A logistics supply chain monitoring method, device, equipment and medium

Abstract

The application discloses a logistics supply chain monitoring method, device, equipment and medium, relates to the technical field of logistics and supply chain informatization monitoring, and comprises the following steps: S1, acquiring real-time data from a transport vehicle and goods through a sensor network, including position coordinates, temperature values and time limit indicators to obtain an initial abnormal signal data set, and performing preliminary screening on humidity fluctuations in a transport scene to determine a humidity-related abnormal subset; S2, grouping data points according to the initial abnormal signal data set to judge abnormal types by adopting a K-means clustering algorithm, and extracting sensitive parameters of the goods type in the humidity-related abnormal subset to determine temperature and humidity combined abnormal types; the logistics supply chain monitoring method, device, equipment and medium significantly improve transport abnormal response efficiency and path optimization capability, and guarantee the safety and timeliness of goods transportation.

Description

Technical Field

[0001] This invention relates to the field of logistics and supply chain information monitoring technology, specifically to a logistics and supply chain monitoring method, device, equipment, and medium. Background Technology

[0002] In modern logistics supply chain management, real-time monitoring of the transportation process is a key technological aspect for ensuring cargo safety and transportation efficiency. Especially in scenarios such as temperature-controlled transportation and cold chain logistics, fluctuations in the temperature and humidity of goods directly impact transportation quality and economic losses. To ensure controllable transportation, existing systems typically rely on sensor nodes to collect environmental parameters of transport vehicles and goods, such as temperature, humidity, and location, to monitor and alert on abnormal conditions. However, with the increasing complexity of supply chain scenarios, existing anomaly management solutions have significant limitations in responding to dynamic environmental changes.

[0003] Currently used anomaly detection methods mainly rely on fixed thresholds or single-rule models. When environmental factors, transportation routes, or cargo types change, the system cannot automatically adjust the judgment conditions. For example, in long-distance transportation in mountainous areas, humidity changes may be affected by multiple factors such as altitude, climate, and cargo type. If fixed threshold identification is still used, it will lead to an increased false alarm rate or missed anomalies. In addition, different goods have different sensitivities to temperature and humidity. For example, the transportation of highly sensitive categories such as fruits and medicines requires dynamic adjustment of the combined temperature and humidity thresholds according to the type, but traditional systems cannot achieve differentiated judgment based on cargo attributes. On the other hand, the response mechanism of existing systems after anomaly confirmation still mainly relies on manual processing, lacking an automated execution chain from anomaly detection to emergency response. When the system detects an anomaly, it usually only triggers an alarm, without automatically matching backup plans or adjusting transportation routes. This not only prolongs the anomaly handling time but also increases the risk of cargo damage. At the same time, existing systems lack a quantitative monitoring mechanism for the entire anomaly response process and cannot optimize threshold settings or response strategies based on historical execution results, resulting in insufficient system adaptability. Summary of the Invention

[0004] The purpose of this invention is to provide a logistics supply chain monitoring method, device, equipment, and medium to solve the problems existing in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a logistics supply chain monitoring method, comprising: S1, acquiring real-time data from transport vehicles and goods through a sensor network, including location coordinates, temperature values, and timeliness indicators to obtain an initial abnormal signal dataset, and performing preliminary screening for humidity fluctuations in the transportation scenario to determine a humidity-related abnormal subset; S2, using a K-means clustering algorithm to group data points based on the initial abnormal signal dataset to determine the abnormal type, and extracting sensitive parameters of the cargo type from the humidity-related abnormal subset to determine the temperature and humidity joint abnormal type; S3, if the abnormal type is determined to be a temperature and humidity joint abnormality, obtaining the corresponding threshold range from a preset cargo type database to obtain dynamic triggering conditions, and incorporating route adjustment instructions as a condition extension part during transportation; S4, for the dynamic triggering conditions, supporting... The measurement and measurement classifier analyzes the deviation between real-time data and threshold range to determine the anomaly confirmation result, and embeds backup plan activation logic into the deviation analysis to generate confirmation output containing sequence execution logic; S5, based on the anomaly confirmation result, obtains the matching automated process script from the preset response rule base to obtain the response execution sequence, and integrates transportation control instructions into the sequence to adapt to path optimization under terrain constraints; S6, sends instructions to the transportation control system through the response execution sequence to adjust the path or activate the backup plan to obtain the updated transportation plan, and records the time difference calculation method in the plan for subsequent time consumption index evaluation; S7, obtains the time difference between the updated transportation plan and the initial anomaly signal dataset to calculate the response time index to judge the system optimization level, and applies the initial dataset comparison in the judgment to quantify the response time and adjust the benchmark for the next anomaly signal.

[0006] Preferably, step S1 includes acquiring real-time data from transport vehicles and goods via a sensor network, including location coordinates, temperature values, timeliness indicators, and humidity values, to generate an initial abnormal signal dataset; processing the initial abnormal signal dataset using a time series analysis method to extract the trend of humidity value changes and obtain humidity fluctuation characteristics; if the humidity fluctuation characteristics exceed a preset threshold range, then using a clustering algorithm to divide the data points into normal and abnormal categories to determine a humidity-related abnormal subset; based on the humidity-related abnormal subset, combined with location coordinates and timeliness indicators, calculating the correlation strength between the abnormal subset and the transportation scenario to obtain an abnormal distribution pattern; using the abnormal distribution pattern, using a decision tree algorithm to classify the abnormal subset, determining whether the abnormality is related to the transportation scenario, and obtaining an abnormal classification result; for the abnormal classification result, acquiring the corresponding location coordinates and timeliness indicators to generate the spatiotemporal distribution characteristics of the abnormal signal and determine the specific scenario in which the abnormality occurred; and updating the acquisition frequency of the sensor network based on the specific scenario in which the abnormality occurred to obtain an optimized data acquisition strategy.

[0007] Preferably, step S2 includes collecting real-time data of transport vehicles and goods through a sensor network, including location coordinates, temperature values, humidity values, and timeliness indicators, to generate an initial abnormal signal dataset; using a K-means clustering algorithm to group the initial abnormal signal dataset to obtain abnormal type grouping results; extracting a humidity abnormal subset from the abnormal type grouping results, analyzing the time series changes of humidity values, and obtaining a humidity abnormal fluctuation pattern; for the humidity abnormal fluctuation pattern, if the fluctuation amplitude of the humidity value exceeds a preset threshold, then combining the temperature value to calculate the joint distribution characteristics of the two, and determining the temperature and humidity joint abnormal type; based on the temperature and humidity joint abnormal type, obtaining the corresponding location coordinates and timeliness indicators, and generating the spatiotemporal distribution pattern of the abnormal signal.

[0008] Preferably, step S3 includes acquiring real-time temperature and humidity values ​​from sensor data collection, matching them with temperature and humidity threshold ranges in the cargo type database using a threshold comparison method to determine if a joint temperature and humidity anomaly exists, and obtaining an anomaly confirmation result; if the anomaly confirmation result is a joint temperature and humidity anomaly, then extracting the corresponding cargo's sensitive parameters from the cargo type database, combining them with real-time data updates to generate dynamic triggering conditions, and obtaining a triggering condition set; based on the triggering condition set, using a path planning algorithm to calculate transportation route adjustment instructions, generating a path instruction set containing time and space constraints; updating the navigation data of the transport vehicle through the path instruction set, adjusting the sensor data acquisition frequency using a real-time data update mechanism, and obtaining an optimized acquisition strategy; extracting real-time data streams from the optimized acquisition strategy, continuously monitoring the joint temperature and humidity anomaly using an anomaly type judgment method, and obtaining an anomaly status update result; if the anomaly status update result exceeds the dynamic triggering conditions, obtaining a backup transportation plan from the cargo type database, and generating an adjusted path instruction set; updating the operating status of the transport vehicle through the adjusted path instruction set, and integrating sensor data acquisition and path planning results using data fusion technology to obtain the final transportation optimization plan.

[0009] Preferably, step S4 includes obtaining temperature and humidity values ​​from real-time data collected by sensors, analyzing the deviation between the real-time data and a preset threshold range using a support vector machine classifier, and determining the confirmation result of the joint anomaly of temperature and humidity; if the confirmation result is anomaly, the startup logic of the backup plan is triggered based on the degree of deviation, generating an anomaly confirmation output containing sequence execution logic; based on the anomaly confirmation output, the corresponding backup transportation plan is obtained from the preset backup plan database, generating an adjusted transportation instruction set; based on the adjusted transportation instruction set, the navigation data of the transportation vehicles is updated using a path planning algorithm, generating a path instruction set containing time and space constraints; real-time navigation data is extracted from the path instruction set, and the sensor-collected data and navigation data are integrated using data fusion technology to generate an optimized transportation data stream; based on the optimized transportation data stream, anomaly detection algorithm is used to continuously monitor the joint anomaly of temperature and humidity, determining the anomaly state update result; if the anomaly state update result exceeds a preset threshold range, the backup transportation plan is updated through sequence execution logic, generating the final optimized transportation instruction set.

[0010] Preferably, step S5 includes: obtaining the anomaly type and deviation data from the anomaly confirmation result; querying a preset response rule base; matching the corresponding automated process script; generating an initial response execution sequence; extracting transportation control instructions based on the initial response execution sequence; combining terrain limitation data; using the A-path planning algorithm to generate an optimized path instruction set; obtaining navigation data from the path instruction set; integrating environmental data and navigation data collected by sensors using data fusion technology to generate a real-time transportation data stream; continuously monitoring the anomaly state using the isolated forest algorithm based on the real-time transportation data stream; judging the state update result; if the state update result exceeds a preset threshold, obtaining a backup automated process script from the response rule base and updating the response execution sequence; adjusting the transportation control instructions based on the updated response execution sequence to generate a final optimized instruction set; extracting path data from the final optimized instruction set; using data verification technology to confirm the integrity of the instruction set and obtain executable transportation instructions.

[0011] Preferably, step S6 includes extracting route adjustment instructions from the response execution sequence, generating a structured instruction dataset using data parsing technology; using the structured instruction dataset, combined with backup plan data, and employing a decision tree algorithm to determine the applicability of the route adjustment instructions, obtaining a filtered instruction set; if the filtered instruction set meets a preset threshold, generating an adjusted transportation plan through a data integration process, obtaining plan data containing route adjustments; extracting time deviation data from the plan data containing route adjustments, generating time deviation records using a time difference calculation method, obtaining a transportation plan containing time differences; using the transportation plan containing time differences, sending route adjustment instructions to the transportation control system using an instruction allocation mechanism, obtaining the updated plan status after execution; obtaining data from the time record module from the updated plan status after execution, using data verification technology to determine the completeness of the updated plan status, obtaining the final transportation plan; generating a real-time monitoring data stream from the final transportation plan, using a random forest algorithm to determine the execution deviation of the transportation plan, obtaining an optimized transportation instruction set;

[0012] S7 includes: obtaining timestamp data from the initial abnormal signal dataset; using time series parsing technology to generate a structured time series dataset; aligning the time series dataset with the updated transportation plan using time alignment and calculating time deviation data using sliding window technology; obtaining a time deviation dataset; if the deviation value in the time deviation dataset meets a preset threshold, matching the time deviation dataset with a preset time consumption standard using data comparison technology to generate a response time consumption index; extracting response time features from the time consumption index dataset; using a linear regression algorithm to analyze the relationship between response time features and system performance, judging the system optimization level, and obtaining an optimization level evaluation result; comparing the optimization level evaluation result with the timestamp data of the initial abnormal signal dataset; using a difference calculation method to quantify the change amplitude of response time, and obtaining a quantized response time dataset; extracting deviation trend values ​​from the quantized response time dataset; using benchmark adjustment technology to adjust the triggering conditions of abnormal signals according to the deviation trend values, generating an adjusted abnormal signal benchmark, and obtaining an updated benchmark dataset; and using data verification technology to verify the integrity of the time series data after benchmark adjustment, and obtaining a verified benchmark dataset.

[0013] A logistics supply chain monitoring device is used to implement the steps of the aforementioned logistics supply chain monitoring method. The device includes a data acquisition module that acquires real-time data from transport vehicles and goods via a sensor network, including location coordinates, temperature values, and timeliness indicators to obtain an initial abnormal signal dataset, and performs preliminary screening based on humidity fluctuations in the transportation scenario to determine a humidity-related abnormal subset; an anomaly analysis module that uses a K-means clustering algorithm to group data points based on the initial abnormal signal dataset to determine the anomaly type, and extracts sensitive parameters of the cargo type from the humidity-related abnormal subset to determine the combined temperature and humidity anomaly type; a dynamic triggering module that, if the anomaly type is determined to be a combined temperature and humidity anomaly, obtains the corresponding threshold range from a preset cargo type database to obtain dynamic triggering conditions, and incorporates route adjustment instructions as an extension of the conditions during transportation; and an anomaly confirmation module that, based on the dynamic triggering conditions... The system uses a support vector machine classifier to analyze the deviation between real-time data and threshold ranges to determine the anomaly confirmation result. It also embeds backup plan activation logic into the deviation analysis to generate a confirmation output containing sequence execution logic. The response execution module retrieves matching automated process scripts from a pre-set response rule base based on the anomaly confirmation result to obtain a response execution sequence. It integrates transportation control instructions into the sequence to adapt to path optimization under terrain constraints. The transportation control module sends instructions to the transportation control system via the response execution sequence to adjust the path or activate backup plans, resulting in an updated transportation plan. It records the time difference calculation method in the plan for subsequent time consumption index evaluation. The optimization evaluation module obtains the time difference between the updated transportation plan and the initial anomaly signal dataset to calculate the response time index and judge the system optimization level. In the judgment, it uses a comparison with the initial dataset to quantify the response time and adjust the benchmark for the next anomaly signal.

