Intelligent logistics supply chain anomaly monitoring method and system based on big data

Abstract

This invention relates to the field of logistics and discloses an intelligent logistics supply chain anomaly monitoring method and system based on big data. The method includes: constructing an operational baseline that dynamically adjusts with time and scenario based on a target operational range using historical operational data; mapping real-time feature data to the scenario corresponding to the operational baseline; organizing feature changes of the same logistics object into multiple operational tracks; continuously evaluating the consistency between operational tracks; gradually accumulating anomaly indices and quantifying the anomaly state by combining the anomaly's growth rate and diffusion range; when the quantified index reaches a trigger condition, performing a time-reverse analysis of the operational process before the anomaly occurs to locate the earliest point of change that caused the deviation in operational track consistency; forming anomaly evolution paths and hypothesis failure types based on operational hypotheses at the point of change; assessing the impact of the anomaly and outputting alarm information. This invention enables early detection, precise location, and alarming of anomalies in the logistics supply chain.

Description

Technical Field

[0001] This invention relates to the field of logistics, and more specifically, to an intelligent logistics supply chain anomaly monitoring method and system based on big data. Background Technology

[0002] The monitoring of the current logistics supply chain operation status mainly relies on manual inspections, empirical rules, or anomaly detection methods based on static thresholds. These methods typically monitor single indicators or localized business processes, lacking the ability to continuously perceive the overall operational status of the logistics supply chain and conduct multi-dimensional correlation analysis.

[0003] When business scale expands, business scenarios switch frequently, or operating modes change, fixed thresholds and static rules are difficult to adapt to complex and ever-changing operating environments, which can easily lead to problems such as delayed anomaly identification, high false alarm rate, and difficulty in anomaly localization.

[0004] In addition, existing anomaly monitoring methods are mostly based on post-event alarms, lacking the ability to analyze the evolution process and root causes of anomalies. This makes it difficult to assess the scope and degree of impact of anomalies on the operation of the logistics supply chain in a timely manner, thus affecting the timeliness and effectiveness of anomaly handling and restricting further improvement of logistics efficiency and refined control of logistics costs.

[0005] There are currently no effective solutions to the problems in the relevant technologies. Summary of the Invention

[0006] In response to the problems in related technologies, this invention proposes an intelligent logistics supply chain anomaly monitoring method and system based on big data to overcome the aforementioned technical problems existing in the existing related technologies.

[0007] Therefore, the specific technical solution adopted by the present invention is as follows: According to one aspect of the present invention, an intelligent logistics supply chain anomaly monitoring method based on big data is provided, comprising: S1. Obtain historical operational data of the multi-source heterogeneous logistics supply chain after standardization processing; S2. Based on historical operational data, target operational ranges are used for sample extraction and analysis of logistics business scenarios to construct operational baselines that are dynamically adjusted with time and scenarios. S3. Map real-time feature data to the scenario corresponding to the operating baseline, and organize the feature changes of the same logistics object in the dimensions of time, space, resources and behavior into multiple operating tracks; S4. By learning the mutual interpretation relationships and co-evolution patterns of each running track in the historical normal state in terms of time sequence, causal constraints and resource consumption, the consistency between running tracks is continuously evaluated; when conflicts, breaks or causal sequence anomalies occur in real-time running tracks, the anomaly index is gradually accumulated, and the anomaly state is quantified by combining the growth rate and diffusion range of the anomaly. S5. When the quantification index reaches the trigger condition, perform a time-reverse analysis of the operation process before the anomaly occurs to locate the earliest point of change that caused the deviation of the operation trace consistency; based on the operation assumptions of the point of change, form the evolution path of the anomaly and the failure type of the assumption; S6. Based on the evolution path of the anomaly and the assumed failure type, assess the impact of the anomaly and output alarm information.

[0008] Furthermore, based on historical operational data and target operational ranges, sample extraction and analysis of logistics business scenarios are performed to construct an operational baseline that dynamically adjusts with time and scenarios, including: Based on historical operational data, determine the target operational range of logistics objects in various logistics business scenarios, including time range, resource occupation range, and behavioral scope; Historical operational data is divided into logistics business scenarios according to business type; within each logistics business scenario, sample data that meets the target operating range is extracted from historical operational data as normal operating samples; Based on normal operation samples, statistical characteristics, including mean, distribution area and fluctuation range, are calculated for the time characteristics, spatial characteristics, resource characteristics and behavioral characteristics of logistics objects to characterize their operation, and corresponding operation baselines are constructed. The operating baseline is periodically updated based on changes in time and scene switching.

[0009] Furthermore, real-time feature data is mapped to the scenario corresponding to the operational baseline, and the feature changes of the same logistics object in the dimensions of time, space, resources, and behavior are organized into multiple operational tracks, including: Acquire real-time characteristic data of logistics objects during operation, including time characteristics, spatial characteristics, resource characteristics, and behavioral characteristics; perform format standardization, time alignment, and outlier handling on the real-time characteristic data; Based on the business status, location status, or operational phase information in real-time feature data, identify the business scenario in which the current logistics object is located, and map the real-time feature data to the operational baseline corresponding to the current logistics business scenario; Real-time feature data is time-series converted, and the time-series converted real-time feature data is divided into time dimension, spatial dimension, resource dimension and behavior dimension, and the change process of features in each dimension over time is extracted respectively; Organize the characteristic changes of the same logistics object in each dimension into corresponding operation tracks; Each track is assigned a track identifier, and the logistics object identifier, business scenario, and time interval are recorded; the association between multiple tracks under the same logistics object is established.

[0010] Furthermore, by learning the mutual explanatory relationships and co-evolutionary patterns of each operational track under normal historical conditions in terms of temporal sequence, causal constraints, and resource consumption, the consistency among operational tracks is continuously evaluated, including: Several corresponding operation traces are extracted from the normal operation samples to form a historical normal operation trace sample set, which covers the operation status under different time periods and different business scenarios. Based on historical normal operation track samples, the temporal sequence relationship between each operation track is statistically analyzed; based on the triggering and response relationship between changes in different operation track characteristics, the causal constraint relationship between operation tracks under normal conditions is established to describe the impact conditions of operation track changes on other operation track changes; based on the timing and mode of resource occupation in the logistics supply chain by multiple operation tracks during operation, the resource occupation relationship of operation tracks is established. Based on temporal sequence, causal constraints, and resource consumption, a collaborative evolution model of multiple running tracks under normal conditions is constructed. The running traces corresponding to real-time feature data are input into the co-evolutionary model to evaluate the consistency of the running traces corresponding to real-time feature data in terms of time sequence, causal constraints, and resource consumption.

