A logistics transportation anomaly event identification and handling system and its usage method

Abstract

This invention relates to a system for identifying and handling abnormal events in logistics transportation and its usage method, belonging to the field of logistics and supply chain management technology. It includes modules for data acquisition, anomaly identification, handling decision-making, handling execution, and feedback learning. The data acquisition module uses an edge-cloud collaborative architecture to perform spatiotemporal alignment and streaming fusion of multi-source heterogeneous logistics data to generate a standardized data stream. The anomaly identification module employs a rule engine and multiple dedicated machine learning models for parallel analysis, and uses a weighted decision fusion algorithm for anomaly judgment and classification. The handling decision-making module integrates historical case matching and multi-objective Pareto optimization to generate a set of trade-off solutions. The handling execution module achieves automated task execution through RPA and collaborative communication APIs. The feedback learning module performs closed-loop iterative optimization of the identification model and decision knowledge base based on quantitative evaluation of handling effects. This invention achieves real-time, accurate, and automated identification and handling of abnormal transportation events.

Description

Technical Field

[0001] This invention relates to the field of logistics and supply chain management technology, and in particular to a system for identifying and handling abnormal events in logistics transportation and its usage method. Background Technology

[0002] With the rapid development of global trade and e-commerce, modern logistics and transportation networks are becoming increasingly complex, placing higher demands on the reliability, timeliness, and transparency of the transportation process. In complex transportation scenarios such as long-distance and multimodal transport, abnormal events such as vehicle breakdowns, cargo damage, traffic delays, and route deviations occur frequently, causing huge economic losses and reputational risks to logistics companies. Promoting the intelligent transformation of logistics and transportation management and achieving rapid and accurate response to abnormal events has become an urgent need for the industry's development.

[0003] Currently, while logistics and transportation anomaly management technology has made some progress, it still has significant limitations, specifically: 1. Fragmented data foundation, making it difficult to support accurate global perception: Existing monitoring systems mostly rely on GPS positioning, video surveillance, or independent sensors. Data sources are scattered and formats are heterogeneous, forming "data silos." For example, information such as vehicle location, cargo temperature and humidity, and road weather is fragmented, lacking an effective spatiotemporal alignment and fusion mechanism. This makes it impossible to construct a unified view reflecting the status of the entire transportation chain, resulting in one-sided and delayed risk perception; 2. Single anomaly identification methods, making it difficult to cope with complex risk scenarios: Mainstream technical solutions mostly rely on preset thresholds or simple rules for anomaly judgment (e.g., timeout or over-temperature alarms). This method is effective for obvious, preset anomalies, but it cannot identify complex anomalies caused by multiple factors coupled and gradual pattern changes (such as hidden damage to goods caused by continuous vibration at a specific frequency, or hidden damage caused by driver behavior). 1. The system suffers from several problems: 1) High rates of false alarms and missed alarms (e.g., route deviation), failing to meet the safety early warning requirements for high-value cargo transportation; 2) High reliance on manual decision-making, lacking scientific optimization and automated execution: When anomalies occur, most systems only provide alarms or simple statistical classification results. The generation of disposal plans is highly dependent on the personal experience of dispatchers, resulting in slow response speeds and difficulty in finding the optimal balance among multiple objectives such as cost, timeliness, and customer impact. From decision-making to execution, a large amount of manual communication and coordination is still required, leading to low automation and significant efficiency bottlenecks; 3) The system lacks self-learning capabilities, making it unable to achieve experience accumulation and continuous optimization: Most existing solutions are designed for handling single events. Effective experiences and lessons learned during the disposal process are not systematically recorded, analyzed, and reused. The system lacks the ability to learn from historical disposal results and optimize identification models and decision-making strategies, resulting in stagnant intelligence levels and an inability to adapt to dynamically changing transportation environments and risk patterns.

[0004] In summary, existing technical solutions for logistics and transportation anomaly management are characterized by "fragmentation, superficiality, passivity, and staticity," failing to form a complete capability system from full-chain perception, intelligent identification, optimized decision-making, automatic execution to closed-loop evolution. Therefore, there is an urgent need for a closed-loop management system that can break down data silos, integrate hybrid intelligence, achieve optimized decision-making and automatic execution, and possess continuous self-learning capabilities, in order to fundamentally improve the resilience, efficiency, and intelligence level of logistics and transportation. Summary of the Invention

[0005] This invention provides a logistics transportation anomaly event identification and handling system and its usage method, which solves the problems of fragmented and passive logistics anomaly management in the prior art. It constructs a technical solution that integrates full-chain perception, intelligent identification, optimized decision-making, automatic execution, and closed-loop learning, and can realize proactive, accurate, and automated handling of anomalies and continuous system evolution.

