Internet of Things distribution terminal sorting data dynamic processing system based on cloud computing

Abstract

The invention relates to the technical field of sorting anomaly detection, in particular to an Internet of Things distribution terminal sorting data dynamic processing system based on cloud computing, which comprises a region modeling unit, a graph structure modeling unit, an anomaly detection unit and an anomaly alarm unit, specifically, the area modeling unit is used for constructing a distribution network based on all distribution points in a target distribution area, and transmitting the distribution network to the graph structure modeling unit; the sorting flow direction of a target distribution object is obtained based on the sorting data of the distribution terminal of the Internet of Things, and a sorting graph structure is constructed based on the distribution network and the sorting flow direction. According to the invention, through the area modeling unit and the graph structure modeling unit, the system can construct the distribution network based on the target distribution area and the sorting flow direction of the distribution network and perform accurate modeling, which is helpful for ensuring that each link in the sorting process can be dynamically tracked and optimized, thereby improving the efficiency and accuracy of the whole distribution chain, and reducing the labor intensity of workers. And human errors and operation errors are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of sorting anomaly detection, and in particular to a cloud computing-based sorting data dynamic processing system for an Internet of Things distribution terminal. Background Art

[0002] Cloud computing is an Internet-based computing model that allows users to access shared computing resources (such as servers, storage, databases, software, etc.) through the Internet without having to manage or maintain them locally. These resources are usually provided and managed by a third party, and users can dynamically increase or decrease resource usage as needed. Distribution terminal equipment refers to hardware equipment in the distribution process, such as smart trucks, automatic sorting machines, sensors, GPS positioning devices, etc. These devices exchange data and communicate through the Internet of Things technology.

[0003] Traditional systems often rely on manual or semi-automated processes to manage distribution networks and sorting flows. Due to the lack of accurate regional modeling and graph structure modeling tools, the links in the sorting process are difficult to track in real time, which may result in sorting errors, low efficiency and waste of resources, and it is difficult to respond to emergencies quickly, which is prone to delays. In traditional systems, the sorting process often relies on manual inspection and intervention. Operators may make mistakes due to fatigue, lack of experience or other factors, such as incorrect classification of items and improper path selection. This not only increases the probability of human errors, but also may cause waste of resources and time delays. Traditional systems usually lack real-time monitoring capabilities based on data analysis, and abnormal situations often rely on manual inspections to discover and handle them. This method is inefficient and easy to miss potential problems, and cannot detect and correct errors in a timely manner. For example, incorrect sorting flows or path deviations may be ignored, leading to the spread of distribution problems. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a cloud computing-based Internet of Things distribution terminal sorting data dynamic processing system.

[0005] The technical solution adopted to solve the above technical problems is: a cloud computing-based IoT distribution terminal sorting data dynamic processing system, including a regional modeling unit, a graph structure modeling unit, an anomaly detection unit and an anomaly alarm unit, specifically:

[0006] The regional modeling unit is used to build a distribution network based on all distribution points in the target distribution area, and transmit the distribution network to the graph structure modeling unit;

[0007] The graph structure modeling unit is used to receive the distribution network transmitted by the regional modeling unit, and obtain the sorting flow direction of the target distribution object based on the sorting data of the Internet of Things distribution terminal, build a sorting graph structure based on the distribution network and the sorting flow direction, and transmit the sorting graph structure to the abnormality detection unit;

[0008] The anomaly detection unit is used to receive the sorting graph structure transmitted by the graph structure modeling unit, and perform anomaly detection on the sorting graph structure based on the trained anomaly detection model to obtain an anomaly score corresponding to the sorting graph structure, that is, to obtain an anomaly score of the sorting flow direction of the target delivery object, and transmit the anomaly score of the sorting flow direction of the target delivery object to the anomaly alarm unit;

[0009] The abnormality alarm unit is used to receive the abnormality score of the sorting flow direction of the target delivery object transmitted by the abnormality detection unit, and compare the abnormality score of the sorting flow direction of the target delivery object with a preset score threshold to determine the abnormality level of the sorting flow direction of the target delivery object, and issue a corresponding abnormality alarm based on the abnormality level of the sorting flow direction of the target delivery object.

[0010] Preferably, the distribution network includes a plurality of distribution nodes and a plurality of distribution edges, wherein the distribution nodes correspond to the distribution points, and the distribution edges correspond to roads between the distribution points.

[0011] Preferably, the sorting graph structure includes a plurality of sorting nodes and a plurality of sorting flow edges, wherein the sorting nodes correspond to the delivery points, and the sorting flow edges correspond to the sorting flows.

