Express sorting error alarm method and system

Abstract

The invention discloses an express sorting error alarm method and system. The method comprises the steps of multi-modal data acquisition and fusion, sorting path dynamic prediction, sorting path dynamic pre-treatment grading alarm and feedback optimization, and construction of a grading early warning mechanism to implement intelligence. The system comprises a distributed edge computing node, a central processing server cluster and a visual alarm terminal. Through cooperative work of data acquisition, path prediction, trajectory tracking and alarm decision, physical characteristics of parcels are dynamically matched with sorting paths, millimeter-level positioning precision is realized by establishing a three-dimensional space coordinate system, misjudgment recognition capability is remarkably improved in combination with an adaptive threshold algorithm, and the accuracy of parcel sorting is improved. And the abnormity detection and early warning functions of the sorting center are further improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent logistics, in particular to a method and system for alarm of express sorting errors. BACKGROUND

[0002] Intelligent logistics is the use of integrated intelligent technology to enable the logistics system to simulate human intelligence, with the ability of thinking, perception, learning, reasoning and judgment, and to solve some problems in logistics. The future development of intelligent logistics will reflect four characteristics: intelligence, integration and hierarchy, flexibility and socialization. The intelligence of a large number of operations and decisions in the logistics process; the integration of transportation, storage, packaging, loading and unloading in the logistics process with the core of logistics management, and the hierarchy of intelligent logistics system. In the express industry, express sorting errors are a common problem in the express industry, which may cause package delays, misdelivery or loss, and the express sorting error alarm system in the intelligent logistics system is an intelligent system for detecting and preventing errors in the express sorting process. The system can accurately determine whether the package has been sorted to the wrong destination through advanced recognition technology, and immediately alarms once the mis-sorting is found, thereby effectively reducing the sorting error rate. However, even intelligent systems may make mistakes when faced with complex situations. Therefore, it is necessary to continuously improve its technical solutions and update new alarm methods to adapt to future logistics market demand. SUMMARY

[0003] (I) Technical problems solved

[0004] In view of the deficiencies of the prior art, the present application provides a method for monitoring express sorting based on digital twinning and machine learning, which establishes a three-dimensional spatial coordinate system to achieve millimeter-level positioning accuracy, and combines an adaptive threshold algorithm to significantly improve the misjudgment recognition ability, especially for abnormal detection and early warning scenarios in automated sorting centers.

[0005] (II) Technical solutions

[0006] To achieve the above purpose, the present application provides the following technical solutions: a method for alarm of express sorting errors, the method comprising: S1, collecting multi-dimensional data of packages by a 3D structured light camera, an RFID reader and a weighing sensor, switching to a millimeter wave radar mode when a metal package is detected, and constructing a digital twinning model; S2, inputting the digital twinning model into a pre-trained path prediction model, training and encrypting the transmission parameters using a federated learning framework, and outputting the optimal sorting path; S3, establishing a three-dimensional spatial coordinate system of the sorting line, using UWB and inertial navigation cooperative positioning, enabling Kalman filter compensation when the signal is lost, calculating the trajectory deviation rate, and triggering abnormal detection when the average deviation rate of the last 5 sampling points exceeds the dynamic threshold. S4, hierarchical alarm, 5-10% deviation starts LED early warning and trajectory correction, 10-20% deviation activates sound and light alarm and reduces speed, >20% deviation stops urgently and generates block chain evidence, and at the same time, the abnormal data is fed back to the path prediction model for online learning optimization.

[0007] As a preferred solution, the method for constructing the path prediction model in S2 comprises: a) constructing a multi-dimensional feature matrix containing package physical characteristics, environmental parameters, and equipment status, wherein the physical characteristics are extracted by wavelet transform to obtain surface texture characteristic parameters; b) adopting an improved K-means++ algorithm for cluster analysis of historical sorting data, and dynamically adjusting the cluster number according to the Bayesian information criterion (BIC) to generate a typical scene sample set; c) automatically adjusting the XGBoost hyperparameters through the Bayesian optimization algorithm, and selecting the optimal parameter combination with the model sparsity as the constraint condition to establish a scene-adaptive prediction model.