[0006] The present invention provides the following solution to the above-mentioned technical problems: a logistics transportation abnormal event identification and handling system and its usage method, comprising a data acquisition module, an abnormality identification module, a handling decision module, a handling execution module, and a feedback learning module, wherein the data acquisition module is used to collect status data of the entire logistics transportation chain from multiple data sources in real time;

[0007] The anomaly identification module is communicatively connected to the data acquisition module and is used to analyze the acquired data based on preset rules and artificial intelligence models in order to identify and classify abnormal events.

[0008] The decision-making module is communicatively connected to the anomaly identification module and is used to generate an optimized handling plan based on the type and level of the abnormal event.

[0009] The disposal execution module is communicatively connected to the disposal decision module and is used to automatically execute tasks in the disposal plan and coordinate all parties.

[0010] The feedback learning module is communicatively connected to the anomaly identification module and the handling decision module, respectively, and is used to optimize the identification model and handling strategy based on the handling result data.

[0011] The data acquisition module includes an IoT sensing unit, a positioning and communication unit, an external environment perception unit, and a data processing and transmission unit. The IoT sensing unit includes temperature and humidity sensors, triaxial acceleration and vibration sensors, light sensors, and door magnetic sensors deployed on the cargo or vehicle to collect physical data on the cargo's status. The positioning and communication unit includes a high-precision GNSS positioning module and a 4G / 5G vehicle-mounted communication module integrated on the transportation vehicle to acquire and upload location, trajectory, and vehicle operation data in real time. The external environment perception unit includes a vehicle-mounted camera and an external data interface to acquire visual environment information and third-party traffic and weather data.

[0012] The data processing and transmission unit includes an edge computing gateway and a cloud data bus;

[0013] The edge computing gateway is connected to the IoT sensing unit through a low-power wireless communication protocol for local aggregation and preliminary preprocessing of sensor data. The edge computing gateway is connected to the positioning and communication unit and the external environment perception unit through the vehicle internal bus to form a vehicle-side dataset.

[0014] The 4G / 5G vehicle communication module uploads the vehicle-side dataset to the cloud data bus. The cloud data bus simultaneously receives third-party data from external data interfaces and uses stream processing technology to perform time synchronization, spatial alignment, and fusion processing on all multi-source heterogeneous data to form a standardized transportation status snapshot data stream, which is then output to the anomaly identification module.

[0015] By integrating cargo status, vehicle trajectory, visual environment, and third-party data, it provides the complete information dimensions required for anomaly identification. The "edge-cloud" collaborative architecture reduces network bandwidth pressure and cloud computing load, improving system response speed. Through time synchronization and spatial alignment, it solves the problem of data silos. Through feature-level fusion, it integrates data from different dimensions into a unified standardized snapshot, greatly improving the analysis accuracy and reliability of subsequent anomaly identification modules. The modular design and cloud bus facilitate the access of new data sources or sensors, adapting to future business development.

[0016] Based on the above technical solution, the present invention can be further improved as follows.

[0017] Furthermore, the anomaly identification module includes a rule engine unit, a machine learning model unit, and an early warning synthesis unit;

[0018] The rule engine unit performs preliminary anomaly judgment on the transportation status snapshot data stream based on predefined business rules, and the machine learning model unit uses multiple trained models to perform in-depth analysis on the transportation status snapshot data stream to identify complex patterns and predict risks.

[0019] The machine learning model unit includes at least: a time-series regression model for predicting transportation delays, a vibration pattern recognition model for identifying the risk of physical damage to goods, and a trajectory similarity analysis model for detecting path deviations;

[0020] The early warning synthesis unit is used to fuse the outputs of the rule engine unit and the machine learning model unit, and determine the final anomaly type and severity level through a weighted decision algorithm;

[0021] Combining the rapid determinism of rule-based judgment with the model's ability to uncover complex patterns, it can quickly capture obvious anomalies (such as overheating), discover potential risks (such as continuous abnormal vibration patterns that may lead to cargo damage), and predict future problems (such as delays). The listed time series, vibration, and trajectory models directly target the core pain points of logistics (delays, cargo damage, deviations), enabling the identification capabilities to penetrate into the essence of business. By integrating multi-source judgments through weighted decision algorithms, it avoids the limitations of single methods, making anomaly classification and grading more objective and accurate, and reducing false alarms and missed alarms.

[0022] Furthermore, the decision-making module includes a case matching unit, a multi-objective optimization unit, and a solution adjustment unit;

[0023] The case matching unit is used to perform similarity retrieval between the current abnormal event features and the historical handling case knowledge base to obtain a basic handling template. The multi-objective optimization unit is used to solve the variables in the basic handling template with the handling cost, time and customer impact as optimization objectives, and generate a Pareto optimal solution set. The solution adjustment unit is used to provide a human-computer interaction interface for operators to select or fine-tune the final handling solution from the solution set.

