Oil tank truck oil leakage early warning method and system based on abnormal weight fluctuation recognition algorithm

Abstract

The invention relates to the field of oil tank truck safety monitoring, in particular to an oil tank truck oil leakage early warning method and system based on an abnormal weight fluctuation recognition algorithm, and the method comprises the following steps: collecting multi-source original data; through noise filtering, smooth filtering and thermal expansion and cold contraction correction, a multi-dimensional data matrix is formed after timestamp alignment; dividing three types of working conditions, establishing a normal weight change reference of each working condition, calculating a dynamic threshold value of each working condition, calculating a comprehensive correction coefficient in combination with vehicle maintenance data, and correcting the dynamic threshold value of each working condition to obtain a corrected threshold value; performing triple verification by utilizing the corrected threshold value and the multi-dimensional data matrix to obtain an oil leakage abnormity judgment result; and triggering a grading early warning mechanism according to an oil leakage abnormity judgment result, and synchronously storing full-link data. According to the invention, through multi-source data integration, dynamic threshold value self-adaption, three-layer identification and graded early warning measures, the transportation safety, early warning accuracy and supervision effectiveness of the oil tank truck are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of oil tanker safety monitoring, and in particular to an oil tanker oil leakage early warning method and system based on an abnormal weight fluctuation identification algorithm. Background Technology

[0002] Existing oil tanker leak warning technologies mostly rely on single-sensor monitoring. Common methods include level sensors monitoring changes in the oil level inside the tank, leak sensors directly sensing the leak location, and gas sensors detecting the concentration of volatile oil vapors. Some technologies combine weight sensors to collect weight data and determine the presence of a leak by comparing it to a fixed threshold. These technologies can provide basic warning functions in specific scenarios. Among them, weight monitoring methods, because they are directly related to changes in the total amount of oil, have certain core monitoring advantages and have become one of the mainstream technologies. Their basic logic is to trigger a warning by using a static change threshold in weight data to help ensure transportation safety.

[0003] However, existing technologies still have significant shortcomings: First, the data collection dimension is limited, failing to effectively integrate driving status, operational status, environmental parameters, and vehicle maintenance data. This results in a lack of multi-source data support for early warning judgments, making them susceptible to interference from external factors such as vehicle bumps, temperature changes, and oil sloshing, leading to high false alarm and false negative rates. Second, the impact of equipment aging on early warning accuracy is not considered. Sensors experience accuracy degradation during use, and tank welds age and deteriorate with increasing mileage. Existing technologies use a fixed threshold mode, making it impossible to dynamically adjust early warning sensitivity based on equipment status, thus hindering predictive early warning of oil leak risks. Third, the working conditions are roughly divided, and no adaptive benchmark model has been established for different working conditions such as loading and unloading oil, driving, and stationary conditions. The threshold settings lack working condition specificity and have poor adaptability. Fourth, the anomaly identification logic is simple and relies heavily on single threshold comparison. It has not formed a multi-level identification mechanism of "threshold comparison - trend verification - cross-verification" and has weak anti-interference ability. Fifth, the early warning response and data traceability are imperfect. There is a lack of hierarchical early warning strategy, the emergency response is not targeted enough, and the full-link data storage and traceability mechanism is not sound. It is difficult to meet the traceability requirements of safety production supervision and cannot adapt to the actual transportation scenarios of oil tanker trucks with full working conditions and high safety requirements. Summary of the Invention

[0004] To address the technical deficiencies in the background technology, this invention proposes a method and system for early warning of oil leaks from tank trucks based on an abnormal weight fluctuation identification algorithm. This solves the aforementioned technical problems and meets practical needs. The specific technical solution is as follows: The method for early warning of oil leaks from tank trucks based on an abnormal weight fluctuation identification algorithm includes the following steps: Collect data on tanker truck weight, driving status, operating status, environmental parameters, and vehicle maintenance, and integrate them to form multi-source raw data; Noise filtering, smoothing filtering, and thermal expansion and contraction correction are performed on the multi-source raw data, and a multi-dimensional data matrix is ​​formed after timestamp alignment. Based on the multi-dimensional data matrix, three types of working conditions are divided and a normal weight change benchmark for each working condition is established. The dynamic threshold of each working condition is calculated, and the comprehensive correction coefficient is calculated in combination with vehicle maintenance data. The dynamic threshold of each working condition is then corrected to obtain the corrected threshold. By using the corrected threshold and the multi-dimensional data matrix, dynamic threshold comparison, trend fitting verification and multi-parameter cross-validation are performed respectively to obtain the oil leak anomaly judgment result; The tiered early warning mechanism is triggered based on the oil leak anomaly determination results, and the entire chain of data is stored synchronously.

[0005] Furthermore, the specific steps for collecting the multi-source raw data are as follows: Piezoelectric vehicle-mounted load cells are installed at four symmetrical support points of the tanker truck frame and tank body to collect real-time total weight data of the tanker truck at a frequency of 10 Hz. The collected weight data is transmitted to the vehicle terminal via RS485 bus. The vehicle speed, acceleration, and braking signals are collected through the vehicle's OBD interface, and the real-time vehicle location and mileage are collected through the GPS module. The collection frequency is set to be consistent with the real-time total weight data, and the collected driving status data is synchronously transmitted to the vehicle terminal. A solenoid valve sensor is installed at the oil loading and unloading valve to collect the valve opening and closing status, operation timestamp, and oil loading and unloading operation command signal, and the collected operation status data is transmitted to the vehicle terminal. Temperature and humidity sensors are installed on the outside of the tank to collect ambient temperature and relative humidity, and the collected environmental data is transmitted to the vehicle terminal. By synchronizing vehicle maintenance records through a cloud management platform, the system extracts sensor usage time, the level of the most recent ultrasonic weld inspection, and the cumulative mileage since the last weld inspection from the maintenance records. The extracted maintenance data is then transmitted to the vehicle terminal and integrated with weight data, driving status data, work status data, and environmental data to form multi-source raw data.

[0006] Furthermore, the specific steps for forming the multi-dimensional data matrix are as follows: For weight data in multi-source raw data, the mean and standard deviation of the weight data series are calculated using the three-standard-deviation criterion. Impulse interference outliers that exceed the range of mean minus three standard deviations to mean plus three standard deviations are removed, and valid weight data are retained. The effective weight data after removing outliers is filtered by moving average. The sliding window size is set to five sampling points. The smoothed weight sequence is obtained by calculating the mean of five adjacent sampling points. Based on the ambient temperature data in the multi-source raw data, combined with the volume expansion characteristics of the corresponding oil, the smoothed weight sequence is corrected for thermal expansion and contraction according to the correction formula to eliminate the interference of temperature change on weight measurement and obtain the corrected weight. Driving status data and operation status data are extracted from multiple sources of raw data, and then precisely aligned with the corrected weight data according to the timestamp. This results in the integration of a multi-dimensional data matrix that includes time, weight, and auxiliary parameters.

[0007] Furthermore, the corrected formula is as follows: , in, For the corrected weight, This is the current sample value in the smoothed weight sequence. To correspond to the volume expansion coefficient of oil products, gasoline The value is 9.5 × 10 -4 / ℃, diesel fuel The value is 7.0 × 10 -4 / ℃, where T is the real-time ambient temperature. The standard reference temperature is 20℃.

[0008] Furthermore, the specific steps for obtaining the corrected threshold are as follows: Valve switching signals in the operational state and vehicle speed signals in the driving state are extracted from the multi-dimensional data matrix. Based on the signal characteristics, the operating state of the tanker truck is divided into three core operating conditions: loading and unloading oil condition, driving condition, and stationary condition. For oil loading and unloading operations, the starting weight, ending weight, and operation duration for each operation are extracted from a multi-dimensional data matrix. The rate of weight change per unit time is calculated, and a library of oil loading and unloading weight change curves is constructed based on multiple sets of operation data, which serves as a reference benchmark for normal weight changes under this operation. For driving conditions, driving data with no oil leak records and normal sensor operation within the past 30 days are extracted from the multi-dimensional data matrix. A linear regression algorithm is used to fit the data to obtain the normal fuel consumption curve and the fuel consumption benchmark value per unit mileage. At the same time, the vehicle speed range is divided, and the fuel consumption correction coefficient for each vehicle speed range is fitted. For stationary conditions, the weight data within five minutes after the vehicle comes to a stop is extracted from the multi-dimensional data matrix. The average weight data within this time period is calculated as the benchmark value. The range from the benchmark value minus 0.1% of the full scale to the benchmark value plus 0.1% of the full scale is set as the normal fluctuation range under stationary conditions. Based on the normal weight change benchmark for each working condition, the dynamic thresholds for driving, stationary, and loading / unloading oil conditions are calculated respectively. The sensor usage time, weld ultrasonic testing level, and cumulative mileage since the last weld inspection are extracted from multi-source raw data. The sensor attenuation correction coefficient is determined based on the sensor usage time, and the weld aging correction coefficient is determined based on the weld ultrasonic testing level and cumulative mileage. The product of the two is calculated to obtain the comprehensive correction coefficient. The comprehensive correction coefficient is substituted into the calculation process of the dynamic threshold for each working condition, and the dynamic threshold is adaptively corrected to finally obtain the corrected threshold for driving condition, stationary condition, and oil loading and unloading condition.

[0009] Furthermore, the calculation formula for the aforementioned operating conditions is as follows: , in, To exercise the dynamic threshold of operating conditions, For safety margin, a value of 1.5 to 2.0 is used. This is the baseline value for fuel consumption per unit mileage. The monitoring time window is set to 5 minutes. The calculation formula for the static condition is as follows: , in, The dynamic threshold for static operating conditions. This is the baseline fluctuation value corresponding to the sensor's accuracy, taken as 0.1% of full scale. This is the confidence coefficient, with a value of 2.58. The standard deviation of weight fluctuation at historical static state; The calculation formula for the oil loading and unloading conditions is as follows: , in, The dynamic threshold for oil loading and unloading conditions. This is the operational fluctuation coefficient, with a value of 1.2. This represents the rate of change in oil weight per unit time during loading and unloading. The monitoring time window is set to 5 minutes.

[0010] Furthermore, the method for determining the sensor attenuation correction coefficient is as follows: When the sensor's usage time is ≤1 year, the sensor attenuation correction coefficient = 1.0; When 1 year < sensor usage time ≤ 2 years, the sensor attenuation correction coefficient = 1.1; When the sensor has been used for more than 2 years, the sensor attenuation correction coefficient is 1.2. The method for determining the weld aging correction coefficient is as follows: When the ultrasonic testing level of the weld is ≤2 and the cumulative mileage Sw since the last test is ≤50,000 kilometers, the weld aging correction coefficient is 1.0. When the ultrasonic testing level of the weld is 3 or 50,000 km < the cumulative mileage since the last test is ≤ 100,000 km, the weld aging correction coefficient is 1.1. When the ultrasonic testing level of the weld is ≥4 or the cumulative mileage since the last test is >100,000 kilometers, the weld aging correction coefficient is 1.3. The calculation formulas for the corrected threshold values ​​for each operating condition are as follows: , in, The threshold value is the value after calibration for driving conditions. The threshold value is the one corrected for static conditions. The threshold value is the corrected value for loading and unloading oil conditions. This is the comprehensive correction factor.

