Methods, devices, equipment, and media for identifying abnormalities in annual inspections of heavy-duty vehicles based on remote OBD monitoring.

By acquiring remote OBD monitoring time-series data, calculating the deviation rate, and combining it with national standards, machine learning and time-series anomaly detection models are used to identify anomalies in the annual inspection of heavy vehicles. This solves the problem that existing technologies cannot accurately identify anomalies in the annual inspection, and achieves high-precision, quantifiable anomaly judgment, thereby improving the intelligence and fairness of the annual inspection supervision.

CN122654926BActive Publication Date: 2026-11-17BEIJING MUNICIPAL ENVIRONMENTAL MONITORING CENT
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611145131.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-07-30
Publication Date
2026-11-17
Estimated Expiration
2046-07-30

AI Technical Summary

Technical Problem

Existing technologies cannot accurately and effectively identify abnormalities in the annual inspection of heavy-duty vehicles. They suffer from problems such as non-standard inspection procedures, emissions cheating, distorted test results, temporary repairs, and OBD power-off shielding. They lack remote and reliable data cross-validation, cannot handle fluctuations in operating conditions and data noise, cannot identify hidden anomalies, and have not adapted algorithms to the structural differences between diesel and gas vehicles, resulting in poor universality of judgments. The anomaly identification results are neither quantifiable nor interpretable.

Method used

By acquiring remote OBD monitoring time-series data, the deviation rate is calculated based on the structural differences between diesel and gas vehicles. Abnormal scenarios are identified in conjunction with national standard annual inspection requirements. Machine learning algorithms and time-series anomaly detection models are used to correct the anomaly score. Multi-layer algorithms are integrated to output the final anomaly score, adapting to the differentiated structural requirements of diesel and gas vehicles, and achieving high-precision identification of both explicit and implicit annual inspection anomalies.

Benefits of technology

It achieves high-precision identification of both explicit and implicit annual inspection anomalies, outputs quantifiable and traceable anomaly judgment results, adapts to the differentiated structural needs of diesel and gas vehicles, improves the intelligence and precision of heavy vehicle annual inspection supervision, and ensures the authenticity and impartiality of annual inspection results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122654926B_ABST
    Figure CN122654926B_ABST
Patent Text Reader

Abstract

This invention provides a method, device, equipment, and medium for identifying annual inspection anomalies in heavy-duty vehicles based on remote OBD monitoring, relating to the field of on-board diagnostic data processing technology. The method involves: using remote OBD monitoring time-series data as the base data, calculating the deviation rate between this data and the corresponding detection values ​​in the annual inspection data of the target heavy-duty vehicle; then obtaining a first annual inspection anomaly score based on anomaly scores for different abnormal inspection scenarios according to national standard annual inspection requirements; correcting the first annual inspection anomaly score using a machine learning algorithm to obtain a second annual inspection anomaly score; inputting the remote OBD monitoring time-series data into a time-series anomaly detection model to output a third annual inspection anomaly score; and fusing the first, second, and third annual inspection anomaly scores to obtain a final annual inspection anomaly score, thereby determining the anomaly level of the annual inspection result and / or the annual inspection process. This solution improves the accuracy and effectiveness of annual inspection anomaly identification.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of motor vehicle emission testing and on-board diagnostic data processing technology, and in particular to a method, device, equipment and medium for identifying abnormalities in the annual inspection of heavy-duty vehicles based on remote OBD monitoring. Background Technology

[0002] Current annual inspections of heavy-duty vehicles commonly suffer from problems such as non-standard inspection procedures, emissions cheating, distorted test results, temporary repairs, and OBD (On-Board Diagnostics) system power failure. Traditional inspections rely solely on on-site test data and lack cross-validation with reliable remote data. Existing technologies have the following shortcomings:

[0003] Using only fixed threshold rules cannot handle fluctuations in operating conditions and data noise, resulting in a high false judgment rate. It lacks in-depth time-series analysis capabilities and cannot identify hidden anomalies such as pre-exam temporary repairs, segmented cheating, and soft cheating. Lack of long-term, continuous, and real-world driving data support; The algorithm did not adapt to the structural differences between diesel and natural gas vehicles, resulting in poor general applicability. Anomaly identification results are neither quantifiable nor interpretable.

[0004] Therefore, existing technologies cannot accurately and effectively identify abnormalities in the annual inspection of heavy vehicles. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a method for identifying abnormalities in the annual inspection of heavy-duty vehicles based on remote OBD monitoring, to solve the technical problem in the prior art that it is impossible to accurately and effectively identify abnormalities in the annual inspection of heavy-duty vehicles. The method includes: The remote OBD monitoring time series data of the target heavy vehicle during the annual inspection period is obtained. Based on the structural differences between diesel vehicles and gas vehicles, the deviation rate between the real-time value of each data item in the remote OBD monitoring time series data and the detection value of the corresponding data item in the annual inspection detection data of the target heavy vehicle is calculated. According to the national standard annual inspection requirements, based on the remote OBD monitoring time series data, the annual inspection data and the deviation rate, different abnormal annual inspection scenarios with abnormal situations are identified and abnormal scores are determined. The first annual inspection abnormal score is obtained based on the abnormal scores of different abnormal annual inspection scenarios. Based on the remote OBD monitoring time series data, the deviation rate, and the abnormal situations that occur in different abnormal annual inspection scenarios, the first annual inspection abnormal score is corrected by machine learning algorithm to obtain the second annual inspection abnormal score. The remote OBD monitoring time-series data is input into the time-series anomaly detection model. The time-series anomaly detection model identifies hidden anomalies in the annual inspection and outputs a third annual inspection anomaly score based on the anomaly scores of different hidden anomalies. The time-series anomaly detection model includes a cascaded TCN (Temporal Convolutional Network) dilated convolution module, a two-layer LSTM (Long Short-Term Memory) network, an AdditiveAttention layer, a Dropout (random deactivation, i.e., network regularization layer) layer, and a fully connected layer. The first annual inspection abnormality score, the second annual inspection abnormality score, and the third annual inspection abnormality score are fused to obtain the final annual inspection abnormality score. Alternatively, when a specified abnormal annual inspection scenario is identified, a preset final annual inspection abnormality score is directly output. The abnormality level of the annual inspection result and / or the annual inspection process is determined based on the magnitude of the final annual inspection abnormality score. If the final annual inspection abnormality score is zero, the annual inspection result and / or the annual inspection process is determined to be normal.