[0024] Case matching can quickly provide validated solutions, shorten decision-making time, inherit best practices, and generate a Pareto optimal solution set through multi-objective optimization. It clearly shows the trade-offs between objectives such as cost, time, and customer impact, supporting more comprehensive and economical decision-making. It leverages the system's advantages in computation and search while retaining the final judgment and fine-tuning capabilities of human experts in complex situations, thereby improving the feasibility and acceptability of the solutions.

[0025] Furthermore, the processing execution module includes a task automation unit, a collaborative communication unit, and a process tracking unit:

[0026] The task automation unit integrates a Robotic Process Automation (RPA) tool to automatically execute cross-system operation instructions. The collaborative communication unit integrates an enterprise-level communication platform's application programming interface (API) to automatically create task groups, assign tasks, and notify relevant responsible persons. The process tracking unit is used to monitor the execution status of each sub-task in real time and update the overall progress of handling abnormal events.

[0027] Furthermore, the feedback learning module includes an effect evaluation unit, a model iteration unit, and a knowledge base update unit;

[0028] The effect evaluation unit is used to quantitatively evaluate the handling effect of completed abnormal events according to the key performance indicator (KPI) system. The model iteration unit is used to use the handling feedback data as training samples to perform online learning or periodic incremental training on the model in the machine learning model unit. The knowledge base update unit is used to structure and store the verified and effective handling cases and evaluation results into the historical handling case knowledge base.

[0029] This transforms the system from a static tool into one that continuously learns from historical practice, becoming increasingly intelligent with use, achieving more accurate identification, and making better decisions. It transforms fragmented handling experience into structured, reusable digital assets (model parameters, case library), continuously accumulating core competitiveness for enterprises. Through quantitative KPI evaluation and model retraining, it ensures the continuous improvement of anomaly identification accuracy and the effectiveness of handling solutions, making the system more robust.

[0030] Furthermore, the high-precision GNSS positioning module is a multi-band receiver that supports real-time dynamic RTK technology; the triaxial accelerometer and vibration sensor is a low-noise, high-resolution MEMS accelerometer.

[0031] RTK technology greatly improves positioning accuracy, making path deviation judgment and geofencing alarms more accurate and reliable; low-noise, high-resolution sensors can capture more subtle changes in physical state and provide early warning of potential cargo damage; multi-band receivers have strong anti-interference capabilities and can still work stably in complex environments such as urban canyons.

[0032] On the other hand, a method for identifying and handling abnormal events in logistics transportation, applied to the system as described in any one of claims 1-6, is characterized by comprising the following steps:

[0033] S1. Real-time collection of multi-source logistics data through the IoT sensing unit, positioning and communication unit and external environment sensing unit;

[0034] S2. The data processing and transmission unit performs time synchronization, spatial alignment and fusion processing on multi-source data to form a standardized transportation status snapshot data stream;

[0035] S3. The anomaly identification module analyzes the data stream to identify and classify abnormal events;

[0036] S4. Generate an optimized disposal plan through the disposal decision module;

[0037] S5. The disposal plan is executed automatically through the disposal execution module;

[0038] S6. The feedback learning module optimizes the system based on the handling feedback.

[0039] Furthermore, the identification and classification of abnormal events includes: rapid filtering based on predefined rules; inputting data into multiple parallel models for prediction and deep pattern recognition; and determining the final anomaly type and level by combining rule judgments and the output results of multiple models through a weighted decision fusion algorithm.

[0040] First, a large amount of normal data is quickly filtered through rules to reduce unnecessary complex calculations and improve system response efficiency. Multiple models are analyzed in parallel to make full use of computing resources and deeply mine data value from different dimensions. The decision fusion algorithm combines the judgments of simple rules and complex models to improve the robustness and accuracy of the final anomaly judgment result.

[0041] Furthermore, the optimized solution generation includes: matching similar scenario solution templates from a historical case library; using a multi-objective evolutionary algorithm to optimize the solution path and resources with cost, time, and customer satisfaction as multiple objectives, generating multiple Pareto optimal solutions; and providing a human-computer interaction interface for decision-makers to select or adjust the final solution.

[0042] It integrates experience-based heuristic search (case matching) and mathematically-based precise optimization (multi-objective evolutionary algorithm) to ensure that the solution is both practical and efficient. It generates a set of Pareto optimal solutions rather than a single solution, allowing decision-makers to make flexible choices based on current business priorities (such as whether to prioritize cost or timeliness). It retains necessary human intervention points on the basis of high automation to ensure the final controllability and scenario adaptability of the system decision.