[0011] Furthermore, when anomalies such as conflicts, breaks, or causal sequence abnormalities occur in the real-time running trace, anomaly indices are gradually accumulated, and the abnormal state is quantified by combining the anomaly's growth rate and spread range, including: When a time sequence conflict, a broken track, an abnormal causal order, or an abnormal resource usage occurs between real-time running tracks, a corresponding running track abnormal event is generated. For each logistics object or business instance, when the first operational anomaly event is detected, the corresponding anomaly index is initialized based on the degree to which the current operational status deviates from the normal operating baseline; when operational anomalies are continuously detected in subsequent operations, the anomaly index is cumulatively updated based on the type, duration, and frequency of the operational anomaly events. Within a continuous time window, the rate of abnormal growth is obtained by calculating the change in the abnormal index. The scope of the anomaly's spread is determined based on the number of operational traces affected by the anomaly event, the number of business scenarios involved, or the number of resources affected. Abnormal states are quantified based on the cumulative update value, growth rate, and diffusion range of the anomaly index.

[0012] Furthermore, when the quantification index reaches the trigger condition, a time-reverse analysis is performed on the operation process before the anomaly occurred to locate the earliest point of change that caused the deviation from the consistency of the operation trace, including: When the anomaly index reaches the preset triggering condition, the triggering time point of the anomaly is determined, and the time reverse analysis window is selected backward from the triggering time point as the endpoint. Within the time-reverse analysis window, multiple operation traces related to the anomaly are retrieved by tracing back, and the feature change data in the operation traces are reorganized in reverse time order to form a reverse time series; Based on inverse time series, a consistent description model reflecting the running trace in the dimensions of time, space, resources and behavior is constructed; Based on the minimum description length criterion, the complexity of the consistent description model corresponding to different time points is evaluated, and the time points at which the complexity of the running trace structure changes to the degree of target change are detected. The time point at which the first change in the target degree is detected is taken as the earliest point of change that causes the consistency of the running trace to deviate, and the running trace, feature dimension and business scenario information corresponding to the point of change are recorded.

[0013] Furthermore, based on the minimum description length criterion, the complexity of the consistent description model at different time points is evaluated, and the time points at which the complexity of the runtime trace structure changes to the target degree are detected include: Determine the number of model parameters required for each consistency description model, and calculate the model description length of the corresponding consistency description model based on the number of model parameters and their value precision, which is used to characterize the structural complexity of the consistency description model; By calculating the fit of the consistency description model to the runtime data, the degree of deviation of the runtime data from the consistency description model is determined, and the degree of deviation is converted into data description length. The model description length is combined with the data description length to obtain the minimum description length value of the consistent description model at the corresponding time point, which is used to characterize the overall complexity level of the running trace structure at that time point. Arrange the minimum description length values ​​corresponding to multiple time points in chronological order to form a description length sequence in which the complexity of the running trace structure changes over time; Calculate the magnitude of change in the complexity of the running trace structure based on the comparison results of the minimum description length values ​​at adjacent time points; If the change reaches the target level, the corresponding time point will be determined as the time point when the complexity of the running trace structure changes to the target level.

[0014] Furthermore, based on operational assumptions at points of change, the resulting abnormal evolution paths and assumed failure types include: Before the point of change, the operating state reflected by multiple operating tracks is taken as the normal reference state; based on the normal reference state, operating hypotheses are extracted. Starting from the point of change, multiple operational traces are obtained along time to reflect the actual operational behavior, and compared with the operational assumptions. When the actual operational behavior does not conform to the operational assumptions, the operational assumptions are considered invalid at that point. The entire process from conforming to the operating assumptions to failing to conform to the operating assumptions is linked together in chronological order, forming an abnormal evolutionary path; Based on the characteristics exhibited when the operating hypothesis fails, the anomalies are classified to obtain the hypothesis failure type.

[0015] Furthermore, based on the evolution path of the anomaly and the assumed failure type, the impact of the anomaly is assessed and alarm information is output, including: Based on the evolution path of the anomaly and the assumed failure type, the scope and degree of the anomaly's impact are assessed to obtain the anomaly impact assessment results; Based on the impact assessment results of the anomaly, determine the alarm level of the anomaly and output the alarm information.

[0016] According to another aspect of the present invention, an intelligent logistics supply chain anomaly monitoring system based on big data is also provided, comprising: The baseline acquisition module is used to acquire historical operational data of multi-source heterogeneous logistics supply chains after standardization processing; based on the target operational range of the historical operational data, it performs sample extraction and analysis of logistics business scenarios to construct an operational baseline that is dynamically adjusted with time and scenario. The operation track acquisition module is used to map real-time feature data to the scene corresponding to the operation baseline, and organize the feature changes of the same logistics object in the dimensions of time, space, resources and behavior into multiple operation tracks; The anomaly quantification module is used to continuously evaluate the consistency between running traces by learning the mutual interpretation relationships and co-evolution patterns of each running trace in terms of time sequence, causal constraints and resource consumption under historical normal conditions; when conflicts, breaks or causal sequence anomalies occur in real-time running traces, the anomaly index is gradually accumulated and the anomaly state is quantified by combining the growth rate and diffusion range of the anomaly. The anomaly analysis and alarm output module is used to perform time-reverse analysis of the running process before the anomaly occurs when the quantization index reaches the trigger condition, to locate the earliest point of change that caused the deviation of the running trace consistency; based on the running assumptions of the point of change, to form the evolution path of the anomaly and the assumed failure type; and based on the evolution path of the anomaly and the assumed failure type, to assess the impact of the anomaly and output alarm information.

[0017] The beneficial effects of this invention are as follows: 1. This invention establishes a collaborative evolution model based on the temporal sequence, causal constraints, and resource utilization relationships among operational traces under normal conditions, enabling consistent and continuous assessment of the operational status of the logistics supply chain. When anomalies occur, through the accumulation and quantification of anomaly indices, precise location of change points, and analysis of anomaly evolution paths and hypothetical failure types, anomalies can be detected, accurately located, and alerted in advance, effectively improving the real-time performance, accuracy, and interpretability of logistics supply chain anomaly monitoring.