[0014] A computer device includes a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the logistics supply chain monitoring method.

[0015] A computer-readable storage medium storing computer instructions for causing a computer to execute the aforementioned logistics supply chain monitoring method.

[0016] As can be seen from the above technical solution, the present invention has the following beneficial effects:

[0017] This logistics supply chain monitoring method, device, equipment, and medium acquires real-time data such as the location, temperature, and humidity of transport vehicles and goods through a sensor network. It uses a K-means clustering algorithm to group the initial abnormal signal dataset, filters humidity-related abnormal subsets, and extracts fruit type-sensitive parameters to identify joint temperature and humidity anomalies. For anomalies, this invention uses a support vector machine classifier to analyze the deviation between real-time data and preset thresholds, generating a confirmation output containing backup plan activation logic. It also obtains automated process scripts from a response rule base and integrates transport control commands to optimize mountainous terrain routes. This invention dynamically adjusts transport plans, records time differences to calculate and evaluate time consumption indicators, quantifies response time by comparing with the initial dataset, and adjusts the anomaly signal benchmark. This invention significantly improves transport anomaly response efficiency and route optimization capabilities, ensuring the safety and timeliness of cargo transportation. Attached Figure Description

[0018] Figure 1 This is a flowchart of the logistics supply chain monitoring method of the present invention;

[0019] Figure 2 This is a connection diagram of the logistics supply chain monitoring device module of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] like Figure 1As shown, this invention provides a technical solution: a logistics supply chain monitoring method, comprising: S1, acquiring real-time data from transport vehicles and goods through a sensor network, including location coordinates, temperature values, and timeliness indicators to obtain an initial abnormal signal dataset, and performing preliminary screening for humidity fluctuations in the transportation scenario to determine a humidity-related abnormal subset; S2, using a K-means clustering algorithm to group data points based on the initial abnormal signal dataset to determine the abnormal type, and extracting sensitive parameters of the cargo type from the humidity-related abnormal subset to determine the temperature and humidity joint abnormal type; S3, if the abnormal type is determined to be a temperature and humidity joint abnormality, obtaining the corresponding threshold range from a preset cargo type database to obtain dynamic triggering conditions, and incorporating route adjustment instructions as a condition extension part during transportation; S4, using a support vector machine to analyze the dynamic triggering conditions. The classifier analyzes the deviation between real-time data and the threshold range to determine the anomaly confirmation result, and embeds backup plan activation logic into the deviation analysis to generate a confirmation output containing sequence execution logic; S5, based on the anomaly confirmation result, the matching automated process script is obtained from the preset response rule base to obtain the response execution sequence, and transportation control instructions are integrated into the sequence to adapt to path optimization under terrain constraints; S6, through the response execution sequence, instructions are sent to the transportation control system to adjust the path or activate the backup plan to obtain the updated transportation plan, and the time difference calculation method is recorded in the plan for subsequent time consumption index evaluation; S7, the time difference between the updated transportation plan and the initial anomaly signal dataset is obtained to calculate the response time index to judge the system optimization level, and the initial dataset is used for comparison to quantify the response time and adjust the benchmark for the next anomaly signal.

[0022] In this embodiment, a sensor network composed of multiple types of sensor nodes distributed across transport vehicles and cargo storage areas enables real-time monitoring of the entire transportation process. The sensor nodes include positioning modules for acquiring vehicle location information, environmental sensors for detecting cargo hold temperature and humidity, and accelerometers for monitoring vehicle operating status. Each sensor node transmits the real-time collected data to a cloud monitoring platform via a wireless communication module (e.g., cellular IoT or short-range wireless communication). Upon receiving the data, the platform first performs noise filtering, packet loss compensation, and time synchronization processing on the raw signals to ensure data integrity and consistency, thereby forming a standardized initial abnormal signal dataset.

[0023] When filtering humidity fluctuations in this dataset, the system continuously analyzes the trend of humidity changes. When the humidity change exceeds a preset range within a short period, the system automatically marks the data segment as a candidate for humidity anomalies. In this way, a subset of humidity-related anomalies is initially extracted. For this subset, the system automatically matches environmental sensitivity parameters based on the cargo type, such as the allowable temperature fluctuation range, humidity stability range, and cargo category code, for subsequent anomaly clustering analysis. In the anomaly detection phase, the system uses a clustering algorithm to automatically group different types of anomaly data. This algorithm identifies different types, such as temperature anomalies, humidity anomalies, or combined temperature and humidity anomalies, by calculating the degree of difference between data points. When the system identifies a combined temperature and humidity anomaly, it calls relevant data from the cargo type database to obtain the optimal environmental threshold range for that type of cargo. Based on this range, the system generates dynamic trigger conditions to determine whether route adjustments or environmental control operations need to be performed during the current transportation process.

[0024] Subsequently, the system analyzes the deviation of real-time data from the threshold range using a classification algorithm to determine whether an anomaly truly exists. When the system confirms an anomaly, the backup plan activation logic is activated. This logic selects a corresponding solution from a pre-set emergency strategy library based on the anomaly type and severity, such as activating backup refrigeration equipment, modifying vehicle transport routes, or transferring goods to an intermediate cold chain storage point. The entire confirmation and response process is completed through a sequence execution logic module, which ensures that each action is executed sequentially according to a set priority using sequential control. During the response execution phase, the system matches an automated processing script from the response rule library based on the confirmed anomaly result. The rule library defines information such as anomaly type, triggering conditions, corresponding actions, and execution order. The system generates a transport control instruction sequence based on this script and implements it through the vehicle control system, including route replanning, temperature control equipment adjustment, and emergency communication activation. The time nodes of all executed actions are recorded for subsequent evaluation. After execution, the system obtains a response time index by comparing the time difference between the transport route and task execution before and after adjustment. The monitoring platform calculates the overall system response efficiency based on this index, combined with the timeliness parameters in the initial anomaly signal dataset. If the system detects that the response time is too long or the exception handling is insufficient, it will automatically correct the subsequent exception identification benchmark and trigger parameters to achieve continuous optimization, thereby ensuring that the handling of the next exception event is more efficient and accurate.

[0025] This method combines multidimensional data acquisition with clustering algorithms to accurately identify anomalies during transportation, demonstrating high sensitivity and reliability, particularly in the joint identification of temperature and humidity anomalies. The introduction of dynamic triggering conditions and route adjustment mechanisms improves the real-time performance and adaptability of anomaly handling. The use of support vector machines enhances the accuracy of anomaly confirmation, while embedded backup scheme logic strengthens the system's fault tolerance. Automated process scripts control transportation route optimization, effectively improving logistics response speed and reducing human intervention costs. Finally, a closed-loop evaluation of the system's optimization level is achieved through time-consuming difference quantification, enabling the logistics monitoring system to possess continuous self-optimization capabilities.

[0026] S1 includes acquiring real-time data from transport vehicles and goods via a sensor network, including location coordinates, temperature values, timeliness indicators, and humidity values, to generate an initial abnormal signal dataset; processing the initial abnormal signal dataset using time series analysis methods to extract the trend of humidity value changes and obtain humidity fluctuation characteristics; if the humidity fluctuation characteristics exceed a preset threshold range, then using a clustering algorithm to divide the data points into normal and abnormal categories to determine a humidity-related abnormal subset; based on the humidity-related abnormal subset, combined with location coordinates and timeliness indicators, calculating the correlation strength between the abnormal subset and the transportation scenario to obtain an abnormal distribution pattern; using the abnormal distribution pattern, using a decision tree algorithm to classify the abnormal subset, determining whether the abnormality is related to a specific transportation scenario, and obtaining an abnormal classification result; for the abnormal classification result, acquiring the corresponding location coordinates and timeliness indicators to generate the spatiotemporal distribution characteristics of the abnormal signal and determine the specific scenario in which the abnormality occurred; and updating the sensor network's acquisition frequency based on the specific scenario in which the abnormality occurred to obtain an optimized data acquisition strategy.