[0043] Furthermore, the system based on handling feedback optimization includes: quantitatively calculating the response time, resolution cycle, and cost deviation of this handling; retraining the identification model using handling process and result data; and converting the effective experience gained from this handling into structured cases and updating them to the historical case knowledge base.

[0044] This enables the "identification-disposal-learning" process to form a closed-loop methodology for sustainable improvement, and clear quantitative evaluation provides data-driven direction and basis for the optimization of models and strategies;

[0045] "Standardized transport status snapshot data stream" refers to a structured data record formed by integrating data from different sources and in different formats under a unified spatiotemporal benchmark. It contains synchronous information from multiple dimensions such as cargo status, vehicle location, and environmental information. "Closed loop" refers to a cyclical improvement mechanism in which the system can use the entire process results of an anomaly handling as feedback to optimize the next identification and decision-making.

[0046] The beneficial effects of this invention are as follows: This invention provides a system for identifying and handling abnormal events in logistics transportation and a method for using it, which has the following advantages:

[0047] 1. Full-chain perception and deep insight capabilities, addressing the problem of "fragmented data foundation": Through the multi-source perception and fusion processing unit in the data acquisition module, especially the "time synchronization, spatial alignment and feature fusion" technology, a "standardized transportation status snapshot data stream" is formed, completely breaking down data silos. By integrating the physical state of goods, vehicle trajectory, visual environment and third-party data, the system has built an unprecedented, multi-dimensional digital perception capability for the transportation process, providing complete information for deep anomaly identification.

[0048] 2. Rapid response and scientific decision-making: The rule engine ensures rapid initial screening at the millisecond level, while parallel model analysis guarantees in-depth risk mining. The disposal decision integrates historical experience and multi-objective optimization, generating not a single solution, but a set of optimized solutions that clearly show the trade-offs, supporting faster, more comprehensive, and more economical scientific decision-making.

[0049] 3. High degree of automation and closed-loop execution: From data fusion, anomaly identification, solution generation to task execution, the system achieves highly automated pipeline operations, significantly reducing reliance on manual labor and operational delays, and ensuring that optimization decisions can be accurately and efficiently implemented.

[0050] 4. The system is self-adaptive and continuously evolving. The core "feedback learning" mechanism makes the system an intelligent agent that can continuously learn from practice. Every action becomes a nutrient for the system's growth, driving the iterative updates of the model and knowledge base. This makes the system more and more intelligent with use, effectively responding to the ever-changing risk environment and building a long-term competitive barrier.

[0051] 5. Flexible architecture and strong scalability: The modular design (especially the cloud data bus) enables the system to easily connect to new data sources (such as new sensors or data platforms) and new analysis models, providing excellent scalability and the ability to flexibly adapt to the needs of future business and technology development.

[0052] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it according to the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Specific embodiments of the present invention are given in detail below with reference to the accompanying drawings. Attached Figure Description

[0053] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0054] Figure 1 A flowchart illustrating a logistics transportation anomaly event identification and handling system and its usage method, provided in an embodiment of the present invention;

[0055] Figure 2 This is a system architecture diagram of a logistics transportation abnormal event identification and handling system and its usage method provided in an embodiment of the present invention. Detailed Implementation

[0056] The following is in conjunction with the appendix Figure 1-2 The principles and features of the present invention are described below. The examples given are for illustrative purposes only and are not intended to limit the scope of the invention. The invention is described more specifically in the following paragraphs by way of example with reference to the accompanying drawings. The advantages and features of the invention will become clearer from the following description. It should be noted that the drawings are in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of the invention.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0058] like Figure 1 As shown, the present invention provides a logistics transportation abnormal event identification and handling system and its usage method, including a data acquisition module, an abnormality identification module, a handling decision module, a handling execution module, and a feedback learning module. Each module is interconnected and works collaboratively through standardized data interfaces and communication protocols.

[0059] The data acquisition module includes the following units:

[0060] 1. IoT Sensing Unit: This unit consists of various miniature sensors deployed inside cargo packaging or on transport vehicles (such as containers and truck compartments). For example, high-precision temperature and humidity sensors are used to monitor the environment of cold chain goods; low-noise, high-resolution triaxial MEMS accelerometers are used to monitor vibrations and impacts on the goods; light sensors are configured to detect whether the cargo packaging has been abnormally opened (such as changes in light); and door magnetic sensors are installed to monitor the opening and closing status of the compartment doors. These sensors communicate via low-power wide area networks (such as LoRa) or short-range wireless protocols (such as ZigBee).