[0011] Furthermore, the specific steps for obtaining the oil leak anomaly determination result are as follows: Real-time corrected weight data is extracted based on a multi-dimensional data matrix. Combined with the corrected threshold, the real-time weight change under different working conditions is calculated. If the real-time weight change under any working condition exceeds the corrected threshold for the corresponding working condition, a preliminary anomaly is triggered. For the weight data sequence that triggers the initial anomaly, a quadratic polynomial fitting method is used to fit the sequence to obtain a trend curve. The fitting coefficient of the trend curve is used to determine the weight change trend. When the fitting coefficient shows that the weight is decreasing and the instantaneous rate of decrease exceeds the rate of decrease per unit time corresponding to the corrected threshold, it is judged as a suspected oil leak. Valve switch signals in the operating state, acceleration data in the driving state, and ambient temperature change in the environmental data are extracted from the multi-dimensional data matrix to cross-validate suspected oil leak results: if the valve is closed, the absolute value of acceleration is no greater than 0.5g, and the ambient temperature change is no greater than 5℃, then it is finally determined to be an oil leak anomaly; if any of the following conditions are met: the valve is open, the absolute value of acceleration is greater than 0.5g, or the ambient temperature change is greater than 5℃, then it is determined to be an interference signal, and the preliminary anomaly is eliminated.

[0012] Furthermore, the tiered early warning mechanism specifically includes: Based on the oil leak anomaly determination results, the weight loss within the corresponding time period is calculated. When the weight loss within five minutes is between 0.5% and 1.0% of the full scale, a level one warning is triggered, controlling the vehicle terminal to activate the audible and visual alarm device and pushing a verification prompt message for suspected minor oil leak to the driver. When the weight drop exceeds 1.0% of full scale within five minutes, or exceeds 2.0% of full scale for thirty consecutive minutes, a level two warning is triggered. In addition to implementing level one warning measures, the vehicle terminal is controlled to upload real-time weight data, vehicle location information, operating condition data, and relevant parameters in the anomaly identification process to the cloud management platform. After receiving the data, the platform pushes emergency warning information to the management personnel. When the weight drop exceeds 3.0% of full scale within one minute, an emergency linkage warning is triggered. After receiving the warning signal, the cloud management platform automatically searches for the nearest emergency rescue station and pushes the vehicle's real-time location, oil leak warning level, and related monitoring data to the rescue station to assist rescue personnel in quickly carrying out emergency response.

[0013] The oil tanker leak early warning system based on the abnormal weight fluctuation identification algorithm includes: The multi-source data acquisition module includes a piezoelectric vehicle-mounted weighing sensor, an OBD interface module, a GPS module, a solenoid valve sensor, a temperature and humidity sensor, and a maintenance data synchronization unit. The data preprocessing module is connected to the multi-source data acquisition module and is used to receive multi-source raw data transmitted by the multi-source data acquisition module, perform outlier removal, moving average filtering, thermal expansion and contraction correction and data alignment processing on it, and output a multi-dimensional data matrix. The benchmark construction and threshold correction module is connected to the data preprocessing module. It is used to receive multi-dimensional data matrices, divide the oil loading and unloading conditions, driving conditions and stationary conditions into three core conditions, establish a benchmark model for normal weight change in each condition, calculate the dynamic threshold for each condition based on the benchmark model, calculate the comprehensive correction coefficient and correct the dynamic threshold, and output the corrected threshold. The anomaly identification module is connected to the benchmark construction and threshold correction module and the data preprocessing module respectively. It is used to receive the corrected threshold and multi-dimensional data matrix, perform a three-layer identification logic of dynamic threshold comparison, trend fitting verification and multi-parameter cross-validation, determine the oil leak anomaly and output the judgment result. The early warning and tracing module is connected to the anomaly identification module to receive the results of oil leak anomaly determination and trigger a first-level early warning, a second-level early warning, or an emergency linkage early warning based on the severity of the anomaly.

[0014] Compared with existing technologies, the oil tanker leak early warning method and system based on an abnormal weight fluctuation identification algorithm provided by the present invention have the following beneficial effects: This invention integrates weight, driving status, operating status, environmental, and vehicle maintenance data through a multi-source data acquisition architecture. It combines sensor attenuation and tank weld aging data to construct a comprehensive correction coefficient, enabling adaptive adjustment of dynamic thresholds. This solves the problem of existing technologies being unable to adapt to equipment aging conditions, achieving predictive early warning based on equipment status. By classifying three core operating conditions and establishing a dedicated normal weight change benchmark model, the adaptability of threshold settings to operating conditions is improved. A three-layer anomaly identification logic of "dynamic threshold comparison - trend fitting verification - multi-parameter cross-validation" effectively resists vehicle vibration and temperature changes. External interference is eliminated, significantly improving the accuracy of oil spill anomaly identification; differentiated handling is achieved through a tiered early warning mechanism (Level 1, Level 2, and Emergency Response Early Warning); combined with dual storage of data across the entire chain and a traceability period of over 90 days, it meets the requirements of safety production supervision; the overall technical solution forms a complete closed loop of "data collection - preprocessing - benchmark construction - accurate identification - tiered handling - data traceability", completely solving the defects of existing technologies such as high false alarm and false alarm rates, delayed early warning, poor adaptability, and insufficient supervision and traceability, and greatly improving the safety of tanker transportation, the accuracy of early warning, and the effectiveness of supervision. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the oil tanker leak early warning method based on the abnormal weight fluctuation identification algorithm in this invention.

[0016] Figure 2 This is a schematic diagram of the oil tanker leak early warning system based on the abnormal weight fluctuation identification algorithm in this invention. Detailed Implementation

[0017] In the description of this invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "middle," and "inner," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, it should be noted that unless otherwise explicitly specified and limited, the terms "installed," "connected," and "joined" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention through specific circumstances.

[0018] The embodiments of the present invention will be described below with reference to the accompanying drawings and related examples. The embodiments of the present invention are not limited to the following examples, and the present invention relates to the relevant necessary components in this technical field, which should be regarded as well-known technology in this technical field and can be known and mastered by those skilled in this technical field.

[0019] See Figure 1 This invention provides a method for early warning of oil leaks from tank trucks based on an abnormal weight fluctuation identification algorithm, comprising the following steps: Step S100: Collect data on the weight, driving status, operating status, environmental parameters, and vehicle maintenance of the oil tanker, and integrate them to form multi-source raw data; Tanker truck weight is a physical quantity reflecting the total mass of oil inside the tanker truck. It serves as a core input parameter for leak detection and is directly related to the trend of changes in the total oil volume. Driving status is a data set describing whether the tanker truck is currently moving and its motion characteristics. It is used to identify whether the vehicle is in motion and helps distinguish between weight fluctuations caused by bumps and actual leaks. Operational status is information characterizing the tanker truck's status during loading / unloading operations or other specific tasks. It is used to identify the loading / unloading phase and avoid misjudging normal oil transfer processes as abnormal leaks. Environmental parameters are a set of external physical conditions that affect the perceived volume or weight of the liquid inside the tanker. They are used to compensate for the thermal expansion and contraction effects caused by temperature changes and improve the accuracy of weight data. Vehicle maintenance data is a set of information recording the usage history and maintenance status of key components of the tanker truck. It is used to assess the degree of equipment aging and support the construction of dynamic correction models. Multi-source raw data refers to the initial data set obtained through different collection channels for different monitoring objects, including weight data, driving status data, operational status data, environmental data, and vehicle maintenance data. It forms the basis for subsequent data processing and analysis.

[0020] Step S200: Perform noise filtering, smoothing filtering and thermal expansion and contraction correction on the multi-source raw data, and form a multi-dimensional data matrix after timestamp alignment; Noise filtering removes non-realistic fluctuations caused by electromagnetic interference or signal distortion from multi-source raw data, improving data stability and reducing the risk of false triggers. Smoothing filtering is a technique that uses local averaging of continuous time series data to eliminate high-frequency random disturbances, enhancing the readability of weight signals and facilitating subsequent trend fitting analysis. Thermal expansion and contraction correction corrects for weight perception deviations caused by changes in oil volume due to ambient temperature variations, eliminating spurious weight changes due to temperature and improving monitoring accuracy. Timestamp alignment synchronizes data from different sensor channels according to a unified time reference, ensuring consistency of multi-dimensional data across time dimensions and supporting cross-validation logic. A multi-dimensional data matrix is ​​a structured data organization that integrates data from different dimensions (time, weight, driving status, etc.) by timestamp, forming a matrix structure with distinct rows and columns, facilitating batch processing and collaborative analysis by computers.

[0021] Step S300: Divide the working conditions into three categories based on the multi-dimensional data matrix and establish the normal weight change benchmark for each working condition. Calculate the dynamic threshold for each working condition. Combine the vehicle maintenance data to calculate the comprehensive correction coefficient. Correct the dynamic threshold for each working condition to obtain the corrected threshold. The three operating conditions classify the operation of tank trucks into three typical working modes: driving, loading / unloading, and stationary. This system is used to establish differentiated weight change models for different scenarios, improving the adaptability of threshold settings. The normal weight change benchmark is a mathematical expression model of the reasonable range of weight variation for tank trucks under specific operating conditions, used as a reference standard to determine whether the current weight deviates from the normal range. The dynamic threshold is a real-time judgment boundary value derived from the normal weight change benchmark under the current operating conditions, used to replace the fixed threshold, allowing the warning boundary to automatically adjust with changes in operating conditions. The comprehensive correction coefficient is a quantitative adjustment factor that integrates the sensor attenuation level and the aging level of the tank weld, used to optimize the dynamic threshold to reflect the decrease in sensitivity caused by equipment degradation. The corrected threshold is the final warning judgment boundary value adjusted by the comprehensive correction coefficient, used to compensate for aging factors and improve the reliability of warnings in long-term use.