[0012] Preferably, the anomaly detection model adopts a graph neural network, and the anomaly detection model includes an autoencoding module and an anomaly score generation module.

[0013] Preferably, the autoencoder module includes an encoder and a decoder, the autoencoder module is used to obtain a low-dimensional space representation of the sorting node from feature data sampling of the normal sorting node, wherein the encoder performs convolution embedding on the sorting node based on the adjacency matrix corresponding to the sorting graph structure based on multi-layer graph convolution to obtain the Gaussian distribution parameters of the embedding vector of the intermediate layer, the decoder reconstructs the embedding vector of the sorting node through graph convolution to obtain a reconstructed vector, and obtains the loss function by minimizing the difference between the input feature vector and the reconstructed vector, and the abnormal score generation module adopts a multi-layer perceptron, and the abnormal score generation module generates an abnormal score according to the feature representation obtained by the autoencoder module.

[0014] Preferably, the expression of the Gaussian distribution parameter of the embedding vector is as follows: ;

[0015] in, represents the Gaussian distribution parameters of the embedding vector, Represents a sorting node Representation in latent space, represents the mean vector of the embedding vector, represents the variance vector of the embedding vector, represents a normal distribution, where , represents the standard deviation vector of the embedding vector, represents sampling of a random variable from a standard normal distribution.

[0016] Preferably, the expression of the decoder is as follows: ;

[0017] in, Represents the sorting node in the reconstruction vector and sorting nodes The existence probability of the middle edge, Represents a neural network.

[0018] Preferably, the loss function is expressed as follows: ;

[0019] in, represents the loss function, represents the total number of edges in the sorting graph structure, Represents the sorting node in the reconstruction vector and sorting nodes The true label of the edge in , if the edge exists, then , otherwise, .

[0020] Preferably, the expression of the abnormal score generation module is as follows: ;

[0021] in, and Represents the first Layer and layer, and for the last layer, becomes the anomaly score, Represents the first The weight of the layer, Represents the first The bias term of the layer, represents an extra dimension, Denotes the weight of the additional dimension.

[0022] The beneficial effects of the present invention are as follows:

[0023] (1) Through the regional modeling unit and the graph structure modeling unit, the system of the present invention can construct a distribution network and its sorting flow based on the target distribution area and perform accurate modeling, which helps to ensure that each link in the sorting process can be dynamically tracked and optimized, thereby improving the efficiency and accuracy of the entire distribution chain, reducing human errors and operational errors, and being able to identify potential problems in real time, reducing delays caused by manual inspections and traditional methods;

[0024] (2) The present invention uses an anomaly detection unit to monitor the sorting flow in real time based on IoT data and a trained anomaly detection model, and can promptly detect abnormal behaviors or deviations, such as incorrect item classification, path deviation, etc. This is an anomaly detection based on intelligent algorithms;

[0025] (3) The present invention can automatically determine the abnormal level of the sorting flow according to the set threshold value through the abnormal alarm unit and trigger an early warning. This automated alarm mechanism can greatly reduce the delay of human judgment, improve the response speed of problem handling, and ensure that measures can be taken immediately when an abnormal situation occurs, thereby ensuring the stability and safety of the distribution process. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A schematic diagram of the system architecture of an overall system in an embodiment of the present invention.

[0027] Figure numerals: 1. Regional modeling unit; 2. Graph structure modeling unit; 3. Anomaly detection unit; 4. Anomaly alarm unit. DETAILED DESCRIPTION

[0028] Embodiment 1, as Figure 1 As shown, the cloud computing-based IoT distribution terminal sorting data dynamic processing system proposed by the present invention includes a regional modeling unit 1, a graph structure modeling unit 2, an anomaly detection unit 3 and an anomaly alarm unit 4, specifically:

[0029] The regional modeling unit 1 is used to build a distribution network based on all distribution points in the target distribution area, and transmit the distribution network to the graph structure modeling unit 2;

[0030] The graph structure modeling unit 2 is used to receive the distribution network transmitted by the regional modeling unit 1, and obtain the sorting flow direction of the target distribution object based on the sorting data of the Internet of Things distribution terminal, build a sorting graph structure based on the distribution network and the sorting flow direction, and transmit the sorting graph structure to the anomaly detection unit 3;

[0031] The anomaly detection unit 3 is used to receive the sorting graph structure transmitted by the graph structure modeling unit 2, and perform anomaly detection on the sorting graph structure based on the trained anomaly detection model to obtain an anomaly score corresponding to the sorting graph structure, that is, to obtain an anomaly score of the sorting flow direction of the target delivery object, and transmit the anomaly score of the sorting flow direction of the target delivery object to the anomaly alarm unit 4;

[0032] The abnormal alarm unit 4 is used to receive the abnormal score of the sorting flow direction of the target delivery object transmitted by the abnormal detection unit 3, and compare the abnormal score of the sorting flow direction of the target delivery object with a preset score threshold to determine the abnormal level of the sorting flow direction of the target delivery object, and issue a corresponding abnormal alarm based on the abnormal level of the sorting flow direction of the target delivery object.