[0061] 2. Positioning and Communication Unit: This unit is integrated into the transportation vehicle. The positioning part adopts a high-precision multi-frequency GNSS receiver that supports RTK technology, which can achieve centimeter-level positioning and accurately depict the vehicle trajectory. The communication part adopts a 4G / 5G vehicle communication module (T-Box), which is responsible for uploading positioning data and vehicle CAN bus data (such as speed and fuel consumption) in real time.

[0062] 3. External Environment Perception Unit: This unit includes vehicle-mounted cameras (such as ADAS cameras and cargo box monitoring cameras) to acquire visual information about road conditions, driving behavior, and cargo stacking status. At the same time, this unit accesses third-party data provided by traffic management departments, meteorological service providers, etc. through API interfaces to obtain real-time road conditions, weather warnings, and other information.

[0063] 4. Data Processing and Transmission Unit: This unit is crucial for data fusion. On the vehicle side, the edge computing gateway aggregates data from the positioning unit and cameras via an internal bus (such as CAN or Ethernet) and receives data from IoT sensors via a wireless network. The gateway performs local preprocessing, such as data cleaning, format standardization, and lightweight computing, to reduce cloud pressure and improve real-time performance. The preprocessed vehicle-side dataset is uploaded to the cloud via a 4G / 5G module. On the cloud, the cloud data bus (such as one built on Kafka or Pulsar) simultaneously receives vehicle-side data streams and third-party data streams. The stream processing engine on the bus (such as Apache Flink) performs timestamp alignment, geographic coordinate matching, and feature fusion on these multi-source heterogeneous data, ultimately generating a standardized transportation status snapshot data stream containing dimensions such as cargo status, vehicle location, and environmental information for subsequent modules to consume.

[0064] As a preferred embodiment of the present invention, the time synchronization, spatial alignment and fusion processing specifically includes the following steps:

[0065] S201. Timestamp Alignment: All accessed data sources (vehicle sensor data, GNSS data, camera video frame metadata, and third-party API returned data) are stamped with a unified BeiDou / GPS time stamp. For non-real-time data, an event-time-based window function is used for alignment.

[0066] S202. Spatial Coordinate Matching: The vehicle latitude and longitude coordinates (WGS-84 coordinate system) obtained by the high-precision GNSS module are matched in real time with the spatial grid of electronic fences, planned route segments, and third-party weather / road condition data. The GeoHash algorithm is used to discretize the continuous coordinates into grid codes to achieve rapid spatial association.

[0067] S203. Feature-level fusion: Taking vehicles and transportation tasks as entities, and using a fixed time window (e.g., 10 seconds) as the unit, all data features within the same spatiotemporal window are concatenated to form a standardized transportation status snapshot vector. This vector includes, but is not limited to: the mean / variance of cargo temperature and humidity, the spectral features of three-axis vibration acceleration (e.g., the first N dominant frequency amplitudes after FFT transformation), the instantaneous speed and position of vehicles, the current road segment congestion index, and weather status codes. This vector serves as the unified input for subsequent anomaly detection modules.

[0068] The anomaly detection module receives the above data stream and performs hybrid intelligent analysis, the implementation of which is as follows:

[0069] 1. Rule Engine Unit: Configures a series of predefined business rules, such as: "If the temperature and humidity exceed the threshold X for 5 consecutive minutes, a temperature abnormality alarm will be triggered"; "If a vehicle stays at an unplanned stop for more than 30 minutes, a stop abnormality will be triggered". This unit performs fast scanning and matching of data streams to achieve millisecond-level primary anomaly judgment.

[0070] 2. Machine Learning Model Unit: Multiple specialized models run in parallel to perform in-depth analysis of the data stream. For example, a time-series regression model analyzes historical travel data and current road conditions to predict the probability of arrival time delays; a vibration pattern recognition model analyzes the spectral characteristics of accelerometers to identify specific resonance patterns that may cause damage to fragile items; and a trajectory similarity analysis model (such as one based on the DTW algorithm) compares the actual trajectory with the planned trajectory to detect hidden path deviations. These models are all trained on historical data and can discover complex patterns that are difficult to define by rules.

[0071] 3. Early Warning Synthesis Unit: This unit receives preliminary judgments from the rule engine and analysis results (such as probabilities and scores) from various models. It integrates all input information through a weighted decision fusion algorithm (e.g., assigning fixed weights to rule results and performing weighted averaging or voting on the outputs of multiple models) to finally determine the type of abnormal event (such as "temperature control failure", "potential cargo damage", "path deviation") and classify the severity level (such as "low", "medium", "high", "urgent") based on confidence level, degree of deviation, etc.

