Intelligent secondary metering warehouse management method

Abstract

The application relates to a kind of intelligent secondary metering warehouse management methods, comprising the following steps: step S1: using three-dimensional laser scanning technology to carry out all-around three-dimensional space scanning to warehouse, obtains the accurate space data inside warehouse, obtains warehouse construction scheme;Step S2: corresponding warehouse three-dimensional model is constructed;Step S3: RFID data is integrated with warehouse three-dimensional model, realizes the real-time positioning and management of goods;Step S4: the data center of warehouse is established, collects and integrates the real-time information and historical data of goods, predicts demand using big data analysis technology, and based on the prediction result, the intelligent scheduling and optimization of inventory are carried out.Step S5: establish inventory level model, combine demand prediction and supply chain information, real-time monitoring inventory level, and according to the predicted demand and inventory condition, intelligent distribution and direct distribution decision are made.The application can effectively improve the utilization rate of warehouse, and improve the utilization efficiency and turnover rate of assets, reduce the backlog.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of resource allocation, in particular to a smart secondary metering warehouse management method. BACKGROUND

[0002] In recent years, with the rapid economic development and the popularization of the policy of optimizing the business environment, the number of new users and small and micro enterprise customers increases annually. In combination with the HPLC project construction and the growth of the industry, the warehouse area is maximized. In particular, the city power supply company needs to bear the distribution of metering equipment in multiple regions according to the intensive scheme, and the warehouse capacity is under great pressure. Limited by the area, the warehouse capacity cannot be increased any more, and there will be a short shortage in case of disaster. SUMMARY

[0003] In order to solve the above problems, the purpose of the present application is to provide a smart secondary metering warehouse management method, which can effectively improve the utilization rate of the warehouse, improve the utilization efficiency and turnover rate of assets, and reduce the backlog.

[0004] In order to achieve the above purpose, the present application adopts the following technical scheme:

[0005] A smart secondary metering warehouse management method comprises the following steps:

[0006] Step S1: using three-dimensional laser scanning technology to conduct full three-dimensional space scanning on the warehouse, obtaining accurate spatial data inside the warehouse, and based on the specific needs of the warehouse and the storage characteristics of the goods, arranging the positions of the shelves and equipment in the warehouse intelligently based on the simulated annealing algorithm, obtaining a warehouse construction scheme;

[0007] Step S2: based on the obtained warehouse construction scheme, constructing a corresponding warehouse three-dimensional model;

[0008] Step S3: based on RFID technology, identifying and tracking the goods, integrating the RFID data with the warehouse three-dimensional model, and realizing real-time positioning and management of the goods;

[0009] Step S4: establishing a data center of the warehouse, collecting and integrating real-time information and historical data of the goods, predicting the demand by using big data analysis technology, and based on the prediction result, intelligently scheduling and optimizing the inventory, improving the utilization efficiency and turnover rate of assets, and reducing the backlog.

[0010] Step S5: establishing an inventory level model, combining demand prediction and supply chain information, monitoring the inventory level in real time, and making intelligent allocation and direct distribution decisions according to the predicted demand and inventory situation.

[0011] Further, the step S1 specifically comprises:

[0012] Step S11: using three-dimensional laser scanning technology to conduct full three-dimensional space scanning on the warehouse, covering the ground, walls, ceiling, corners and obstacles in the warehouse, obtaining three-dimensional point cloud data of the warehouse, representing accurate spatial information inside the warehouse;

[0013] Step S12: repairing and optimizing the generated three-dimensional point cloud data to remove noise points and invalid data

[0014] Step S13: based on point cloud registration algorithm, registering and splicing the point cloud data obtained from different scanning positions to obtain a complete three-dimensional model of the warehouse;

[0015] Step S14: based on the design requirements of the warehouse, modeling the shelves and equipment in the warehouse;

[0016] Step S15: based on the simulated annealing algorithm, intelligently arranging the shelves and equipment in the warehouse to obtain a warehouse construction scheme.