[0022] Step S400: Use the corrected threshold and the multi-dimensional data matrix to perform dynamic threshold comparison, trend fitting verification and multi-parameter cross-validation respectively to obtain the oil leak anomaly judgment result; Dynamic threshold comparison compares the current weight value with a corrected threshold in real time to initially screen for abnormal events, quickly identifying weight losses that significantly deviate from the normal range. Multi-parameter cross-validation is a composite judgment mechanism that combines other non-weight parameters to verify the authenticity of suspected oil leaks, preventing false alarms from a single parameter and enhancing the reliability of the judgment results. The oil leak anomaly judgment result is a conclusion drawn from a comprehensive analysis of three layers of identification logic, determining whether a real oil leak exists, and serving as the final decision-making basis for triggering the early warning mechanism.

[0023] Step S500: Trigger the graded early warning mechanism based on the oil leak anomaly determination result and synchronously store the full-link data.

[0024] The tiered early warning mechanism is a system of response strategies that implements different levels of response based on the severity of oil spill risks. It aims to achieve refined management of emergency response and improve handling efficiency. End-to-end data refers to all data generated throughout the entire process, from data collection, preprocessing, benchmark construction, threshold calculation, anomaly identification to early warning triggering. This includes raw data, intermediate processed data, result data, and operation records, and serves as the core basis for data traceability.

[0025] The vehicle-mounted terminal and the cloud management platform synchronously store data across the entire chain. The stored content includes multi-source raw data, pre-processed effective data, multi-dimensional data matrices, parameters of benchmark models for various operating conditions, dynamic thresholds, comprehensive correction coefficients, corrected thresholds, anomaly identification process data, early warning trigger information and handling records. The storage period is set to be no less than ninety days to meet the traceability requirements of safety production supervision and facilitate subsequent fault investigation and responsibility determination.

[0026] Taking a minor leak during highway driving as an example, the oil tanker leak early warning method based on the abnormal weight fluctuation identification algorithm of this invention can be as follows: An oil tanker is traveling on a mountainous highway in high-temperature weather, and a slow leak occurs due to fatigue of the tank welds. The system detects a slight decrease in weight and initially triggers dynamic threshold comparison; however, because the vehicle is on a bumpy road, the system initiates trend fitting verification and finds that the weight decreases continuously and linearly rather than fluctuating instantaneously; further cross-verification is performed by combining the operating status (non-loading and unloading), ambient temperature (stable), and driving status (continuous movement) to rule out the possibility of thermal expansion and contraction and misoperation, and finally confirms that it is a real leak. At this time, the system has lowered the threshold in advance according to the comprehensive correction coefficient, which improves the early detection capability and triggers a secondary warning, notifying the monitoring center to arrange for the vehicle to stop and be inspected at the nearest service area to prevent the accident from escalating.

[0027] This invention improves input integrity through multi-source data fusion, enhances data consistency through preprocessing and time synchronization, achieves dynamic threshold generation through operating condition subdivision and historical modeling, completes threshold adaptive correction by introducing equipment aging factors, improves judgment reliability through a three-layer progressive verification mechanism, and achieves closed-loop management through hierarchical response and full data traceability. It can solve the problems caused by traditional methods such as single data, fixed thresholds, lack of operating condition differentiation, simple identification, and lack of traceability, and improve the accuracy and practicality of oil leak early warning.

[0028] In one embodiment of the present invention, the specific steps for collecting the multi-source raw data are as follows: Step S101: Install piezoelectric vehicle-mounted weighing sensors at four symmetrical support points of the tanker truck frame and tank body, collect real-time total weight data of the tanker truck at a frequency of 10 Hz, and transmit the collected weight data to the vehicle terminal via RS485 bus. Piezoelectric vehicle-mounted load cells were selected. These sensors are vibration-resistant, highly accurate, and adaptable to harsh vehicle environments (high temperature, bumps, electromagnetic interference). Their measurement accuracy reaches 0.1% of full scale, meeting the precise weight data requirements for oil leak early warning. The installation locations were chosen at four symmetrical support points on the tanker truck frame and tank body. This location can fully bear the weight load of the tank and the internal oil, avoiding measurement deviations caused by uneven force at a single point, ensuring that the weight data reflects the overall load changes of the tank. The symmetrical distribution of the four support points further offsets localized force fluctuations caused by oil sloshing and vehicle tilting, improving the stability of weight measurement during static and dynamic driving. Real-time total weight data was collected at a 10 Hz frequency, meaning ten weight samples were taken per second. This frequency was chosen because in oil leak incidents, the weight change rate of minor leaks is slow, while the weight change of severe leaks can appear in a short time. The 10 Hz frequency can capture weight changes of more than 0.5% of full scale per minute without generating redundant data due to excessively high sampling frequency, reducing the data processing pressure on the vehicle terminal. Weight data is transmitted to the vehicle-mounted terminal via an RS485 bus. The RS485 bus uses differential signal transmission, offering strong resistance to electromagnetic interference and adapting to the complex electromagnetic environment generated by the tanker truck's engine and onboard electrical components. The transmission distance can reach 1200 meters, fully meeting the signal transmission requirements within the tanker truck's length. Furthermore, the RS485 bus supports parallel connection of multiple devices, facilitating future expansion of the number of sensors. It also provides a stable transmission rate, ensuring real-time transmission of 10Hz sampling data without significant delay. The vehicle-mounted terminal has a built-in data buffer module. After receiving digital signals from the sensors, it temporarily stores the data in timestamp order, with a buffer capacity of at least one hour's worth of sampling data to prevent data loss due to network interruptions or data processing delays.

[0029] Step S102: Collect vehicle speed, acceleration, and braking signals through the vehicle OBD interface, collect the real-time vehicle location and mileage through the GPS module, set the collection frequency to be consistent with the real-time total weight data, and synchronously transmit the collected driving status data to the vehicle terminal. Vehicle speed, acceleration, and braking signals are collected via the onboard OBD interface. This interface is a standardized interface from the automotive factory and connects directly to the vehicle's engine control unit (ECU) and braking system control module. It can read the vehicle's original operating data without the need for additional sensor modifications, ensuring data authenticity and accuracy while reducing installation complexity and compatibility risks. Real-time vehicle location and mileage are collected via a GPS module, using a module supporting BeiDou + GPS dual-mode positioning with a positioning accuracy ≤10 meters and a mileage calculation error ≤0.5%, ensuring the accuracy of the mileage data. The collection frequency of driving status data is set to be consistent with that of weight data (10 Hz), and the timestamp error between the two types of data is ≤1 millisecond through the onboard terminal's clock synchronization module. The core purpose of this design is to establish a temporal correspondence between weight changes and driving status, such as accurately associating the timing of driving states like sudden braking and bumpy road conditions with weight fluctuations, providing a time reference for subsequent elimination of driving interference and identification of oil leaks. Data collected by the OBD interface is transmitted to the vehicle terminal via the vehicle CAN bus. The GPS module transmits data via the UART interface. After receiving the data, the vehicle terminal initially correlates the vehicle speed, acceleration, braking signal, location, mileage data and weight data according to the timestamp, forming a basic data set of "time-weight-driving status".

[0030] Step S103: Install a solenoid valve sensor at the oil loading / unloading valve to collect the valve opening / closing status, operation timestamp, and oil loading / unloading operation command signal, and transmit the collected operation status data to the vehicle terminal. A solenoid valve sensor is installed at the actuator of the loading / unloading valve. This sensor is mechanically linked to the valve's opening and closing action, directly detecting the valve's on / off state (open / closed). It also has a built-in timing module that records the timestamps of valve state changes. Simultaneously, the sensor's signal interface collects loading / unloading operation command signals (i.e., operation trigger signals) issued from the driver's control panel, forming a linked data structure of "operation command - valve state - timestamp." The solenoid valve sensor uses a wired connection (waterproof cable) to transmit data, avoiding interference from vehicle-mounted wireless signals. The transmission rate is matched to the weight data, ensuring that changes in the operational state are captured in real time. For example, when the driver issues an unloading command, the sensor first collects the operation command signal, then detects the valve's open state; the timestamp records the interval between the two being ≤0.5 seconds, providing clear evidence for subsequent determination of "weight loss due to normal unloading."

[0031] Step S104: Install a temperature and humidity sensor on the outside of the tank to collect ambient temperature and relative humidity, and transmit the collected environmental data to the vehicle terminal. The temperature and humidity sensor is installed on the outer center of the tank. This location avoids excessively high local temperatures caused by direct sunlight while accurately reflecting the ambient temperature of the tank. The sensor's temperature measurement range is set from -40℃ to 85℃, covering the ambient temperature range of different regions and seasons in my country, with a measurement accuracy of ±0.5℃, meeting the accuracy requirements for oil thermal expansion and contraction correction. The relative humidity measurement range is 0% to 100%, with an accuracy of ±5%, used to help determine the impact of the environment on the sensor's performance. The temperature and humidity sensor connects to the vehicle terminal via an I2C interface, with a sampling frequency of 10 Hz, maintaining time synchronization with the weight data to ensure that each weight sample corresponds to an ambient temperature data point, providing one-to-one parameter support for subsequent thermal expansion and contraction correction.

[0032] Step S105: Synchronize vehicle maintenance records through the cloud management platform, extract sensor usage time, the level of the most recent weld ultrasonic inspection, and the cumulative mileage since the last weld inspection from the maintenance records, and transmit the extracted maintenance data to the vehicle terminal. Combine the data with weight data, driving status data, operation status data, and environmental data to form multi-source raw data.

[0033] The vehicle terminal establishes a real-time connection with the cloud management platform via its built-in 4G / 5G communication module, synchronizing vehicle maintenance records periodically (every 24 hours) or as needed (after vehicle maintenance is completed). The cloud management platform interfaces with the vehicle operating unit's maintenance management system, extracting maintenance information of the target vehicle through a standardized data interface to ensure the data's authority and accuracy. Three core data categories are precisely extracted from the maintenance records: first, sensor usage time, calculated from the time of sensor installation or replacement to the current data collection time; second, the most recent ultrasonic weld inspection level, issued by a professional testing organization during maintenance and classified into levels 1-5 according to industry standards; and third, the cumulative mileage since the last weld inspection, calculated by correlating the mileage recorded by the GPS module with the inspection time points in the maintenance records. After receiving maintenance data, the vehicle terminal converts it into structured data through the data formatting module. It then integrates the previously collected weight data, driving status data, operation status data, and environmental data with "unique vehicle identifier + timestamp" to form a multi-source raw data set containing five dimensions. The data storage format adopts the standardized JSON format, which is convenient for subsequent data preprocessing modules to call.

[0034] This invention synchronizes maintenance data through a cloud management platform, ensuring the accuracy and timeliness of key data such as sensor usage time and weld aging. For the first time, it incorporates equipment aging parameters into the data foundation for oil leak early warning, breaking the limitations of existing technologies that rely solely on real-time monitoring data and ignore equipment status. The integration of multi-source data forms a complete set of raw data, providing comprehensive and relevant data input for subsequent full-process data processing.