[0072] As a preferred embodiment of the present invention, the machine learning model unit is constructed and operated as follows:

[0073] M1. Temporal Regression Model (for Delay Prediction): Employs a Long Short-Term Memory (LSTM) network model. Input features include vehicle location sequences and average speed sequences at 5-minute intervals over the past hour, as well as road condition prediction data for the next 2 hours. Output is the estimated delay time (in minutes) to reach the next critical node and the confidence interval.

[0074] M2. Vibration Pattern Recognition Model (for cargo damage risk identification): It adopts a one-dimensional convolutional neural network (1D-CNN). The input features are the raw data of the three-axis accelerometer. After being segmented by a sliding window (such as a 2-second window with 50% overlap), the spectrum of each window's data is obtained by performing a fast Fourier transform (FFT). The output is a 'cargo damage risk probability' between 0 and 1. When a continuous high-energy vibration at a specific frequency (such as the cargo's natural resonant frequency) is identified, the probability value increases.

[0075] M3. Trajectory Similarity Analysis Model (for Path Deviation Detection): Employs Dynamic Time Warping (DTW) algorithm. The input consists of the actual driving trajectory coordinate sequence and the pre-planned path coordinate sequence. The output is the DTW distance between the two trajectories. The larger the distance, the more severe the deviation.

[0076] The weighted decision fusion algorithm of the early warning synthesis unit is as follows: For each snapshot of the transportation status to be determined, the rule engine outputs a binary judgment set {R_i} (0 or 1), and each machine learning model outputs a risk score set {S_j} (0-1 continuous values). First, the model scores are normalized to a range comparable to the rule judgments using the sigmoid function, and finally, the anomaly confidence is calculated.

[0077]

[0078] Where α and β are the mixed weight coefficients (e.g., α=0.3, β=0.7), W_i and V_j are the weights within the rule and the model, respectively (which can be dynamically adjusted based on historical accuracy). The anomaly level is determined based on the threshold range where the Confidence value is located (e.g., [0,0.3] low, [0.3,0.7) medium, [0.7,0.9) high, [0.9,1.0] urgent).

[0079] Upon receiving an abnormal event and its severity, the decision-making module initiates an intelligent decision-making process:

[0080] 1. Case Matching Unit: The current anomaly's feature vector (such as type, level, location, and cargo value) is compared with the historical disposal case knowledge base using similarity retrieval (such as using cosine similarity or more complex graph matching algorithms) to find the most similar historical cases and use their disposal solutions as basic templates.

[0081] 2. Multi-objective optimization unit: With the optimization objectives of minimizing handling costs, minimizing handling time, and minimizing customer impact (such as complaint probability), a multi-objective optimization model is established. Optimization variables may include: backup vehicle dispatching routes, maintenance personnel dispatching plans, customer communication strategies, etc. A multi-objective evolutionary algorithm (such as NSGA-II) is used to solve the variables in the basic template to generate a set of Pareto optimal solutions. This set of solutions intuitively shows the trade-offs between different objectives.

[0082] 3. Solution Adjustment Unit: Through a visual human-computer interaction interface, the system displays a set of Pareto optimal solutions (e.g., a scatter plot with cost-time as the coordinate axis) to the dispatcher. The dispatcher can select a solution based on the current business priority and use the interface tools to fine-tune the solution details (e.g., priority customers to contact) to form the final handling solution.

[0083] As a preferred embodiment of the present invention, the mathematical model of the multi-objective optimization unit is constructed as follows: Assuming that a backup vehicle needs to be dispatched due to vehicle failure, the decision variable x_{vk} is defined as a Boolean variable, indicating whether the backup vehicle v is dispatched to task k;

[0084] Objective function:

[0085] 1. Minimize disposal costs:

[0086] ;

[0087] 2. Minimize the maximum completion time:

[0088] ;

[0089] in This is the estimated arrival time;

[0090] 3. Minimize customer impact:

[0091] ;

[0092] in The priority weights are set based on the value of the goods and the customer level, and delay_k is the expected delay duration;

[0093] Constraints include vehicle load capacity, distance between the vehicle's current location and the task point, and driver working time regulations. The multi-objective optimization unit uses an improved NSGA-II algorithm to solve the above model. The algorithm uses the basic disposal template (such as the suggested backup vehicle pool and task list) output by the case matching unit as part of the initial population to accelerate convergence and finally output a Pareto optimal solution set. Each solution corresponds to a non-dominated equilibrium point of cost (C), time (T), and impact (I).

[0094] The execution module is responsible for automating the implementation of the solution:

[0095] 1. Task Automation Unit: Integrates RPA robots. For example, based on the final plan, RPA automatically logs into the Transportation Management System (TMS) to create a dispatch order, logs into the Customer Relationship Management System (CRM) to send delay notification SMS messages, and logs into the financial system to apply for reserve funds.