[0008] As a preferred solution, the method for establishing the three-dimensional space coordinate system in S3 comprises: deploying 4 groups of UWB positioning base stations equidistantly on both sides of the sorting line, realizing three-dimensional space positioning through the TDOA algorithm, and achieving a positioning accuracy of ±3 mm and a sampling frequency of 100 Hz.

[0009] As a preferred solution, the method for determining the dynamic threshold in S3 is: based on the real-time throughput of the sorting line, environmental temperature and humidity, equipment vibration amplitude, and other working condition parameters, a hybrid prediction model of LSTM and GBDT is constructed to output a nonlinear threshold reference value of multi-dimensional working condition characteristics, and the dynamic threshold is set at 120% of the reference value.

[0010] As a preferred solution, the online learning optimization in S4 specifically comprises: establishing an abnormal data buffer area, triggering an incremental learning mechanism when the cumulative abnormal sample amount reaches a preset threshold, updating the model parameters using the FTRL optimizer, and at the same time, retaining important historical samples to prevent catastrophic forgetting.

[0011] A fast sorting error alarm system, comprising: a distributed edge computing node for real-time processing of sensor data; a central processing server cluster adopting a Kubernetes architecture, including a path prediction engine, a trajectory analysis engine, and an alarm decision engine; a visual alarm terminal integrated with an AR display device, which can superimpose and display the deviation heat map of the predicted trajectory and the actual trajectory of the package.

[0012] As a preferred solution, the optimal sorting path in S3 needs to meet: the joint angular velocity of the robot arm does not exceed the rated value, and the sorting action time is ≤1.5 seconds.

[0013] As a preferred solution, the distributed edge computing node comprises: A multi-protocol access module supports RS485, ZigBee, LoRa communication protocols, and is compatible with sensing devices of different manufacturers; A data cleaning module adopts a sliding window mechanism for outlier elimination, and the window size is 10 sampling periods; A feature extraction module calculates physical features such as package volume weight ratio, centroid position, and surface curvature in real time; A trusted execution environment (TEE) module adds Laplace noise (ε=0.7) using a differential privacy algorithm during data cleaning, and an adaptive surface curvature calculation unit is built-in the feature extraction module to support centroid correction for irregular packages.

[0014] As a preferred solution, the visual alarm terminal integrates a multi-physical field display layer, which can synchronously superimpose and transmit belt vibration modal analysis and air flow field simulation data, and prompt potential interference risks and issue alarms through color coding.

[0015] (Three) beneficial effects

[0016] Compared with the prior art, the present application provides an express sorting error alarm method and system, which has the following beneficial effects: Firstly, the method of the present application includes multi-source sensing data acquisition to establish a package digital twin model, machine learning-based sorting path dynamic prediction, establishment of a three-dimensional coordinate system to realize real-time tracking of package trajectory, and construction of a hierarchical early warning mechanism to implement intelligent alarm. The system comprises a distributed edge computing node, a central processing server, and a visual alarm terminal. Through the cooperative work of data acquisition, path prediction, trajectory tracking, and alarm decision, the package physical features are dynamically matched with the sorting path, millimeter-level positioning accuracy is achieved by establishing a three-dimensional coordinate system, the misjudgment recognition ability is significantly improved by combining an adaptive threshold algorithm, and the abnormal detection and early warning function of the sorting center is further improved. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The figure is a flowchart of the alarm method of the present application; Figure 2 The figure is a flowchart of the construction method of the path prediction model of the present application; Figure 3 The figure is a schematic diagram of the alarm system module of the present application. DETAILED DESCRIPTION

[0018] In order to better understand the purpose, structure and function of the present application, the present application of an express sorting error alarm method and system will be further described below in combination with the drawings and specific embodiments. Example 1