[0017] Further, the step S12 first uses statistical filtering to remove outliers, and then uses Gaussian filtering to smooth the data, as follows:

[0018] A spherical neighborhood with a radius of r is used for statistical filtering, and for each point P i (x i ,y i ,z i ), the average value and standard deviation of the points in its neighborhood are calculated, and if the deviation of the distance of the point from the average value exceeds a predetermined threshold, it is determined to be a noise point, as follows:

[0019]

[0020]

[0021] If |P i -mu|>n·σ, then P i is a noise point; wherein P j represents the points in the neighborhood, k represents the number of points in the neighborhood, and n is the threshold value;

[0022] After statistical filtering to remove outliers, Gaussian filtering is used to smooth the data and remove remaining noise points:

[0023]

[0024] Wherein I(x,y,z) represents the filtered point cloud data, P(x,y,z) represents the original point cloud data, and σ g represents the standard deviation of the Gaussian kernel, and represents the convolution operation.

[0025] Further, the step S13 is specifically:

[0026] (1) Feature extraction is performed on each point cloud data, SIFT algorithm is used to extract key points, and descriptors are calculated;

[0027] (2) RANSAC algorithm is used to realize point cloud registration by matching the descriptors of the key points;

[0028] First, match the key point descriptors in the two point cloud data sets, and use the nearest neighbor algorithm to find the best match of each key point in the other point cloud;

[0029] A certain number of pairs of matched key points are randomly selected to estimate the transformation model;

[0030] The selected point pairs are used to estimate the transformation model, and the rotation matrix and translation vector are calculated by least squares method;

[0031] Apply the estimated transformation model to all point pairs and calculate their distances to the corresponding points. According to the set threshold, judge which point pairs belong to the inliers, i.e. the point pairs consistent with the estimated model;

[0032] Repeat the above steps several times, and select the estimated model with the most inliers as the final point cloud registration result;

[0033] (5) Based on the registered point cloud data, convert the registered point cloud data to voxel representation, after obtaining the attributes of each voxel, perform voxel-level splicing on multiple registered point cloud data; based on the spliced voxel representation, use voxel grid data to generate a continuous three-dimensional surface;

[0034] (6) Use a triangular mesh reconstruction algorithm to perform surface reconstruction on the registered and spliced point cloud data to obtain a complete warehouse three-dimensional model.

[0035] Further, the use of a triangular mesh reconstruction algorithm to perform surface reconstruction on the registered and spliced point cloud data to obtain a complete warehouse three-dimensional model is specifically:

[0036] Use Delaunay triangulation to triangulate the registered and spliced point cloud data, i.e. convert the point cloud data to a mesh composed of triangles;

[0037] After obtaining the initial triangular mesh, perform mesh optimization, including mesh simplification, smoothing and subdivision, to optimize the topological structure of the mesh;

[0038] Based on the optimized triangular mesh, use Poisson reconstruction algorithm to obtain a complete three-dimensional model:

[0039] Use the normal vector estimation algorithm to calculate the normal vector f{N} of each point;

[0040] Construct the gradient field and convert the normal vector field into the gradient field;

[0041] Using the gradient field as input, solve the Poisson equation to obtain the surface S:

[0042] nabla 2 (S) = nabla*f{N}

[0043] Where S is the surface to be reconstructed, nabla^2 is the Laplace operator, and nabla is the gradient operator.

[0044] Furthermore, step S3 is specifically as follows: setting a data interface for transmitting the data collected by the RFID system to the three-dimensional model system in real time, setting a data point for the location of each item in the three-dimensional model, and this data point is directly associated with the RFID tag, thereby realizing real-time update of the item location; when the item moves in the warehouse, the tag information is read by the RFID reader / writer, and then the item location in the three-dimensional model is updated through the data interface, and the warehouse management personnel use the visual interface of the three-dimensional model.

[0045] Furthermore, the step S4 is specifically as follows:

[0046] Step S41: Based on RFID technology and the three-dimensional warehouse model, the inbound, outbound, in-stock status, and location of items are collected in real time; historical data on historical usage, seasonal factors, and loss are obtained, and the real-time and historical data are stored in the Hadoop database;

[0047] Step S42: constructing an inventory demand forecasting model based on historical data, and forecasting inventory demand based on real-time data;

[0048] Step S43: Based on the demand forecast results, a genetic algorithm is used to optimize inventory scheduling.

[0049] Furthermore, the inventory demand forecasting model is constructed as follows:

[0050] (1) Obtain historical usage, seasonal factors, and loss conditions based on historical data as feature inputs to construct training and test sets;

[0051] (2) Train the LSTM model, Lasso regression model, and ARIMA time series model based on the training set;

[0052] (3) Input the test set data into the trained LSTM model, Lasso regression model and ARIMA time series model respectively to obtain the prediction result y of the neural network modelnn a prediction result y of the regression model Reg a prediction result y of the time series model TS a final prediction result is obtained:

[0053]

[0054] wherein a1, a2, a3 are weight coefficients;

[0055] (4) by cross-validation, the weight coefficients a1, a2, a3 that can make the performance of the combined model optimal are selected, and a final prediction model is obtained.