[0018] 2. This invention can identify potential anomalies from the perspective of overall consistency between running tracks, avoiding the limitations of relying on a single indicator or a single track to detect anomalies, thereby improving the ability to identify complex and hidden anomalies.

[0019] 3. This invention can accurately locate the key time point that first causes the deviation of the consistency of the running trace. Compared with the method of analyzing only after the anomaly has become apparent, it can realize the forward location of the root cause of the anomaly, and improve the accuracy and efficiency of anomaly tracing.

[0020] 4. This invention constructs a complete evolution path of an anomaly from initial deviation to hypothesis failure, and classifies anomalies based on the characteristics when the hypothesis fails, thereby elevating the analysis of anomalies from a judgment of whether they occur to a process analysis of how they occur and how they evolve, thus enhancing the interpretability and reusability of anomaly analysis results. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of an intelligent logistics supply chain anomaly monitoring method based on big data according to an embodiment of the present invention; Figure 2 This is a block diagram of an intelligent logistics supply chain anomaly monitoring system based on big data, according to an embodiment of the present invention.

[0023] In the picture: 1. Baseline acquisition module; 2. Running trace acquisition module; 3. Anomaly quantification module; 4. Anomaly analysis and alarm output module. Detailed Implementation

[0024] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0025] According to embodiments of the present invention, an intelligent logistics supply chain anomaly monitoring method and system based on big data are provided.

[0026] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, according to an embodiment of the present invention, an intelligent logistics supply chain anomaly monitoring method based on big data is provided, including: S1. Obtain standardized historical operational data from multi-source heterogeneous logistics supply chains. Specifically, this involves the unified collection and organization of data generated during the operation of the logistics supply chain. Historical operational data originates from logistics companies' own systems, authorized business platforms, or publicly available data interfaces, including operational data such as order processing, transportation scheduling, warehousing operations, resource usage, and business status. Addressing the differences in data format, time granularity, naming rules, and sampling frequency among different data sources, the acquired historical operational data undergoes standardization processing, including cleaning, noise reduction, format unification, and time alignment, enabling multi-source heterogeneous data to be fused and represented within the same timeline and feature system.

[0027] S2. Based on historical operational data, target operational ranges are used to extract samples and analyze logistics business scenarios, and an operational baseline is constructed that is dynamically adjusted with time and scenario.

[0028] In one embodiment, based on historical operational data and a target operational range, sample extraction and analysis of logistics business scenarios are performed to construct an operational baseline that dynamically adjusts with time and scenarios, including: Based on historical operational data, the target operational range of logistics objects in each logistics business scenario is determined, including time range, resource occupancy range, and behavioral scope. Historical operational data is then divided into different logistics business scenarios according to business type. Within each logistics business scenario, sample data that meets the target operational range is extracted from historical operational data as normal operation samples. Based on these normal operation samples, statistical characteristics, including mean, distribution area, and fluctuation range, are calculated for the time, space, resource, and behavioral characteristics used to characterize the operation of logistics objects, and corresponding operational baselines are constructed. The operational baselines are periodically updated according to time changes and scenario switching.

[0029] Based on historical operational data, the target ranges for logistics objects in various logistics business scenarios are determined. For example, in different business scenarios such as trunk transportation, warehouse loading and unloading, and last-mile delivery, the vehicle arrival time range, loading and unloading resource occupation range, and operation sequence range are determined respectively. Historical operational data are divided into logistics business scenarios according to business type. Within each logistics business scenario, data samples that meet the corresponding target operational range are extracted from historical operational data as normal operation samples. In the warehouse loading and unloading scenario, only historical records where vehicles arrive on time and loading and unloading resource occupation does not exceed the limit are selected.

[0030] Based on normal operation samples, statistical characteristics of logistics objects are calculated for time characteristics, spatial characteristics, resource characteristics, and behavioral characteristics under different time periods (same time period, different seasons) and business scenarios, including mean, distribution range, and fluctuation range, and an operation baseline is constructed accordingly; for example, different vehicle waiting time baselines are formed for peak and off-peak periods.

[0031] This invention establishes corresponding operational baselines by differentiating between different business scenarios and time periods, so that anomaly detection no longer relies on a unified static threshold, but is based on a dynamic reference standard that matches the current business scenario, thereby reducing the probability of false alarms and false negatives.

[0032] S3. Map real-time feature data to the scenario corresponding to the operating baseline, and organize the feature changes of the same logistics object in the dimensions of time, space, resources and behavior into multiple operating tracks.

[0033] In one embodiment, mapping real-time feature data to the scenario corresponding to the operational baseline, and organizing the feature changes of the same logistics object in the dimensions of time, space, resources, and behavior into multiple operational tracks includes: Acquire real-time feature data of logistics objects during operation, including time features, spatial features, resource features, and behavioral features; standardize the format, align the time, and handle outliers of the real-time feature data to form real-time feature data that can be used for analysis; based on the business status, location status, or operational stage information in the real-time feature data, identify the business scenario in which the current logistics object is located, and map the real-time feature data to the operational baseline corresponding to the current logistics business scenario; serialize the real-time feature data over time, and divide the time-serialized real-time feature data into time dimension, spatial dimension, resource dimension, and behavioral dimension, and extract the change process of features in each dimension over time; organize the feature change process of the same logistics object in each dimension into corresponding operation tracks; assign an operation track identifier to each operation track, and record the logistics object identifier, business scenario, and time interval; establish the correlation between multiple operation tracks under the same logistics object.

[0034] Among them, the time dimension refers to the characteristic dimension reflecting the rhythm and temporal characteristics of the logistics object's operation, including the start and end time of the operation, the arrival and departure time, the waiting time, the processing time, and the time interval between each operation stage; the spatial dimension refers to the characteristic dimension reflecting the spatial movement status of the logistics object's geographical location, mainly including real-time location coordinates, driving or movement trajectory, spatial area affiliation, and spatial migration distance and direction; the resource dimension refers to the characteristic dimension reflecting the logistics object's occupation and status of system resources, mainly including the occupation status, quantity, occupation duration, and resource utilization rate of resources such as vehicles, warehouses, loading and unloading equipment, and manpower; the behavioral dimension refers to the characteristic dimension reflecting the specific execution and operation sequence of the logistics object in the business process, mainly including the behavioral types such as loading, unloading, picking, handover, and signing, as well as their occurrence sequence, frequency, and status changes.

[0035] Based on the business status, location status, or operational phase information reflected in real-time feature data, this invention identifies the current logistics business scenario of a logistics object and maps the real-time feature data to the operational baseline corresponding to that logistics business scenario, ensuring the consistency between the real-time data and the reference baseline in the scenario dimension.