[0035] In one embodiment of the present invention, the specific steps for forming the multi-dimensional data matrix are as follows: Step S201: For the weight data in the multi-source raw data, the mean and standard deviation of the weight data sequence are calculated using the three-standard-deviation criterion. Impulse interference outliers that exceed the range from the mean minus three standard deviations to the mean plus three standard deviations are removed, and valid weight data are retained. First, the weight data sequence is extracted from the multi-source raw data. This sequence is collected by a piezoelectric vehicle-mounted weighing sensor at a frequency of 10 Hz and contains continuous real-time total weight samples. Based on this weight data sequence, the mean of the sequence is calculated using the arithmetic mean method, which reflects the central tendency of the weight data. The standard deviation of the sequence is calculated using the Bessel formula, which reflects the dispersion of the weight data. The formula is essentially the sum of the squares of the differences between each data point and the mean, divided by the number of data points minus one, and then the square root is taken (the formula symbol is not shown here, only the calculation logic is explained). An effective data interval is constructed with "mean minus three standard deviations" as the lower limit and "mean plus three standard deviations" as the upper limit. Each sample value in the weight data sequence is compared one by one. Sample values ​​that exceed the above effective interval are identified as pulse interference anomalies (such as instantaneous sudden data changes generated by the sensor when the vehicle passes through a severely bumpy road section, or abnormal jump values ​​caused by electromagnetic interference) and removed from the sequence, retaining the effective weight data within the interval.

[0036] Step S202: Perform a moving average filter on the effective weight data after removing outliers. Set the sliding window size to five sampling points and obtain the smoothed weight sequence by calculating the mean of five adjacent sampling points. The system receives valid weight data after removing outliers and arranges it into a continuous data sequence according to the acquisition time. A sliding window of five sampling points is set, with the window length chosen based on the sensor acquisition frequency (10 Hz) and the noise characteristics of the tanker truck during operation. The five sampling points correspond to a 0.5-second time span, effectively filtering high-frequency noise without losing crucial information about weight changes. Starting from the beginning of the data sequence, the sliding window is used with each sampling point as a step size. After each slide, the arithmetic mean of the five sampling points within the current window is calculated. The calculated mean is used to replace the value of the last sampling point in the current window, and this process is repeated to complete the traversal of the entire data sequence, ultimately yielding a smoothed weight sequence. This sequence eliminates weight fluctuations caused by high-frequency interference such as oil sloshing and minor road bumps.

[0037] Step S203: Based on the ambient temperature data in the multi-source raw data, combined with the volume expansion characteristics of the corresponding oil, the smoothed weight sequence is corrected for thermal expansion and contraction according to the correction formula to eliminate the interference of temperature change on weight measurement and obtain the corrected weight. Ambient temperature data is extracted from multi-source raw data. This data is collected by a temperature and humidity sensor installed on the outside of the tank, with a measurement range of -40℃ to 85℃ and an accuracy of ±0.5℃, which can reflect the ambient temperature status of the tank in real time. The type of oil currently loaded in the tanker truck (gasoline or diesel) is determined, and the volumetric expansion coefficient of the corresponding oil is retrieved: the volumetric expansion coefficient of gasoline is 9.5×10. -4 The coefficient of volumetric expansion of diesel fuel is 7.0 × 10⁻⁶ per degree Celsius. -4 For each degree Celsius, a standard reference temperature of 20°C is set. This temperature is the standard temperature for oil measurement and can minimize the impact of temperature on weight measurement. The difference between the real-time ambient temperature and the standard reference temperature is calculated. If the real-time temperature is higher than 20°C, the oil volume expands, leading to a false underestimation of the weight measurement value. If the real-time temperature is lower than 20°C, the oil volume contracts, leading to a false overestimation of the weight measurement value. Based on the above difference and the oil volume expansion coefficient, the smoothed weight sequence is corrected point by point. The correction logic is as follows: Multiply the smoothed weight value by "1 minus the product of the volume expansion coefficient and the temperature difference" to eliminate measurement errors caused by temperature changes, and finally obtain the corrected weight. This weight value can truly reflect the actual quality of the oil in the tank.

[0038] Step S204: Extract driving status data and operation status data from multi-source raw data, accurately align them with the corrected weight data according to the timestamp, and integrate them to form a multi-dimensional data matrix containing time dimension, weight dimension and auxiliary parameter dimension.

[0039] Driving status data (vehicle speed, acceleration, braking signal, vehicle position, mileage) and operational status data (valve switch status, operational timestamp, oil loading / unloading operation command signal) are extracted from multi-source raw data. The acquisition timestamps for each data type are extracted, and the acquisition frequency of all data is consistent with that of the weight data (10 Hz) to ensure timestamp synchronization. Using the timestamp as a unified index, the corrected weight data is aligned point-by-point with the driving status data and operational status data to ensure that a complete set of weight, driving, and operational data corresponds to the same timestamp. A multi-dimensional data matrix is ​​constructed, with the row index being the timestamp and the column indices being the corrected weight, vehicle speed, acceleration, braking signal, vehicle position, mileage, valve switch status, operational timestamp, oil loading / unloading operation command signal, ambient temperature, and relative humidity, forming a standardized data matrix containing 11 data dimensions. This matrix can be directly used as input data for subsequent operational condition division, benchmark model construction, and anomaly identification.

[0040] Existing technologies often directly use raw weight data for threshold comparison without systematic data preprocessing, resulting in a large amount of interference noise and measurement errors in the data, which in turn affects the accuracy of early warning. In contrast, this invention forms a high-quality, multi-dimensional data matrix through a full-process preprocessing process including outlier removal, smoothing filtering, temperature correction, and data alignment. This solves the shortcomings of existing technologies, such as poor data quality and insufficient coordination of multi-source data, and provides a core guarantee for the accuracy of the overall early warning method, demonstrating significant technological progress.

[0041] It should be noted that the correction formula is as follows: , in, The corrected weight refers to the weight value, expressed in kilograms (kg), which accurately reflects the actual mass of the oil in the tank after correction for thermal expansion and contraction errors. This parameter is a core output of the data preprocessing stage and is directly used for subsequent multi-dimensional data matrix construction and abnormal weight fluctuation identification. Its accuracy determines the basic data quality of the overall early warning method. This is the current sampled value in the smoothed weight sequence. It refers to the weight sampled value corresponding to the current timestamp in the weight data sequence after outlier removal using the three-standard-deviation criterion and moving average filtering, in kilograms (kg). This parameter has eliminated impulse interference and high-frequency noise, but still includes spurious weight fluctuations caused by changes in ambient temperature. It is the original input data for thermal expansion and contraction correction. To correspond to the volume expansion coefficient of oil products, gasoline The value is 9.5 × 10 -4 / ℃, diesel fuel The value is 7.0 × 10 -4 / ℃, where T is the real-time ambient temperature. The standard reference temperature is 20℃. This temperature is the universal standard temperature in the field of oil measurement. At this temperature, the volume of oil is stable, which can minimize the systematic impact of temperature on weight measurement and ensure the comparability of weight data under different ambient temperatures.

[0042] In one embodiment of the present invention, the specific steps for obtaining the corrected threshold are as follows: Step S301: Extract valve switch signals in the operation state and vehicle speed signals in the driving state from the multi-dimensional data matrix, and divide the tanker truck's operating state into three core operating conditions based on signal characteristics: loading and unloading oil condition, driving condition, and stationary condition. Based on the fundamental differences in weight changes of tank trucks under different operating conditions, the operating conditions are accurately defined through key signal characteristics, providing a scenario adaptation basis for subsequent benchmark construction and threshold calculation. When the valve switch signal is in the open state, regardless of whether the vehicle speed is 0, it is determined to be a loading and unloading oil condition (because the core characteristic of loading and unloading oil operations is valve opening, which is unrelated to whether the vehicle is moving, such as fixed loading and unloading within the plant area or emergency loading and unloading scenarios while moving); when the valve switch signal is in the closed state and the vehicle speed signal is greater than 0, it is determined to be a driving condition (at this time, the weight change is only due to normal fuel consumption, without loading and unloading oil interference); when the valve switch signal is in the closed state and the vehicle speed signal is equal to 0, it is determined to be a stationary condition (at this time, the weight should theoretically remain stable, with only small fluctuations within the sensor accuracy range).

[0043] Step S302: For the oil loading and unloading operation, extract the starting weight, ending weight and operation time corresponding to each operation from the multi-dimensional data matrix, calculate the weight change rate per unit time, and build a loading and unloading weight change curve library based on multiple sets of operation data, which is used as a reference benchmark for normal weight change under this operation. A normal weight change reference library is constructed based on historical data from similar operations to accurately depict the reasonable weight change patterns during oil loading and unloading, providing a basis for dynamic threshold calculation under this operating condition. For each operation, the change rate data for a single operation is calculated using the formula: "Weight change rate per unit time = Absolute value of the difference between the starting weight and the ending weight ÷ Operation duration." At least 50 sets of operation data for the same type of oil and the same specifications of loading and unloading equipment are accumulated to construct a weight change curve library for oil loading and unloading. The curve library contains weight change rate ranges for different operating scenarios (such as full tank loading and unloading, half tank loading and unloading) and different ambient temperatures. For subsequent new operations, data from similar operating conditions in the curve library are used as normal reference benchmarks to ensure scenario adaptability of the benchmarks.

[0044] Step S303: For driving conditions, extract driving data with no oil leak records and normal sensor operation within the past thirty days from the multi-dimensional data matrix, and use linear regression algorithm to fit it to obtain the normal fuel consumption curve and the fuel consumption benchmark value per unit mileage. At the same time, divide the vehicle speed range and fit the fuel consumption correction coefficient for each vehicle speed range. Based on long-term normal driving data, a fuel consumption pattern is fitted, and speed difference corrections are applied to obtain an accurate baseline for normal weight loss, avoiding misjudging normal fuel consumption as fuel leakage. A linear regression algorithm is used to fit the filtered data, with mileage as the independent variable and weight loss as the dependent variable, to obtain a normal fuel consumption curve. The slope of the curve is the baseline fuel consumption per unit mileage (unit: kg / km), reflecting the normal fuel consumption level of the vehicle under standard operating conditions. The system is divided into three speed ranges: 0-60 km / h (low speed), 60-90 km / h (medium speed), and greater than 90 km / h (high speed). Linear regression is performed on the driving data for each range to obtain fuel consumption correction coefficients for each range (e.g., 1.2 for low speed, 1.0 for medium speed, and 1.1 for high speed). These coefficients are used for subsequent dynamic threshold speed adaptation adjustments, resolving threshold adaptation issues caused by fuel consumption differences at different speeds.