[0056] Further, the step S43 is specifically:

[0057] A target function is defined for evaluating the fitness of the chromosome, while considering minimizing the total inventory cost and maximizing the asset turnover rate, and is specifically as follows:

[0058] M = a * minC + b * maxP;

[0059] C = C1 + C2 + C3;

[0060]

[0061] Wherein, a and b are weight coefficients; C is the holding cost, and P is the turnover rate; C1, C2 and C3 are the holding cost, ordering cost and shortage cost respectively; N' is the usage quantity, and N is the total inventory quantity.

[0062] The selection, crossover and mutation operations of the genetic algorithm are used to iteratively optimize the chromosome, and the optimal inventory scheduling scheme is searched.

[0063] Further, the step S5 realizes real-time monitoring and management of the inventory level based on the economic order quantity model and the safety inventory model, and is specifically as follows:

[0064]

[0065] Wherein, Q* is the economic order quantity, D is the annual demand, S is the ordering cost, and C is the holding cost; the safety inventory model is:

[0066]

[0067] Wherein, SS is the safety inventory, LT is the supply cycle, is the standard deviation of demand, is the average demand, is the standard deviation of the supply cycle, and z is the safety factor.

[0068] The present application has the following beneficial effects:

[0069] 1、The application uses three-dimensional laser scanning technology combined with simulated annealing algorithm to intelligently arrange the positions of shelves and equipment in the warehouse, can construct the optimal scheme for the warehouse, thereby improving the utilization rate and efficiency of the warehouse, and based on the integration of RFID technology and three-dimensional model, realizes real-time positioning and efficient management of the goods in the warehouse, greatly improves the intelligent level of warehouse operation;

[0070] 2、The application uses genetic algorithm to optimize the scheduling of inventory, takes minimizing the total inventory cost and maximizing the asset turnover rate as the objective function, comprehensively considers the optimization method of inventory cost and asset turnover rate, realizes more effective inventory management and asset utilization;

[0071] 3、The application establishes an inventory level model, combines demand prediction and supply chain information, realizes real-time monitoring of inventory level, and makes intelligent allocation and direct distribution decision according to the predicted demand and inventory situation, effectively optimizes inventory control, improves supply chain efficiency, and reduces inventory cost. BRIEF DESCRIPTION OF DRAWINGS

[0072] Figure 1 The method flowchart of the application. DETAILED DESCRIPTION

[0073] The application will be further described in detail below in combination with the drawings and specific embodiments:

[0074] Reference Figure 1 The application provides a kind of intelligent secondary measurement warehouse management method, it is characterized in that, include the following steps:

[0075] Step S1: using three-dimensional laser scanning technology to carry out all-around three-dimensional space scanning to warehouse, obtains accurate spatial data inside warehouse, and according to the specific demand of warehouse and the characteristics of goods storage, based on simulated annealing algorithm, the positions of shelves and equipment in the warehouse are intelligently arranged, and the warehouse construction scheme is obtained;

[0076] Step S2: based on the obtained warehouse construction scheme, the corresponding warehouse three-dimensional model is constructed;

[0077] Step S3: based on RFID technology, the goods are identified and tracked, RFID data is integrated with warehouse three-dimensional model, and the real-time positioning and management of goods are realized;

[0078] Step S4: establish the data center of warehouse, collect and integrate the real-time information and historical data of goods, predict demand using big data analysis technology, and based on the prediction result, intelligently schedule and optimize inventory, improve the utilization efficiency and turnover rate of assets, and reduce the backlog situation.

[0079] Step S5: Establish an inventory level model, combine demand forecasts and supply chain information, monitor inventory levels in real time, and make intelligent allocation and direct delivery decisions based on forecasted demand and inventory status.

