random number generator to generate a random number within that interval;

[0025] Match the randomly selected point in each interval with the value of the corresponding dimension in the neighbor node set;

[0026] On multiple dimensions, select the nodes that match the interval where the randomly selected point is located multiple times.

[0027] As a further solution of the present invention, use the HDBSCAN density clustering algorithm to cluster the commodities:

[0028] After n commodity data are processed by GAT, the feature vector of each commodity is zi, and the dimension is dout. Concatenate them into an n×dout matrix Y, where Ypq represents the value of the pth commodity on the qth feature dimension;

[0029] For any two commodities i and j, calculate the similarity according to the formula d(Yi,Yj) = ||Yi - Yj||;

[0030] Determine a neighborhood radius ε and the minimum number of points MinPts that should be included at least in this neighborhood;

[0031] Find all the points within the neighborhood radius ε, that is, the points j that satisfy d(Yi,Yj) <= ε, and denote them as the set N ε (i);

[0032] Calculate the set N ε (i) The distances from the MinPts points closest to i to i in the set N are calculated. After sorting these distances, take the MinPts-th largest distance as the core distance coredist(i). If the number of points in the set is less than MinPts, the core distance is defined as infinity;

[0033] Based on the core distance, judge the direct density reachability, density reachability, and density connection between commodities;

[0034] For any core point i, all the points that are density-connected to it form a cluster. Traverse all the core points until all the points are divided into the corresponding clusters or are marked as noise points (those points that neither belong to any cluster core point nor are density-connected to any point in the cluster).

[0035] As a further solution of the present invention, the steps for each cluster Ci to confirm the best storage location model QTt are as follows:

[0036] Construct the dataset E1 = {(Y1, U1), (Y2, U2),..., (Yn, Un)}, where Yi ∈ R dout is the input feature vector extracted by GAT, and Ui is the data in the optimal storage location database;

[0037] Divide the dataset into a 70% training set Etrain and a 30% test set Etest;

[0038] Set the particle swarm size to N, the inertia weight w, the learning factors c1, c2, and the position of particle i can be expressed as Q i =(λ i , σ1 i ), and the velocity is expressed as V i =(v λ,i , v σ1,i ). Initialize the velocity Vi of the particle to the 0 vector, and Qi is randomly generated within the set parameter range. The initial individual optimal position pbesti = Qi, and the global optimal position is gbest;

[0039] For [0.7*n] ([] is the rounding symbol) samples in the training set Etrain, according to the selected Gaussian kernel function, that is obtain the kernel matrix K ∈ R [0.7*n]×[0.7*n] ;

[0040] For each particle, substitute its position parameter Qi into the kernel ridge regression model. The objective function of this kernel ridge regression model is (K T K + λ i I)β = K T U to calculate the coefficient vector β;

[0041] Use k-fold cross-validation, use the obtained kernel ridge regression model to train and validate on the training set, and calculate the mean squared error (MSE) of the model as the fitness value Fi, that is where MESj is the mean squared error of the j-th fold of validation, nj is the number of samples in the j-th fold of the validation set, Us true is the true value, and Us ypred is the predicted value;

[0042] Compare the current fitness value of each particle, and update pbesti and gbest;

[0043] Update the velocity and position of the particle:

[0044]

[0045] Use the obtained global optimal position gbest as the optimal regularization parameter and kernel parameter of the kernel ridge regression model, substitute it into the training set Etrain to retrain the model, and obtain QTt;

[0046] Use the test set Etest to evaluate QTt. If the effect is not good, continue iterative optimization until the best QTt is found.