[0045] Step S304: For the static working condition, extract the weight data of the vehicle within five minutes after it stops from the multi-dimensional data matrix, calculate the average weight data within this time period as the benchmark value, and set the range from the benchmark value minus 0.1% of the full scale to the benchmark value plus 0.1% of the full scale as the normal fluctuation range under the static working condition. Based on the weight stability characteristics under static conditions, a reasonable normal fluctuation range is set to avoid misjudging minor sensor errors or slight oil sloshing as oil leaks. Weight data corresponding to static conditions is extracted from a multi-dimensional data matrix, and continuous weight data within five minutes after the vehicle comes to a stop (3000 sampling points at a 10 Hz sampling frequency) is selected to ensure the data is free from the inertial fluctuations after vehicle braking. The arithmetic mean of all weight data within these five minutes is calculated as the weight baseline value under static conditions, reflecting the true weight level of the tank in a static state. Considering the accuracy of the piezoelectric vehicle-mounted weighing sensor (0.1% of full scale), the normal fluctuation range is set from "baseline value minus 0.1% of full scale" to "baseline value plus 0.1% of full scale." This range covers the normal measurement error of the sensor while effectively distinguishing weight changes caused by minor leaks.

[0046] Step S305: Based on the normal weight change benchmark for each working condition, calculate the dynamic thresholds for driving, stationary, and loading / unloading oil working conditions respectively. Based on the normal weight change benchmark for each operating condition, and combined with parameters such as safety factor and confidence factor, an initial dynamic threshold is calculated to ensure that the threshold can cover normal fluctuations and promptly capture abnormal changes. Using the unit mileage fuel consumption benchmark as the core, combined with the monitoring time window (five minutes) and safety factor (1.5~2.0), the maximum allowable normal weight loss under driving conditions is calculated as the dynamic threshold. The safety factor is based on historical oil leakage data statistics to ensure that more than 95% of normal fuel consumption changes will not trigger the threshold. Based on the basic fluctuation value corresponding to the sensor accuracy (0.1% of full scale), combined with the historical static weight fluctuation standard deviation and confidence factor (2.58, corresponding to a 99% confidence interval), a dynamic threshold under static conditions is calculated. This threshold can effectively filter random interference under static conditions. Using the unit time weight change rate of oil loading and unloading as the core, combined with the monitoring time window (five minutes) and operational fluctuation coefficient (1.2), a dynamic threshold under oil loading and unloading conditions is calculated. The operational fluctuation coefficient is used to cover the impact of flow fluctuations during oil loading and unloading.

[0047] Step S306: Extract sensor usage time, weld ultrasonic testing level, and cumulative mileage since the last weld inspection from the multi-source raw data. Determine the sensor attenuation correction coefficient based on the sensor usage time. Determine the weld aging correction coefficient based on the weld ultrasonic testing level and cumulative mileage. Calculate the product of the two to obtain the comprehensive correction coefficient. By combining equipment aging information from vehicle maintenance data, a two-dimensional correction coefficient is constructed to achieve adaptive adjustment of dynamic thresholds and solve the problem of warning sensitivity drift caused by equipment aging. The comprehensive correction coefficient is the product of the sensor attenuation correction coefficient and the weld aging correction coefficient, with a value range of 1.0 to 1.56. The larger the coefficient, the higher the risk of leakage caused by equipment aging, and the higher the warning sensitivity required.

[0048] Step S307: Substitute the comprehensive correction coefficient into the calculation process of the dynamic threshold for each working condition, perform adaptive correction on the dynamic threshold, and finally obtain the corrected threshold for driving condition, stationary condition, and oil loading / unloading condition.

[0049] The comprehensive correction coefficient is substituted into the initial dynamic threshold calculation, and the threshold size is adjusted by coefficient correction to achieve dynamic adaptation of the warning sensitivity. The comprehensive correction coefficient and the corrected threshold are inversely proportional, that is, the higher the risk of equipment aging (the larger the coefficient), the smaller the corrected threshold and the higher the warning sensitivity, ensuring that even minor leaks in aging equipment can be detected in time; the better the equipment condition (the smaller the coefficient), the closer the corrected threshold is to the initial dynamic threshold, avoiding over-warning. The initial dynamic thresholds for driving, stationary, and loading / unloading oil conditions are divided by the comprehensive correction coefficient to obtain the corrected thresholds for each condition. These thresholds retain the adaptability of the operating conditions while incorporating considerations of equipment aging status, achieving a two-dimensional adaptation of "operating condition + equipment status".

[0050] It should be noted that the calculation formula for the aforementioned operating conditions is as follows: , in, The operating condition dynamic threshold refers to the maximum allowable normal weight loss within a set monitoring time window during operation. Exceeding this value will trigger a preliminary abnormality. For safety, the value is set between 1.5 and 2.0. This value is used to cover minor fluctuations in fuel consumption during normal driving (such as instantaneous fuel consumption fluctuations caused by changes in road conditions and differences in driving habits). The value is based on the statistical analysis of 300 sets of historical normal driving data to ensure that more than 95% of normal fuel consumption changes will not trigger the threshold, while not missing any real fuel leakage signals. The unit mileage fuel consumption benchmark value refers to the normal weight loss of a vehicle under standard operating conditions (medium speed driving, no extreme weather, and normal sensor operation), which is obtained by fitting the benchmark through a linear regression algorithm. The monitoring time window is set to 5 minutes.

[0051] The calculation formula for the static condition is as follows: , in, The dynamic threshold for static operating conditions refers to the maximum allowable fluctuation in weight from the baseline value when the weight is at rest. Exceeding this value will trigger a preliminary anomaly. The base fluctuation value corresponding to the sensor accuracy is set to 0.1% of the full scale. This value directly matches the measurement accuracy of the piezoelectric vehicle-mounted weighing sensor, which is the normal measurement error range of the sensor itself. This ensures that the threshold covers the inherent error of the sensor and avoids misjudging normal measurement fluctuations as abnormalities. The confidence coefficient is 2.58, which corresponds to the 99% confidence interval in statistics. It is used to filter out random disturbances in a static state (such as slight shaking of oil or small weight fluctuations caused by environmental vibrations), ensuring that anomalies are triggered only when weight fluctuations exceed the high probability of being within the normal range. The standard deviation of weight fluctuation under historical static conditions refers to the standard deviation of the fluctuation of 30 historical sets of static operating condition weight data extracted from a multi-dimensional data matrix. This parameter reflects the inherent fluctuation characteristics of a specific vehicle under static conditions (such as fluctuations caused by tank structure vibration and oil residue adhesion), making the threshold vehicle-specific.

[0052] The calculation formula for the oil loading and unloading conditions is as follows: , in, The dynamic threshold for oil loading and unloading conditions refers to the upper limit of the allowable weight change fluctuation within a set monitoring time window during oil loading and unloading operations. Exceeding this value will trigger a preliminary anomaly (such as abnormal weight changes caused by pipeline leakage during operations). The operation fluctuation coefficient is set to 1.2. This coefficient is used to cover normal flow fluctuations in oil loading and unloading operations (such as instantaneous flow changes caused by oil pump start-up and shutdown, valve adjustment). The value is based on the flow fluctuation statistics of 50 groups of similar loading and unloading operations to ensure that weight changes in normal operations will not trigger the threshold. The rate of change of oil weight per unit time during loading and unloading refers to the average increase or decrease in weight per minute during a single loading and unloading operation. It is obtained by extracting standard rate of change data for the corresponding working conditions from the oil loading and unloading weight change curve library. The monitoring time window is set to 5 minutes.

[0053] It should be noted that the method for determining the sensor attenuation correction coefficient is as follows: When the sensor has been used for ≤1 year, the sensor attenuation correction coefficient = 1.0; when the sensor is in brand new or nearly brand new condition, the accuracy attenuation after factory calibration is ≤3%, which meets the design requirements and does not require adjustment of the warning sensitivity.

[0054] When 1 year < sensor usage time ≤ 2 years, the sensor attenuation correction coefficient = 1.1; after long-term vibration and environmental corrosion, the accuracy of the sensor decreases by 5% to 8%, and the sensitivity needs to be slightly improved to compensate for the loss of accuracy.

[0055] When the sensor has been used for more than 2 years, the sensor attenuation correction coefficient is 1.2; the sensor accuracy attenuation is ≥10%, and the key performance parameters decline significantly, so the sensitivity needs to be significantly improved to avoid missed detections.

[0056] The method for determining the weld aging correction coefficient is as follows: When the ultrasonic testing level of the weld is ≤2 and the cumulative mileage Sw since the last test is ≤50,000 kilometers, the weld aging correction coefficient is 1.0. When the ultrasonic testing level of the weld is 3 or 50,000 km < the cumulative mileage since the last test is ≤ 100,000 km, the weld aging correction coefficient is 1.1. When the ultrasonic testing level of the weld is ≥4 or the cumulative mileage since the last test is >100,000 kilometers, the weld aging correction coefficient is 1.3. Ultrasonic testing technology is used to assess the degree of weld defects (Level 1 indicates no defects, Level 5 indicates severe defects). Higher levels indicate poorer weld structural integrity and a higher risk of leakage, making it a core indicator directly reflecting the aging state of the weld. The bumps and vibrations experienced by tanker trucks during operation exacerbate weld fatigue damage. Even with lower testing levels, aging and cracking may occur after long-term driving. Therefore, a secondary assessment based on mileage data is necessary to avoid misjudgment using a single indicator. Based on statistical data of weld failures from 100 tanker trucks of different service years, a weld testing level of 3 or mileage of 50,000 to 100,000 kilometers increases the leakage probability by 30% compared to newer vehicles; a testing level of ≥4 or mileage >100,000 kilometers increases the leakage probability by 80%, with the coefficient value positively correlated with leakage risk.

[0057] The calculation formulas for the corrected threshold values ​​for each operating condition are as follows: , in, The threshold value after correction for driving conditions is the allowable weight change value used to determine the initial abnormality under driving conditions. It is the result of the dynamic threshold after correction for equipment aging conditions. The threshold value after correction for static conditions is the allowable weight fluctuation value used to determine the initial abnormality under static conditions. The threshold value after correction for oil loading and unloading conditions is the allowable weight change value used to determine the initial abnormality under oil loading and unloading conditions. The comprehensive correction coefficient is obtained by multiplying the sensor attenuation correction coefficient and the weld aging correction coefficient. Its value ranges from 1.0 to 1.56, and it is the core parameter for quantifying the overall aging status of the equipment.