[0080] In this embodiment, step S1 is specifically as follows:

[0081] Step S11: Using 3D laser scanning technology to perform a full-scale 3D spatial scan of the warehouse, covering the floor, walls, ceiling, corners, and obstacles inside the warehouse, to obtain 3D point cloud data of the warehouse, representing the precise spatial information inside the warehouse;

[0082] Step S12: Repair and optimize the generated 3D point cloud data to remove noise points and invalid data

[0083] Step S13: registering and stitching the point cloud data obtained from different scanning positions based on the point cloud registration algorithm to obtain a complete three-dimensional model of the warehouse;

[0084] Step S14: Modeling the shelves and equipment in the warehouse based on the warehouse design requirements;

[0085] Step S15: Intelligently arrange the shelves and equipment in the warehouse based on the simulated annealing algorithm to obtain a warehouse construction plan.

[0086] In this embodiment, step S12 first uses statistical filtering to remove outliers, and then uses Gaussian filtering to smooth the data, as follows:

[0087] Use a spherical neighborhood with a radius of r to perform statistical filtering. For each point P i (x i ,y i ,z i ), calculate the mean and standard deviation of the points in its neighborhood. If the deviation between the distance of the point and the mean exceeds the preset threshold, it is determined to be a noise point, as follows:

[0088]

[0089]

[0090] If |P i -mu|>n·σ, then P i is a noise point; where Pj represents the point in the neighborhood, k represents the number of points in the neighborhood, and n is the threshold;

[0091] After statistical filtering removes outliers, Gaussian filtering is used to smooth the data and remove the remaining noise points:

[0092]

[0093] where I(x, y, z) represents the filtered point cloud data, P(x, y, z) represents the original point cloud data, and σ g represents the standard deviation of the Gaussian kernel, and represents the convolution operation.

[0094] In this embodiment, step S13 is specifically:

[0095] (1) Feature extraction is performed on each point cloud data, SIFT algorithm is used to extract key points, and descriptors are calculated;

[0096] (2) The RANSAC algorithm is used to realize the registration of point clouds by matching the descriptors of key points;

[0097] First, the key point descriptors in the two point cloud data sets are matched, and the nearest neighbor algorithm is used to find the best match of each key point in the other point cloud;

[0098] A certain number of pairs of matched key points are randomly selected to estimate the transformation model;

[0099] The selected point pairs are used to estimate the transformation model, and the rotation matrix and translation vector are calculated by the least squares method;

[0100] The estimated transformation model is applied to all point pairs, and the distances between them and the corresponding points are calculated. According to the set threshold, it is judged which point pairs belong to the inliers, i.e. the point pairs consistent with the estimated model;

[0101] The above steps are repeated several times, and the estimated model with the most inliers is selected as the final point cloud registration result;

[0102] (7) Based on the registered point cloud data, the registered point cloud data is converted into a voxel representation. The voxel representation is a uniform voxel grid that divides the three-dimensional space into a grid of voxels, each representing a volume unit. For each voxel, the properties of the point cloud data inside or nearby can be calculated, such as the number of points, the average normal, etc. This can be achieved by projecting the point cloud data onto the voxel grid and calculating the properties of each voxel. After obtaining the properties of each voxel, the voxel-level splicing of multiple registered point cloud data is performed; This usually involves merging or merging multiple voxel grids to obtain the overall voxel representation. During the splicing process, problems such as overlapping and inconsistency between voxels may need to be solved, such as through weighted fusion or other methods. Based on the spliced voxel representation, the voxel grid data is used to generate a continuous three-dimensional surface. This usually involves extracting isosurfaces from voxel data and performing triangulation and other operations to obtain a continuous three-dimensional surface model;

[0103] (8) The registered and spliced point cloud data is surface reconstructed using a triangular mesh reconstruction algorithm to obtain a complete warehouse three-dimensional model.

[0104] In this embodiment, the point cloud data after registration and splicing is surface reconstructed using a triangular mesh reconstruction algorithm to obtain a complete warehouse three-dimensional model, specifically:

[0105] The point cloud data after registration and splicing is triangulated using Delaunay triangulation, that is, the point cloud data is converted into a mesh composed of triangles;

[0106] After obtaining the initial triangular mesh, mesh optimization is performed, including mesh simplification, smoothing and subdivision, to optimize the topological structure of the mesh;

[0107] Based on the optimized triangular mesh, a Poisson reconstruction algorithm is used to obtain a complete three-dimensional model:

[0108] A normal vector estimation algorithm is used to calculate the normal vector f{N} of each point;

[0109] A gradient field is constructed to convert the normal vector field into a gradient field;

[0110] The gradient field is used as input to solve the Poisson equation to obtain the surface S:

[0111] nabla 2 (S)=nabla*f{N}

[0112] Where S is the surface to be reconstructed, nabla^2 is the Laplacian operator, and nabla is the gradient operator.