[0033] In the present invention, the target distribution area refers to a specific geographical area, usually refers to all places where goods need to be delivered in the distribution network within a period of time; it can be a city, region, warehouse network or other spatial scope of logistics management; the distribution point refers to a node in the distribution network, usually refers to the place where the goods need to be delivered, such as customer address, warehouse, sorting center, etc.; the distribution network refers to the network architecture connecting all distribution points to ensure that the goods are delivered from the supplier or warehouse to the end user along a certain path; the distribution network contains the following basic information: distribution point data: the location (coordinates) and basic attributes of each distribution point (for example: distribution demand, priority, service time window, etc.), distribution route data: the connection relationship (i.e. path) between distribution points, each path has a certain distance, transportation time or cost and other information, road network topology: based on the topological structure of the distribution network, the relationship between nodes (distribution points) and edges (paths connecting distribution points) is constructed; when constructing a sorting graph, the direction and attributes of the edges need to be defined according to the sorting flow direction of the items. For example, if an item flows from distribution point A to distribution point B, a directed edge will be drawn from node A to node B in the sorting graph; the sorting path of each delivery item will generate a directed edge in the sorting graph, indicating that the item flows from one node (distribution point) to another node; if the item needs to pass through multiple distribution points or sorting stations to reach the target distribution point, multiple edges are formed in the sorting graph to connect these nodes in sequence; the graph structure modeling unit refers to the nodes and connections (such as distribution paths) in the distribution network represented in the form of a graph structure; the graph structure is a mathematical model, usually composed of nodes and edges, used to describe the relationship between things; in this scenario, it is used to describe the various distribution points in the distribution network and the connections between them; the Internet of Things distribution terminal refers to the equipment used to perform physical distribution tasks, usually an intelligent device embedded with sensors, communication modules, etc., used to collect sorting data, transmit information and control equipment in real time; the Internet of Things (IoT) device The equipment can provide data support, such as real-time sorting data, location monitoring and other information; sorting data refers to the relevant data of classification, sorting and packaging of items according to specified order or conditions in the process of logistics distribution; it usually contains information such as the sorting path, target location, and sorting time of each item; sorting flow refers to the flow path of logistics items in a certain order and rule during the sorting process; the sorting flow describes the path of items from the source to the final destination, which is usually affected by factors such as the warehousing system, sorting rules, and transportation conditions; the sorting graph structure refers to a network structure constructed using graph theory based on sorting data and sorting flow, which is used to describe the sorting path of items; the nodes in the graph structure may be sorting sites or distribution points, and the edges are the paths for item transmission; the anomaly detection unit refers to a module in the system that is used to identify and analyze abnormal or unexpected behaviors in the sorting graph structure; for example, anomaly detection can detect whether there is an error in the sorting flow of items, whether the distribution route is abnormal, etc.This unit usually uses machine learning or statistical methods to detect abnormal patterns; the anomaly score is the result of quantitatively scoring the anomalies that may exist in the sorting graph structure; the data is usually analyzed through a model to obtain a numerical value to indicate the degree of anomaly. The higher the score, the greater the possibility of anomaly; the abnormal alarm unit refers to the module in which the system automatically generates an alarm to notify relevant personnel after detecting an anomaly; it will compare the anomaly score with the preset threshold. If the score exceeds the threshold, it will trigger an alarm to remind the management personnel to handle it; the score threshold refers to a standard preset in anomaly detection to determine whether the sorting flow is abnormal; if the anomaly score exceeds this threshold, it will be considered an anomaly and trigger an alarm; the threshold can be set based on historical data, empirical rules or machine learning models; the anomaly level is based on the anomaly score. The system divides the anomaly into different levels; for example, a score within a certain range can be considered a minor anomaly, while a higher score may indicate a serious anomaly; this level helps guide subsequent responses and handling measures; the anomaly alarm is the process in which the system notifies relevant personnel through an alarm mechanism after detecting an anomaly; the alarm can be in different forms, such as SMS, email or APP notification, to help managers take quick action. ;

[0034] Embodiment 2, the cloud computing-based Internet of Things distribution terminal sorting data dynamic processing system proposed in the present invention, compared with embodiment 1, this embodiment also includes: the distribution network includes multiple distribution nodes and multiple distribution edges, wherein the distribution nodes correspond to the distribution points, and the distribution edges correspond to the roads between the distribution points.