[0096] 2. Collaborative Communication Unit: By calling the APIs of platforms such as WeChat Work, DingTalk, or Slack, a temporary group containing relevant responsible persons such as dispatchers, drivers, customer service representatives, and maintenance workers is automatically created, and the task list and assignments are @mentioned to specific personnel;

[0097] 3. Process Tracking Unit: Monitors the status of each sub-task in real time (such as "driver has accepted the order" or "repair completed"), and updates the overall processing progress bar on the command center's large screen or mobile terminal to achieve transparent process management;

[0098] The feedback learning module is activated after the processing is completed, enabling system evolution:

[0099] 1. Effectiveness Evaluation Unit: Based on preset KPIs (such as "alarm response time", "anomaly resolution cycle", "actual cost vs. budget cost deviation"), the handling of this incident is quantitatively scored.

[0100] 2. Model Iteration Unit: The complete data of this event (from occurrence to resolution) is used as a new training sample and added to the training set to incrementally train or learn online the machine learning model in the anomaly recognition module, optimize the model parameters, and improve the future recognition accuracy.

[0101] 3. Knowledge Base Update Unit: The complete process, final solution, and effect evaluation results of this handling are structured and packaged into a new handling case, which is then stored in the historical handling case knowledge base to enrich the system's experience for future case matching.

[0102] like Figure 2 The specific working principle and usage method of this invention are as follows:

[0103] Step S1: System initialization, all sensors, positioning devices and cameras start up, and begin to collect multi-source logistics data such as the physical status of goods, vehicle trajectory and visual environment in real time;

[0104] In step S2, the edge computing gateway locally aggregates and preprocesses multi-source data from the vehicle and uploads it to the cloud via the vehicle communication module. The cloud data bus integrates vehicle data and third-party data, and after time synchronization, spatial alignment and feature fusion, a standardized transportation status snapshot data stream is formed.

[0105] In step S3, the anomaly detection module synchronously receives the data stream, the rule engine performs rapid filtering, and identifies obvious anomalies; at the same time, multiple machine learning models perform in-depth analysis and prediction on the data stream in parallel, and the early warning synthesis unit integrates all results and determines the anomaly type and level through a weighted decision fusion algorithm.

[0106] In step S4, the decision-making module is triggered. First, it matches similar cases in the historical case library to obtain a basic template. Then, it generates a Pareto optimal solution set through a multi-objective optimization algorithm. Finally, the decision-maker confirms the final solution through the human-computer interaction interface.

[0107] In step S5, the execution module breaks down the plan into specific tasks, the RPA tool automatically executes system operations, the collaborative communication unit notifies and coordinates relevant personnel, and the process tracking unit monitors the execution status in real time.

[0108] Step S6: After the treatment is completed, the feedback learning module is activated to quantitatively evaluate the treatment effect, use the feedback data to iteratively optimize the machine learning model, and store the effective experience as a new case in the knowledge base to complete the closed-loop learning.

[0109] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Content not described in detail in this specification is prior art known to those skilled in the art.

[0110] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Those skilled in the art can readily implement the present invention based on the accompanying drawings and the above description. However, any modifications, alterations, or variations made by those skilled in the art without departing from the scope of the present invention, utilizing the disclosed technical content, are equivalent embodiments of the present invention. Furthermore, any modifications, alterations, or variations made to the above embodiments based on the essential technology of the present invention are still within the protection scope of the present invention.

Claims

1. A logistics transportation anomaly event identification and handling system, comprising a data acquisition module, an anomaly identification module, a handling decision module, a handling execution module, and a feedback learning module, characterized in that, The data acquisition module is used to collect real-time status data of the entire logistics and transportation chain from multiple data sources; the anomaly identification module is connected to the data acquisition module and analyzes the collected data based on preset rules and artificial intelligence models to identify and classify abnormal events; the handling decision module is connected to the anomaly identification module and generates optimized handling plans according to the type and level of abnormal events; the handling execution module is connected to the handling decision module and automatically executes the tasks in the handling plan and coordinates all parties; the feedback learning module is connected to both the anomaly identification module and the handling decision module and optimizes the identification model and handling strategy based on the handling result data. The data acquisition module includes an IoT sensing unit for collecting physical data of cargo status, a positioning and communication unit for acquiring and uploading location, trajectory and vehicle operation data in real time, an external environment perception unit for acquiring visual environment information and third-party data, and a data processing and transmission unit. The data processing and transmission unit includes an edge computing gateway and a cloud data bus. It integrates heterogeneous data from multiple sources, such as IoT sensing units, positioning and communication units, and external environment perception units, to form a standardized transportation status snapshot data stream, which is then output to the anomaly identification module. The IoT sensing unit includes a temperature and humidity sensor, a triaxial acceleration and vibration sensor, a light sensor, and a door magnetic sensor deployed on the cargo or vehicle. The positioning and communication unit includes a high-precision GNSS positioning module and a 4G / 5G vehicle communication module integrated on the transportation vehicle. The external environment perception unit includes a vehicle camera and an external data interface.