[0113] In this embodiment, step S3 is specifically: setting a data interface for real-time transmission of data collected by the RFID system to the three-dimensional model system, and setting a data point for the position of each item in the three-dimensional model. This data point is directly associated with the RFID tag, thereby realizing real-time updating of the position of the item; when the item moves in the warehouse, the RFID reading and writing device reads the tag information, and then updates the position of the item in the three-dimensional model through the data interface, and the warehouse manager views the three-dimensional model through the visual interface.

[0114] In this embodiment, step S4 is specifically:

[0115] Step S41: based on the RFID technology and the warehouse three-dimensional model, real-time collection of the warehousing, de-warehousing, in-warehouse state and position of the item; acquisition of historical data of historical usage, seasonal factors and loss, and storage of real-time data and historical data in a Hadoop database;

[0116] Step S42: based on the historical data, constructing and building a warehouse demand prediction model, and predicting the warehouse demand according to the real-time data;

[0117] Step S43: based on the demand prediction result, the genetic algorithm is used to optimize the scheduling of the inventory.

[0118] In the embodiment, the inventory demand prediction model is constructed as follows:

[0119] (1) According to the historical data, the historical usage, the seasonal factor and the loss are obtained as the characteristic input, and the training set and the test set are constructed;

[0120] (2) According to the training set, the LSTM model, the Lasso regression model and the ARIMA time series model are trained respectively;

[0121] (3) The test set data are respectively input into the trained LSTM model, the Lasso regression model and the ARIMA time series model, and the prediction result y nn of the neural network model, the prediction result y Reg of the regression model and the prediction result y TS of the time series model are obtained, and the final prediction result is obtained:

[0122]

[0123] Where a1, a2, a3 are weight coefficients;

[0124] (4) Through cross-validation, the weight coefficients a1, a2, a3 that can make the performance of the combined model optimal are selected, and the final prediction model is obtained.

[0125] In the embodiment, the step S43 is specifically:

[0126] A target function is defined for evaluating the fitness of the chromosome, while considering minimizing the total inventory cost and maximizing the asset turnover rate, which is specifically as follows:

[0127] M = a min C + b max P;

[0128] C = C1 + C2 + C3;

[0129]

[0130] Where a and b are weight coefficients; C is the holding cost, P is the turnover rate; C1, C2, C3 are the holding cost, ordering cost and shortage cost respectively; N' is the usage quantity, and N is the total inventory quantity.

[0131] The selection, crossover and mutation operations of the genetic algorithm are used to iteratively optimize the chromosome, and the optimal inventory scheduling scheme is searched.

[0132] In the present embodiment, the step S5 realizes the real-time monitoring and management of the inventory level based on the economic order quantity model and the safety stock model, in particular as follows:

[0133]

[0134] wherein Q* is the economic order quantity, D is the annual demand, S is the ordering cost, and C is the holding cost; the safety stock model:

[0135]

[0136] wherein SS is the safety stock, LT is the lead time, is the standard deviation of demand, is the average demand, is the standard deviation of lead time, and z is the safety factor. Preferably, in the present embodiment, z takes 1.96, corresponding to a 95% confidence level.

[0137] Those skilled in the art will understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code.

[0138] The present application is described in reference to the flowcharts and / or block diagrams of the methods, apparatus (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, a special purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowcharts and / or block diagrams block or blocks. Figure 1 The flowcharts and / or block diagrams can also be implemented by one or more computer program instructions implemented in computer readable program code, software, firmware, or a combination thereof. Figure 1 The computer program instructions can be stored in a computer readable storage medium, which can include any medium that can be read and / or written by a machine, such as a general purpose computer, a special purpose computer, an embedded processor, or other programmable data processing apparatus.

[0139] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or a computer readable storage medium to cause a series of operational steps to be performed on the computer or other programmable data processing apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable data processing apparatus implement the functions specified in the flowcharts and / or block diagrams block or blocks. Figure 1 The flowcharts and / or block diagrams can also be implemented by one or more computer program instructions implemented in computer readable program code, software, firmware, or a combination thereof. Figure 1 The computer program instructions can be stored in a computer readable storage medium, which can include any medium that can be read and / or written by a machine, such as a general purpose computer, a special purpose computer, an embedded processor, or other programmable data processing apparatus.