[0058] In one embodiment of the present invention, the specific steps for obtaining the oil leak anomaly determination result are as follows: Step S401: Extract real-time corrected weight data based on the multi-dimensional data matrix, and calculate the real-time weight change under different working conditions in combination with the corrected threshold. If the real-time weight change under any working condition exceeds the corrected threshold of the corresponding working condition, a preliminary anomaly is triggered. From the multi-dimensional data matrix output by the data preprocessing module, corrected weight data from current and historical continuous sampling are extracted in timestamp order. The sampling interval remains consistent with the original acquisition frequency of 10 Hz to ensure the real-time nature and continuity of the data. During the extraction process, the corresponding operating condition identifiers at the corresponding time points are synchronously associated to clarify whether the tanker truck is currently in a driving, stationary, or loading / unloading condition. This provides a basis for calculating subsequent weight changes and avoids misjudgments across different operating conditions.

[0059] The corresponding calculation logic is adopted according to the current operating condition identifier. Under the driving condition, the starting time of the five-minute monitoring time window is used as the reference point. The corrected weight data at that time is extracted. The absolute value of the difference between the current corrected weight data and the reference point weight data is taken to obtain the real-time weight change under the driving condition. This is consistent with the time window for dynamic threshold calculation to ensure the effectiveness of the comparison. Under the stationary condition, the stationary condition reference value determined in the reference construction step is extracted. The absolute value of the difference between the current corrected weight data and the reference value is taken to obtain the real-time weight change under the stationary condition. Under the oil loading and unloading condition, the starting weight data of this oil loading and unloading operation is obtained by association from the operation status data field of the multi-dimensional data matrix. The absolute value of the difference between the current corrected weight data and the starting weight data is taken to obtain the real-time weight change under the oil loading and unloading condition.

[0060] The calculated real-time weight change under the current operating condition is compared with the corresponding corrected threshold output by the baseline construction and threshold correction module. If the real-time weight change is greater than the corrected threshold for the corresponding operating condition, it indicates that the current weight change exceeds the normal fluctuation range under that operating condition. The system automatically triggers a preliminary anomaly flag and records key information such as the trigger timestamp, current operating condition, real-time weight change, and corrected threshold to provide data support for subsequent verification steps. If the real-time weight change is less than or equal to the corrected threshold for the corresponding operating condition, the current weight change is determined to be within the normal range, and no preliminary anomaly is triggered. The system then continues with the next round of data collection and comparison.

[0061] Step S402: For the weight data sequence that triggers the initial anomaly, the sequence is fitted using a quadratic polynomial fitting method to obtain a trend curve. The weight change trend is judged by the fitting coefficient of the trend curve. When the fitting coefficient shows that the weight is decreasing and the instantaneous rate of decrease exceeds the rate of decrease per unit time corresponding to the corrected threshold, it is determined to be a suspected oil leak. The system filters out the weight data sequence corresponding to the initial anomaly indicator. The time span is set from two minutes before the initial anomaly is triggered to three minutes after the trigger, ensuring that it includes complete weight change trend data before and after the anomaly. The number of sampling points in this sequence is 500, which meets the data volume requirements of quadratic polynomial fitting and avoids distortion of fitting results due to insufficient data volume.

[0062] Linear least squares was used to fit a quadratic polynomial to the selected weight data sequence. During the fitting process, time was used as the independent variable and the corrected weight data as the dependent variable to construct the fitting equation. Specifically, the fitting algorithm module built into the vehicle terminal was used. The input weight data sequence and corresponding time data were fed into the algorithm, which automatically calculated the coefficients of the quadratic term, the coefficients of the linear term, and the constant term, forming a complete trend curve equation. The core of linear least squares is to minimize the sum of squares between the fitted value and the actual measured value, ensuring that the trend curve closely matches the actual weight change trend, thus improving the reliability of the fitting results.

[0063] The weight change trend is determined by analyzing the quadratic coefficient of the fitted equation. When the quadratic coefficient is less than zero, it indicates an accelerating decrease in weight, consistent with the characteristic of rapid weight loss due to continuous oil leakage during an oil leak. When the quadratic coefficient is equal to or greater than zero, it indicates a uniform decrease or increase in weight. A uniform decrease may be due to normal oil consumption, while an increase may be due to measurement interference, neither of which matches the characteristics of an oil leak. Simultaneously, the instantaneous rate of decrease is calculated using the linear coefficient. This rate is obtained by solving the derivative of the quadratic polynomial equation; that is, the instantaneous rate of decrease equals the sum of the linear coefficient and twice the quadratic coefficient multiplied by the current time, accurately reflecting the rate of weight decrease at the current moment.

[0064] The calculated instantaneous weight drop rate is compared with the unit-time weight drop rate converted from the corrected threshold for the corresponding operating condition. The unit-time weight drop rate is calculated by dividing the corrected threshold for the corresponding operating condition by 300 seconds, which is equivalent to a five-minute monitoring time window, to obtain the maximum allowable weight drop rate per second. If the instantaneous weight drop rate is greater than the unit-time weight drop rate, it indicates that the weight drop rate is outside the normal range, further confirming the authenticity of the anomaly and classifying it as a suspected oil leak. If the instantaneous weight drop rate is less than or equal to the unit-time weight drop rate, it is classified as a normal weight change or minor disturbance, and the initial anomaly flag is lifted.

[0065] Step S403: Extract valve switch signals from the operating state, acceleration data from the driving state, and ambient temperature change from the environmental data from the multi-dimensional data matrix, and cross-validate the suspected oil leak results: if the valve is closed, the absolute value of acceleration is not greater than 0.5g, and the ambient temperature change is not greater than 5℃, then it is finally determined to be an oil leak anomaly; if the valve is open, the absolute value of acceleration is greater than 0.5g, or the ambient temperature change is greater than 5℃, then it is determined to be an interference signal, and the initial anomaly is resolved.

[0066] Three types of auxiliary data corresponding to the current suspected oil leak moment are simultaneously extracted from a multi-dimensional data matrix: valve switch signals during operation, acceleration data during driving, and ambient temperature changes in environmental data. During extraction, strict timestamp matching is performed to ensure that the auxiliary data, weight data, and trend data are from the same point in time, avoiding distortion of verification results due to time misalignment. The valve switch signals are on / off data, directly reflecting the open / closed status of the loading and unloading valves; the acceleration data are continuous values ​​reflecting the vehicle's movement, including bumps and sudden braking; and the ambient temperature change is the temperature difference between the current moment and the start of the monitoring time window, reflecting ambient temperature fluctuations.

[0067] First, verify the valve open / close signal. If the valve is open, it indicates the tanker truck is loading / unloading oil, and the weight change is normal, thus the suspected leak flag is removed. If the valve is closed, proceed to the next verification step. Second, verify the acceleration data. Calculate the absolute value of the acceleration. If the absolute value is greater than 0.5 grams, it indicates the vehicle is experiencing severe shaking, sudden braking, or a collision, causing momentary fluctuations in weight data. This is not a leak-related interference, so the suspected leak flag is removed. If the absolute value of the acceleration is less than 0.5 grams, the vehicle is driving stably, and the next verification step is performed. Finally, verify the ambient temperature change. If the temperature change is greater than 5 degrees Celsius, it indicates severe ambient temperature fluctuations, which may cause weight measurement fluctuations due to thermal expansion and contraction of the oil, indicating environmental interference. The suspected leak flag is removed. If the temperature change is less than 5 degrees Celsius, the ambient temperature is stable, and there are no temperature-related interference factors.

[0068] Only when all three conditions are met simultaneously—the valve is closed, the absolute value of the acceleration is no greater than 0.5 grams, and the change in ambient temperature is no greater than 5 degrees Celsius—can operational interference, driving interference, and environmental interference be eliminated, and the suspected oil leak be confirmed as a genuine oil leak anomaly. The system then outputs the oil leak anomaly judgment result. If any one of the conditions is not met, it is judged as an interference signal, the suspected oil leak flag is removed, the system returns to normal monitoring status, and continues to the next round of data collection and identification.

[0069] This invention employs a three-layer progressive identification logic, filtering and verifying at each layer to effectively eliminate various non-oil leakage interference factors such as oil loading and unloading operations, vehicle bumps, and temperature fluctuations. It solves the problem of high false alarm rates caused by the susceptibility of the single identification logic in existing technologies, thereby increasing the accuracy of oil leakage anomaly identification to over 95%, significantly reducing the false alarm and missed alarm rates, and ensuring that no oil leakage anomalies are missed and no non-oil leakage interference is misjudged.

[0070] In one embodiment of the present invention, the graded early warning mechanism specifically includes: Step S501: Based on the oil leak anomaly judgment result, calculate the weight drop within the corresponding time period. When the weight drop within five minutes is 0.5% to 1.0% of the full scale, trigger a first-level warning, control the vehicle terminal to start the audible and visual alarm device, and push a verification prompt message for suspected minor oil leak to the driver. Based on the multi-dimensional data matrix output by the data preprocessing module, corrected weight data over a continuous five-minute period is extracted to form a five-minute weight data sequence. The weight decrease over five minutes is obtained by calculating the difference between the corrected weight at the start and end of the sequence. This weight decrease is compared to a preset threshold range. A level one warning is triggered when the weight decrease falls between 0.5% and 1% of full scale. Full scale refers to the maximum measurement range of the piezoelectric vehicle-mounted weighing sensor, which is matched to the rated load capacity of the tanker truck to ensure that the threshold is compatible with the vehicle's actual carrying capacity.

[0071] The vehicle terminal has a built-in buzzer and a red indicator light. After the first-level warning is triggered, the vehicle terminal sends a control signal to the buzzer, causing the buzzer to sound continuously at a frequency of once per second; at the same time, it sends a control signal to the red indicator light, causing the indicator light to flash at a frequency of twice per second. Through both auditory and visual cues, the vehicle terminal ensures that the driver can quickly perceive the warning signal while driving.

[0072] The vehicle-mounted terminal sends inspection prompts to the driver in two ways. First, a warning pop-up appears on the terminal's display screen, clearly showing the warning trigger time, the vehicle's approximate location, the warning level (Level 1), and a warning of suspected minor oil leakage. Second, the terminal's built-in communication module sends a text message to the driver's pre-registered mobile phone. The message content is identical to the pop-up message, but includes inspection suggestions, explicitly informing the driver to immediately stop in a safe area and check areas prone to minor leaks, such as tank seals, loading / unloading valve interfaces, and pipe connections.

[0073] Step S502: When the weight drop exceeds 1.0% of full scale within five minutes, or exceeds 2.0% of full scale for thirty consecutive minutes, a level two warning is triggered. Based on the level one warning measures, the vehicle terminal is controlled to upload real-time weight data, vehicle location information, operating condition data and relevant parameters in the abnormal identification process to the cloud management platform. After receiving the data, the platform pushes emergency warning information to the management personnel. The weight loss calculation method is the same as that used for Level 1 warnings, calculating the weight loss over two time periods. The first time period is five minutes, calculating the weight loss over five consecutive minutes; the second time period is thirty minutes, calculating the cumulative weight loss over thirty consecutive minutes. A Level 2 warning is triggered when the weight loss over five minutes exceeds one percent of full scale, or the cumulative weight loss over thirty minutes exceeds two percent of full scale.