[0019] Reference Figure 1 The application discloses a package sorting error alarm method, which comprises the following steps: S1, collect multi-dimensional data of the package by a 3D structured light camera, an RFID reader and a weighing sensor, switch to a millimeter wave radar mode when a metal package is detected, and construct a digital twin model; S2, input the digital twin model into a pre-trained path prediction model, train the model using a federated learning framework, encrypt and transmit the parameters, and output an optimal sorting path; S3, establish a three-dimensional space coordinate system of the sorting line, use UWB and inertial navigation to cooperatively locate, enable Kalman filter compensation when the signal is lost, calculate the trajectory deviation rate, and trigger abnormal detection when the average deviation rate of five consecutive sampling points exceeds the dynamic threshold; S4, grade alarm, start LED warning and trajectory correction when the deviation is 5-10%, activate sound and light alarm and reduce speed when the deviation is 10-20%, and stop urgently and generate a blockchain record when the deviation is greater than 20%, and feedback the abnormal data to the path prediction model for online learning and optimization.

[0020] Specifically, S1 is multi-modal data acquisition and fusion: a 3D structured light camera deployed at the entrance of the sorting line is used to acquire three-dimensional size data of the package, an RFID reader is used to read electronic tag information, a weighing sensor is used to acquire quality data, and a fusion framework resistant to metal interference is constructed, and when a metal package is detected, the system automatically switches to a millimeter wave radar assisted identification mode, and a package digital twin model containing physical characteristics and logistics information is constructed; S2 is dynamic prediction of the sorting path: the digital twin model is input into a pre-trained path prediction model, the model fuses historical sorting data, current sorting line state and target bin distribution based on the XGBoost algorithm, and the model training uses a federated learning framework, the edge node uploads gradient parameters through homomorphic encryption after local training, and outputs an optimal sorting path and an expected motion trajectory; S3 is dynamic prediction of the sorting path: a three-dimensional space coordinate system of the sorting line is established, a cooperative positioning system of UWB and MEMS inertial navigation modules is deployed, when the UWB signal is lost, an inertial navigation compensation algorithm based on Kalman filter is enabled, the space coordinates of the package are acquired in real time through the UWB positioning tag, the Euclidean distance deviation rate of the actual motion trajectory and the predicted trajectory is calculated, and when the average deviation rate of five consecutive sampling points exceeds the dynamic threshold, abnormal detection is triggered; Its S4 is hierarchical alarm and feedback optimization: according to the deviation level, the corresponding alarm mechanism is started: the first level deviation (5-10%) triggers the LED indicator light warning and starts the pneumatic fine adjustment device to correct the trajectory deviation, the second level deviation (10-20%) activates the sound and light alarm and reduces the conveyor belt speed to 80%, the third level deviation (>20%) executes mechanical emergency stop and generates a time-stamped block chain storage record, while the abnormal data is fed back to the path prediction model for online learning optimization.

[0021] More specifically, as shown in Figure 2 the construction method of the path prediction model in the present embodiment S2 includes: a) Construct a multi-dimensional feature matrix containing package physical characteristics, environmental parameters and equipment status, wherein the surface texture feature parameters are extracted by wavelet transform; b) Cluster analysis is performed on historical sorting data using an improved K-means++ algorithm, and the number of clusters is dynamically adjusted according to the Bayesian Information Criterion (BIC), generating a typical scene sample set; c) Automatically adjust the XGBoost hyperparameters through the Bayesian optimization algorithm, and select the optimal parameter combination with the model sparsity as the constraint condition, and establish a scene-adaptive prediction model.

[0022] Based on the prior art based on fixed rules or static historical data, the present embodiment adopts a dynamic spatio-temporal coupled XGBoost model, the feature matrix contains 28 dimensions, and innovatively adds package centroid offset, surface curvature coefficient and other physical characteristics, combined with Bayesian optimization and improved K-means++ clustering, realizes scene-adaptive prediction, and the initial cluster center is determined by density peak detection to avoid falling into local optimum.