[0140] These computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable devices to generate a computer implemented process, so that the instructions executed on the computer or other programmable devices provide a process for implementing the flowchart Figure 1 one flowchart or multiple flowcharts and / or blocks Figure 1 one flowchart or multiple flowcharts and / or blocks

[0141] The above description is only the preferred embodiment of the present application, not other forms of the present application, any skilled in the art can use the above disclosed technical content to change or modify as equivalent embodiments of equivalent changes. But any simple modification, equivalent change and modification of the above embodiments without departing from the technical solution of the present application, according to the technical essence of the present application, still belongs to the protection scope of the technical solution of the present application.

Claims

1. An intelligent secondary metering warehouse management method, characterized in that: The following steps are involved: Step S1: Use 3D laser scanning technology to perform a comprehensive 3D spatial scan of the warehouse to obtain accurate spatial data inside the warehouse. Based on the specific needs of the warehouse and the storage characteristics of the items, the simulated annealing algorithm is used to intelligently arrange the positions of the shelves and equipment in the warehouse to obtain a warehouse construction plan; Step S2: Based on the obtained warehouse construction plan, construct a corresponding warehouse three-dimensional model; Step S3: Identify and track items based on RFID technology, integrate RFID data with the warehouse 3D model to achieve real-time positioning and management of items; Step S4: Establish a warehouse data center to collect and integrate real-time information and historical data of items, use big data analysis technology to predict demand, and intelligently schedule and optimize inventory based on the prediction results; Step S5: Establish an inventory level model, combine demand forecasts and supply chain information, monitor inventory levels in real time, and make intelligent allocation and direct delivery decisions based on forecasted demand and inventory status; The step S1 is specifically as follows: Step S11: Using 3D laser scanning technology to perform a full-scale 3D spatial scan of the warehouse, covering the floor, walls, ceiling, corners, and obstacles inside the warehouse, to obtain 3D point cloud data of the warehouse, representing the precise spatial information inside the warehouse; Step S12: Repair and optimize the generated 3D point cloud data to remove noise points and invalid data Step S13: registering and stitching the point cloud data obtained from different scanning positions based on the point cloud registration algorithm to obtain a complete three-dimensional model of the warehouse; Step S14: Modeling the shelves and equipment in the warehouse based on the warehouse design requirements; Step S15: Intelligently arrange the shelves and equipment in the warehouse based on the simulated annealing algorithm to obtain a warehouse construction plan; The step S12 first uses statistical filtering to remove outliers, and then uses Gaussian filtering to smooth the data, as follows: Use a spherical neighborhood with a radius of r to perform statistical filtering. For each point P i (x i ,y i ,z i ), calculate the mean and standard deviation of the points in its neighborhood. If the deviation between the distance of the point and the mean exceeds the preset threshold, it is determined to be a noise point, as follows: If |P i -mu|>n·σ, then P i is a noise point; among them, P j represents the points in the neighborhood, k represents the number of points in the neighborhood, and n is the threshold; After statistical filtering removes outliers, Gaussian filtering is used to smooth the data and remove the remaining noise points: Among them, I(x, y, z) represents the filtered point cloud data, P(x, y, z) represents the original point cloud data, σ g represents the standard deviation of the Gaussian kernel, and ⊙ represents the convolution operation.

2. The intelligent secondary metering warehouse management method according to claim 1 is characterized in that: The step S13 is specifically as follows: (1) Perform feature extraction on each point cloud data, use SIFT algorithm to extract key points, and calculate descriptors; (2) Use the RANSAC algorithm to achieve point cloud registration by matching the descriptors of key points; First, the key point descriptors in the two point cloud datasets are matched, and the nearest neighbor algorithm is used to find the best match of each key point in the other point cloud; Randomly select a certain number of pairs from the matched key point pairs to estimate the transformation model; Use the selected point pairs to estimate the transformation model and calculate the rotation matrix and translation vector by the least squares method; Apply the estimated transformation model to all point pairs and calculate their distances to corresponding points. Based on the set threshold, determine which point pairs are inliers, i.e., point pairs that are consistent with the estimated model. Repeat the above steps multiple times and select the estimated model with the most inliers as the final point cloud registration result; (3) Based on the registered point cloud data, the registered point cloud data is converted into voxel representation. After obtaining the attributes of each voxel, multiple registered point cloud data are spliced ​​at the voxel level; based on the spliced ​​voxel representation, the voxel grid data is used to generate a continuous three-dimensional surface; (4) Use the triangular mesh reconstruction algorithm to reconstruct the surface of the aligned and spliced ​​point cloud data to obtain a complete three-dimensional model of the warehouse.