[0036] Real-time feature data is processed using time-series analysis, and features are categorized according to time, space, resource, and behavioral dimensions. The process of feature changes over time is extracted for each dimension, and the feature processes of the same logistics object at each dimension are organized into corresponding operational tracks. Each operational track is assigned a unique track identifier, and the logistics object identifier, business scenario, and time interval information are recorded to establish the correlation between multiple operational tracks under the same logistics object, providing structured input for subsequent operational track consistency analysis. This approach not only helps reveal the dynamic evolution characteristics of logistics objects during operation but also provides a foundation for identifying anomalies from the perspective of consistency between operational tracks.

[0037] S4. By learning the mutual interpretation relationships and co-evolution patterns of each running track in the historical normal state in terms of time sequence, causal constraints and resource occupation, the consistency between running tracks is continuously evaluated; when conflicts, breaks or causal sequence anomalies occur in the real-time running tracks, the anomaly index is gradually accumulated, and the anomaly state is quantified by combining the growth rate and diffusion range of the anomaly.

[0038] In one embodiment, the consistency among running tracks is continuously evaluated by learning the mutual explanatory relationships and co-evolution patterns of each running track in terms of time sequence, causal constraints, and resource consumption under historical normal conditions, including: Several operational traces are extracted from the normal operation samples to form a historical normal operation trace sample set, which covers the operational status under different time periods and business scenarios. Based on the historical normal operation trace samples, the temporal sequence relationship between each operational trace is statistically analyzed. According to the triggering and response relationship between different operational trace feature changes, the causal constraint relationship between operational traces under normal conditions is established to describe the impact conditions of operational trace changes on other operational trace changes. According to the timing and mode of occupation of the same resources in the logistics supply chain by multiple operational traces during operation, the resource occupation relationship of operational traces is established. Based on the temporal sequence relationship, causal constraint relationship, and resource occupation relationship, a collaborative evolution model of multiple operational traces under normal conditions is constructed. The operational traces corresponding to real-time feature data are input into the collaborative evolution model to evaluate the consistency of the operational traces corresponding to real-time feature data in terms of temporal sequence, causal constraint, and resource occupation.

[0039] This invention extracts multiple operational tracks from historical normal operation samples, forming a sample set of historical normal operation tracks covering different time periods and logistics business scenarios. Based on this, it statistically analyzes the temporal sequence relationship between each operational track under normal operation conditions, the causal constraint relationship under normal conditions, and the timing and method of multiple operational tracks occupying the same resource in the logistics supply chain during operation, establishing resource occupation relationships between operational tracks. In other words, by comprehensively considering temporal sequence relationships, causal constraint relationships, and resource occupation relationships, a co-evolutionary model reflecting the collaborative change patterns of multiple operational tracks under normal conditions is constructed. The operational tracks corresponding to real-time feature data are input into the co-evolutionary model, continuously evaluating the consistency between real-time operational tracks at the levels of temporal sequence, causal constraint, and resource occupation.

[0040] Among them, the temporal sequence relationship refers to the order in which key feature changes occur in different operation tracks under normal operating conditions and their stability constraints. For example, in the operation of the same logistics object, the arrival at the warehouse event in the vehicle arrival operation track should occur before the start of loading and unloading event in the loading and unloading behavior operation track. The causal constraint relationship refers to the triggering condition or influence relationship of a specific feature change in one operation track on the feature change in another operation track under normal conditions. For example, the completion of vehicle arrival and check-in is a prerequisite for the loading and unloading resource operation track to change from idle to occupied. The resource occupation relationship refers to the timing, occupation method, and mutual exclusion or sharing constraints of multiple operation tracks occupying the same logistics resource during operation. For example, the loading and unloading behavior operation tracks of multiple vehicles have a sequential occupation relationship for the same loading and unloading platform resource, which cannot be parallelized.

[0041] The temporal sequence, causal constraints, and resource occupancy relationships remain coordinated and consistent throughout the time process. The state changes of each track mutually explain and support each other without conflict or breakage. The co-evolution model is a structured description of this stable cooperative relationship. It is used to characterize how, under normal operating conditions, multiple tracks trigger in a predetermined order, mutually constrain each other, and rationally share resources during the time process to jointly form an interpretable and predictable consistent operating state.

[0042] This invention can evaluate the consistency of real-time operation status based on the overall correlation between operation tracks, rather than being limited to deviation detection of a single operation track or a single feature, thereby effectively identifying complex anomalies caused by the disruption of the cooperative relationship between operation tracks.

[0043] In one embodiment, when anomalies such as conflicts, breaks, or causal sequence abnormalities occur in the real-time runtime trace, an anomaly index is gradually accumulated, and the anomaly state is quantified by combining the anomaly's growth rate and spread range, including: When real-time operational traces exhibit time sequence conflicts, trace breaks, causal sequence anomalies, or resource usage anomalies, corresponding operational trace anomaly events are generated. These events are categorized into time sequence anomalies, causal constraint anomalies, and resource usage anomalies based on their corresponding dimensions. The relationships between anomaly events are identified to determine whether anomalies propagate or overlap across multiple operational traces. For each logistics object or business instance, upon detecting the first operational trace anomaly event, a corresponding anomaly index is initialized based on the degree to which the current operational state deviates from the normal operating baseline. As subsequent operational trace anomalies are continuously detected, the anomaly index is cumulatively updated based on the type, duration, and frequency of the anomaly events. Within a continuous time window, the growth rate of the anomaly is obtained by calculating the change in the anomaly index. The scope of anomaly propagation is determined based on the number of operational traces affected, the number of business scenarios involved, or the number of resources affected by the anomaly event. The anomaly state is quantified based on the cumulative update value, growth rate, and scope of propagation of the anomaly index.

[0044] In the process of consistency evaluation of real-time operation tracks, this invention generates corresponding operation track anomaly events when time sequence conflicts, operation track breaks, causal sequence anomalies, or resource occupation anomalies are detected between operation tracks. Based on the feature dimensions involved in the anomaly events, the anomaly events are classified into time sequence anomalies, causal constraint anomalies, and resource occupation anomalies. For example, under normal circumstances, the spatial operation track of the vehicle arriving at the warehouse should be completed first, and then the behavior operation track of loading completion should be triggered. If the real-time data shows that the loading behavior operation track precedes the vehicle arrival operation track, it is determined to be a time sequence anomaly.