[0047] As a further solution of the present invention, the specific steps of using the Pareto algorithm to optimize the alternative paths are as follows:

[0048] Organize the values of factors such as the path length, number of turns, labor cost, equipment loss cost, waiting time, etc. of each alternative path into a matrix form. If there are m1 alternative paths in total, construct matrix A1 as where li represents the path length of the i-th path, ci represents the number of turns, hi represents the labor cost, ei represents the equipment loss cost, fi represents the waiting time, and i ∈ [1, m];

[0049] Compare each path. If path i is not worse than path j in all factors and at least one factor is better than path j, then path i dominates path j. Find all paths that are not dominated by other paths to form the first-level non-dominated set F1, and repeat this process from the remaining paths until all paths are divided into the corresponding non-dominated layers;

[0050] For the paths in each non-dominated layer (Fs, s = 1, 2,...), calculate their crowding distance;

[0051] Give priority to the paths in the non-dominated layer with a higher order. If there are a large number of selectable paths in the same non-dominated layer, select the path with a larger crowding distance.

[0052] An RFID warehouse logistics management system, which includes: a historical information database construction module, a feature extraction module, a commodity clustering module, an optimal storage location construction module, a new commodity storage location determination module, and an alternative path selection module;

[0053] The historical information database construction module constructs a commodity feature database and an optimal storage location database according to the warehouse historical information and transmits them to the feature extraction module;

[0054] The feature extraction module uses a graph attention network (GAT) to extract features from the data in the commodity feature database and transmits them to the commodity clustering module;

[0055] The commodity clustering module performs clustering operations on the commodities based on the features extracted by GAT using the HDBSCAN density clustering algorithm to obtain r clusters and transmits the clusters to the optimal storage location construction module;

[0056] Optimal storage location construction module, which trains the commodities in each cluster Ct (t ∈ [1, r]) using the kernel ridge regression model, and optimizes its regularization parameter and kernel parameter through the particle swarm optimization algorithm to obtain the optimal storage location model QTt corresponding to the commodities in Ct, and transmits QTt to the new commodity storage location confirmation module;

[0057] New commodity storage location confirmation module. When a new commodity enters the warehouse, this module scans the RFID tag of the new commodity through an RFID reader to obtain all its information, then uses the trained GAT model to extract its features, calculates the distances between it and the feature centers of each cluster, finds the cluster Cmin with the smallest distance, substitutes the features into the optimal storage location model Qmin corresponding to Cmin to obtain the optimal storage location of the new commodity, and transmits it to the alternative path selection module;

[0058] Alternative path selection module. Based on the optimal storage location of the new commodity, this module uses the backtracking algorithm to generate all alternative paths from the warehouse entrance to this storage location, and uses the Pareto algorithm to perform multi-objective optimization on the alternative paths to obtain the optimal alternative path.

[0059] The present invention provides a method and system for RFID warehouse logistics management, which has the following beneficial effects compared with the prior art:

[0060] (1) When the present invention uses GAT to extract the key features of commodity data, to avoid redundant calculations caused by high overlap of neighbor nodes, LHS is innovatively introduced to sample the neighbor nodes of the current node, which can better ensure that the selected neighbor nodes are random and have good representativeness in the feature space;

[0061] (2) The present invention uses HDBSCAN to cluster historical commodities, constructs different optimal storage location models for different clusters, and optimizes them using the particle swarm optimization algorithm to obtain the final relatively accurate optimal storage location model;

[0062] (3) Based on the optimal storage location of the new commodity, the present invention uses the backtracking algorithm to find several alternative paths to the warehouse door, and uses the Pareto algorithm to optimize these alternative paths to find the best alternative path, effectively saving transportation time and improving the outbound efficiency of commodities. Description of the Drawings

[0063] Figure 1 is the flowchart of the steps of the present invention;

[0064] Figure 2 is the principle block diagram of the system of the present invention. Detailed Embodiments

[0065] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0066] As Figure 1 , the present invention provides an RFID warehouse logistics management method, including:

[0067] S1. Construct a comprehensive commodity feature database and an optimal storage location database. The commodity feature database covers multi-dimensional information such as size, weight, sales quantity, inventory turnover rate, gross profit margin, and favorable comment rate. These features reflect the attributes and market performance of commodities from different perspectives. For example, for a smart phone, its size may be 160 mm in length, 75 mm in width, 8 mm in height, weight 200 g, price 3000 yuan, sales quantity reaching 5000 units in the past quarter, inventory turnover rate 0.8, gross profit margin 40%, and favorable comment rate 92%;

[0068] The optimal storage location database is the position coordinates on the shelf where the commodity is located. A specific corner of the warehouse can be selected as the origin. For example, select the ground position in the lower left corner of the warehouse, determine the horizontal right direction as the positive direction of the X-axis, which is used to represent the horizontal position of the commodity in the warehouse plane. Set the direction perpendicular to the X-axis and in the same plane (usually horizontal forward) as the positive direction of the Y-axis to represent the vertical position of the commodity in the warehouse plane. And define the direction vertically upward away from the ground as the positive direction of the Z-axis to reflect the shelf layer or height position where the commodity is located. When measuring, the X-axis and Y-axis are scaled with appropriate length units (such as meters, decimeters, etc.) to accurately determine the position of the commodity in the plane. For the Z-axis, if the height of each shelf layer is fixed, it can be marked according to the number of layers. For example, the bottom layer is layer 0, and it can be successively layer 1, layer 2, etc. upward, so as to clearly represent the position information of the commodity in the vertical direction.

[0069] S2. Use the Graph Attention Network (GAT) to deeply extract the data in the commodity feature database. The GAT model can automatically capture the complex correlation relationships between commodities by learning the attention weights between nodes (commodities), so as to find out the key features. The specific steps are as follows:

[0070] Perform normalization processing on the data in the commodity feature database. The normalization formula is Among them, \(x\) is the original feature, \(x'\) is the feature after standardization, \(\mu\) is the mean, and \(\sigma\) is the standard deviation. This step helps to improve the training effect and stability of the GAT model, making different features comparable and avoiding adverse effects on the model caused by differences in data dimensions.

[0071] When constructing the commodity graph \(G=(V, E)\), in addition to constructing edge connections based on certain obvious similarities of commodities (such as commodities of the same category), factors such as the sales association of commodities can also be considered. For example, if two commodities are often purchased simultaneously, an edge can be established between them. Let the node set \(H = \{v_1, v_2, \ldots, v_n\}\), the edge set \(E=\{(v_i, v_j)|i\neq j\}\), and an initial weight is assigned to each edge (if necessary), and the weight can be set based on a quantitative index of the degree of association.

[0072] Determine the hyperparameters of the GAT layer, including the number of attention heads \(K\), the output feature dimension \(d_{out}\), etc. By setting different numbers of attention heads, the neighbor information of nodes can be focused on from different perspectives. Each attention head will generate an independent feature representation, and finally these representations are fused.

[0073] Initialize the weight matrix of each attention head and the attention coefficient vector \(\alpha\) k \(\in\mathbb{R}\) 2dout / K , and these parameters are usually initialized by the random initialization method, but it is necessary to ensure that the value range of the parameters is within a suitable interval to avoid the problems of gradient explosion or disappearance. Therefore, the Xavier or He initialization method can be used to initialize the weight matrix.

[0074] In the process of GAT calculation, the neighbor node information of nodes is crucial, but the high overlap of neighbor nodes will cause redundant calculations and reduce the calculation efficiency. To solve this problem, the Latin hypercube sampling (LHS) method is introduced to sample the neighbor nodes of each node.

[0075] For each node \(v_i\), before calculating the attention coefficient, first use LHS to select some nodes from its neighbor node set \(N_i\), and let the number of neighbor nodes to be sampled be \(m\) (\(m < |N_i|\)).