[0027] In this embodiment, the sensor network first collects and records the time, location coordinates, temperature value, humidity value, and timeliness index of each data point in chronological order, forming an initial abnormal signal dataset. The location coordinates are the combination of longitude and latitude of the geographical location, and the timeliness index is the difference between the planned arrival time and the actual arrival time of the corresponding transportation task at that moment, or the sum of the duration of the vehicle being en route and the duration of its stationary state at that moment. To ensure the usability of the dataset, it is processed in the following fixed order: Step 1, sort all records by time; Step 2, delete records with missing time, missing location coordinates, missing temperature values, or missing humidity values; Step 3, sort records by time... Step 4: For duplicate records within a given time period, retain the single record whose time is closest to a whole second; Step 5: Delete records of temperature and humidity values ​​that exceed the physically measurable range; After completion, a standard dataset for calculation is obtained; Subsequently, time series analysis is performed on the humidity values, clarifying the calculation process as follows: Determine a fixed-length sliding time window. The window length is determined by the following method: statistically analyze the duration of all humidity anomaly events during historical normal transportation phases, sort these durations from smallest to largest, and take the value in the first 1 / 4 of all data as the window length. If this value is less than 20 seconds, the window length is set to 20 seconds; if the value is greater than 300 seconds, the window length is increased. The window length is set to 300 seconds to avoid misjudgments caused by windows that are too short or too long. Within each window, three humidity fluctuation characteristics are calculated sequentially: the first is the short-term fluctuation amplitude, calculated by subtracting the minimum humidity value from the maximum humidity value within the window and recording the difference; the second is the rate of change, calculated as the average of the absolute values ​​of the differences between two adjacent humidity samples within the window; the third is the duration, calculated as the total time within the window that continuously exceeds the stable interval. The stable interval is derived from the distribution of humidity values ​​during historical normal transportation periods. The humidity values ​​during historical normal transportation periods are sorted from smallest to largest, and the values ​​in the top 10% are taken as the lower limit of the stable interval, while the values ​​in the bottom 10% are taken as the lower limit. The value at the location is used as the upper limit of the stable interval; the three humidity fluctuation features are bound to the window timestamp and location coordinates and written into the temporary result set; then the threshold range of humidity fluctuation features is set and the determination method is given: calculate the three humidity fluctuation features one by one for the temporary result set of the historical normal transportation phase, and obtain the distribution of short-term fluctuation amplitude, the distribution of change rate and the distribution of duration respectively. The value at 95% position of each distribution is used as the upper limit of the threshold for that feature, and the value at 5% position is used as the lower limit of the threshold for that feature. If any one humidity fluctuation feature in the current window exceeds the corresponding upper limit of the threshold or is lower than the corresponding lower limit of the threshold, then the window is marked as a humidity fluctuation exceeding limit window.When a humidity fluctuation exceeds the limit window, a clustering algorithm is activated to divide all data points within the window into two categories. The division dimensions and order are fixed as follows: humidity value, temperature value, absolute value of adjacent humidity sampling differences, absolute value of adjacent temperature sampling differences, timeliness index, location coordinates, and the aforementioned three humidity fluctuation characteristics. When using location coordinates for distance calculations, the difference between longitude and latitude is converted to meters. The initial cluster centers are determined as follows: first, the data point with the humidity value closest to the window mean is selected as the normal initial center; then, the data point with both humidity and temperature values ​​far from the window mean and whose absolute value of adjacent humidity sampling differences is in the highest 10% within the window is selected as the abnormal initial center. Subsequently, allocation and updates are performed. The iterative process is set with an upper limit of 100 iterations. It stops immediately when the sum of the intra-class distances after two iterations falls below a pre-defined stopping threshold. After completion, the set of data points classified as anomalies is identified as a humidity-related anomaly subset. After obtaining the humidity-related anomaly subset, the correlation strength between this subset and the transportation scenario is calculated. The transportation scenario is defined by the geographical segment corresponding to the location coordinates and the on-the-go and stop-the-way states corresponding to the timeliness indicators. The correlation strength calculation consists of three steps: Step 1 calculates the spatial overlap ratio, which is the ratio of the number of anomalies within the coverage area of ​​a certain scenario to the total number of anomalies in that anomaly subset, and converts it to a percentage; Step 2 calculates the temporal overlap ratio, which is the ratio of the number of anomalies occurring within a key time period of a certain scenario to the total number of anomalies in that anomaly subset. The ratio of all anomalies is calculated and converted to a percentage. The key time period is determined by the hourly intervals in historical data where anomalies occur most frequently in this scenario. If historical data is insufficient, the vehicle dwell time period is used. The third step is to calculate the consistency score, which is the degree of similarity between the humidity fluctuation characteristics of anomalies in this scenario and the humidity fluctuation characteristics of historically confirmed anomalies in this scenario. Specifically, each of the three humidity fluctuation characteristics is compared to see if it falls within the main interval of its respective historical anomaly distribution. One point is awarded for each characteristic, and the total score of the three characteristics is between 0 and 3. The spatial overlap ratio, temporal overlap ratio, and consistency score are weighted and summed according to a fixed ratio to obtain a correlation strength score between 0 and 100. The ratio is determined before deployment using the following method: based on historical data... Using labeled data as a baseline, the system iterates through integer ratio combinations from 1:1:1 to 4:4:4, calculates the misclassification rate of each combination on the validation data, and selects the combination with the lowest misclassification rate as a fixed ratio. After obtaining the correlation strength of each scenario, all scenarios are sorted from high to low scores, and the scenario corresponding to the highest score is recorded. Then, decision-making is made based on the abnormal distribution pattern. The abnormal distribution pattern generation process is as follows: the trajectory is divided into several continuous segments at a fixed distance along the transportation path. The number of abnormal points in each segment is counted, the start and end times of the abnormal points are counted, the average interval time between abnormal points in the segment is calculated, and the density level is divided into three levels: high, medium, and low. Finally, a pattern set consisting of spatial segments, time ranges, and density levels is formed.Based on this, a decision tree algorithm is used to classify the abnormal subset. The root node of the decision tree is judged by whether the highest association strength is not less than 70 and the density level of the segment is high. If the conditions are met, it is determined to be related to the specific transportation scenario. If not, it proceeds to the next level node for judgment. The second level node is judged by whether the time overlap ratio is not less than 50 and the consistency score is not less than 2. If the conditions are met, it is determined to be related to the specific transportation scenario; otherwise, it is determined to be unrelated to the specific transportation scenario. The maximum depth of the decision tree is fixed at 3 levels, the minimum number of leaf node samples is fixed at 20, and the splitting stops when the classification accuracy improvement is less than 1% after further splitting. After the decision is completed, the abnormal classification results are generated, and the corresponding location information and timeliness indicators are extracted to generate anomaly classification results. The spatiotemporal distribution characteristics of normal signals are determined by: First, the earliest and latest times of outliers in the identified relevant scenarios are used as the time range. The proportion of outliers within this range to all sampled points is used as the time density. Second, the minimum and maximum longitudes, and minimum and maximum latitudes covered by the scenario are calculated as the spatial range. The proportion of outliers within this range to all sampled points is used as the spatial density. Simultaneously, the transportation status corresponding to this scenario is recorded as either en route or stopped. Based on these spatiotemporal distribution characteristics, the specific scenario where the anomaly occurred is determined. The specific determination rule is: if the correlation strength of a certain scenario is the highest among all scenarios and not less than 70, and the time density and spatial density of this scenario are not less than 30, then this scenario is considered a relevant scenario. Once the specific scenario of the anomaly is identified, otherwise, return to the previous process to expand the window or extend the observation time for recalculation. Finally, update the sensor network's sampling frequency based on the specific scenario of the anomaly. The sampling frequency update method is a tiered adjustment: if the identified specific scenario is a high-risk scenario, set the sampling period to 20 seconds; if it is a medium-risk scenario, set the sampling period to 60 seconds; if it is a low-risk scenario, set the sampling period to 300 seconds. The rules for determining high, medium, and low risk are as follows: when the correlation strength is not less than 70 and the decision tree output is related to a specific transportation scenario, it is judged as high-risk; when the correlation strength is between 40 and 70 or the decision tree only meets the correlation condition at the second layer, it is judged as medium-risk; when the correlation strength is less than 40 or the decision tree output is not related to a specific transportation scenario, it is judged as low-risk. When the risk is unrelated to a specific transportation scenario, it is considered low-risk. The aforementioned thresholds of 70 and 40 were determined through optimization using historical data. Specifically, the thresholds were tested step-by-step in increments of 5 from 10 to 90 on historical labeled data, and the value that minimized the combined false alarm and false negative rates was selected as the fixed threshold. After the frequency update, the new sampling period was written into the sensor network configuration and took effect immediately. Simultaneously, the anomaly detection rate and data transmission load were continuously evaluated over the next hour. If the anomaly detection rate was below 95%, the sampling period for the corresponding risk level was shortened by 20%. If the data transmission load exceeded the predetermined upper limit, it was extended by 20%. This process was repeated a maximum of three times within a 24-hour rolling window to ensure that monitoring effectiveness and resource consumption remained within a stable range.

[0028] S2 includes collecting real-time data on transport vehicles and goods through a sensor network, including location coordinates, temperature values, humidity values, and timeliness indicators, to generate an initial abnormal signal dataset; using a K-means clustering algorithm to group the initial abnormal signal dataset to obtain abnormal type grouping results; extracting a humidity abnormal subset from the abnormal type grouping results, analyzing the time series changes of humidity values, and obtaining humidity abnormal fluctuation patterns; for humidity abnormal fluctuation patterns, if the fluctuation amplitude of humidity values ​​exceeds a preset threshold, combining the temperature value with the temperature value to calculate the joint distribution characteristics of the two, and determining the temperature and humidity joint abnormality type; based on the temperature and humidity joint abnormality type, obtaining the corresponding location coordinates and timeliness indicators, and generating the spatiotemporal distribution pattern of the abnormal signals.

[0029] In this embodiment, the sensor network first collects and records the time, location coordinates, temperature value, humidity value, and timeliness index of each data point at a fixed sampling period, generating an initial abnormal signal dataset. To ensure data availability, the data cleaning and standardization process is performed sequentially as follows: Step 1: Sort all records in ascending order of time; Step 2: Delete records with missing time, location coordinates, temperature values, or humidity values; Step 3: For duplicate records at the same time, retain only the record with the time closest to a whole second; Step 4: Delete records with temperature and humidity values ​​that exceed the nominal measurable range of the equipment; Step 5: Standardize the dimensions of the temperature and humidity values ​​and unify the location coordinates to decimal longitude and latitude to obtain standard numbers for calculation. The dataset is then processed using the mean clustering algorithm to group outlier types. The specific calculation process is as follows: Grouping features are constructed for each record. These features are fixed as temperature, humidity, the absolute value of the difference between adjacent temperature samples, the absolute value of the difference between adjacent humidity samples, timeliness indicators, and the distance quantization value formed by the sum of the absolute values ​​of the longitude and latitude differences between adjacent records converted to meters. Each feature is zero-centered and scaled by unit dimension to eliminate the influence of dimensions. The candidate set for the number of clusters is set from 2 to 6, and the sum of intra-cluster distances is calculated for each cluster. If the decrease in the sum of intra-cluster distances from one candidate to the next is less than 10, then the previous candidate is taken as the final number of clusters. This parameter controls the balance between grouping accuracy and computational cost. To avoid instability caused by randomness at the starting point... For each candidate cluster, 10 random initializations with different starting points are performed, and the one with the smallest sum of intra-cluster distances is retained as the final starting point. During the iteration phase, a "distribution-update" loop is used. Each time, the record is assigned to the cluster with the smallest distance, and the cluster center is updated to the arithmetic mean of the cluster's features. The iteration limit is set to 100 times. If the change in the sum of intra-cluster distances between two iterations is less than 1, convergence is considered achieved and the process stops. The final output is the anomaly type grouping result. The specific method for extracting the humidity anomaly subset from the anomaly type grouping result is as follows: For each cluster, the concentration and fluctuation of the humidity values ​​within the cluster are calculated. The concentration is represented by the absolute value of the difference between the mean and median humidity values ​​within the cluster, and the fluctuation is represented by the distribution width of the humidity within the cluster and the difference between adjacent samples. The average absolute value is used to represent the fluctuation level of all clusters. The fluctuation level of all clusters is sorted from smallest to largest, and clusters ranked at or above the 80th percentile are selected as high fluctuation candidates. Then, clusters with a concentration level lower than the 50th percentile of all clusters are identified as humidity anomaly clusters. Records from all humidity anomaly clusters are merged to form a humidity anomaly subset. Time series analysis is performed on the humidity anomaly subset to obtain the humidity anomaly fluctuation pattern. The calculation process is as follows: humidity fluctuation is calculated segment by segment using a sliding time window. The window length is determined by sorting the duration of confirmed humidity anomaly events in the historical normal transportation phase and taking the value ranked in the first 1 / 4 of all data as the window length. At the same time, a lower limit of 20 seconds and an upper limit of 300 seconds are set for the window length to avoid misjudgment due to being too short or too long.Within each window, three humidity fluctuation characteristics are calculated: the first is the short-term fluctuation amplitude, defined as the difference between the maximum and minimum humidity values ​​within the window; the second is the rate of change, defined as the average of the absolute values ​​of the humidity differences between two adjacent samples within the window; and the third is the duration, defined as the total time within the window that continuously exceeds the stable range. The upper and lower limits of the stable range are determined by taking the 10th percentile as the lower limit and the 90th percentile as the upper limit from the humidity value distribution during historical normal transportation periods. The humidity fluctuation threshold is set based on the statistical distribution of the above three characteristics during historical normal transportation periods, using the 95th percentile as the upper limit threshold for each characteristic and the 5th percentile as the lower limit threshold for each characteristic. If any feature of a window exceeds the upper threshold or falls below the lower threshold, the humidity value fluctuation is considered to exceed the preset threshold, and the window and its records are marked as out-of-bounds windows. When an out-of-bounds window occurs, the calculation of the joint distribution features of temperature and humidity is initiated to determine the joint anomaly type. Specifically, the records within the out-of-bounds window are divided into a fixed number of bins in both the temperature and humidity dimensions using an equal-width method. The number of bins is determined to be 20 based on the sensor's effective resolution, as determined through playback evaluation before deployment, to ensure a balance between distribution characterization accuracy and computational cost. The number of records in each temperature and humidity bin combination is counted and divided by the total number of records in the window to obtain the occurrence ratio of that combination. Subsequently... During normal transportation phases, the same container settings are used to statistically analyze the occurrence ratio of each combination as a reference distribution. If the occurrence ratio of any combination within the out-of-bounds window is lower than the 5th percentile of the reference distribution for that combination, and the temperature and humidity values ​​within that window are simultaneously outside their respective stable ranges, then the window is determined to have a combined temperature and humidity anomaly. This anomaly is then categorized into one of four types: high temperature / high humidity, low temperature / high humidity, high temperature / low humidity, or low temperature / low humidity, for subsequent interpretation and handling. After confirming the combined temperature and humidity anomaly, a spatiotemporal distribution pattern of the anomaly signal is generated based on the records corresponding to the determined combined anomaly type. The calculation process is as follows: For each trajectory in the time dimension... The earliest and latest joint anomaly times are taken as the time range, and the proportion of joint anomaly records to all records within this range is calculated as the time density. At the same time, the distribution of joint anomalies at the hourly scale and during key intraday periods is statistically analyzed. Spatially, the transportation path is divided into continuous segments at fixed distances, and the number of joint anomaly records in each segment and the proportion of that number to all records in the segment are calculated as the spatial density. At the same time, the minimum longitude, maximum longitude, minimum latitude, and maximum latitude of the joint anomaly records are extracted as the spatial range. The time density and spatial density are comprehensively labeled to output a spatiotemporal distribution pattern that includes type, time range, time density, spatial range, and spatial density.All parameters and thresholds in the above process have been given clear meanings and determination methods: the number of clusters comes from a candidate set of 2 to 6 and is selected according to the rule of the decreasing ratio of the sum of intra-cluster distances; the number of random initializations is fixed at 10 to improve stability; the iteration upper limit is 100 to limit computational latency; the convergence threshold is that the change ratio of the sum of intra-cluster distances is less than 1 to ensure stable results; the screening threshold for humidity anomaly subsets is that the fluctuation degree is not lower than the 80th percentile and the concentration degree is lower than the 50th percentile to ensure that only clusters with strong fluctuations and unstable centers are selected; the sliding window length comes from the first 1 / 4 of the historical anomaly duration and is set with a boundary of 20 to 300 seconds to balance sensitivity and robustness; the stable interval uses the 10th and 90th percentiles of the historical normal distribution to cover common fluctuation ranges; the humidity fluctuation feature threshold uses the 5th and 95th percentiles of the historical normal distribution to limit false alarms; the joint distribution feature is divided into fixed 20 bins to ensure stable repeatability; the reference low probability threshold uses the 5th percentile of each combined reference distribution to highlight the abnormal area.