[0045] Identify the correlations between different operational track anomalies to determine whether anomalies propagate or overlap across multiple operational tracks. For example, vehicle delays may cause anomalies in transportation time tracks, further triggering anomalies in warehousing resource occupancy tracks. For each logistics object or business instance, upon the first detection of an operational track anomaly, initialize a corresponding anomaly index based on the degree of deviation of the current operational status from the normal operating baseline; for example, when a single time-series anomaly occurs with a small deviation, initialize the anomaly index to a lower value. If operational track anomalies are continuously detected in subsequent operations, the anomaly index is cumulatively updated based on the type, duration, and frequency of the anomaly events; for example, when the same logistics object experiences multiple causal constraint anomalies or resource occupancy anomalies within a continuous time window, the anomaly index gradually increases with the duration and frequency of the anomaly. Calculate the change in the anomaly index within the continuous time window to characterize the rate of anomaly growth; and determine the scope of anomaly spread based on the number of operational tracks affected by the anomaly event, the number of business scenarios involved, or the number of resources affected, such as an anomaly expanding from a single transportation link to multiple warehousing and delivery links. Based on the cumulative value, growth rate, and diffusion range of the anomaly index, the abnormal state of logistics objects is quantitatively described: ; In the formula, Different weighting coefficients are used to balance the degree of abnormal accumulation. Abnormal growth rate and the range of abnormal diffusion Impact on the overall anomaly score. The three weighting coefficients are summed to 1, and the overall anomaly score is adjusted according to the degree of importance attached to the accumulation, growth rate, and diffusion range of balanced anomalies.

[0046] S5. When the quantification index reaches the trigger condition, perform a time-reverse analysis of the operation process before the anomaly occurs to locate the earliest point of change that caused the deviation of the operation trace consistency; based on the operation assumptions of the point of change, form the evolution path of the anomaly and the failure type of the assumption.

[0047] In one embodiment, when the quantization index reaches the trigger condition, a time-reverse analysis is performed on the running process before the anomaly occurs to locate the earliest point of change that caused the deviation in the consistency of the running trace, including: When the anomaly index reaches the preset triggering condition, the triggering time point of the anomaly is determined, and a time-reverse analysis window is selected backward from the triggering time point as the endpoint. Within the time-reverse analysis window, multiple operation traces related to the anomaly are retrieved back, and the feature change data in the operation traces are reorganized in reverse time order to form a reverse time series. Based on the reverse time series, a consistency description model reflecting the operation trace in the dimensions of time, space, resources, and behavior is constructed. Based on the minimum description length criterion, the complexity of the consistency description model corresponding to different time points is evaluated, and the time point when the structural complexity of the operation trace changes to the target degree is detected. The time point when the target degree change is detected for the first time is taken as the earliest change point that causes the consistency of the operation trace to deviate, and the operation trace, feature dimension, and business scenario information corresponding to the change point are recorded.

[0048] In one embodiment, based on the minimum description length criterion, the complexity of the consistent description model at different time points is evaluated, and the time points at which the target degree of change in the complexity of the runtime trace structure is detected include: The number of model parameters required for each consistency description model is determined, and the model description length of the corresponding consistency description model is calculated based on the number of model parameters and their accuracy, which is used to characterize the structural complexity of the consistency description model. The degree of deviation of the running trace data from the consistency description model is determined by calculating the fit of the consistency description model to the running trace data, and this deviation is converted into a data description length. The model description length is combined with the data description length to obtain the minimum description length value of the consistency description model at the corresponding time point, which is used to characterize the overall complexity level of the running trace structure at that time point. The minimum description length values ​​corresponding to multiple time points are arranged in chronological order to form a description length sequence showing the change in running trace structural complexity over time. The change amplitude of the running trace structural complexity is calculated based on the comparison results of the minimum description length values ​​of adjacent time points. If the change amplitude reaches the target level, the corresponding time point is determined as the time point when the running trace structural complexity changes to the target level.

[0049] This invention determines the trigger time point of an anomaly and selects a reverse time analysis window backward from that trigger time point. For example, during the execution of a transportation order, a time window covering loading, transportation, and warehousing is selected backward from the trigger time point. Within the reverse analysis window, multiple operational traces related to the anomaly are retrieved, and the characteristic change data in the operational traces are reorganized in reverse chronological order to form a reverse time series. For example, operational trace data such as vehicle location, loading / unloading status, and warehouse resource occupancy are rearranged in chronological order from most recent to oldest.

[0050] Based on the aforementioned inverse time series, a descriptive model is constructed to characterize the consistency of the operational trajectory across dimensions such as time, space, resources, and behavior. Based on the minimum description length criterion, the complexity of the consistency description model corresponding to different time points is evaluated. Specifically, the model description length is calculated to characterize the model's structural complexity by statistically analyzing the number of model parameters required and the precision of parameter values. Furthermore, the deviation of the consistency description model from the operational trajectory data is converted into a data description length. Finally, the model description length and the data description length are combined to obtain the minimum description length value for the consistency description model corresponding to each time point. For example, when vehicles are operating normally and resource scheduling is coordinated, the model structure is stable and the description length is small. However, when scheduling relationships are disrupted, more parameters need to be introduced to explain the operational trajectory data, leading to a significant increase in description length.

[0051] The minimum description length criterion is used to quantify the overall complexity of the consistency description model at different time points, thereby detecting whether abrupt changes occur in the runtime trace structure. The runtime state description at a given time point is divided into two parts: the amount of information required to describe the consistency description model itself, and the amount of information required to describe the runtime trace data under the model's conditions. The model description length is determined by the number of model parameters, parameter types, and parameter value precision. For example, the number of temporal constraints, causal constraints, and resource constraints involved in the consistency description model will all increase the model description length. The data description length, given the consistency description model, is the amount of information required to encode the actual runtime trace data; its size is determined by the fitting error of the consistency description model to the runtime trace data.

[0052] Minimum description length value:

[0053] In the formula, Describe the length of the model. The length of the data description.

[0054] Furthermore, the consistency description model is a structured model used to characterize whether multiple tracks maintain coordination and consistency in terms of temporal order, causal constraints, and resource consumption within a certain point in time or time window. Essentially, it is a formalized expression of a normal co-evolutionary state. The consistency description model consists of three types of constraint sub-models: a temporal order constraint sub-model, used to describe the sequential relationship of key events in different tracks; a causal constraint sub-model, used to describe the triggering conditions of feature changes in one track on feature changes in another track; and a resource consumption constraint sub-model, used to describe the mutual exclusion, sharing, and temporal consumption relationships of multiple tracks for the same resource.