[0035] In an optional embodiment, the sorting graph structure includes a plurality of sorting nodes and a plurality of sorting flow edges, wherein the sorting nodes correspond to the delivery points, and the sorting flow edges correspond to the sorting flows.

[0036] In an optional embodiment, the anomaly detection model adopts a graph neural network, and the anomaly detection model includes an autoencoding module and an anomaly score generation module.

[0037] It should be noted that graph neural network is a deep learning model based on graph structured data. It can effectively learn the relationship between nodes in the graph structure. Unlike traditional neural networks (such as convolutional neural networks (CNN), GNN can process graph data, where each node represents a data point and each edge represents the relationship between nodes.

[0038] In an optional embodiment, the autoencoder module includes an encoder and a decoder, and the autoencoder module is used to obtain a low-dimensional space representation of the sorting node from feature data sampling of a normal sorting node, while avoiding the overfitting problem caused by the reconstruction of the autoencoder module, wherein the encoder performs convolution embedding on the sorting node based on the adjacency matrix corresponding to the sorting graph structure based on multi-layer graph convolution to obtain the Gaussian distribution parameters of the embedding vector of the intermediate layer, and the decoder reconstructs the embedding vector of the sorting node through graph convolution to obtain a reconstructed vector, and obtains a loss function by minimizing the difference between the input feature vector and the reconstructed vector, and avoids the overfitting problem caused by the reconstruction of the autoencoder module based on the loss function, and the abnormal score generation module adopts a multi-layer perceptron, and the abnormal score generation module generates an abnormal score according to the feature representation obtained by the autoencoder module.

[0039] It should be noted that the autoencoder module is an unsupervised learning model, which consists of an encoder and a decoder. Its purpose is to learn the low-dimensional representation (encoding) of the data and then reconstruct the input data through decoding. The autoencoder module is widely used in tasks such as feature learning, dimensionality reduction and anomaly detection. Graph convolution is a technology for processing graph data. It can consider the relationship between a node and its neighboring nodes and update the representation of the node through convolution operations. In this context, the encoder processes the features of the sorting nodes based on the graph convolution network (GCN). The feature vector of each node not only contains its own information, but also integrates the features of its neighboring nodes. Gaussian distribution (or normal distribution) is a common probability distribution, which is widely used to describe the distribution pattern of data. In the autoencoder module, the embedding vector of each node is usually represented as a Gaussian distribution, and the parameters of the distribution (such as mean and variance) are learned. This practice is common in the variational autoencoder module (VAE), which enables the model to learn the potential probability distribution of the data. Multilayer perceptron is a feedforward neural network, which is usually used to generate nonlinear mapping of features. Here, MLP is used to generate anomaly scores based on the node features learned by the autoencoder module. MLP can capture complex data patterns through multiple layers of nonlinear transformations.

[0040] In an optional embodiment, the expression of the Gaussian distribution parameter of the embedding vector is as follows: ;

[0041] in, represents the Gaussian distribution parameters of the embedding vector, Represents a sorting node Representation in latent space, represents the mean vector of the embedding vector, represents the variance vector of the embedding vector, represents a normal distribution, where , represents the standard deviation vector of the embedding vector, represents sampling of a random variable from a standard normal distribution.

[0042] In an optional embodiment, the decoder is expressed as follows: ;

[0043] in, Represents the sorting node in the reconstruction vector and sorting nodes The existence probability of the middle edge, Represents a neural network.

[0044] In an optional embodiment, the loss function is expressed as follows: ;

[0045] in, represents the loss function, represents the total number of edges in the sorting graph structure, Represents the sorting node in the reconstruction vector and sorting nodes The true label of the edge in , if the edge exists, then , otherwise, .

[0046] In an optional embodiment, the expression of the abnormality score generation module is as follows: ;

[0047] in, and Represents the first Layer and layer, and for the last layer, becomes the anomaly score, Represents the first The weight of the layer, Represents the first The bias term of the layer, represents an extra dimension, Denotes the weight of the additional dimension.