[0030] S3 includes acquiring real-time temperature and humidity values ​​from sensor data, matching them with temperature and humidity threshold ranges in the cargo type database using a threshold comparison method to determine if a joint temperature and humidity anomaly exists, and obtaining an anomaly confirmation result. If the anomaly confirmation result is a joint temperature and humidity anomaly, the sensitive parameters of the corresponding cargo are extracted from the cargo type database, and dynamic triggering conditions are generated by combining them with real-time data updates, resulting in a triggering condition set. Based on the triggering condition set, a route planning algorithm is used to calculate transportation route adjustment instructions, generating a route instruction set containing time and space constraints. The navigation data of the transport vehicles is updated through the route instruction set, and the sensor data acquisition frequency is adjusted using a real-time data update mechanism to obtain an optimized acquisition strategy. Real-time data streams are extracted from the optimized acquisition strategy, and the joint temperature and humidity anomaly is continuously monitored using an anomaly type judgment method to obtain an anomaly status update result. If the anomaly status update result exceeds the dynamic triggering conditions, an alternative transportation plan is obtained from the cargo type database, generating an adjusted route instruction set. The operating status of the transport vehicles is updated through the adjusted route instruction set, and data fusion technology is used to integrate the sensor data acquisition and route planning results to obtain the final transportation optimization scheme.

[0031] In this embodiment, real-time temperature and humidity values ​​are first acquired from sensor data. Sensitive parameters corresponding to the cargo type are then read from the cargo type database, including the lower and upper limits of temperature and humidity thresholds, the maximum permissible duration of exceeding limits, the maximum permissible rate of change, the target recovery time, the maximum acceptable delay for route adjustment, and the sampling period level boundary. The sensitive parameters are determined by statistically analyzing the stable distribution of temperature and humidity in historical normal transportation data in chronological order. The 5th percentile is used as the lower limit, the 95th percentile as the upper limit, and one-tenth of the historical normal deviation is added to each limit as a safety margin to obtain four thresholds. The values ​​per unit time during the historical normal period are then used as the thresholds. The distribution of temperature and humidity changes is calculated by taking the 95th percentile and adding the maximum historical measurement error to obtain the maximum allowable rate of change. The median time from triggering to the regression threshold for historically successfully handled samples is determined as the target recovery time. The shipper's timeliness agreement requirements and contractual stipulations are determined as the maximum acceptable delay for route adjustments. The sampling period level boundaries obtained from historical capture rates and load assessments are fixed as 20 seconds for high risk, 60 seconds for medium risk, and 300 seconds for low risk. When the system executes the threshold comparison method, the anomaly confirmation result for each real-time record is determined according to the following rules: when both the real-time temperature and humidity values ​​are within the corresponding threshold range and the rate of change is lower than the maximum allowable rate of change, no output is generated. In case of anomalies, an alert is issued when any parameter exceeds the limit and the duration does not reach the maximum permissible over-limit duration; a combined temperature and humidity anomaly is issued when any parameter exceeds the limit and the duration reaches or exceeds the maximum permissible over-limit duration. When the anomaly confirmation result is a combined temperature and humidity anomaly, sensitive parameters of this batch of goods are extracted from the cargo type database and used with the current real-time data to generate a dynamic trigger condition set. The condition set contains four types of definite conditions that must be met simultaneously to determine triggering: one type is the over-limit amplitude condition, that is, the deviation of the real-time temperature value from the upper or lower limit of the temperature threshold is not less than the permissible temperature deviation value, or the deviation of the real-time humidity value from the upper or lower limit of the humidity threshold is not less than the permissible humidity deviation value, and the permissible temperature deviation value and permissible humidity deviation value are both considered. The first category is a fixed number that minimizes the combined false alarm rate and false negative rate through historical playback tests before deployment. The second category is a duration condition, that is, the cumulative duration of the out-of-limit state is not less than the longest allowable out-of-limit duration. The third category is a change rate condition, that is, the temperature change rate or humidity change rate within the most recent sliding time window is not less than the maximum allowable change rate. The length of the sliding time window is determined by taking the quartiles of the time distribution from the start of the treatment of historical successful treatment samples to the regression threshold and limiting it to 20 to 300 seconds. The fourth category is a treatment feasibility and timeliness constraint condition, that is, the temperature and humidity must be able to return to the threshold range within the target recovery time and the expected arrival time should not be delayed by more than the maximum acceptable delay of the path adjustment relative to the original plan.After the trigger condition set is generated, the route planning algorithm is entered to calculate the transportation route adjustment instruction. The route calculation does not introduce external modules but adopts a deterministic candidate and evaluation process: the system generates a set of candidate routes on the current planned route according to the stations passed through and the alternative road segments. For each candidate route, the estimated arrival time, the over-limit exposure time, and the recovery accessibility are calculated in the order of road segments. The estimated arrival time is obtained by summing the travel time of the road segment length and the speed limit of the road segment, plus the loading, unloading, or waiting time. The over-limit exposure time is the current duration of the over-limit exposure plus the estimated time required to complete the handling during the execution of the candidate route. The recovery accessibility is whether the over-limit exposure time can be completed before the end of the candidate route. The target recovery time is determined by bringing both temperature and humidity back to their respective threshold ranges. Then, unqualified candidates are eliminated based on the hard screening criteria of meeting recovery accessibility requirements and ensuring the estimated delay does not exceed the maximum acceptable delay for route adjustments. The optimal route is then selected in order of minimum exposure time followed by estimated arrival time, forming a route instruction set that includes both time and spatial constraints. The time constraints specify the latest arrival time and the upper limit of the handling completion time, while the spatial constraints specify the sequence of road segments that must be traversed and the sequence of road segments that must be avoided. After the system writes the route instruction set into the vehicle navigation data, it immediately activates a real-time data update mechanism to adjust sensors in conjunction with the risk level. The sampling frequency and risk level are determined as follows: High risk is defined as a 20-second sampling period when all three conditions (amplitude exceeding limit, duration, and rate of change) are met simultaneously; medium risk is defined as a 60-second sampling period when only the amplitude and duration conditions are met; and low risk is defined as a 300-second sampling period in other cases. Closed-loop calibration is performed within a one-hour evaluation window after navigation updates, based on the anomaly capture rate and communication load. When the anomaly capture rate is below 95%, the sampling period for the corresponding risk level is shortened by 20%; when the communication load exceeds a predetermined upper limit, the sampling period for the corresponding risk level is extended by 20%. The system can be adjusted a maximum of 3 times within a 4-hour rolling window. The system continuously extracts real-time data streams from the optimized acquisition strategy and performs continuous monitoring to determine the anomaly type. Within a fixed-length sliding window, the system recalculates the over-limit amplitude, duration, and rate of change and compares them item by item with the dynamic trigger condition set to output the anomaly status update result. The update result is one of three values: mitigation, stabilization, or deterioration. Mitigation means that the three indicators do not increase compared to the previous window and at least one returns to the threshold range. Stabilization means that the changes in the three indicators do not exceed the set micro-fluctuation threshold, which is determined by the 10th percentile of the historical normal fluctuation distribution. Deterioration means that any one of them increases compared to the previous window but is still outside the threshold.When the abnormal state update result indicates deterioration and any triggering condition is met or exceeded again, the system retrieves the backup transportation plan bound to the cargo from the cargo type database and generates an adjusted route instruction set. The backup transportation plan includes a pre-approved sequence of alternative routes and operational adjustment entries. The alternative route sequence is generated by performing calculations of estimated arrival time, over-limit exposure time, and recovery accessibility, and selecting the optimal value according to the aforementioned rules. The operational adjustment entries are deterministic modifications to the route execution order and waiting time to shorten the exposure time. Finally, the system uses data fusion technology to deterministically merge the sensor data acquisition results with the current route planning results, with the fusion priority fixed at a safety threshold. Prioritizing the following criteria: safety threshold, target recovery time, timeliness, and shortest path length. Specifically, when a safety threshold or target recovery time cannot be met, a forced switch to a verified candidate path that satisfies both is implemented, and the risk level is increased by one level to simultaneously increase the sampling frequency. When both safety and recovery are met, but timeliness is not, a candidate with a shorter arrival time is selected, provided it does not exceed the maximum acceptable delay for path adjustment. When all four criteria are met, the current path and sampling period remain unchanged. The final transportation optimization scheme output includes the latest path instruction set, latest arrival time, upper limit for handling completion time, sequence of necessary and avoided road segments, current sampling period, risk level, and abnormal status.

[0032] S4 includes acquiring temperature and humidity values ​​from real-time data collected by sensors, using a support vector machine classifier to analyze the deviation between the real-time data and a preset threshold range, and determining the confirmation result of joint temperature and humidity anomalies. If the confirmation result is anomaly, the startup logic of the backup plan is triggered based on the degree of deviation, generating an anomaly confirmation output containing sequence execution logic. Based on the anomaly confirmation output, the corresponding backup transportation plan is retrieved from the preset backup plan database, generating an adjusted transportation instruction set. Using the adjusted transportation instruction set, the navigation data of the transportation vehicles is updated using a path planning algorithm, generating a path instruction set containing time and space constraints. Real-time navigation data is extracted from the path instruction set, and data fusion technology is used to integrate the sensor-collected data and navigation data to generate an optimized transportation data stream. Based on the optimized transportation data stream, anomaly detection algorithms are used to continuously monitor joint temperature and humidity anomalies, determining the anomaly status update result. If the anomaly status update result exceeds a preset threshold range, the backup transportation plan is updated through sequence execution logic, generating the final optimized transportation instruction set.

[0033] In this embodiment, real-time temperature and humidity values ​​are first collected from sensors. These values ​​are then compared against preset threshold ranges established during system initialization. These preset threshold ranges are determined before implementation based on historical normal transportation data and quality standards for similar goods. The lower and upper limits of the temperature threshold are taken as the 5th and 95th percentiles of historical normal temperature distributions, respectively, with a safety margin added. Similarly, the lower and upper limits of the humidity threshold are taken as the 5th and 95th percentiles of historical normal humidity distributions, respectively, with a safety margin added. This safety margin is one-tenth of the median of historical normal deviations. The value should be at least twice the sensor resolution. At each sampling time, two deviations are calculated: the absolute difference between the real-time temperature value and the nearest temperature threshold boundary, and the absolute difference between the real-time humidity value and the nearest humidity threshold boundary. Simultaneously, the temperature change rate and humidity change rate are obtained by dividing the difference between adjacent samples by the sampling interval. The duration of the out-of-limit state is continuously accumulated on the time axis to obtain the out-of-limit duration. The temperature deviation, humidity deviation, temperature change rate, humidity change rate, out-of-limit duration, and whether the value is within the threshold are combined as features and input into a support vector machine classifier for anomaly detection. The support vector machine classifier uses a radial basis function kernel as the kernel type. The penalty coefficient and kernel width are selected from historical labeled samples through grid traversal and cross-validation to minimize the combined index of false positive and false negative rates. The classification output is either normal or a joint anomaly of temperature and humidity, along with a confidence score. The system sets the confidence threshold to 0.8. An anomaly is confirmed when the output confidence score is not lower than 0.8; otherwise, it is not confirmed. Once an anomaly is confirmed, the system determines whether to trigger the backup scheme activation logic based on the degree of deviation. The degree of deviation is quantified by three categories: the first category is the temperature deviation being wider than the temperature threshold range. The first category is the ratio of the degree or humidity deviation to the relative width of the humidity threshold range, whichever is larger. The second category is the ratio of the duration of exceeding the limit to the longest permissible duration of exceeding the limit. The third category is the ratio of the rate of temperature change or the rate of humidity change to the maximum permissible rate of change, whichever is larger. The longest permissible duration of exceeding the limit and the maximum permissible rate of change are determined as fixed values ​​during the initialization phase by adding a safety margin to the 95th percentile of the historical normal phase time distribution and the change distribution. The triggering rule is triggered when any two of the three categories have a ratio of not less than 1 and the support vector machine confidence score is not less than 0.9, or when any one category has a ratio of not less than 1.The system is triggered immediately at 2 o'clock. Upon triggering, it generates an anomaly confirmation output containing sequence execution logic. The fixed order of the sequence execution logic is: path evaluation first, sampling linkage second, and continuous monitoring follow-up. Subsequently, the system retrieves alternative transportation plans from the preset backup plan database using cargo type, anomaly type, and current location as search keys. If multiple candidates exist, it calculates three indicators for each plan: estimated arrival time, over-limit exposure time, and threshold recovery accessibility. The estimated arrival time is the sum of the travel time and loading / unloading waiting time for each segment in the plan. The over-limit exposure time is the current accumulated over-limit duration plus the estimated time required to complete the disposal during the execution of this plan. Threshold recovery accessibility is the achievable judgment of whether temperature and humidity can be simultaneously restored to their respective threshold ranges within the target recovery time. The target recovery time is determined in the initialization phase by the median disposal time of historical successful disposal samples. Threshold recovery accessibility is a mandatory condition that must be met, and the estimated arrival time must not exceed the maximum acceptable path adjustment time. Delay is the second hard condition for filtering out unqualified schemes. The maximum acceptable delay for route adjustment is given as a fixed value before going online according to the shipping agreement and planned timeliness. Among the remaining schemes, they are sorted and selected according to the minimum exposure time of exceeding the limit and the estimated arrival time, and an adjusted transportation instruction set is generated. The system inputs the adjusted transportation instruction set into the route planning algorithm to update the vehicle navigation data and outputs a route instruction set containing time constraints and spatial constraints. The time constraints specify the latest arrival time and the upper limit of the handling completion time. The spatial constraints specify the sequence of road segments that must be passed and must be avoided. Subsequently, the system extracts real-time navigation data from the route instruction set and performs data fusion with the sensor-collected data to generate an optimized transportation data stream. The data fusion adopts a weighted fusion method. The weights are determined as a set of fixed values ​​through historical playback before going online. The default sensor data weight is 0.7 and the navigation prediction weight is 0.3. When detours or speed limit adjustments occur, the navigation prediction weight is increased to 0 in the next evaluation window.5. To enhance responsiveness to path changes, an evaluation window length of 10 minutes is used. Based on the optimized transportation data stream, the system executes an anomaly detection algorithm for continuous monitoring. Monitoring employs a fixed-length sliding time window, the length of which is determined before going live by the quartiles of the historical successful handling sample trigger-to-regression time distribution and limited to a range of 20 to 300 seconds. Within each window, temperature deviation, humidity deviation, the rates of change for both parameters, and the duration of exceeding limits are recalculated and compared with their respective upper limit thresholds. The four upper limit thresholds are the upper limit for temperature deviation, humidity deviation, rate of change, and duration. These upper limit thresholds are fixed during the initialization phase by adding a safety margin to the 95th percentile of the historical normal phase distribution. After calculating the direction of indicator change in the current window compared to the previous window, the anomaly status update result is output as one of three: mitigation, stabilization, or deterioration. Mitigation means that all four indicators do not rise and at least one returns to within the threshold. Stabilization means that the changes in all four indicators fall within a small fluctuation range, which is fixed by the 10th percentile of the historical normal fluctuation distribution. Deterioration means that any indicator rises but remains within the threshold. In addition, when the abnormal status update result exceeds the preset threshold range, the sequence execution logic is triggered to update the backup transportation plan. The judgment condition for exceeding the threshold is that at least two out of three consecutive windows are deteriorating and any indicator in the current window exceeds 1.2 times the corresponding upper limit threshold or the support vector machine confidence is not less than 0.9. Under this condition, the system re-retrieves a higher priority plan from the backup plan database and restores the maximum acceptable delay constraints for reachability and path adjustment with the same threshold to generate the final transportation optimization instruction set. At the same time, the sampling period is linked to the risk level and adjusted to 20 seconds for high risk, 60 seconds for medium risk, and 300 seconds for low risk. The risk level judgment rule is that if all three categories have a ratio of not less than 1, it is high risk; if any two categories have a ratio of not less than 1, it is medium risk; and others are low risk. Within the 1-hour evaluation window after the navigation update, closed-loop calibration is performed based on the anomaly capture rate and communication load. When the anomaly capture rate is lower than 95%, the sampling period for the corresponding risk level is shortened by 20; when the load exceeds the upper limit, the sampling period is extended by 20. The system adjusts a maximum of 3 times within the 24-hour rolling window to avoid frequent oscillations.