[0055] The configuration consistency description model includes the set of directed edges for time-sequence constraints and their allowed time deviation thresholds, the triggering conditions and response probabilities of causal constraints, and resource capacity, occupancy duration, and concurrency limits in resource occupancy relationships. Under normal conditions, these constraint parameters are obtained through statistical learning from historical normal operation trace samples.

[0056] The consistency description model is represented as: ; In the formula, Θ is the set of model parameters. The graph is a time-sequence constraint graph, consisting of a set of event nodes representing the execution trace and time-sequence constraint edges. It is a causal constraint graph, including the set of event nodes of the running trace and the edges of causal triggering relationships. This is a resource occupancy constraint graph, including a set of resource nodes and the occupancy relationship between the running trace and the resources. In other words, the consistency description model M is composed of three types of constraint sub-models. The time sequence constraint sub-model is used to describe the sequential relationship of key events in different running tracks. It takes the form of a directed graph structure, with nodes representing key state changes or business events in the running track, directed edges representing the sequential constraint relationship between events, and parameters representing the allowable time deviation threshold, the confidence level of sequence stability, or the probability of occurrence.

[0057] The causal constraint sub-model is used to describe the trigger-response relationship between running traces. It takes the form of a causal relationship graph, where nodes are state change events in the running trace, directed edges are causal trigger relationships, and parameters are trigger conditions, response probability, and maximum response time window.

[0058] The resource occupancy constraint sub-model is used to describe the occupancy relationship of multiple running tracks on shared resources. It takes the form of a two-part or multi-part graph structure. Resource nodes: resources such as vehicles, platforms, warehouses, and manpower; running track nodes: running tracks corresponding to the occupied resources; parameters: resource capacity, maximum concurrency, occupancy duration constraint, mutual exclusion or sharing rules.

[0059] The consistency description model first extracts and aligns key events and state changes in the running trace, and then substitutes them into the time sequence constraint sub-model, causal constraint sub-model, and resource usage constraint sub-model, respectively. It checks whether the running trace meets the normal co-evolution constraints in terms of event sequence, trigger response, and resource usage. At the same time, it quantifies the degree of violation of various constraints and performs weighted fusion. The output of the consistency description model is the consistency structure description of the running trace at the current time point or time window and its consistency deviation measurement result. This result is further used to calculate the fitting of the consistency description model to the running trace data and the corresponding data description length.

[0060] The minimum description length values ​​at multiple time points are arranged in chronological order to form a description length sequence showing the change in the complexity of the running trace structure over time. The magnitude of the change in description length at adjacent time points is compared. When the magnitude of the change in description length reaches the target level for the first time, for example, when a vehicle delay causes a conflict in loading and unloading resources, resulting in a sudden increase in the consistency structure complexity, this moment is identified as the earliest point of change that causes the running trace consistency to deviate, and the corresponding running trace, feature dimension, and business scenario information are recorded.

[0061] In one embodiment, based on the operational assumptions at the point of change, the resulting abnormal evolution paths and assumption failure types include: Before the point of change, the operational state reflected by multiple operational traces is taken as the normal reference state. Based on the normal reference state, operational hypotheses are extracted. Starting from the point of change, the actual operational behavior reflected by multiple operational traces is obtained along time and compared with the operational hypotheses. When the actual operational behavior does not conform to the operational hypothesis, the operational hypothesis is considered to have failed at this point. The entire process from conforming to the operational hypothesis to not conforming to the operational hypothesis is linked together in chronological order to form an abnormal evolution path. Based on the characteristics exhibited when the operational hypothesis fails, the anomalies are classified to obtain the hypothesis failure type.

[0062] This invention determines the operating state reflected by multiple operating tracks before the point of change as the normal reference state, and extracts the corresponding operating hypothesis based on the normal reference state. For example, in a logistics transportation scenario, if multiple operating tracks before the point of change show that the vehicle arrival time, loading and unloading operation sequence, and warehouse resource occupancy remain stable and consistent, then the operating hypothesis can be extracted: after the vehicle arrives, loading and unloading are completed within a predetermined time window, and the loading and unloading resource occupancy does not exceed the rated capacity.

[0063] Starting from the point of change, multiple operational tracks are continuously acquired along time, reflecting the actual operational behavior. This actual behavior is then compared with the operational assumptions. Specifically, it monitors whether the vehicle arrives at the loading / unloading area within the agreed timeframe, whether loading / unloading operations are performed in the predetermined order, and whether resource usage meets the constraints. When actual operational behavior is detected as not conforming to the operational assumptions, the operational assumption is deemed invalid at the corresponding time point. For example, if a vehicle arrives on time but waits for a long time due to occupied loading / unloading resources, causing the loading / unloading operation to not be completed within the assumed time window, the operational assumption is considered invalid at that moment.

[0064] The entire process from actual operational behavior conforming to the operational assumptions to failing to conform to the operational assumptions is linked together in chronological order to form an evolutionary path of anomalies from their inception to their manifestation. Based on the characteristics exhibited when the operational assumptions fail, such as time delay, disruption of causal order, or breach of resource constraints, the anomalies are classified to obtain the corresponding assumption failure type.

[0065] S6. Based on the evolution path of the anomaly and the assumed failure type, assess the impact of the anomaly and output alarm information.

[0066] In one embodiment, assessing the impact of the anomaly and outputting alarm information based on the anomaly's evolution path and assumed failure type includes: Based on the evolution path of the anomaly and the assumed failure type, assess the scope and degree of the anomaly's impact to obtain the anomaly impact assessment results; based on the anomaly impact assessment results, determine the anomaly's alarm level and output alarm information.

[0067] A comprehensive analysis of the propagation of anomalies across time, space, business, and resources is conducted to assess the scope and degree of their impact, and to generate anomaly impact assessment results. For example, in a logistics and transportation scenario, if the anomaly evolution path shows that vehicle waiting times gradually increase due to loading and unloading resource conflicts, thereby affecting subsequent batches of delivery tasks, the scope of the anomaly's impact can be determined based on the duration of the anomaly's extension over time, the number of affected transportation tracks, the number of warehousing and delivery business scenarios involved, and the number of resources occupied or blocked.