[0076] The specific sampling process of the LHS is as follows: The nodes in the neighbor node set Ni are numbered in a certain order (random order or ascending / descending order of similarity between nodes); According to the Latin Hypercube Design principle, each dimension (which can be simply understood as a feature dimension of each neighbor node or a quantization index of the comprehensive feature) is divided into m equally probable intervals; A point is randomly selected within each interval, and the neighbor node corresponding to the interval where the selected point is located is the sampled node. Through LHS sampling, while ensuring a certain degree of randomness, the sampled neighbor nodes have good representativeness in the feature space, reducing redundant calculations that may be caused by highly overlapping neighbor nodes;

[0077] Calculate the attention coefficient eij of each node vi and its neighbor node vj ∈ Ni after LHS sampling sampled (Ni sampled is the set of neighbor nodes after sampling) under each attention head k k :

[0078] where LeakyReLU is the activation function, and its formula is LeakyReLU(a) = max(0.01a, a);

[0079] Calculate the aggregated feature of node vi under each attention head k where, is the normalized attention weight, and each node vi obtains a new feature representation zi under each attention head k k , which synthesizes the information of neighbor nodes and performs weighted summation according to the attention weights;

[0080] Represent the feature under each attention head k as zi k Perform a concatenation operation to obtain the final output feature of node vi as zi = concat(zi 1 , zi 2 ,..., zi K ), and the dimension of the final output feature is dout. Among them, concat() is the concatenation function, and its role is to connect multiple vectors in sequence to form a new, higher-dimensional vector.

[0081] S3. Based on the key features extracted by GAT, use the HDBSCAN density clustering algorithm to cluster the commodities to obtain r clusters:

[0082] After being processed by GAT, there are n commodities, and the feature vector of each commodity is zi, with a dimension of dout. It can be assembled into an n×dout matrix Y, where Ypq represents the value of the p-th commodity on the q-th feature dimension;

[0083] The similarity between two commodity feature vectors is measured by calculating the Euclidean distance between them. That is, for any two commodities i and j, the distance is calculated according to the formula d(Yi, Yj) = ||Yi - Yj||.

[0084] Determine a neighborhood radius ε and the minimum number of points MinPts that should be included at least in this neighborhood. The steps to calculate the core distance coredist(i) of commodity i include:

[0085] Find all points within the neighborhood radius ε, that is, points j that satisfy d(Yi, Yj) <= ε, and denote them as set N ε (i);

[0086] Calculate the distances from the MinPts points closest to i in set N ε (i) to i. After sorting these distances, take the MinPts-th largest distance as the core distance coredist(i). If the number of points in the set is less than MinPts, the core distance is defined as infinity;

[0087] Based on the core distance, judge the direct density reachability between commodities. For two commodities i and j, if the following two conditions are satisfied, it is said that j is directly density reachable from i:

[0088] j ∈ N ε (i), that is, j is within the neighborhood radius ε of i;

[0089] d(Yi, Yj) <= coredist(i), that is, the distance from j to i does not exceed the core distance of i;

[0090] For two commodities u and v, if there exists a commodity g such that both u and v are density reachable from g, then it is said that u and v are density connected;

[0091] For any core point i, all points that are density connected to it form a cluster. Traverse all core points until all points are assigned to the corresponding clusters or marked as noise points (those points that neither belong to any cluster core point nor are density connected to any point in a cluster).

[0092] S4. For the r clusters after clustering, use the kernel ridge regression model to train the commodities in each cluster Ct (t ∈ [1, r]) one by one, and optimize its regularization parameter and kernel parameter through the particle swarm algorithm to obtain the corresponding optimal storage location model QTt of the commodities:

[0093] Construct the dataset E1 = {(Y1, U1), (Y2, U2),..., (Yn, Un)}, where Yi ∈ R doutis the input feature vector extracted by GAT, and Ui is the data in the optimal storage location database;

[0094] The dataset is divided into a 70% training set Etrain and a 30% test set Etest;