[0034] S5 includes: obtaining anomaly type and deviation data from anomaly confirmation results; querying a preset response rule base; matching the corresponding automated process script; generating an initial response execution sequence; extracting transportation control instructions based on the initial response execution sequence; combining terrain constraint data; using the A-path planning algorithm to generate an optimized path instruction set; obtaining navigation data from the path instruction set; using data fusion technology to integrate environmental data and navigation data collected by sensors to generate a real-time transportation data stream; continuously monitoring anomaly states using the isolated forest algorithm based on the real-time transportation data stream and judging the state update results; if the state update results exceed a preset threshold, obtaining a backup automated process script from the response rule base and updating the response execution sequence; adjusting transportation control instructions based on the updated response execution sequence to generate a final optimized instruction set; extracting path data from the final optimized instruction set; using data verification technology to confirm the integrity of the instruction set and obtain executable transportation instructions.

[0035] In this implementation, the anomaly type and deviation data are first read from the anomaly confirmation results. The anomaly type is limited to three categories: single temperature anomaly, single humidity anomaly, and combined temperature and humidity anomaly. The deviation data includes five items: temperature deviation, humidity deviation, temperature change rate, humidity change rate, and duration of exceeding the limit. Temperature deviation and humidity deviation are the absolute differences between the real-time values ​​and the nearest boundary of the corresponding threshold, respectively. Temperature change rate and humidity change rate are the rates of change obtained by dividing the difference between adjacent samples by the sampling interval, respectively. The duration of exceeding the limit is the cumulative time spent continuously outside the threshold. These five values ​​are calculated in real-time by the preceding process and carried with the record. The system uses anomalies as the basis for classification. The type and deviation level are used as search keys to access the response rule base and generate the initial response execution sequence. The deviation level is divided into three levels: low, medium, and high. The segmentation thresholds for each of the five deviation indicators are determined based on the joint distribution of historical normal data and confirmed abnormal data. Specifically, the 50th and 85th percentiles of each indicator in the historical normal data are calculated. The low level is defined as the 50th percentile, the medium level as the 85th percentile, and the intervals exceeding the 85th percentile are classified as high level. To avoid misclassification due to sampling noise, a safety margin equal to twice the sensor resolution is added to each threshold before it goes live and stored as a fixed value in the rule base. Matching is performed in a strict priority order: first, anomaly type; then, deviation level; and finally, terrain level. Terrain level is provided by terrain restriction data and is fixed at three levels—smooth, moderate, and complex—after aggregation based on slope, altitude change, road surface type, and prohibition rules. Simultaneously, the action sequence and parameter boundaries in the rule entries are read, and the initial response execution sequence is output. If multiple rules are matched, the one with the highest priority value is selected; if they are still tied, the one with the smallest script number is selected. Subsequently, transportation control instructions are extracted based on the initial response execution sequence, and the optimized path instruction set is generated by calling the path planning algorithm in conjunction with the terrain restriction data. Path calculation does not introduce external models from the rule base. The system generates a candidate set on the current planned route. The candidate generation constraint is that it can only replace the route on road segments with a terrain level no higher than complex and that do not trigger the prohibition rule. The system calculates three types of deterministic costs for each candidate route: mileage cost is the sum of the lengths of all road segments, time cost is the sum of the travel time of all road segments under speed limit conditions and the expected loading and unloading waiting time, and terrain penalty is the sum of the slope level, altitude change range and road surface type of each road segment according to fixed weights and then superimposed with the infinite penalty of prohibition conflict. The fixed weights are determined by historical playback test before going online. The goal is to minimize the time delay rate and the abnormal exposure time at the same time and use the weight combination with the minimum comprehensive cost as the fixed value.Route selection is based on two hard constraints: time constraints and spatial constraints. The time constraints are the latest arrival time and the maximum completion time for handling the issue. The spatial constraints are the sequence of road segments that must be traversed and those that must be avoided. Candidates meeting the hard constraints are sorted by their overall cost from smallest to largest; if there is a tie, the one with the smaller terrain penalty is prioritized. The optimized route instruction set is output and written into the vehicle navigation data. Next, the system extracts navigation data from the route instruction set and fuses it with environmental data collected by sensors to generate a real-time transportation data stream. The fusion uses a weighted, deterministic method, with a default number of sensors. The navigation prediction weight is 0.7, and the navigation prediction weight is 0.3. These two weights are fixed through historical playback optimization. When a detour or speed limit adjustment occurs, the navigation prediction weight is temporarily increased to 0.5 in the subsequent evaluation window to enhance the response to route changes. The evaluation window length is fixed at 10 minutes and is determined after verification by the median and upper quantile interval of the completion time distribution of historical route switching events. The real-time transportation data stream includes fields such as timestamp, location coordinates, temperature value, humidity value, mileage progress, difference between planned arrival time and estimated arrival time, terrain level, and rule hit number. If any key field is missing, it is discarded. This record ensures complete input for subsequent algorithms. The system employs an isolated forest algorithm to continuously monitor abnormal states on the real-time transportation data stream. Model parameters are optimized using historical normal sample replay before deployment. The number of trees is fixed at 100, the subsample size at 256, and the splitting stopping depth at 8. The anomaly scoring threshold is the 95th percentile of the historical normal sample score distribution, plus a safety margin obtained from sensor noise replay. This safety margin is one-quarter of the difference between the upper bound of the short-term fluctuation range of historical normal scores and this percentile. Online monitoring is performed using a fixed-length sliding time window. The quartiles of the time distribution from triggering to regression of historical successful treatment samples are limited to the range of 20 to 300 seconds. Within each window, the anomaly score corresponding to the standardized path length is calculated for each record. The maximum and average values ​​of the window are then used as the comprehensive score for that window and compared with the threshold. If the comprehensive score is lower than the threshold and decreases compared to the previous window, it is judged as mitigation. If the comprehensive score is near the threshold and the fluctuation falls within the 10th to 90th percentile of the short-term fluctuation range of the historical normal score, it is judged as stagnant. If the comprehensive score exceeds the threshold, it is judged as deterioration. The window judgment result is output as the status update result.When the status update result exceeds the preset threshold, the system retrieves the backup automated process script from the response rule base according to the sequence execution logic and updates the response execution sequence. The conditions for exceeding the threshold are: at least two out of three consecutive windows are judged as deteriorating and the current window's overall score exceeds 1.2 times the threshold; or both mileage progress lag and estimated arrival time difference being positive occur simultaneously. The mileage progress lag threshold is taken as the 10th percentile of the historical normal travel progress distribution. A positive estimated arrival time difference indicates that the plan has been exceeded. The backup script has the same fields as the main script. The update strategy is to retain executed actions, replace unexecuted actions with backup actions, and tighten the time constraint to 0.8 times the original value to accelerate processing. The system adjusts the transportation control instructions based on the updated response execution sequence and generates the final optimized instruction set. Changes to control instructions are limited to three categories: route switching, detour, and stop order adjustment. Each change requires re-verification. The system checks for spatial and temporal constraints, ensuring the overall cost does not exceed 1.1 times the original solution. If this is not met, the system reverts to the second-best candidate. Finally, the system extracts path data from the final optimized instruction set and performs data verification to confirm the integrity of the instruction set. Verification includes four types of fixed checks: field integrity check requires each instruction to include five fields: origin number, destination number, road segment sequence, latest arrival time, and upper limit of disposal completion time; logical consistency check requires the road segment sequence to be continuous without breaks or loops; range validity check requires all road segment terrain levels to be no higher than the terrain upper limit determined in the optimization stage and not to violate prohibition rules; conflict detection requires all time windows to not overlap and not to conflict with the buffer of the planned timeliness. The buffer size is taken as the 95th percentile of the historical delay distribution. If all four checks pass, an executable transportation instruction is generated. If any check fails, the corresponding instruction is located and the system returns to the path optimization step for recalculation.

[0036] S6 includes extracting route adjustment instructions from the response execution sequence, using data parsing technology to generate a structured instruction dataset; using the structured instruction dataset, combined with backup plan data, and employing a decision tree algorithm to determine the applicability of the route adjustment instructions, resulting in a filtered instruction set; if the filtered instruction set meets a preset threshold, then through a data integration process, generating an adjusted transportation plan, resulting in plan data containing route adjustments; extracting time deviation data from the plan data containing route adjustments, using a time difference calculation method to generate time deviation records, resulting in a transportation plan containing time differences; using the transportation plan containing time differences, employing an instruction allocation mechanism to send route adjustment instructions to the transportation control system, resulting in a post-execution plan update status; obtaining data from the time record module from the post-execution plan update status, using data verification technology to determine the completeness of the plan update status, resulting in the final transportation plan; and using the final transportation plan to generate a real-time monitoring data stream, using a random forest algorithm to determine the execution deviation of the transportation plan, resulting in an optimized transportation instruction set.