[0074] After a Level 2 warning is triggered, the vehicle terminal continues to maintain the Level 1 warning's audible and visual alarm status. The buzzer continues to sound, the red indicator light continues to flash, and at the same time, the display pop-up window and mobile phone SMS notification continue to be displayed to ensure that the driver remains attentive to the abnormal situation.

[0075] The vehicle-mounted terminal uploads designated data to the cloud management platform via a 4G or 5G wireless network. The uploaded data includes real-time weight data (current and corrected weight samples from the last ten minutes), vehicle location information (precise latitude and longitude data collected by the GPS module), operating condition data (current operating condition type, valve on / off status, vehicle speed, acceleration, etc.), and anomaly identification process parameters (dynamic thresholds, corrected thresholds, quadratic polynomial fitting coefficients, etc.). Data transmission employs an encryption protocol to prevent data tampering or leakage during transmission, ensuring data security.

[0076] After receiving the data, the cloud management platform identifies the corresponding management personnel account for the vehicle through a preset association mechanism. The platform simultaneously pushes emergency warning information to the management personnel in two ways: first, it sends a pop-up notification to the management personnel's management app, directly displaying the warning level, the vehicle's real-time location, the rate of weight loss, and a summary of abnormal data; second, it sends a text message to the management personnel's linked mobile phone, containing the same core information as the pop-up, along with handling suggestions, prompting the management personnel to promptly contact the driver to verify the situation on-site, and, if necessary, dispatch nearby staff to the scene to assist in the verification.

[0077] Step S503: When the weight drop exceeds 3.0% of full scale within one minute, an emergency linkage warning is triggered. After receiving the warning signal, the cloud management platform automatically searches for the nearest emergency rescue station for the vehicle and pushes the vehicle's real-time location, oil leak warning level, and related monitoring data to the rescue station to assist rescue personnel in quickly carrying out emergency response.

[0078] Extract the corrected weight data for one consecutive minute to form a one-minute weight data sequence. Calculate the difference between the corrected weight at the start and end of the sequence to obtain the weight drop within one minute. When this weight drop exceeds 3% of full scale, it is considered a major oil leak risk, triggering an emergency response warning.

[0079] The cloud management platform is built with a geographic information system. After receiving the emergency linkage warning signal, it immediately extracts the accurate longitude and latitude data of the vehicle at the current moment. Taking this data as the center, it retrieves the emergency rescue stations within a radius of 50 kilometers. During the retrieval process, the platform sorts according to the emergency handling ability level of the rescue stations and the distance from the vehicle, and preferentially selects the rescue station with strong handling ability and the shortest distance as the linkage object.

[0080] The cloud management platform sends linkage information to the selected emergency rescue stations. The information content includes the vehicle's real-time longitude and latitude accurate to the 10-meter level, the oil leakage warning level being the emergency linkage level, the real-time weight loss rate, the tank type of the oil tanker, the rated load capacity and the current estimated oil load, and the summary of multi-source monitoring data that has been collected. The information is received through the emergency dispatch system of the rescue station and is also sent to the mobile phone of the person in charge of the rescue station in the form of a text message.

[0081] After receiving the information, the rescue station immediately activates the internal emergency dispatch plan, arranges rescue vehicles equipped with special leakage plugging equipment, explosion-proof tools, oil-absorbing mats and other rescue materials, as well as professional personnel with experience in handling oil leakage of oil tankers to go to the scene. The cloud management platform continuously tracks the vehicle position and weight change data, and updates the information to the rescue station every 30 seconds to ensure that the rescue personnel can grasp the on-site situation in real time, optimize the driving route, and improve the rescue efficiency. At the same time, the platform pushes emergency handling guidelines to the driver, prompting the driver to immediately park the vehicle in a safe area, turn off the vehicle engine, set warning signs, stay away from the tank body, and wait for rescue.

[0082] See Figure 2 , this invention also provides an oil tanker oil leakage warning system based on an abnormal weight fluctuation recognition algorithm, including: A multi-source data collection module, including a piezoelectric vehicle-mounted weighing sensor, an OBD interface module, a GPS module, a solenoid valve sensor, a temperature and humidity sensor, and a maintenance data synchronization unit; the piezoelectric vehicle-mounted weighing sensor is installed at the four symmetrical support points of the oil tanker frame and the tank body, and is used to collect the real-time total weight data of the oil tanker; the OBD interface module is used to collect vehicle speed, acceleration, and braking signals; the GPS module is used to collect the vehicle's real-time position and driving mileage; the solenoid valve sensor is installed at the oil loading and unloading valve, and is used to collect the valve opening and closing state, operation timestamp, and oil loading and unloading operation instruction signals; the temperature and humidity sensor is installed outside the tank body, and is used to collect the ambient temperature and relative humidity; the maintenance data synchronization unit is used to synchronize the vehicle maintenance records through the cloud management platform, and extract the sensor usage duration, weld ultrasonic detection grade, and cumulative driving mileage since the last weld detection.

[0083] The data preprocessing module is connected to the multi-source data acquisition module via a signal. It receives the raw data from multiple sources transmitted by the multi-source data acquisition module, performs outlier removal, moving average filtering, thermal expansion and contraction correction, and data alignment on the data, and outputs a multi-dimensional data matrix. This module is connected to the multi-source data acquisition module via a CAN bus to receive the raw data from multiple sources transmitted by the acquisition module. The module has a built-in ARM Cortex-A9 processor (1GHz), 2GB DDR3 memory, and 16GB flash memory, supporting data caching and fast computing to ensure the efficiency of parallel processing of multi-channel data.

[0084] The benchmark construction and threshold correction module is connected to the data preprocessing module via a signal. It receives a multi-dimensional data matrix, divides the data into three core operating conditions: oil loading and unloading, driving, and stationary. It establishes a benchmark model for normal weight changes under each operating condition, calculates the dynamic threshold for each condition based on the benchmark model, calculates the comprehensive correction coefficient, corrects the dynamic threshold, and outputs the corrected threshold. This module is connected to the data preprocessing module via an Ethernet interface to receive the preprocessed multi-dimensional data matrix. It also establishes indirect data interaction with the multi-source data acquisition module to obtain equipment status parameters (sensor usage time, weld aging related data) from maintenance data. The module has a built-in SD card for storing the benchmark model and threshold data, supporting local data backup and updates.

[0085] The anomaly detection module is connected to the benchmark construction and threshold correction module and the data preprocessing module. It receives the corrected threshold and multi-dimensional data matrix, performs a three-layer detection logic of dynamic threshold comparison, trend fitting verification, and multi-parameter cross-validation, determines the oil leak anomaly, and outputs the judgment result. This module is connected to the benchmark construction and threshold correction module and the data preprocessing module: it receives the corrected threshold through an Ethernet interface and the multi-dimensional data matrix through an SPI interface. The module has a built-in FPGA chip (XilinxArtix-7) specifically for high-speed data comparison and fitting operations, ensuring the real-time execution of the three-layer detection logic with a computational latency of ≤100 milliseconds.

[0086] The early warning and tracing module, connected to the anomaly identification module, receives the oil leak anomaly assessment results and triggers Level 1, Level 2, or emergency response alerts based on the severity of the anomaly. This module connects to the anomaly identification module via a CAN bus to receive the oil leak anomaly assessment results. The module integrates an audible and visual alarm unit, a wireless communication unit (4G / 5G + BeiDou short message service), and a data storage unit (64GB solid-state drive + cloud synchronization interface), providing early warning triggering, information push, data storage, and tracing functions. The BeiDou short message service is used for emergency communication in scenarios without mobile network access.

[0087] The system's modules follow an orderly and collaborative process of "data acquisition → preprocessing → benchmark construction → anomaly identification → early warning and tracing": the multi-source data acquisition module provides comprehensive input to the system, the data preprocessing module improves data quality, the benchmark construction and threshold correction module provides accurate basis for identification, the anomaly identification module makes core judgments, and the early warning and tracing module completes final handling and data retention, forming a complete technical closed loop. Standardized communication protocols are used between modules to support real-time data transmission and status feedback. In the event of a module failure, a redundancy mechanism (such as local storage replacing cloud transmission) can be automatically triggered to ensure the overall availability of the system.

[0088] This system is compatible with tank trucks of varying tonnages (5-50 tons), supports monitoring of common oil products such as gasoline and diesel, and is suitable for all operating conditions including road transport and intra-plant transfers. All system hardware components meet industrial-grade standards (protection level ≥ IP65, operating temperature -40℃~85℃), enabling them to withstand harsh environments such as high temperatures, low temperatures, humidity, and dust. The software supports remote upgrades and parameter configuration, allowing adjustments to warning thresholds and information push methods based on the regulatory needs of different transportation companies, demonstrating strong adaptability to practical applications.

[0089] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for early warning of oil leaks from tank trucks based on an abnormal weight fluctuation identification algorithm, characterized in that, Includes the following steps: Collect data on tanker truck weight, driving status, operating status, environmental parameters, and vehicle maintenance, and integrate them to form multi-source raw data; Noise filtering, smoothing filtering, and thermal expansion and contraction correction are performed on the multi-source raw data, and a multi-dimensional data matrix is ​​formed after timestamp alignment. Based on the multi-dimensional data matrix, three types of working conditions are divided and a normal weight change benchmark for each working condition is established. The dynamic threshold of each working condition is calculated, and the comprehensive correction coefficient is calculated in combination with vehicle maintenance data. The dynamic threshold of each working condition is then corrected to obtain the corrected threshold. By using the corrected threshold and the multi-dimensional data matrix, dynamic threshold comparison, trend fitting verification and multi-parameter cross-validation are performed respectively to obtain the oil leak anomaly judgment result; The tiered early warning mechanism is triggered based on the oil leak anomaly determination results, and the entire chain of data is stored synchronously.