[0095] Set the particle swarm size to N, the inertia weight w, the learning factors c1 and c2. The position of particle i can be represented as Q i = (λ i , σ1 i ), and the velocity is represented as V i = (v λ,i , v σ1,i ). Initialize the velocity Vi of the particle to the 0 vector, and Qi is randomly generated within the set parameter range. The initial individual optimal position pbesti = Qi, and the global optimal position is gbest;

[0096] For [0.7*n] ([] is the rounding symbol) samples in the training set Etrain, according to the selected Gaussian kernel function, that is obtain the kernel matrix K ∈ R [0.7*n]×[0.7*n] ;

[0097] For each particle, substitute its position parameter Qi into the kernel ridge regression model. The objective function of this kernel ridge regression model is From (K T K + λ i I)β = K T U, calculate the coefficient vector β;

[0098] Use k-fold cross-validation, use the obtained kernel ridge regression model to train and validate on the training set, and calculate the mean squared error (MSE) of the model as the fitness value Fi, that is where MESj is the mean squared error of the j-th fold of validation, nj is the number of samples in the j-th fold of the validation set, Us true is the true value, and Us ypred is the predicted value;

[0099] For each particle i, compare its current fitness value Fi with the fitness value corresponding to the individual optimal position pbesti. If Fi is smaller, then update pbesti = Qi. Compare the current fitness values of all particles, find the particle position corresponding to the smallest fitness value among them, and update the global optimal position gbest;

[0100] Update the velocity and position of the particle:

[0101]

[0102]

[0103] Check whether the maximum number of iterations is reached, or the change in the fitness value in several consecutive iterations is less than a certain threshold. If the termination condition is met, stop the optimization; otherwise, continue the iteration.

[0104] Take the obtained global optimal position gbest as the optimal regularization parameter and kernel parameter of the kernel ridge regression model, substitute it into the training set Etrain to retrain the model, and obtain QTt.

[0105] Use the test set Etest to evaluate QTt, and the evaluation metrics are mean squared error (MSE) and coefficient of determination (R 2 )

[0106] S5. When a new commodity enters the warehouse, use the trained GAT model to extract its features, obtain the feature vector as xnew, calculate the distances between it and the feature centers of each cluster. For cluster Ct, its feature center vector is hi, and the distance is

[0107] Find the cluster Cmin with the smallest distance, which indicates that the commodity has the highest similarity with the commodities in this cluster. Use the best storage position model Qmin corresponding to Cmin to calculate the best storage position of the new commodity.

[0108] S6. Based on the best storage position of the new commodity, use the backtracking algorithm to generate all alternative paths from the warehouse entrance to this storage position. In the warehouse layout diagram, abstract the shelves, aisles, etc. as the nodes and edges of the graph, and traverse all possible paths through the backtracking search algorithm. For each alternative path, record its path length, number of turns, labor cost, equipment loss cost, waiting time, and use the Pareto algorithm to perform multi-objective optimization on these alternative paths:

[0109] Organize the values of factors such as the path length, number of turns, labor cost, equipment loss cost, and waiting time of each alternative path into a matrix form. If there are m1 alternative paths in total, construct matrix A1 as where li represents the path length of the i-th path, ci represents the number of turns, hi represents the labor cost, ei represents the equipment loss cost, fi represents the waiting time, and i ∈ [1, m];

[0110] For any two paths i and j (i ≠ j), if for all objectives, there is A1 k,i <= A1 k,j (k ∈ [1, 5]), and there is at least one objective l such that A l,i < A l,j, then it is said that path i dominates path j. Find all paths that are not dominated by other paths. These paths form the first-layer non-dominated set, denoted as F1. From the remaining paths (excluding the paths in F1), continue to find the non-dominated paths according to the above rules to form the second-layer non-dominated set F2, and so on, until all paths are divided into the corresponding non-dominated layers;

[0111] For the paths in each non-dominated layer (Fs, s = 1, 2,...), calculate their crowding distance to measure the distribution of solutions around the path, so as to further screen paths in the same non-dominated layer later:

[0112] For each target column (a total of 5 columns, corresponding to 5 factors respectively), sort the paths in Fs in ascending order according to the values corresponding to this column;

[0113] For the first and last paths after sorting, their crowding distances are assigned as infinity;

[0114] For the middle path i, calculate its crowding distance where Ak max and Ak min are the maximum and minimum values of the target values in the k-th column in the non-dominated layer Fs respectively;

[0115] Give priority to the paths in the non-dominated layer that are earlier. If there are more paths to choose from in the same non-dominated layer, further screening can be combined with the crowding distance, and give priority to choosing the paths with larger crowding distances;

[0116] If only one path is obtained after screening, this path is the better path considering various factors, and it can be determined as the final transportation path of the commodity;

[0117] If there are still multiple paths after screening by the crowding distance, the final path is determined according to the focus of the actual warehouse operation. For example, if the current focus is on reducing costs, then the total labor cost and equipment loss cost of these paths can be compared, and the path with lower cost can be selected. If more emphasis is placed on transportation efficiency, the path length and waiting time can be focused on, and the path that can make the overall transportation time shorter can be selected.

[0118] Such as Figure 2 , an RFID warehouse logistics management system, including: a historical information database construction module, a feature extraction module, a commodity clustering module, an optimal storage location construction module, a new commodity storage location determination module, and an alternative path selection module;

[0119] The historical information database construction module constructs a commodity feature database and an optimal storage location database according to the warehouse historical information, and transmits them to the feature extraction module;

[0120] A feature extraction module that uses a graph attention network (GAT) to extract features from the data in the commodity feature database and transmits them to the commodity clustering module;

[0121] A commodity clustering module that, based on the features extracted by the GAT, uses the HDBSCAN density clustering algorithm to cluster the commodities, obtains r clusters, and transmits the clusters to the optimal storage location construction module;

[0122] An optimal storage location construction module that uses a kernel ridge regression model to train the commodities in each cluster Ct (t ∈ [1, r]), and uses a particle swarm optimization algorithm to optimize its regularization parameter and kernel parameter, obtains the optimal storage location model QTt corresponding to the commodities in Ct, and transmits QTt to the new commodity storage location confirmation module;

[0123] A new commodity storage location confirmation module that, when a new commodity enters the warehouse, scans the RFID tag of the new commodity through an RFID reader to obtain all its information, then uses the trained GAT model to extract its features, calculates the distances between it and the feature centers of each cluster, finds the cluster Cmin with the smallest distance, substitutes the features into the optimal storage location model Qmin corresponding to Cmin to obtain the optimal storage location of the new commodity, and transmits it to the alternative path selection module;

[0124] An alternative path selection module that, based on the optimal storage location of the new commodity, uses a backtracking algorithm to generate all alternative paths from the warehouse entrance to this storage location, and uses the Pareto algorithm to perform multi-objective optimization on the alternative paths to obtain the optimal alternative path.

[0125] Some of the data in the above formulas are numerically calculated after removing their dimensions, and the content not described in detail in this specification belongs to the prior art well-known to those skilled in the art.

[0126] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An RFID warehouse logistics management method, characterized in that: The following steps are involved: S1. Construct a commodity feature database and an optimal storage location database based on warehouse historical information; S2. Use GAT to extract features from the data in the product feature database. GAT is a graph attention network model. S3. Based on the features extracted by GAT, the HDBSCAN density clustering algorithm is used to cluster the products and obtain r clusters; S4. Use the kernel ridge regression model to train the commodities in each cluster Ct(t∈[1,r]), and optimize its regularization parameters and kernel parameters through the particle swarm optimization algorithm to obtain the optimal storage location model QTt of Ct; S5. When a new product enters the warehouse, the RFID tag of the new product is scanned by the RFID reader to obtain all its information. Then, the trained GAT model is used to extract its features, and the distance between it and the feature center of each cluster is calculated. The cluster Cmin with the smallest distance is found, and the features are substituted into the optimal storage location model Qmin corresponding to Cmin to obtain the optimal storage location of the new product. S6. Based on the optimal storage location of the new product, a backtracking algorithm is used to generate all alternative paths from the warehouse entrance to the storage location, and the Pareto algorithm is used to perform multi-objective optimization on the alternative paths to obtain the optimal alternative path.