[0037] In this implementation, route adjustment instructions are first extracted one by one from the response execution sequence. Data parsing converts the semi-structured text into a structured instruction dataset. Fixed fields include instruction number, start point location, end point location, road segment list, planned departure time, planned arrival time, latest allowed arrival time, priority, route adjustment action, prerequisite dependency number, prohibited area identifier, spare area identifier, and safety buffer duration. The parsing rule is to first separate key-value pairs by colons, then separate them by commas to obtain the field list. Time is uniformly converted to seconds since the task started, and location is uniformly set to decimal longitude and latitude. The road segment list is expanded into a road segment number sequence. If any key field is missing, the instruction is marked as unavailable and... In subsequent judgments, these are eliminated; then, using structured instructions as the main table and backup plan data as the reference table, they are linked one by one to form sample records for decision tree judgment. Features used for splitting are calculated for each record, and their sources and calculation methods are explained: Expected delay is the sum of the candidate segment's travel time and loading / unloading waiting time minus the sum of the originally planned times for the same period. Travel time is calculated from the segment length and speed limit, while waiting time is determined as a fixed value based on the historical average loading / unloading time; No-traffic conflict count is the number of candidate segments that trigger no-traffic rules, which are derived from the route control list; Area access window matching degree is the statistical proportion of each area traversed by the candidate segment that is in a permitted passage state during the current time period. The area window is determined by the historical release time and the current... The preceding list is determined; the prerequisite dependency satisfaction is determined by whether all prerequisite dependencies of this instruction are arranged before it on the timeline and whether their actual arrival is no later than the planned departure of this instruction. If satisfied, it is marked as available; otherwise, it is marked as unavailable. The road segment continuity is determined by whether the road segment list is connected end to end and without any breakpoints or loops. If any breakpoint or loop exists, it is marked as unavailable. The planned arrival over-limit risk is determined by whether the expected arrival time exceeds the latest allowed arrival time. If it exceeds, it is marked as yes; if not, it is marked as no. The safety buffer sufficiency is determined by whether the safety buffer duration is greater than the expected delay. If it is greater, it is marked as sufficient; if it is less than or equal to, it is marked as insufficient. The decision tree uses the splitting threshold and splitting order frozen during the training phase. The threshold and order are optimized based on the playback of labeled historical samples before going online. The first layer is set as a fixed value. The root node splits based on the expected delay threshold. The threshold is determined by iterating through the original planned duration in increments of 5% to 30%, calculating the weighted sum of the delay rate and the abnormal exposure duration, and selecting the minimum as the fixed threshold. The second layer splits based on the prohibited conflict count threshold and the area access window matching degree threshold, where the prohibited conflict count threshold is fixed at zero and the area access window matching degree threshold is fixed at 80%. The third layer splits based on two Boolean conditions: the satisfaction of prerequisite dependencies and the continuity of road segments. The leaf nodes use the planned arrival over-limit risk and the adequacy of the safety buffer as the final screening conditions. The execution is carried out in the above fixed order from top to bottom to obtain the available record set, which is the filtered instruction set.The system performs threshold checks on the filtered instruction set to determine whether to generate an adjusted transportation plan. The preset threshold requires that at least 80% of the instructions be qualified, all qualified instructions have no risk of exceeding the planned arrival limit, and the safety buffer is sufficient. The qualified instruction percentage threshold is fixed by the 90th percentile lower bound of the historical plan achievement rate distribution. When the preset threshold is met, the system enters the data integration process to generate plan data including route adjustments. The integration rule is to arrange instructions in order of priority from high to low, based on time, and satisfying prerequisite dependencies. Necessary waiting time is inserted between adjacent instructions to ensure that the actual completion time of the previous instruction is no later than the planned departure time of the next instruction. Repeated actions on the same route segment are handled accordingly. Line merging is performed to eliminate conflicts and reduce the number of handovers. Then, time deviation data is extracted segment by segment from the plan data containing route adjustments, and a time difference calculation method is executed to generate time deviation records. The time difference calculation outputs three types of time differences: starting time difference (the difference between the planned departure and the original planned departure), arrival time difference (the difference between the planned arrival and the original planned arrival), and margin time difference (the difference between the allowed latest arrival and the planned arrival). All are recorded in seconds and accompanied by instruction number, segment number, and recording timestamp, generating a transportation plan containing time differences. Based on the transportation plan containing time differences, the system initiates an instruction allocation mechanism to send route adjustment instructions to the transportation control system. The allocation order is from high to low priority and satisfies pre-dependencies. Within the same priority range... Within the priority list, urgent instructions are prioritized by sorting them by arrival time difference from smallest to largest. After sending, an acknowledgment is required within a fixed acknowledgment time limit (30 seconds). If no acknowledgment is received, the instruction is resent with a maximum of 3 retries at 30-second intervals. If the retry still fails, the instruction is marked as failed and enters a rollback process. The system generates a post-execution plan update status based on the control system's feedback results and reads fields such as actual departure and arrival times from the time recording module. Data verification technology is then used to determine the integrity of the plan update status. Verification includes four fixed checks with provided judgment rules and threshold sources: Field integrity check requires each record to include the instruction number, departure time, arrival time, and departure time. The system requires four checkpoints: route segment, actual departure time, actual arrival time, and execution result. Missing any of these fields results in a failure. For time continuity, the system requires that the actual departure not be earlier than the planned departure and that the actual arrival not be earlier than the actual departure. Violation of this rule results in a failure. For spatial continuity, the system requires that the route segment sequence be continuous without breaks or loops. Finding any breaks or loops results in a failure. For conflict detection, the system requires that no contradictory operations be performed on the same route segment within the same time interval and that there be no conflict with the planned time buffer. The buffer size is a fixed value determined by the 95th percentile of the historical delay distribution. Passing all four checks generates the final transportation plan; otherwise, the problem is recorded and returned to the data integration process for re-arranging the relevant instructions.Real-time monitoring data streams are generated based on the final transportation plan. Fixed fields include timestamp, instruction number, route number, current location, current speed, cumulative start time difference, cumulative arrival time difference, current margin time difference, and execution status flag. The system uses a random forest algorithm on this data stream to determine execution deviations and provides the source and determination method for all parameters and thresholds. The number of random forest trees is fixed at 100, the maximum depth at 8, and the subsample ratio at 50. These three parameters were optimized and frozen before deployment through playback tests with labeled historical samples. The feature set is fixed as cumulative start time difference, cumulative arrival time difference, current margin time difference, speed change amplitude in the previous time period, and failure count in the previous time period. The model output is categorized into three types: normal, slight deviation, and severe deviation. The deviation threshold is fixed and determined by traversing the labeled sample grid. Minor deviations are defined as follows: a cumulative arrival time difference of no less than 5% and less than 15% of the original planned time, or a current margin time difference less than 20% of the original value. Severe deviations are defined as a cumulative arrival time difference of no less than 15% of the original planned time, or a current margin time difference less than 10% of the original value. During the online phase, a fixed window is used to recalculate the judgment, with a fixed window length of 10 minutes, aligned with the aforementioned confirmation and retry cycle for coordinated handling. When a minor deviation is determined, the system adjusts the execution order of priorities based on the final transportation plan to eliminate minor delays. When a severe deviation is determined, the system triggers the pre-positioning of high-priority instructions and shortens their safety buffer time to 80% of the original value, while replacing high-risk sections in the original route with backup area identifiers to generate an optimized transportation instruction set.

[0038] S7 includes: obtaining timestamp data from the initial abnormal signal dataset; using time series parsing technology to generate a structured time series dataset; aligning the time series dataset with the updated transportation plan using a sliding window technique to calculate time deviation data; obtaining a time deviation dataset; if the deviation value in the time deviation dataset meets a preset threshold, matching the time deviation dataset with a preset time consumption standard using data comparison technology to generate a response time consumption index; extracting response time features from the time consumption index dataset; using a linear regression algorithm to analyze the relationship between response time features and system performance, judging the system optimization level, and obtaining an optimization level evaluation result; comparing the optimization level evaluation result with the timestamp data of the initial abnormal signal dataset; using a difference calculation method to quantify the change in response time, and obtaining a quantized response time dataset; extracting deviation trend values ​​from the quantized response time dataset; using benchmark adjustment technology to adjust the triggering conditions of abnormal signals based on the deviation trend values, generating an adjusted abnormal signal benchmark, and obtaining an updated benchmark dataset; and using data verification technology to verify the integrity of the time series data after benchmark adjustment, and obtaining a verified benchmark dataset.

[0039] In this implementation, timestamps are first extracted from the initial abnormal signal dataset, and time series parsing is performed. Specifically, all records are sorted in ascending order by time; records with missing times are deleted; for duplicate records within the same second, only the earliest one is retained; each time is converted to the number of seconds since the task started; and each record is appended with an anomaly type and source marker corresponding to its generation time for subsequent matching, resulting in a structured time series dataset. Subsequently, the time series dataset is aligned with the updated transportation plan. The alignment anchor points are the planned departure time and the planned arrival time—two key time points. During alignment, a sliding window technique is used to calculate the time deviation point by point. The dynamic window length is candidateed based on the larger of the updated transportation plan time granularity and the quartiles of the historical path switch completion time distribution, and is limited to a lower limit of 20 seconds and an upper limit of 300 seconds to avoid under-sampling and over-smoothing. Within each window, an anomaly timestamp with the smallest time difference from the anchor point is searched, and the difference between the two in seconds is taken as the single-point time deviation. A positive value indicates that the anomaly occurred later than the anchor point, and a negative value indicates that it occurred earlier than the anchor point. To improve stability, three derived quantities are calculated for the single-point time deviation in continuous windows: maximum window deviation, average window deviation, and number of consecutive exceedances. The number of consecutive exceedances is the percentage of single-point time deviations exceeding the threshold in adjacent windows. Continuous counting generates a time deviation dataset. The system performs threshold checks on the time deviation dataset to initiate time consumption index matching. The thresholds are derived from the percentile statistics of historical normal execution intervals and are frozen as fixed values ​​before going live. Specifically, the warning threshold for a single-point time deviation is the 90th percentile of the historical normal distribution, the alarm threshold is the 95th percentile of the historical normal distribution, and the threshold for consecutive exceedances is 3 times. Reaching any of these thresholds triggers the matching process. The matching process uses data comparison technology to compare the time deviation dataset with preset time consumption standards item by item and generates response time consumption indicators. The preset time consumption standards are divided into allowable response initiation time, allowable confirmation time, and allowable execution time, based on the type of goods and the disposal stage. The total allowable duration consists of four items. Before going live, these four items are based on labeled historical samples. The combination that maximizes the plan achievement rate and minimizes the anomaly residual rate is selected as the fixed value by iterating through candidate second values. The actual calculation method for the four durations is as follows: the time difference from the first occurrence of an anomaly to the first issuance of a path adjustment instruction is recorded as the response start time; the time difference from the first issuance to the completion of anomaly confirmation is recorded as the confirmation time; the time difference from the completion of anomaly confirmation to the actual effectiveness of the first path instruction is recorded as the execution time; and the time difference from the first occurrence of an anomaly to the actual effectiveness of the first path instruction is recorded as the total duration. The four durations are compared with the four allowable durations one by one to generate a compliance mark and the extent of exceeding the limit. The output is a time consumption index dataset.The system extracts response time features from the time consumption index dataset to evaluate the system's optimization level. The fixed feature set includes seven items: average single-point time deviation, maximum single-point time deviation, number of consecutive exceedances, response start-up time, confirmation time, execution time, and overall time. A linear regression algorithm is used to analyze the relationship between these features and the system performance score to output the optimization level evaluation result. Before deployment, the system performance score is uniformly defined as a weighted combination of the plan achievement rate and the abnormal residual rate, with the plan achievement rate having a weight of 0.7 and the abnormal residual rate having a weight of 0.3. These two weights are determined and fixed through historical playback optimization. The linear regression model is trained on multiple historical batches using the least squares error principle to obtain coefficients, which are then fixed and used to output the predicted score for the current batch. The predicted score is compared with the actual score for the current batch to obtain the score difference. The judgment rule is that when both the predicted score and the actual score are not lower than the historical median, and the score difference is... When the absolute value of the difference is not higher than the 90th percentile of the historical score fluctuation, it is judged as optimized and stable. When both the predicted score and the actual score are higher than the historical median and the actual score is at least 10 times higher than the historical median, it is judged as optimized and improved. When the actual score is lower than the historical median or the absolute value of the score difference is higher than the 90th percentile of the historical score fluctuation, it is judged as optimized and insufficient. After completing the optimization level assessment, the system compares the optimization level assessment result with the timestamp of the initial abnormal signal dataset window by window and performs difference calculation to quantify the response time change. The quantification caliber is the total duration of the current window minus the total duration of the historical benchmark to obtain the window change value. The total duration of the historical benchmark is the median of the total duration distribution during the historical normal period and is frozen before going online. At the same time, the average change of the last 3 windows, the maximum change of the last 3 windows, and the number of times the change direction of the last 3 windows are consistent are counted to obtain the quantified response time dataset.The system extracts deviation trend values ​​from the quantified response time dataset and performs baseline adjustments. The deviation trend value is composed of the directional consistency and amplitude level of the window change values. The amplitude level is divided into three categories based on the ratio of the change value to the total duration of the historical baseline: mild (not less than 5 and less than 15), moderate (not less than 15 and less than 30), and severe (not less than 30). These three boundaries are selected as fixed values ​​on historical data through grid traversal, with the goal of minimizing both false alarm and false negative rates. The baseline adjustment technique modifies the abnormal signal triggering conditions according to the following rules: when the deviation trend is continuously improving and the amplitude is mild, the single-point time deviation warning threshold is relaxed to 95% of its original value, and the alarm threshold is relaxed to 98% of its original value. When the deviation trend is continuously improving and the magnitude is moderate, the single-point time deviation warning threshold is relaxed to 90% of the original value and the alarm threshold is relaxed to 95% of the original value, while the consecutive exceedance threshold is increased by 1 time. When the deviation trend is continuously worsening and the magnitude is moderate, the single-point time deviation warning threshold is tightened to 90% of the original value and the alarm threshold is tightened to 90% of the original value, while the consecutive exceedance threshold is decreased by 1 time. When the deviation trend is continuously worsening and the magnitude is severe, the single-point time deviation warning threshold is tightened to 80% of the original value and the alarm threshold is tightened to 85% of the original value, while the consecutive exceedance threshold is decreased by 1 time. When the deviation trend is fluctuating or... When the amplitude is mild, the threshold remains unchanged. All relaxations and tightenings have upper and lower protection limits. The upper limit is the 98th percentile of the single-point time deviation distribution during historical normal periods, and the lower limit is the 80th percentile of the single-point time deviation distribution during historical normal periods. The protection limits are fixed before implementation. After the above adjustments are performed, the data is written to the updated benchmark dataset. Finally, the system performs data validation on the updated benchmark dataset to verify the integrity of the adjusted time series data. The validation includes four fixed checks with clear judgment criteria: Field integrity check requires each benchmark record to include five fields: threshold type, current value, adjustment direction, adjustment amplitude, and effective time. Missing any field is considered unqualified. The monotonicity and rationality check requires that the warning threshold not be higher than the alarm threshold and the alarm threshold not be lower than the protection lower limit within the same effective period; otherwise, it is considered unqualified. The historical consistency check requires that the adjustment direction be consistent with the deviation trend and that the adjustment magnitude match the magnitude level; inconsistency is considered unqualified. The playback effectiveness check requires that the false alarm rate generated by the updated threshold in historical similar scenario playback not be higher than the 90th percentile of the historical false alarm rate and the missed alarm rate not be higher than the 90th percentile of the historical missed alarm rate; otherwise, it is considered unqualified. If all four checks pass, the verified benchmark dataset is output as the new execution benchmark. If any check fails, it reverts to the previous qualified benchmark and recalculates in the next evaluation window.