[0023] Further, the establishment method of the three-dimensional space coordinate system in the present embodiment S3 includes: deploying 4 groups of UWB positioning base stations equidistantly on both sides of the sorting line, realizing three-dimensional space positioning through TDOA algorithm, positioning accuracy reaching ±3mm, sampling frequency being 100Hz. Its 4 UWB base stations (similar to WiFi hotspots) are installed on both sides of the sorting line, the package position is located by calculating the time difference of signal arrival (similar to judging your position with multiple sound sources), the base stations have high-precision clock synchronization (error less than 5ns), further improving the three-dimensional positioning accuracy, and not afraid of sorting line vibration or multi-metal interference signal; The determination method of the dynamic threshold is: based on the real-time throughput of the sorting line, the environmental temperature and humidity, the equipment vibration amplitude and other working condition parameters, a mixed prediction model of LSTM and GBDT is constructed, the nonlinear threshold reference value of the multi-dimensional working condition characteristics is output, and the dynamic threshold is set according to 120% of the reference value. Two AI models: LSTM (good at analyzing time data, such as sorting line speed change) and GBDT (good at processing complex relationships) are used to dynamically adjust the alarm threshold, such as the threshold is 10% normally, if the sorting line vibrates greatly, it is automatically increased to 12%, which can reduce false alarms (such as normal vibration of the conveyor belt will not trigger an alarm, and the threshold adjustment is completed within 2 seconds, the response is faster.

[0024] The online learning optimization in S4 specifically includes: establishing an abnormal data buffer area, triggering an incremental learning mechanism when the cumulative abnormal sample amount reaches a preset threshold, updating the model parameters using the FTRL optimizer, and simultaneously retaining important historical samples to prevent catastrophic forgetting. When encountering abnormal data, store them in the "learning pool" first, and trigger model updating after accumulating a certain amount, so that previous tasks are not affected by model updating. Embodiment 2

[0025] Reference Figure 3 The express sorting error alarm system of the application comprises: Distributed edge computing nodes for real-time processing of sensor data; A central processing server cluster using a Kubernetes architecture, including a path prediction engine, a trajectory analysis engine, and an alarm decision engine; A visual alarm terminal integrated with an AR display device, which can superimpose and display the deviation heat map of the predicted trajectory and the actual trajectory of the package, and workers wearing AR glasses can see the deviation of the predicted path and the actual path of the package.

[0026] Among them, the edge node is a small computer distributed on the sorting line, which can process sensor data in real time, reduce network delay, dynamically allocate cloud resources, and avoid lag during peak periods.

[0027] Specifically, the distributed edge computing node comprises: A multi-protocol access module supporting RS485, ZigBee, and LoRa communication protocols, compatible with different manufacturers' sensing devices; A data cleaning module using a sliding window mechanism to remove outliers, with a window size of 10 sampling periods; A feature extraction module for real-time calculation of physical characteristics such as package volume-weight ratio, centroid position, and surface curvature; A trusted execution environment (TEE) module, differential privacy algorithm is used to add Laplace noise (epsilon=0.7) during data cleaning, and an adaptive surface curvature calculation unit is built in the feature extraction module, supporting centroid correction of irregularly wrapped packages.

[0028] The visual alarm terminal integrates a multi-physical field display layer, can synchronously superimpose and transmit the vibration modal analysis and air flow field simulation data of the conveying belt, and can prompt potential interference risks and alarm through color coding. The vibration and air flow data (for example, high vibration areas are represented by red) are superimposed on the AR interface, and early warning is performed through colors, and the visual guidance can enable workers to quickly locate problems and discover hidden dangers in advance.

[0029] Compared with the traditional technical solution, the false positive rate is reduced, the express sorting monitoring technology based on digital twinning and machine learning is used, a three-dimensional space coordinate system is established to realize millimeter-level positioning accuracy, a self-adaptive threshold algorithm is combined to significantly improve the misjudgment recognition ability, and the abnormal detection and early warning function of the sorting center is further improved, and the express sorting monitoring technology is especially suitable for the abnormal detection and early warning scene of the automatic sorting center.

[0030] It can be understood that the present application is described through some embodiments, and those skilled in the art know that various changes or equivalent replacements can be made to the features and embodiments without departing from the spirit and scope of the present application. In addition, under the guidance of the present application, the features and embodiments can be modified to adapt to specific conditions and materials without departing from the spirit and scope of the present application. Therefore, the present application is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of the present application belong to the scope of protection of the present application.