2. The RFID warehouse logistics management method according to claim 1, characterized in that: The specific steps for using GAT to extract features from commodity data are as follows: Standardize product data; Construct the association graph G = (V, E) between each product; Determine the hyperparameters of the GAT layer, including the number of attention heads K and the output feature dimension dout; Initialize the weight matrix and attention coefficient vector of each attention head; The neighbor nodes of each node are sampled using the Latin Hypercube (LHS) sampling method; Calculate each node vi and its neighbor node vj∈Ni after LHS sampling sampled (Ni sampled is the set of sampled neighbor nodes) the attention coefficient eij under each attention head k k : Among them, LeakyReLU is the activation function, and its formula is LeakyReLU(a)=max(0.01a,a); Get the aggregated features of node vi under each attention head k in, is the normalized attention weight; The feature zi under each attention head k k Perform concatenation operation to obtain the final output feature of node vi as zi=concat(zi 1 ,zi 2 ,...,zi K ), where concat() is the concatenation function.

3. The RFID warehouse logistics management method according to claim 1, characterized in that: The specific process of LHS sampling includes: Number the nodes in the neighbor node set Ni in random order; Divide each dimension (here it can be understood as a feature dimension of each neighbor node) into m equal probability intervals; For each divided interval, a random number generator is used to generate a random number in the interval; Match the randomly selected points in each interval with the values ​​of the corresponding dimensions in the neighbor node set; In multiple dimensions, select nodes that match the interval of the randomly selected point multiple times.

4. The RFID warehouse logistics management method according to claim 1, characterized in that: Use HDBSCAN density clustering algorithm to cluster products: After the n product data are processed by GAT, the feature vector of each product is zi, and the dimension is dout, which is assembled into an n×dout matrix Y, where Ypq represents the value of the pth product in the qth feature dimension; For any two products i and j, the similarity is calculated according to the formula d(Yi,Yj)=||Yi-Yj||; Determine a neighborhood radius ε and the minimum number of points MinPts contained in the neighborhood; Find all points within the neighborhood radius ε, that is, point j that satisfies d(Yi,Yj)<=ε, and record it as set N ε (i); Calculate the set N ε (i) The distances from the MinPts points closest to i to i. After sorting these distances, take the MinPts largest distance as the core distance coredist(i). If the number of points in the set is less than MinPts, the core distance is defined as infinity; Based on the core distance, the direct density accessibility, density accessibility and density connection between commodities are determined; For any core point i, all points densely connected to it constitute a cluster, and all core points are traversed until all points are divided into corresponding clusters or marked as noise points (those points that neither belong to any cluster core point nor are densely connected to any cluster point).