[0040] like Figure 2As shown, a logistics supply chain monitoring device is also provided to implement the steps of the logistics supply chain monitoring method. The device includes a data acquisition module, which acquires real-time data from transport vehicles and goods through a sensor network, including location coordinates, temperature values, and timeliness indicators to obtain an initial abnormal signal dataset, and performs preliminary screening for humidity fluctuations in the transportation scenario to determine a humidity-related abnormal subset; an anomaly analysis module, which uses a K-means clustering algorithm to group data points based on the initial abnormal signal dataset to determine the anomaly type, and extracts sensitive parameters of the cargo type from the humidity-related abnormal subset to determine the temperature and humidity joint anomaly type; a dynamic triggering module, which, if the anomaly type is determined to be a temperature and humidity joint anomaly, obtains the corresponding threshold range from a preset cargo type database to obtain dynamic triggering conditions, and incorporates route adjustment instructions as an extension of the conditions during transportation; and an anomaly confirmation module, which, in response to the dynamic triggering... The system uses a support vector machine classifier to analyze the deviation between real-time data and threshold range to determine the anomaly confirmation result. It also embeds backup plan activation logic into the deviation analysis to generate a confirmation output containing sequence execution logic. The response execution module retrieves a matching automated process script from a pre-set response rule base based on the anomaly confirmation result to obtain a response execution sequence. It integrates transportation control instructions into the sequence to adapt to path optimization under terrain constraints. The transportation control module sends instructions to the transportation control system via the response execution sequence to adjust the path or activate backup plans to obtain an updated transportation plan. It records the time difference calculation method in the plan for subsequent time consumption index evaluation. The optimization evaluation module obtains the time difference between the updated transportation plan and the initial anomaly signal dataset to calculate the response time index and judge the system optimization level. In the judgment, it uses a comparison with the initial dataset to quantify the response time and adjust the benchmark for the next anomaly signal.

[0041] In this embodiment, the data acquisition module is equipped with a sensor interface unit, a time synchronization unit, and a data cache unit. It acquires location coordinates, temperature values, humidity values, and timeliness indicators at a fixed sampling period and generates an initial abnormal signal dataset. To ensure data validity, the acquisition module performs four processes locally: ascending time sorting, missing field removal, retention of the earliest record for duplicates within the same second, and validity verification of temperature and humidity ranges. Subsequently, it continuously calculates three types of features in the cache: short-term humidity fluctuation amplitude, rate of change, and duration of the ultra-stable zone using a sliding time window. The window length is set based on the quartile values ​​of the duration distribution of historically confirmed anomalies and limited to a range of 20 to 300 seconds. The upper and lower limits of the stable zone are determined based on the 10th and 90th percentiles of the historical normal distribution. When any feature exceeds the 95th percentile upper limit or falls below the 5th percentile lower limit of its historical normal distribution, the data belonging to that window is marked as a humidity candidate anomaly and written into the humidity-related anomaly subset. The anomaly analysis module receives standardized data and humidity-related anomaly subsets output by the acquisition module. First, it constructs a feature set for each record for clustering and type determination. The features are fixed and include temperature value, humidity value, absolute value of the difference between adjacent temperature and humidity samples, timeliness index, and distance quantification value obtained by converting location coordinates. Then, it performs K-means clustering grouping. The number of clusters is selected from a discrete set of 2 to 6 using the intra-cluster distance decrease criterion. Initialization is repeated several times in a multi-starting point random manner to select the minimum intra-cluster distance. The iteration limit is set to 100 times, and convergence is achieved when the intra-cluster distance change ratio between two iterations is less than 1. After grouping, humidity anomaly clusters are marked based on the criteria of high humidity fluctuation and low concentration. Within this cluster, combined with the sensitive parameters in the cargo type database and the generated humidity fluctuation pattern, the joint temperature and humidity anomaly type is determined according to four rules: high temperature and high humidity, low temperature and high humidity, high temperature and low humidity, and low temperature and low humidity. Upon receiving a combined anomaly type, the dynamic triggering module reads sensitive parameters from the cargo type database, including the lower and upper limits of the temperature and humidity thresholds, the allowable temperature and humidity deviation, the longest allowable over-limit duration, the maximum allowable rate of change, the target recovery time, and the maximum acceptable delay for route adjustment. The thresholds and parameters are set based on the quantile statistics of historical normal transportation data plus a safety margin, the time statistics of historical successful handling samples, and contractual constraints, and are fixed as values ​​during device initialization. The dynamic triggering condition set follows the rule of simultaneously satisfying four conditions: the over-limit range reaches the allowable deviation, the over-limit duration reaches the longest allowable over-limit duration, the rate of change reaches or exceeds the maximum allowable rate of change, and the handling can be completed without exceeding the target recovery time and the maximum acceptable delay for route adjustment. At the same time, the route adjustment instruction is output as an extension of the conditions along with the triggering condition set.The anomaly confirmation module takes a set of triggering conditions and real-time features as input and uses a support vector machine classifier for anomaly confirmation. The input features are temperature deviation, humidity deviation, two types of change rates, duration of exceeding limits, and threshold / inside / outside markers. The kernel type, penalty coefficient, and kernel width of the classifier are optimized and set to fixed values ​​through cross-validation before going live. The classification output includes two categories: normal and jointly abnormal, as well as a confidence score. The device sets the confidence score threshold to 0.8. If the confidence score is not lower than the threshold, an anomaly is confirmed. At the same time, the module determines whether to immediately activate the backup plan based on the deviation ratio and duration ratio. The rule is that if any two of the three categories have a ratio of not less than 1 and a confidence score of not less than 0.9, or if any one category has a ratio of not less than 1.2, it will be triggered. After triggering, the output includes a confirmation result containing the sequence execution logic. The response execution module accesses the response rule base to match automated process scripts based on the confirmation results. Rule base entries include anomaly type, deviation level, terrain level, action sequence, and parameter boundaries. The matching strategy prioritizes anomaly type, followed by deviation level, and then terrain level. If multiple matches are found, the script with the highest priority and smallest script number is selected. The module parses the action sequence to obtain a set of transportation control instructions and adds path optimization constraints under terrain limitations, forming the response execution sequence for this round. The transportation control module receives the response execution sequence and sends path adjustment and backup plan activation instructions to the transportation control system. The sending order is from highest to lowest priority and satisfies prerequisite dependencies; within the same priority, instructions are sorted by arrival time difference from smallest to largest. The control module synchronously records the time difference calculation method in the plan, specifically recording three types of second-level values: single-segment start time difference, single-segment arrival time difference, and overall margin time difference for evaluation. The optimization evaluation module takes the updated transportation plan and initial abnormal signal dataset as input, generates a response time index according to the aforementioned time difference definition, and compares it with the historical benchmark. The benchmark is the median of the total duration of the historical normal period. The module outputs an evaluation vector consisting of the average and maximum values ​​of single-point time deviation, the number of consecutive exceedances, and the duration of four types of stages. Then, linear regression is used to determine the relationship between the evaluation vector and the system performance score and to give the optimization level result. When the results show continuous improvement or continuous deterioration, the module relaxes or tightens the warning and alarm thresholds for single-point time deviation and the threshold for the number of consecutive exceedances according to the benchmark adjustment technique. The relaxation and tightening range is set to a fixed ratio during initialization and is protected by upper and lower bounds. The upper bound is the value corresponding to the 98th percentile of the historical normal period, and the lower bound is the value corresponding to the 80th percentile of the historical normal period. After completion, it is written into the new abnormal signal benchmark for use in the next round.

[0042] A computer device includes a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the logistics supply chain monitoring method.

[0043] After the device powers on, the processor loads instructions and pre-defined parameters from memory, establishing a raw data area, a feature data area, a judgment result area, and a planning area. It receives location coordinates, temperature, humidity, and timeliness indicators at a fixed sampling period, performing time sorting, missing data removal, deduplication within the same second, and range verification. Within a sliding window, it calculates the humidity fluctuation amplitude, rate of change, and duration, forming candidate humidity anomalies. The processor performs K-means clustering on the standardized data, labeling humidity anomaly clusters based on intra-cluster fluctuation and concentration, and categorizing the joint performance of temperature and humidity into a preset joint anomaly type based on cargo type thresholds. When a joint anomaly occurs, the processor reads the temperature threshold, humidity threshold, allowable deviation, maximum over-limit duration, maximum rate of change, target recovery time, and path delay constraints from the cargo type database, generating dynamic triggering conditions that must be simultaneously met. Subsequently, the temperature deviation, humidity deviation, two types of rate of change, and over-limit duration are input into a support vector machine classifier for anomaly confirmation, and a backup plan is initiated based on the deviation and duration ratios. The confirmation results are used to retrieve the response rule base, parsing the action sequence and parameter boundaries, and generating a path instruction set containing time and space constraints. The processor sends instructions to the transportation control system and receives confirmation within a specified time limit; if the timeout occurs, it resends the instructions a predetermined number of times; successful confirmations are summarized into an updated transportation plan. During operation, the processor generates a real-time data stream from path instructions and sensor data, calculates execution deviations according to a fixed window, and outputs normal, slight, or severe deviations using a random forest; for slight deviations, it adjusts the order of priority; for severe deviations, it performs path switching or replacement of backup entries, and links the risk classification sampling cycle, ultimately outputting an executable transportation plan and the latest control instructions.

[0044] A computer-readable storage medium, characterized in that the computer-readable storage medium stores computer instructions, the computer instructions being used to cause a computer to execute the logistics supply chain monitoring method.

[0045] When the computer reads the storage medium, it first executes the loading instructions in the medium's boot area to complete version identification, integrity verification, and permission verification. After successful verification, the instruction segments are mapped into memory in a preset order, establishing four types of operating areas: raw data area, feature data area, judgment result area, and plan evaluation area. Subsequently, parameter constants and resource files are loaded. The parameter constants include fixed values ​​such as the upper and lower bounds of the sliding window duration, steady-state quantiles, fluctuation feature threshold quantiles, the range of candidate clusters for clustering, the upper limit of clustering iterations and the convergence threshold, the support vector machine confidence threshold, the backup trigger confidence threshold, the deviation ratio trigger coefficient, the rule matching priority order, the path delay constraint, the sampling period classification, the number and depth of trees in isolated forests and random forests, the time alignment quantile threshold, the threshold for consecutive exceedances, four time consumption standards, and the upper and lower bounds of the benchmark protection, which are fixed before going live. The resource files include a mirror of the response rule base, a cargo type threshold table, a path terrain restriction table, and the trained parameters of the classifier and regressor.After loading, the processor executes each instruction segment sequentially: the data acquisition instruction segment receives location, temperature, humidity, and timeliness data at a fixed sampling period, performs time sorting, missing data removal, deduplication within the same second, and range verification, and calculates the short-term fluctuation amplitude, rate of change, and duration of the ultra-stable zone of humidity within a sliding window, writing it into the feature data area; the anomaly analysis instruction segment performs mean clustering on the standardized features, selects the number of clusters based on the decrease in intra-cluster distance, and generates anomaly type groups, identifying the joint anomaly type of temperature and humidity by combining cargo type thresholds; the dynamic trigger instruction segment reads the upper and lower limits of temperature and humidity, allowable deviation, maximum over-limit duration, maximum rate of change, target recovery time, and path delay upper limit from the threshold table, generating a set of trigger conditions that must be met simultaneously; the anomaly confirmation instruction segment inputs temperature deviation, humidity deviation, two types of rate of change, and over-limit duration into a support vector machine classifier, obtains labels and confidence scores, confirms anomalies according to the confidence score threshold, and determines whether to activate the backup plan based on the deviation ratio and duration ratio, forming a confirmation output containing sequential execution logic; the response execution instruction segment accesses the appropriate database according to anomaly type, deviation level, and terrain level. The rule base mirrors and parses action sequences and parameter boundaries to synthesize a path instruction set containing time and space constraints. The transportation control instruction segment sends the path instruction set to the transportation control system, receives acknowledgments within the confirmation time limit, resends according to the maximum number of retries, summarizes successful instructions into an updated transportation plan, and records the start time difference, arrival time difference, and margin time difference. The optimization evaluation instruction segment aligns the initial abnormal timestamp with the key time points of the updated transportation plan, calculates the single-point time deviation, the maximum and average deviations within a fixed window, and the number of consecutive over-limits. After triggering, it compares each item with the preset time consumption standard to obtain the response initiation, confirmation, execution, and overall duration, as well as whether it meets the standard and the over-limit range. It extracts deviation and duration features, outputs the optimization level evaluation results through linear regression, and forms quantitative data based on the difference between the current overall duration and the historical baseline overall duration. According to the baseline adjustment rules, it relaxes or tightens the warning threshold, alarm threshold, and consecutive over-limit threshold, updates the abnormal signal baseline, and completes four checks: field integrity, monotonic rationality, historical consistency, and playback effectiveness. After passing the checks, it becomes the effective baseline for the next cycle.