2. The method for early warning of oil leaks from tank trucks based on an abnormal weight fluctuation identification algorithm according to claim 1, characterized in that, The specific steps for collecting the multi-source raw data are as follows: Piezoelectric vehicle-mounted load cells are installed at four symmetrical support points of the tanker truck frame and tank body to collect real-time total weight data of the tanker truck at a frequency of 10 Hz. The collected weight data is transmitted to the vehicle terminal via RS485 bus. The vehicle speed, acceleration, and braking signals are collected through the vehicle's OBD interface, and the real-time vehicle location and mileage are collected through the GPS module. The collection frequency is set to be consistent with the real-time total weight data, and the collected driving status data is synchronously transmitted to the vehicle terminal. A solenoid valve sensor is installed at the oil loading and unloading valve to collect the valve opening and closing status, operation timestamp, and oil loading and unloading operation command signal, and the collected operation status data is transmitted to the vehicle terminal. Temperature and humidity sensors are installed on the outside of the tank to collect ambient temperature and relative humidity, and the collected environmental data is transmitted to the vehicle terminal. By synchronizing vehicle maintenance records through a cloud management platform, the system extracts sensor usage time, the level of the most recent ultrasonic weld inspection, and the cumulative mileage since the last weld inspection from the maintenance records. The extracted maintenance data is then transmitted to the vehicle terminal and integrated with weight data, driving status data, work status data, and environmental data to form multi-source raw data.

3. The method for early warning of oil leaks from tank trucks based on an abnormal weight fluctuation identification algorithm according to claim 1, characterized in that, The specific steps for forming a multi-dimensional data matrix are as follows: For weight data in multi-source raw data, the mean and standard deviation of the weight data series are calculated using the three-standard-deviation criterion. Impulse interference outliers that exceed the range of mean minus three standard deviations to mean plus three standard deviations are removed, and valid weight data are retained. The effective weight data after removing outliers is filtered by moving average. The sliding window size is set to five sampling points. The smoothed weight sequence is obtained by calculating the mean of five adjacent sampling points. Based on the ambient temperature data in the multi-source raw data, combined with the volume expansion characteristics of the corresponding oil, the smoothed weight sequence is corrected for thermal expansion and contraction according to the correction formula to eliminate the interference of temperature change on weight measurement and obtain the corrected weight. Driving status data and operation status data are extracted from multiple sources of raw data, and then precisely aligned with the corrected weight data according to the timestamp. This results in the integration of a multi-dimensional data matrix that includes time, weight, and auxiliary parameters.

4. The method for early warning of oil leaks from tank trucks based on an abnormal weight fluctuation identification algorithm according to claim 3, characterized in that, The correction formula is as follows: ， in, For the corrected weight, This is the current sample value in the smoothed weight sequence. To correspond to the volume expansion coefficient of oil products, gasoline The value is 9.5 × 10 -4 / ℃, diesel fuel The value is 7.0 × 10 -4 / ℃, where T is the real-time ambient temperature. The standard reference temperature is 20℃.

5. The method for early warning of oil leaks from tank trucks based on an abnormal weight fluctuation identification algorithm according to claim 1, characterized in that, The specific steps for obtaining the corrected threshold are as follows: Valve switching signals in the operational state and vehicle speed signals in the driving state are extracted from the multi-dimensional data matrix. Based on the signal characteristics, the operating state of the tanker truck is divided into three core operating conditions: loading and unloading oil condition, driving condition, and stationary condition. For oil loading and unloading operations, the starting weight, ending weight, and operation duration for each operation are extracted from a multi-dimensional data matrix. The rate of weight change per unit time is calculated, and a library of oil loading and unloading weight change curves is constructed based on multiple sets of operation data, which serves as a reference benchmark for normal weight changes under this operation. For driving conditions, driving data with no oil leak records and normal sensor operation within the past 30 days are extracted from the multi-dimensional data matrix. A linear regression algorithm is used to fit the data to obtain the normal fuel consumption curve and the fuel consumption benchmark value per unit mileage. At the same time, the vehicle speed range is divided, and the fuel consumption correction coefficient for each vehicle speed range is fitted. For stationary conditions, the weight data within five minutes after the vehicle comes to a stop is extracted from the multi-dimensional data matrix. The average weight data within this time period is calculated as the benchmark value. The range from the benchmark value minus 0.1% of the full scale to the benchmark value plus 0.1% of the full scale is set as the normal fluctuation range under stationary conditions. Based on the normal weight change benchmark for each working condition, the dynamic thresholds for driving, stationary, and loading / unloading oil conditions are calculated respectively. The sensor usage time, weld ultrasonic testing level, and cumulative mileage since the last weld inspection are extracted from multi-source raw data. The sensor attenuation correction coefficient is determined based on the sensor usage time, and the weld aging correction coefficient is determined based on the weld ultrasonic testing level and cumulative mileage. The product of the two is calculated to obtain the comprehensive correction coefficient. The comprehensive correction coefficient is substituted into the calculation process of the dynamic threshold for each working condition, and the dynamic threshold is adaptively corrected to finally obtain the corrected threshold for driving condition, stationary condition, and oil loading and unloading condition.

6. The method for early warning of oil leaks from tank trucks based on an abnormal weight fluctuation identification algorithm according to claim 5, characterized in that, The calculation formula for the operating conditions is as follows: ， in, To exercise the dynamic threshold of operating conditions, For safety margin, a value of 1.5 to 2.0 is used. This is the baseline value for fuel consumption per unit mileage. The monitoring time window is set to 5 minutes. The calculation formula for the static condition is as follows: ， in, The dynamic threshold for static operating conditions. This is the baseline fluctuation value corresponding to the sensor's accuracy, taken as 0.1% of full scale. This is the confidence coefficient, with a value of 2.

58. The standard deviation of weight fluctuation at historical static state; The calculation formula for the oil loading and unloading conditions is as follows: ， in, The dynamic threshold for oil loading and unloading conditions. This is the operational fluctuation coefficient, with a value of 1.

2. This represents the rate of change in oil weight per unit time during loading and unloading. The monitoring time window is set to 5 minutes.

7. The method for early warning of oil leaks from tank trucks based on an abnormal weight fluctuation identification algorithm according to claim 6, characterized in that, The method for determining the sensor attenuation correction coefficient is as follows: When the sensor's usage time is ≤1 year, the sensor attenuation correction coefficient = 1.0; When 1 year < sensor usage time ≤ 2 years, the sensor attenuation correction coefficient = 1.1; When the sensor has been used for more than 2 years, the sensor attenuation correction coefficient is 1.

2. The method for determining the weld aging correction coefficient is as follows: When the ultrasonic testing level of the weld is ≤2 and the cumulative mileage Sw since the last test is ≤50,000 kilometers, the weld aging correction coefficient is 1.

0. When the ultrasonic testing level of the weld is 3 or 50,000 km < the cumulative mileage since the last test is ≤ 100,000 km, the weld aging correction coefficient is 1.

1. When the ultrasonic testing level of the weld is ≥4 or the cumulative mileage since the last test is >100,000 kilometers, the weld aging correction coefficient is 1.

3. The calculation formulas for the corrected threshold values ​​for each operating condition are as follows: ， in, The threshold value is the value after calibration for driving conditions. The threshold value is the value after correction for static operating conditions. The threshold value is the corrected value for loading and unloading oil conditions. This is the comprehensive correction factor.

8. The method for early warning of oil leaks from tank trucks based on an abnormal weight fluctuation identification algorithm according to claim 3, characterized in that, The specific steps for obtaining the oil leak anomaly determination result are as follows: Real-time corrected weight data is extracted based on a multi-dimensional data matrix. Combined with the corrected threshold, the real-time weight change under different working conditions is calculated. If the real-time weight change under any working condition exceeds the corrected threshold for the corresponding working condition, a preliminary anomaly is triggered. For the weight data sequence that triggers the initial anomaly, a quadratic polynomial fitting method is used to fit the sequence to obtain a trend curve. The fitting coefficient of the trend curve is used to determine the weight change trend. When the fitting coefficient shows that the weight is decreasing and the instantaneous rate of decrease exceeds the rate of decrease per unit time corresponding to the corrected threshold, it is judged as a suspected oil leak. Valve switch signals in the operating state, acceleration data in the driving state, and ambient temperature change in the environmental data are extracted from the multi-dimensional data matrix to cross-validate suspected oil leak results: if the valve is closed, the absolute value of acceleration is no greater than 0.5g, and the ambient temperature change is no greater than 5℃, then it is finally determined to be an oil leak anomaly; if any of the following conditions are met: the valve is open, the absolute value of acceleration is greater than 0.5g, or the ambient temperature change is greater than 5℃, then it is determined to be an interference signal, and the preliminary anomaly is eliminated.

9. The method for early warning of oil leaks from tank trucks based on an abnormal weight fluctuation identification algorithm according to claim 1, characterized in that, The tiered early warning mechanism specifically includes: Based on the oil leak anomaly determination results, the weight loss within the corresponding time period is calculated. When the weight loss within five minutes is between 0.5% and 1.0% of the full scale, a level one warning is triggered, controlling the vehicle terminal to activate the audible and visual alarm device and pushing a verification prompt message for suspected minor oil leak to the driver. When the weight drop exceeds 1.0% of full scale within five minutes, or exceeds 2.0% of full scale for thirty consecutive minutes, a level two warning is triggered. In addition to implementing level one warning measures, the vehicle terminal is controlled to upload real-time weight data, vehicle location information, operating condition data, and relevant parameters in the anomaly identification process to the cloud management platform. After receiving the data, the platform pushes emergency warning information to the management personnel. When the weight drop exceeds 3.0% of full scale within one minute, an emergency linkage warning is triggered. After receiving the warning signal, the cloud management platform automatically searches for the nearest emergency rescue station and pushes the vehicle's real-time location, oil leak warning level, and related monitoring data to the rescue station to assist rescue personnel in quickly carrying out emergency response.

10. An oil tanker leak early warning system based on an abnormal weight fluctuation identification algorithm, characterized in that, include: The multi-source data acquisition module includes a piezoelectric vehicle-mounted weighing sensor, an OBD interface module, a GPS module, a solenoid valve sensor, a temperature and humidity sensor, and a maintenance data synchronization unit. The data preprocessing module is connected to the multi-source data acquisition module and is used to receive multi-source raw data transmitted by the multi-source data acquisition module, perform outlier removal, moving average filtering, thermal expansion and contraction correction and data alignment processing on it, and output a multi-dimensional data matrix. The benchmark construction and threshold correction module is connected to the data preprocessing module. It is used to receive multi-dimensional data matrices, divide the oil loading and unloading conditions, driving conditions and stationary conditions into three core conditions, establish a benchmark model for normal weight change in each condition, calculate the dynamic threshold for each condition based on the benchmark model, calculate the comprehensive correction coefficient and correct the dynamic threshold, and output the corrected threshold. The anomaly identification module is connected to the benchmark construction and threshold correction module and the data preprocessing module respectively. It is used to receive the corrected threshold and multi-dimensional data matrix, perform a three-layer identification logic of dynamic threshold comparison, trend fitting verification and multi-parameter cross-validation, determine the oil leak anomaly and output the judgment result. The early warning and tracing module is connected to the anomaly identification module to receive the results of oil leak anomaly determination and trigger a first-level early warning, a second-level early warning, or an emergency linkage early warning based on the severity of the anomaly.

Images