5. The RFID warehouse logistics management method according to claim 1, characterized in that: The steps for each cluster Ci to determine the best storage location model QTt are: Construct a data set E1 = {(Y1, U1), (Y2, U2), ..., (Yn, Un)}, where Yi∈R dout is the input feature vector extracted by GAT, and Ui is the data in the optimal storage location database; The data set is divided into 70% training set Etrain and 30% test set Etest; Assuming the particle swarm size is N, the inertia weight is w, and the learning factors are c1 and c2, the position of particle i can be expressed as Q i =(λ i ,σ1 i ), the speed is expressed as V i =(v λ,i ,v σ1,i ), the velocity Vi of the initial particle is 0 vector, Qi is randomly generated within the set parameter range, the initial individual optimal position pbesti = Qi, and the global optimal position is gbest; For the [0.7*n] ([] is the integer symbol) samples in the training set Etrain, according to the selected Gaussian kernel function, that is, Get the kernel matrix K∈R [0.7*n]×[0.7*n] ; For each particle, its position parameter Qi is substituted into the kernel ridge regression model. The objective function of the kernel ridge regression model is: By (K T K+λ i I) β=K T U calculates the coefficient vector β; Using k-fold cross validation, the obtained kernel ridge regression model is trained and validated on the training set, and the mean square error (MSE) of the model is calculated as the fitness value Fi, that is, Where MESj is the mean square error of the j-fold validation, nj is the number of samples in the j-fold validation set, Us true is the true value, Us ypred is the predicted value; Compare the current fitness value of each particle and update pbesti and gbest; Update the particle's velocity and position: The obtained global optimal position gbest is used as the optimal regularization parameter and kernel parameter of the kernel ridge regression model, and the training set Etrain is substituted into the retrained model to obtain QTt; Use the test set Etest to evaluate QTt. If the effect is not good, continue to iterate and optimize until the best QTt is found.

6. The RFID warehouse logistics management method according to claim 1, characterized in that: The specific steps of using the Pareto algorithm to select the optimal path are: The values ​​of factors such as the path length, number of turns, labor cost, equipment loss cost, and waiting time of each alternative path are organized into a matrix form. If there are m1 alternative paths in total, the matrix A1 is constructed as Where li represents the path length of the i-th path, ci represents the number of turns, hi represents the labor cost, ei represents the equipment loss cost, fi represents the waiting time, and i∈[1,m]; Compare the paths. If path i is not worse than path j in all factors and has at least one factor better than path j, then path i dominates path j. Find all paths that are not dominated by other paths to form the first-level non-dominated set F1. Repeat the process from the remaining paths until all paths are divided into the corresponding non-dominated layers. For each path in the non-dominated layer (Fs, s = 1, 2, ...), calculate its congestion distance; Prioritize the paths at the front of the non-dominated layer. If there are many selectable paths in the same non-dominated layer, choose the path with a larger congestion distance.

7. An RFID warehouse logistics management system according to claim 1, used to execute an RFID warehouse logistics management method according to any one of claims 1 to 6, characterized in that: The system includes: a historical information database construction module, a feature extraction module, a commodity clustering module, an optimal storage location construction module, a new commodity storage location determination module, and an alternative path selection module; A historical information database construction module, which constructs a commodity feature database and an optimal storage location database based on warehouse historical information and transmits them to a feature extraction module; The feature extraction module uses the graph attention network (GAT) to extract features from the data in the product feature database and transmits them to the product clustering module; Commodity clustering module: This module uses the HDBSCAN density clustering algorithm to cluster commodities based on the features extracted by GAT, obtains r clusters, and transfers the clusters to the optimal storage location construction module; The optimal storage location construction module uses the kernel ridge regression model to train the products in each cluster Ct (t∈[1,r]), and optimizes its regularization parameters and kernel parameters through the particle swarm optimization algorithm to obtain the optimal storage location model QTt corresponding to the products in Ct, and transmits QTt to the new product storage location confirmation module; New product storage location confirmation module: When a new product enters the warehouse, the module scans the RFID tag of the new product through the RFID reader to obtain all its information, and then uses the trained GAT model to extract its features, calculates the distance between it and the feature center of each cluster, finds the cluster Cmin with the smallest distance, substitutes the features into the optimal storage location model Qmin of the product corresponding to Cmin, obtains the optimal storage location of the new product, and transmits it to the alternative path selection module; The alternative path selection module uses a backtracking algorithm to generate all alternative paths from the warehouse entrance to the storage location based on the optimal storage location of the new product, and uses the Pareto algorithm to perform multi-objective optimization on the alternative paths to obtain the optimal alternative path.

Images