2. The logistics transportation abnormal event identification and handling system according to claim 1, characterized in that, The anomaly detection module includes a rule engine unit, a machine learning model unit, and an early warning synthesis unit; The rule engine unit performs preliminary anomaly judgment on the transportation status snapshot data stream based on predefined business rules, and the machine learning model unit uses multiple trained models to perform in-depth analysis on the transportation status snapshot data stream to identify complex patterns and predict risks. The machine learning model unit includes at least: a time-series regression model for predicting transportation delays, a vibration pattern recognition model for identifying the risk of physical damage to goods, and a trajectory similarity analysis model for detecting path deviations; The early warning synthesis unit is used to fuse the outputs of the rule engine unit and the machine learning model unit, and determine the final anomaly type and severity level through a weighted decision algorithm.

3. The logistics transportation abnormal event identification and handling system according to claim 1, characterized in that, The decision-making module includes a case matching unit, a multi-objective optimization unit, and a solution adjustment unit. The case matching unit is used to perform similarity retrieval between the current abnormal event characteristics and the historical handling case knowledge base to obtain a basic handling template. The multi-objective optimization unit is used to solve the variables in the basic handling template with the handling cost, time and customer impact as optimization objectives, and generate a Pareto optimal solution set. The solution adjustment unit is used to provide a human-computer interaction interface for operators to select or fine-tune the final handling solution from the solution set.

4. The logistics transportation abnormal event identification and handling system according to claim 1, characterized in that, The processing execution module includes a task automation unit, a collaborative communication unit, and a process tracking unit. The task automation unit integrates a Robotic Process Automation (RPA) tool to automatically execute cross-system operation instructions. The collaborative communication unit integrates an enterprise-level communication platform's application programming interface (API) to automatically create task groups, assign tasks, and notify relevant responsible persons. The process tracking unit is used to monitor the execution status of each sub-task in real time and update the overall progress of handling abnormal events.

5. The logistics transportation abnormal event identification and handling system according to claim 1, characterized in that, The feedback learning module includes an effect evaluation unit, a model iteration unit, and a knowledge base update unit. The effect evaluation unit is used to quantitatively evaluate the handling effect of completed abnormal events according to the key performance indicator (KPI) system. The model iteration unit is used to use the handling feedback data as training samples to perform online learning or periodic incremental training on the model in the machine learning model unit. The knowledge base update unit is used to structure and store the verified and effective handling cases and evaluation results into the historical handling case knowledge base.

6. The logistics transportation abnormal event identification and handling system according to claim 1, characterized in that, The high-precision GNSS positioning module is a multi-band receiver that supports real-time dynamic RTK technology; the triaxial accelerometer and vibration sensor is a low-noise, high-resolution MEMS accelerometer.

7. A method for identifying and handling abnormal events in logistics transportation, applied to the system described in any one of claims 1-6, characterized in that, Includes the following steps: S1. Real-time collection of multi-source logistics data through the IoT sensing unit, positioning and communication unit and external environment sensing unit; S2. The data processing and transmission unit performs fusion processing on the multi-source data to form a standardized transportation status snapshot data stream; S3. The anomaly identification module analyzes the data stream to identify and classify abnormal events; S4. Generate an optimized disposal plan through the disposal decision module; S5. The disposal plan is executed automatically through the disposal execution module; S6. The feedback learning module optimizes the system based on the handling feedback.

8. The method for identifying and handling abnormal events in logistics transportation according to claim 7, characterized in that, The identification and classification of abnormal events includes: rapid filtering based on predefined rules; inputting data into multiple parallel models for prediction and deep pattern recognition; and determining the final anomaly type and level by combining rule judgments and the output results of multiple models through a decision fusion algorithm.

9. The method for identifying and handling abnormal events in logistics transportation according to claim 7, characterized in that, The optimized solution generation includes: matching similar scenario solution templates from a historical case library; using a multi-objective evolutionary algorithm to optimize the solution path and resources with multiple objectives such as cost, time, and customer satisfaction, generating multiple non-dominated solutions; and providing a human-computer interaction interface for decision-makers to select or adjust the final solution.

10. The method for identifying and handling abnormal events in logistics transportation according to claim 7, characterized in that, The system based on feedback analysis includes: quantitatively calculating the response time, resolution cycle, and cost deviation of this action; retraining the identification model using the data from the action process and results; and converting the effective experience gained from this action into structured cases and updating them to the historical case knowledge base.

Images