[0068] The nature of the anomaly is determined by combining the assumed failure type. For example, resource constraint-related assumption failures have a greater impact on overall operational stability than single-time delay-related assumption failures. The scope and degree of impact are combined and matched with preset alarm judgment rules to determine the alarm level corresponding to the anomaly. For example, when an anomaly only affects a single vehicle and lasts for a short period, it is determined as a low-level alarm, while when the anomaly spreads along the evolution path to multiple warehouses and delivery links, resulting in long-term occupation of critical resources, it is determined as a high-level alarm. The alarm information includes the source of the anomaly, the evolution path, and the impact of the assumed failure type on the assessment results.

[0069] Table 1. Comparison of Experimental Results on Anomaly Monitoring Effectiveness in Logistics Supply Chain

[0070] Method A is an anomaly monitoring method based on a fixed threshold, while Method B is based on single-index statistical anomaly detection. Overall, this invention offers higher accuracy, earlier detection capabilities, and stronger location capabilities in complex and multi-scenario logistics supply chain environments, thereby enhancing the intelligence level and practical application value of logistics supply chain anomaly monitoring.

[0071] likeFigure 2 As shown, according to another embodiment of the present invention, an intelligent logistics supply chain anomaly monitoring system based on big data is also provided, including: Baseline acquisition module 1 is used to acquire historical operational data of multi-source heterogeneous logistics supply chains after standardization processing; based on the target operational range of historical operational data, sample extraction and analysis of logistics business scenarios are performed to construct an operational baseline that is dynamically adjusted with time and scenario. The operation track acquisition module 2 is used to map real-time feature data to the scene corresponding to the operation baseline, and organize the feature changes of the same logistics object in the dimensions of time, space, resources and behavior into multiple operation tracks; Anomaly quantification module 3 is used to continuously evaluate the consistency between running traces by learning the mutual interpretation relationships and co-evolution patterns of each running trace in terms of time sequence, causal constraints and resource consumption under historical normal conditions; when conflicts, breaks or causal sequence anomalies occur in real-time running traces, the anomaly index is gradually accumulated and the anomaly state is quantified by combining the growth rate and diffusion range of the anomaly. The anomaly analysis and alarm output module 4 is used to perform time-reverse analysis of the running process before the anomaly occurs when the quantization index reaches the trigger condition, locate the earliest change point that caused the deviation of the running trace consistency; based on the running hypothesis of the change point, form the evolution path of the anomaly and the assumed failure type; according to the evolution path of the anomaly and the assumed failure type, evaluate the impact of the anomaly and output alarm information.

[0072] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent logistics supply chain anomaly monitoring based on big data, characterized in that, include: S1. Obtain historical operational data of the multi-source heterogeneous logistics supply chain after standardization processing; S2. Based on historical operational data, target operational ranges are used for sample extraction and analysis of logistics business scenarios to construct operational baselines that are dynamically adjusted with time and scenarios. S3. Map real-time feature data to the scenario corresponding to the operating baseline, and organize the feature changes of the same logistics object in the dimensions of time, space, resources and behavior into multiple operating tracks; S4. By learning the mutual interpretation relationships and co-evolution patterns of each running track in the historical normal state in terms of time sequence, causal constraints and resource consumption, the consistency between running tracks is continuously evaluated; when conflicts, breaks or causal sequence anomalies occur in real-time running tracks, the anomaly index is gradually accumulated, and the anomaly state is quantified by combining the growth rate and diffusion range of the anomaly. S5. When the quantification index reaches the trigger condition, perform a time-reverse analysis of the operation process before the anomaly occurs to locate the earliest point of change that caused the deviation of the operation trace consistency; based on the operation assumptions of the point of change, form the evolution path of the anomaly and the failure type of the assumption; S6. Based on the evolution path of the anomaly and the assumed failure type, assess the impact of the anomaly and output alarm information.

2. The intelligent logistics supply chain anomaly monitoring method based on big data according to claim 1, characterized in that, The target operating range based on historical operating data involves sample extraction and analysis of different logistics business scenarios to construct an operating baseline that dynamically adjusts with time and scenarios, including: Based on historical operational data, determine the target operational range of logistics objects in various logistics business scenarios, including time range, resource occupation range, and behavioral scope; Historical operational data is divided into logistics business scenarios according to business type; within each logistics business scenario, sample data that meets the target operating range is extracted from historical operational data as normal operating samples; Based on normal operation samples, statistical characteristics, including mean, distribution area and fluctuation range, are calculated for the time characteristics, spatial characteristics, resource characteristics and behavioral characteristics of logistics objects to characterize their operation, and corresponding operation baselines are constructed. The operating baseline is periodically updated based on changes in time and scene switching.

3. The intelligent logistics supply chain anomaly monitoring method based on big data according to claim 1, characterized in that, The process of mapping real-time feature data to the scenario corresponding to the operational baseline, and organizing the feature changes of the same logistics object in the dimensions of time, space, resources, and behavior into multiple operational tracks, includes: Acquire real-time characteristic data of logistics objects during operation, including time characteristics, spatial characteristics, resource characteristics, and behavioral characteristics; perform format standardization, time alignment, and outlier handling on the real-time characteristic data; Based on the business status, location status, or operational phase information in real-time feature data, identify the business scenario in which the current logistics object is located, and map the real-time feature data to the operational baseline corresponding to the current logistics business scenario; Real-time feature data is time-series converted, and the time-series converted real-time feature data is divided into time dimension, spatial dimension, resource dimension and behavior dimension, and the change process of features in each dimension over time is extracted respectively; Organize the characteristic changes of the same logistics object in each dimension into corresponding operation tracks; Each track is assigned a track identifier, and the logistics object identifier, business scenario, and time interval are recorded; the association between multiple tracks under the same logistics object is established.

4. The intelligent logistics supply chain anomaly monitoring method based on big data according to claim 3, characterized in that, The continuous evaluation of consistency among running tracks by learning the mutual explanatory relationships and co-evolution patterns of each running track in terms of time sequence, causal constraints, and resource consumption under normal historical conditions includes: Several corresponding operation traces are extracted from the normal operation samples to form a historical normal operation trace sample set, which covers the operation status under different time periods and different business scenarios. Based on historical normal operation track samples, the temporal sequence relationship between each operation track is statistically analyzed; based on the triggering and response relationship between changes in different operation track characteristics, the causal constraint relationship between operation tracks under normal conditions is established to describe the impact conditions of operation track changes on other operation track changes; based on the timing and mode of resource occupation in the logistics supply chain by multiple operation tracks during operation, the resource occupation relationship of operation tracks is established. Based on temporal sequence, causal constraints, and resource consumption, a collaborative evolution model of multiple running tracks under normal conditions is constructed. The running traces corresponding to real-time feature data are input into the co-evolutionary model to evaluate the consistency of the running traces corresponding to real-time feature data in terms of time sequence, causal constraints, and resource consumption.