[0046] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A logistics supply chain monitoring method, characterized in that, include: S1. Obtain real-time data from transport vehicles and goods through sensor networks, including location coordinates, temperature values, and timeliness indicators, to obtain an initial abnormal signal dataset, and perform preliminary screening for humidity fluctuations in the transportation scenario to determine a humidity-related abnormal subset. S2. Based on the initial abnormal signal dataset, the K-means clustering algorithm is used to group the data points to determine the abnormality type, and the sensitive parameters of the cargo type are extracted from the humidity-related abnormality subset to determine the joint temperature and humidity abnormality type. S3. If the anomaly type is determined to be a combined temperature and humidity anomaly, the corresponding threshold range is obtained from the preset cargo type database to obtain dynamic triggering conditions, and route adjustment instructions are incorporated into the transportation as an extended condition. S4. For dynamic triggering conditions, the deviation between real-time data and threshold range is analyzed by a support vector machine classifier to determine the anomaly confirmation result, and the backup scheme startup logic is embedded in the deviation analysis to generate a confirmation output containing sequence execution logic. S5. Based on the anomaly confirmation result, obtain the matching automated process script from the preset response rule base to obtain the response execution sequence, and integrate transportation control instructions into the sequence to adapt to path optimization under terrain constraints. S6. By sending instructions to the transportation control system through the response execution sequence to adjust the route or activate the backup plan, an updated transportation plan is obtained, and the time difference calculation method is recorded in the plan for subsequent time consumption index evaluation. S7. Obtain the time difference between the updated transportation plan and the initial abnormal signal dataset to calculate the response time index and judge the system optimization level. In the judgment, the initial dataset is compared to quantify the response time and adjust the benchmark for the next abnormal signal. S1 includes: Real-time data, including location coordinates, temperature values, timeliness indicators, and humidity values, are acquired from transport vehicles and goods through sensor networks to generate an initial dataset of abnormal signals. The initial abnormal signal dataset was processed using time series analysis methods to extract the trend of humidity values ​​and obtain humidity fluctuation characteristics. If the humidity fluctuation characteristics exceed the preset threshold range, the data points are divided into normal and abnormal categories by clustering algorithm to determine the humidity-related abnormal subset; Based on the humidity-related anomaly subset, combined with location coordinates and timeliness indicators, the correlation strength between the anomaly subset and the transportation scenario is calculated to obtain the anomaly distribution pattern. By identifying abnormal distribution patterns, a decision tree algorithm is used to classify abnormal subsets, determine whether the abnormalities are related to the transportation scenario, and obtain the abnormal classification results. Based on the anomaly classification results, obtain the corresponding location coordinates and timeliness indicators, generate the spatiotemporal distribution characteristics of the anomaly signal, and determine the specific scenario in which the anomaly occurred; By analyzing the specific scenarios in which anomalies occur, the sampling frequency of the sensor network is updated, resulting in an optimized data acquisition strategy.

2. The logistics supply chain monitoring method according to claim 1, characterized in that: S2 includes: Real-time data of transport vehicles and goods are collected through sensor networks, including location coordinates, temperature values, humidity values ​​and timeliness indicators, to generate an initial abnormal signal dataset. The K-means clustering algorithm was used to group the initial abnormal signal dataset to obtain the abnormality type grouping results. Extract the humidity anomaly subset from the anomaly type grouping results, analyze the time series changes of humidity values, and obtain the humidity anomaly fluctuation pattern; For abnormal humidity fluctuation patterns, if the fluctuation range of humidity value exceeds the preset threshold, the joint distribution characteristics of temperature and humidity are calculated in combination to determine the joint anomaly type of temperature and humidity. Based on the combined anomaly type of temperature and humidity, the corresponding location coordinates and timeliness indicators are obtained to generate the spatiotemporal distribution pattern of the anomaly signal.

3. The logistics supply chain monitoring method according to claim 1, characterized in that: S3 includes: Real-time temperature and humidity values ​​are acquired from sensor data. A threshold comparison method is used to match the temperature and humidity threshold ranges in the cargo type database to determine whether a joint temperature and humidity anomaly exists, and an anomaly confirmation result is obtained. If the anomaly confirmation result is a combined temperature and humidity anomaly, then the sensitive parameters of the corresponding goods are extracted from the goods type database, and dynamic triggering conditions are generated by combining them with real-time data updates to obtain a set of triggering conditions. Based on the set of triggering conditions, a route planning algorithm is used to calculate the transportation route adjustment instructions, generating a set of route instructions that includes time and space constraints; The navigation data of the transport vehicle is updated by updating the path instruction set, and the sensor data acquisition frequency is adjusted by adopting a real-time data update mechanism to obtain an optimized acquisition strategy. Real-time data streams are extracted from the optimized acquisition strategy, and anomaly type judgment method is used to continuously monitor the joint anomalies of temperature and humidity to obtain anomaly status update results. If the abnormal status update result exceeds the dynamic triggering condition, then the alternative transportation plan is obtained from the cargo type database, and an adjusted route instruction set is generated; The operational status of transport vehicles is updated by adjusting the route instruction set, and the results of sensor data acquisition and route planning are integrated using data fusion technology to obtain the final transportation optimization scheme.

4. The logistics supply chain monitoring method according to claim 1, characterized in that: S4 includes: Temperature and humidity values ​​are obtained from real-time data collected by sensors. A support vector machine classifier is used to analyze the deviation between the real-time data and the preset threshold range to confirm the joint anomaly of temperature and humidity. If the confirmation result is abnormal, the startup logic of the backup plan is triggered based on the degree of deviation, and an abnormal confirmation output containing the sequence execution logic is generated. Based on the anomaly confirmation output, retrieve the corresponding backup transportation plan from the preset backup plan database and generate the adjusted transportation instruction set; By using the adjusted transportation instruction set, the navigation data of the transportation vehicles is updated using a path planning algorithm to generate a path instruction set that includes time and space constraints; Real-time navigation data is extracted from the path instruction set, and data fusion technology is used to integrate sensor-collected data and navigation data to generate an optimized transportation data stream; Based on the optimized transportation data stream, an anomaly detection algorithm is used to continuously monitor the joint anomalies of temperature and humidity, and to determine the anomaly status update results. If the abnormal status update result exceeds the preset threshold range, the backup transportation plan is updated through sequence execution logic to generate the final transportation optimization instruction set.

5. The logistics supply chain monitoring method according to claim 1, characterized in that: S5 includes: Obtain the anomaly type and deviation data from the anomaly confirmation results, query the preset response rule base, match the corresponding automated process script, and generate the initial response execution sequence; Based on the initial response execution sequence, transport control instructions are extracted, and combined with terrain constraint data, the A-path planning algorithm is used to generate an optimized path instruction set. Navigation data is obtained from the path instruction set, and data fusion technology is used to integrate environmental data and navigation data collected by sensors to generate a real-time transportation data stream. Based on real-time transportation data streams, the isolated forest algorithm is used to continuously monitor abnormal states and determine the state update results. If the status update result exceeds the preset threshold, a backup automated process script is retrieved from the response rule base to update the response execution sequence; Based on the updated response execution sequence, adjust the transport control instructions to generate the final optimized instruction set; Path data is extracted from the final optimized instruction set, and data verification technology is used to confirm the integrity of the instruction set, resulting in executable transportation instructions.

6. The logistics supply chain monitoring method according to claim 1, characterized in that: S6 includes: Path adjustment instructions are extracted from the response execution sequence, and a structured instruction dataset is generated using data parsing techniques. By combining a structured instruction dataset with backup plan data, and using a decision tree algorithm, the applicability of path adjustment instructions is determined, resulting in a filtered instruction set. If the filtered instruction set meets the preset threshold, the adjusted transportation plan is generated through the data integration process, resulting in plan data that includes route adjustments. Time deviation data is extracted from the planning data that includes route adjustments. Time deviation calculation method is used to generate time deviation records and obtain a transportation plan that includes time difference. By using a transportation plan that incorporates time differences and employing an instruction allocation mechanism, route adjustment instructions are sent to the transportation control system to obtain the updated plan status after execution. Data from the time record module is obtained from the post-execution plan update status. Data verification technology is used to determine the completeness of the plan update status and obtain the final transportation plan. Based on the final transportation plan, a real-time monitoring data stream is generated. The random forest algorithm is used to determine the execution deviation of the transportation plan and obtain an optimized transportation instruction set. S7 includes: Timestamp data is obtained from the initial abnormal signal dataset, and time series parsing technology is used to generate a structured time series dataset. By aligning the time series dataset with the updated transportation plan, and using the sliding window technique, the time deviation data is calculated to obtain the time deviation dataset. If the deviation value in the time deviation dataset meets the preset threshold, then the time deviation dataset is matched with the preset time consumption standard through data comparison technology to generate response time consumption index and obtain time consumption index dataset. Response time features are extracted from the time consumption index dataset. A linear regression algorithm is used to analyze the relationship between response time features and system performance, determine the system optimization level, and obtain the optimization level evaluation results. By comparing the optimized level assessment results with the timestamp data of the initial abnormal signal dataset, the difference calculation method is used to quantify the change in response time and obtain the quantified response time dataset. Deviation trend values ​​are extracted from the quantized response time dataset. A benchmark adjustment technique is used to adjust the triggering conditions of abnormal signals based on the deviation trend values, generating an adjusted abnormal signal benchmark and obtaining an updated benchmark dataset. By updating the benchmark dataset and employing data verification techniques, the integrity of the adjusted time series data is verified, resulting in a verified benchmark dataset.

7. A logistics supply chain monitoring device, used to implement the steps of the logistics supply chain monitoring method according to any one of claims 1-6, characterized in that, The device includes: The data acquisition module obtains real-time data from transport vehicles and goods through a sensor network, including location coordinates, temperature values, and timeliness indicators to obtain an initial abnormal signal dataset, and performs preliminary screening for humidity fluctuations in the transportation scenario to determine a humidity-related abnormal subset. The anomaly analysis module uses the K-means clustering algorithm to group data points based on the initial anomaly signal dataset to determine the anomaly type, and extracts sensitive parameters of cargo type from the humidity-related anomaly subset to determine the joint temperature and humidity anomaly type. The dynamic triggering module, if the abnormality type is determined to be a combined temperature and humidity abnormality, obtains the corresponding threshold range from the preset cargo type database to obtain the dynamic triggering conditions, and incorporates route adjustment instructions as an extended condition part during transportation. The anomaly confirmation module uses a support vector machine classifier to analyze the deviation between real-time data and threshold range under dynamic triggering conditions to determine the anomaly confirmation result. It also embeds backup scheme startup logic in the deviation analysis to generate confirmation output containing sequence execution logic. The response execution module obtains the matching automated process script from the preset response rule base based on the anomaly confirmation result to obtain the response execution sequence, and integrates transportation control instructions into the sequence to adapt to path optimization under terrain constraints. The transportation control module sends instructions to the transportation control system to adjust the route or activate the backup plan by responding to the execution sequence, and obtains the updated transportation plan. The plan also records the time difference calculation method for subsequent time consumption index evaluation. The optimization evaluation module obtains the time difference between the updated transportation plan and the initial abnormal signal dataset to calculate the response time index and judge the system optimization level. In the judgment, the initial dataset is compared to quantify the response time and adjust the benchmark for the next abnormal signal.

8. A computer device, characterized in that, The system includes a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the logistics supply chain monitoring method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the logistics supply chain monitoring method according to any one of claims 1-6.

Images