Claims

1. A method for alarming express sorting errors, characterized in that: The method comprises: S1. Collect multi-dimensional package data through a 3D structured light camera, RFID reader, and weighing sensor. When a metal package is detected, the millimeter-wave radar mode is switched to build a digital twin model. S2. Input the digital twin model into the pre-trained path prediction model, use the federated learning framework to train and encrypt the transmission parameters, and output the optimal sorting path; S3. Establish a three-dimensional spatial coordinate system for the sorting line, use UWB and inertial navigation for collaborative positioning, enable Kalman filter compensation when the signal is lost, calculate the trajectory deviation rate, and trigger anomaly detection when the average deviation rate of five consecutive sampling points exceeds the dynamic threshold; S4, graded alarm: 5-10% deviation activates LED warning and trajectory correction, 10-20% deviation activates sound and light alarm and slows down, >20% deviation makes emergency stop and generates blockchain evidence, and at the same time feeds abnormal data into the path prediction model for online learning optimization.

2. The express sorting error alarm method according to claim 1, characterized in that: The method for constructing the path prediction model in S2 includes: a) Construct a multi-dimensional feature matrix containing the package's physical characteristics, environmental parameters, and device status. The physical characteristics are used to extract surface texture feature parameters through wavelet transform. b) Use the improved K-means++ algorithm to perform cluster analysis on the historical sorting data. The number of clusters is dynamically adjusted according to the Bayesian Information Criterion (BIC) to generate a typical scenario sample set. c) Automatically adjust XGBoost hyperparameters through the Bayesian optimization algorithm, select the optimal parameter combination based on model sparsity as a constraint, and establish a scenario-adaptive prediction model.

3. The express sorting error alarm method according to claim 1, characterized in that: The method for establishing the three-dimensional spatial coordinate system in S3 includes: deploying four groups of UWB positioning base stations equidistantly on both sides of the sorting line, realizing three-dimensional spatial positioning through the TDOA algorithm, with a positioning accuracy of ±3mm and a sampling frequency of 100Hz.

4. The express sorting error alarm method according to claim 1, characterized in that: The method for determining the dynamic threshold in S3 is as follows: based on operating parameters such as the real-time throughput of the sorting line, ambient temperature and humidity, and equipment vibration amplitude, a hybrid prediction model of LSTM and GBDT is constructed to output a nonlinear threshold reference value of multi-dimensional operating condition characteristics, and the dynamic threshold is set at 120% of the reference value.

5. The express sorting error alarm method according to claim 1, characterized in that: The online learning optimization in S4 specifically includes: establishing an abnormal data buffer, triggering an incremental learning mechanism when the accumulated abnormal sample volume reaches a preset threshold, using the FTRL optimizer to update the model parameters, and retaining important historical samples to prevent catastrophic forgetting.

6. A courier sorting error alarm system, characterized in that: include: Distributed edge computing nodes for real-time processing of sensor data; The central processing server cluster uses the Kubernetes architecture and includes a path prediction engine, a trajectory analysis engine, and an alarm decision engine; The visual alarm terminal integrates an AR display device, which can overlay and display a heat map showing the deviation between the predicted and actual trajectory of the package.

7. The express sorting error alarm system according to claim 6, characterized in that: The distributed edge computing node includes: Multi-protocol access module, supports RS485, ZigBee, and LoRa communication protocols, and is compatible with sensor devices from different manufacturers; The data cleaning module uses a sliding window mechanism to remove outliers, with a window size of 10 sampling periods; Feature extraction module, which calculates physical features such as package volume-to-weight ratio, center of mass position, and surface curvature in real time; The Trusted Execution Environment (TEE) module uses a differential privacy algorithm to add Laplace noise (ε=0.7) during data cleaning, and the feature extraction module has a built-in adaptive surface curvature calculation unit that supports center of mass correction for irregular packages.

8. The express sorting error alarm system according to claim 6, characterized in that: The visual alarm terminal integrates a multi-physics field display layer, which can simultaneously superimpose conveyor belt vibration modal analysis and air flow field simulation data, and prompt potential interference risks through color coding and issue an alarm.

Images