3. The intelligent secondary metering warehouse management method according to claim 2 is characterized in that: The triangular mesh reconstruction algorithm is used to reconstruct the surface of the registered and spliced ​​point cloud data to obtain a complete three-dimensional model of the warehouse, specifically: Use Delaunay triangulation to triangulate the registered and stitched point cloud data, that is, convert the point cloud data into a mesh composed of triangles; After obtaining the initial triangular mesh, mesh optimization is performed, including mesh simplification, smoothing and subdivision, to optimize the topological structure of the mesh; Based on the optimized triangular mesh, the Poisson reconstruction algorithm is used to obtain a complete 3D model: the normal vector estimation algorithm is used to calculate the normal vector f{N} of each point; Construct the gradient field and convert the normal vector field into the gradient field; Using the gradient field as input, solve the Poisson equation to obtain the surface S: nabla 2 (S)=nabla*f{N}; Where S is the surface to be reconstructed, nabla^2 is the Laplace operator, and nabla is the gradient operator.

4. The intelligent secondary metering warehouse management method according to claim 1 is characterized in that: The step S3 specifically includes: setting a data interface for transmitting the data collected by the RFID system to the three-dimensional model system in real time; setting a data point for the location of each item in the three-dimensional model, and this data point is directly associated with the RFID tag, thereby realizing real-time update of the item location; when the item moves in the warehouse, the tag information is read by the RFID reader / writer, and then the item location in the three-dimensional model is updated through the data interface, and the warehouse management personnel can view the visualization interface of the three-dimensional model.

5. The intelligent secondary metering warehouse management method according to claim 1 is characterized in that: The step S4 is specifically as follows: Step S41: Based on RFID technology and the three-dimensional warehouse model, the real-time data of the items entering, leaving, and in-stock status and location are collected; historical data on historical usage, seasonal factors, and loss is obtained, and the real-time and historical data are stored in the Hadoop database; Step S42: constructing an inventory demand forecasting model based on historical data, and forecasting inventory demand based on real-time data; Step S43: Based on the demand forecast results, a genetic algorithm is used to optimize inventory scheduling.

6. The intelligent secondary metering warehouse management method according to claim 5, characterized in that: The inventory demand forecasting model is constructed as follows: (1) Obtain historical usage, seasonal factors, and loss conditions based on historical data as feature inputs to construct training and test sets; (2) Train the LSTM model, Lasso regression model, and ARIMA time series model based on the training set; (3) Input the test set data into the trained LSTM model, Lasso regression model and ARIMA time series model respectively to obtain the prediction result y of the neural network model nn , the predicted result y of the regression model Reg , the prediction result y of the time series model TS , and get the final prediction result: Among them, a1, a2, a3 are weight coefficients; (4) Through cross-validation, the weight coefficients a1, a2, and a3 that can make the performance of the combined model reach the optimal level are selected to obtain the final prediction model.

7. The intelligent secondary metering warehouse management method according to claim 5, characterized in that: The step S43 is specifically as follows: Define an objective function to evaluate the fitness of the chromosome while minimizing the total inventory cost and maximizing the asset turnover rate, as follows: M = α·min C + β·max P; C=C1+C2+C3; Among them, α and β are weight coefficients; C is holding cost, P is turnover rate; C1, C2, C3 are holding cost, ordering cost and out-of-stock cost respectively; N' is the number of units used, N is the total inventory; The selection, crossover and mutation operations of the genetic algorithm are used to iteratively optimize the chromosomes and search for the optimal inventory scheduling solution.

8. The intelligent secondary metering warehouse management method according to claim 1, characterized in that: The step S5 implements real-time monitoring and management of inventory levels based on the economic order quantity model and the safety stock model, as follows: Where Q* is the economic order quantity, D is the annual demand, S is the ordering cost, and C is the holding cost; Safety stock model: Among them, SS is safety stock, LT is supply cycle, is the standard deviation of demand, is the average demand, is the standard deviation of the lead time, and z is the safety factor.

Images