5. The intelligent logistics supply chain anomaly monitoring method based on big data according to claim 1, characterized in that, When conflicts, breaks, or causal sequence anomalies occur in the real-time running trace, the anomaly index is gradually accumulated, and the anomaly state is quantified by combining the growth rate and spread range of the anomaly. This includes: When a time sequence conflict, a broken track, an abnormal causal order, or an abnormal resource usage occurs between real-time running tracks, a corresponding running track abnormal event is generated. For each logistics object or business instance, when the first operational anomaly event is detected, the corresponding anomaly index is initialized based on the degree to which the current operational status deviates from the normal operating baseline; when operational anomalies are continuously detected in subsequent operations, the anomaly index is cumulatively updated based on the type, duration, and frequency of the operational anomaly events. Within a continuous time window, the rate of abnormal growth is obtained by calculating the change in the abnormal index. The scope of the anomaly's spread is determined based on the number of operational traces affected by the anomaly event, the number of business scenarios involved, or the number of resources affected. Abnormal states are quantified based on the cumulative update value, growth rate, and diffusion range of the anomaly index.

6. The intelligent logistics supply chain anomaly monitoring method based on big data according to claim 1, characterized in that, When the quantification index reaches the trigger condition, a time-reverse analysis is performed on the running process before the anomaly occurs to locate the earliest point of change that caused the deviation in the consistency of the running trace, including: When the anomaly index reaches the preset triggering condition, the triggering time point of the anomaly is determined, and the time reverse analysis window is selected backward from the triggering time point as the endpoint. Within the time-reverse analysis window, multiple operation traces related to the anomaly are retrieved by tracing back, and the feature change data in the operation traces are reorganized in reverse time order to form a reverse time series; Based on inverse time series, a consistent description model reflecting the running trace in the dimensions of time, space, resources and behavior is constructed; Based on the minimum description length criterion, the complexity of the consistent description model corresponding to different time points is evaluated, and the time points at which the complexity of the running trace structure changes to the degree of target change are detected. The time point at which the first change in the target degree is detected is taken as the earliest point of change that causes the consistency of the running trace to deviate, and the running trace, feature dimension and business scenario information corresponding to the point of change are recorded.

7. The intelligent logistics supply chain anomaly monitoring method based on big data according to claim 6, characterized in that, The process of evaluating the complexity of the consistent description model at different time points based on the minimum description length criterion, and detecting the time points where the complexity of the runtime trace structure changes to the target degree, includes: Determine the number of model parameters required for each consistency description model, and calculate the model description length of the corresponding consistency description model based on the number of model parameters and their value precision, which is used to characterize the structural complexity of the consistency description model; By calculating the fit of the consistency description model to the runtime data, the degree of deviation of the runtime data from the consistency description model is determined, and the degree of deviation is converted into data description length. The model description length is combined with the data description length to obtain the minimum description length value of the consistent description model at the corresponding time point, which is used to characterize the overall complexity level of the running trace structure at that time point. Arrange the minimum description length values ​​corresponding to multiple time points in chronological order to form a description length sequence in which the complexity of the running trace structure changes over time; Calculate the magnitude of change in the complexity of the running trace structure based on the comparison results of the minimum description length values ​​at adjacent time points; If the change reaches the target level, the corresponding time point will be determined as the time point when the complexity of the running trace structure changes to the target level.

8. The intelligent logistics supply chain anomaly monitoring method based on big data according to claim 1, characterized in that, The operational assumptions based on the points of change lead to abnormal evolution paths and assumption failure types, including: Before the point of change, the operating state reflected by multiple operating tracks is taken as the normal reference state; based on the normal reference state, operating hypotheses are extracted. Starting from the point of change, multiple operational traces are obtained along time to reflect the actual operational behavior, and compared with the operational assumptions. When the actual operational behavior does not conform to the operational assumptions, the operational assumptions are considered invalid at that point. The entire process from conforming to the operating assumptions to failing to conform to the operating assumptions is linked together in chronological order, forming an abnormal evolutionary path; Based on the characteristics exhibited when the operating hypothesis fails, the anomalies are classified to obtain the hypothesis failure type.

9. The intelligent logistics supply chain anomaly monitoring method based on big data according to claim 1, characterized in that, The process of assessing the impact of an anomaly and outputting alarm information based on its evolution path and assumed failure type includes: Based on the evolution path of the anomaly and the assumed failure type, the scope and degree of the anomaly's impact are assessed to obtain the anomaly impact assessment results; Based on the impact assessment results of the anomaly, determine the alarm level of the anomaly and output the alarm information.

10. A big data-based intelligent logistics supply chain anomaly monitoring system, used to implement any one of the big data-based intelligent logistics supply chain anomaly monitoring methods according to claims 1-9, characterized in that, include: The baseline acquisition module is used to acquire historical operational data of a multi-source heterogeneous logistics supply chain after standardization. Based on historical operational data, target operational ranges are used for sample extraction and analysis of logistics business scenarios to construct operational baselines that are dynamically adjusted over time and according to scenarios. The operation track acquisition module is used to map real-time feature data to the scene corresponding to the operation baseline, and organize the feature changes of the same logistics object in the dimensions of time, space, resources and behavior into multiple operation tracks; The anomaly quantification module is used to continuously evaluate the consistency between running traces by learning the mutual interpretation relationships and co-evolution patterns of each running trace in terms of time sequence, causal constraints and resource consumption under historical normal conditions; when conflicts, breaks or causal sequence anomalies occur in real-time running traces, the anomaly index is gradually accumulated and the anomaly state is quantified by combining the growth rate and diffusion range of the anomaly. The anomaly analysis and alarm output module is used to perform time-reverse analysis of the running process before the anomaly occurs when the quantization index reaches the trigger condition, to locate the earliest point of change that caused the deviation of the running trace consistency; based on the running assumptions of the point of change, to form the evolution path of the anomaly and the assumed failure type; and based on the evolution path of the anomaly and the assumed failure type, to assess the impact of the anomaly and output alarm information.

Images