[0048] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. The cloud computing-based IoT distribution terminal sorting data dynamic processing system is characterized by: The system comprises a regional modeling unit (1), a graph structure modeling unit (2), an anomaly detection unit (3) and an anomaly alarm unit (4), specifically: The regional modeling unit (1) is used to construct a distribution network based on all distribution points in the target distribution area, and transmit the distribution network to the graph structure modeling unit (2); The graph structure modeling unit (2) is used to receive the distribution network transmitted by the regional modeling unit (1), and obtain the sorting flow direction of the target distribution object based on the sorting data of the Internet of Things distribution terminal, build a sorting graph structure based on the distribution network and the sorting flow direction, and transmit the sorting graph structure to the abnormality detection unit (3); The anomaly detection unit (3) is used to receive the sorting graph structure transmitted by the graph structure modeling unit (2), and perform anomaly detection on the sorting graph structure based on the trained anomaly detection model to obtain an anomaly score corresponding to the sorting graph structure, that is, to obtain an anomaly score of the sorting flow direction of the target delivery object, and transmit the anomaly score of the sorting flow direction of the target delivery object to the anomaly alarm unit (4); The abnormality alarm unit (4) is used to receive the abnormality score of the sorting flow direction of the target delivery object transmitted by the abnormality detection unit (3), and compare the abnormality score of the sorting flow direction of the target delivery object with a preset score threshold to determine the abnormality level of the sorting flow direction of the target delivery object, and issue a corresponding abnormality alarm based on the abnormality level of the sorting flow direction of the target delivery object.

2. According to the cloud computing-based Internet of Things distribution terminal sorting data dynamic processing system according to claim 1, it is characterized in that: The distribution network includes a plurality of distribution nodes and a plurality of distribution edges, wherein the distribution nodes correspond to the distribution points, and the distribution edges correspond to roads between the distribution points.

3. The cloud computing-based Internet of Things distribution terminal sorting data dynamic processing system according to claim 2 is characterized in that: The sorting graph structure includes a plurality of sorting nodes and a plurality of sorting flow edges, wherein the sorting nodes correspond to the delivery points, and the sorting flow edges correspond to the sorting flows.

4. The cloud computing-based Internet of Things distribution terminal sorting data dynamic processing system according to claim 3 is characterized in that: The anomaly detection model adopts a graph neural network, and the anomaly detection model includes an autoencoding module and an anomaly score generation module.

5. The cloud computing-based Internet of Things distribution terminal sorting data dynamic processing system according to claim 4 is characterized in that: The autoencoder module includes an encoder and a decoder. The autoencoder module is used to obtain a low-dimensional space representation of the sorting node from feature data sampling of the normal sorting node, wherein the encoder performs convolution embedding on the sorting node based on the adjacency matrix corresponding to the sorting graph structure based on multi-layer graph convolution to obtain the Gaussian distribution parameters of the embedding vector of the intermediate layer, and the decoder reconstructs the embedding vector of the sorting node through graph convolution to obtain a reconstructed vector, and obtains the loss function by minimizing the difference between the input feature vector and the reconstructed vector. The abnormal score generation module adopts a multi-layer perceptron, and the abnormal score generation module generates an abnormal score according to the feature representation obtained by the autoencoder module.

6. The cloud computing-based Internet of Things distribution terminal sorting data dynamic processing system according to claim 5 is characterized in that: The expression of the Gaussian distribution parameter of the embedding vector is as follows: ； in, represents the Gaussian distribution parameters of the embedding vector, Represents a sorting node Representation in latent space, represents the mean vector of the embedding vector, represents the variance vector of the embedding vector, represents a normal distribution, where , represents the standard deviation vector of the embedding vector, represents sampling of a random variable from a standard normal distribution.

7. The cloud computing-based Internet of Things distribution terminal sorting data dynamic processing system according to claim 6 is characterized in that: The expression of the decoder is as follows: ； in, Represents the sorting node in the reconstruction vector and sorting nodes The existence probability of the middle edge, Represents a neural network.

8. The cloud computing-based Internet of Things distribution terminal sorting data dynamic processing system according to claim 7 is characterized in that: The expression of the loss function is as follows: ； in, represents the loss function, represents the total number of edges in the sorting graph structure, Represents the sorting node in the reconstruction vector and sorting nodes The true label of the edge in , if the edge exists, then , otherwise, .

9. The cloud computing-based Internet of Things distribution terminal sorting data dynamic processing system according to claim 5 is characterized in that: The expression of the abnormal score generation module is as follows: ； in, and Represents the first Layer and The output of the layer, and for the last layer, becomes the abnormality score, Represents the first The weight of the layer, Represents the first The bias term of the layer, represents an extra dimension, Denotes the weight of the additional dimension.

Images