[0006] This invention also provides a device for identifying abnormalities in the annual inspection of heavy-duty vehicles based on remote OBD monitoring, to solve the technical problem in the prior art that it is impossible to accurately and effectively identify abnormalities in the annual inspection of heavy-duty vehicles. The device includes: The deviation rate calculation module is used to acquire remote OBD monitoring time series data of the target heavy vehicle during the annual inspection period. Based on the structural differences between diesel vehicles and gas vehicles, it calculates the deviation rate between the real-time value of each data item in the remote OBD monitoring time series data and the detection value of the corresponding data item in the annual inspection detection data of the target heavy vehicle. The first identification module is used to identify different abnormal annual inspection scenarios and determine the abnormal score according to the remote OBD monitoring time series data, the annual inspection data and the deviation rate, in accordance with the national standard annual inspection requirements, and to obtain the first annual inspection abnormal score based on the abnormal score of different abnormal annual inspection scenarios. The second identification module is used to correct the first annual inspection abnormality score based on the remote OBD monitoring time series data, the deviation rate, and the abnormal situations that occur in different abnormal annual inspection scenarios, and to obtain the second annual inspection abnormality score by using a machine learning algorithm. The third identification module is used to input the remote OBD monitoring time series data into the time series anomaly detection model. The time series anomaly detection model identifies hidden anomalies in the annual inspection and outputs a third annual inspection anomaly score based on the anomaly scores of different hidden anomalies. The time series anomaly detection model includes a cascaded TCN dilated convolution module, a two-layer LSTM network, an AdditiveAttention layer, a Dropout layer, and a fully connected layer. An anomaly identification module is used to fuse the first annual inspection anomaly score, the second annual inspection anomaly score, and the third annual inspection anomaly score to obtain a final annual inspection anomaly score. Alternatively, when a specified abnormal annual inspection scenario is identified, a preset final annual inspection anomaly score is directly output. The anomaly level of the annual inspection result and / or the annual inspection process is determined based on the magnitude of the final annual inspection anomaly score. If the final annual inspection anomaly score is zero, the annual inspection result and / or the annual inspection process is determined to be normal.

[0007] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-mentioned methods for identifying abnormalities in the annual inspection of heavy vehicles based on remote OBD monitoring, thereby solving the technical problem in the prior art that it is impossible to accurately and effectively identify abnormalities in the annual inspection of heavy vehicles.

[0008] This invention also provides a computer-readable storage medium storing a computer program that executes any of the above-described methods for identifying abnormalities in the annual inspection of heavy vehicles based on remote OBD monitoring, in order to solve the technical problem in the prior art that it is impossible to accurately and effectively identify abnormalities in the annual inspection of heavy vehicles.

[0009] Compared with the prior art, the beneficial effects that at least one of the above-mentioned technical solutions adopted in the embodiments of this specification can achieve include at least the following: using remote OBD real-time / historical time-series data as the sole basic data source, and through a three-layer heterogeneous fusion algorithm that combines abnormal annual inspection scenario identification according to national standard annual inspection requirements with machine learning algorithms and time-series anomaly detection models, high-precision identification of explicit and hidden annual inspection anomalies is achieved. This adapts to the differentiated structural needs of diesel vehicles and gas vehicles, strictly follows the relevant national standard annual inspection requirements such as GB3847-2018, and outputs quantifiable, traceable, and directly data-based evidence for law enforcement. This fills the gap in the existing technology, helps to improve the intelligence and precision of heavy vehicle annual inspection supervision, and helps to ensure the authenticity and fairness of annual inspection results. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart of a method for identifying abnormalities in the annual inspection of heavy vehicles based on remote OBD monitoring, provided by an embodiment of the present invention; Figure 2 This is a structural block diagram of a computer device provided in an embodiment of the present invention; Figure 3 This is a structural block diagram of a heavy vehicle annual inspection abnormality identification device based on remote OBD monitoring provided in an embodiment of the present invention. Detailed Implementation

[0012] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0013] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] In this embodiment of the invention, a method for identifying abnormalities in the annual inspection of heavy-duty vehicles based on remote OBD monitoring is provided, such as... Figure 1 As shown, the method includes: Step S101: Obtain remote OBD monitoring time series data of the target heavy vehicle during the annual inspection period. Based on the structural differences between diesel vehicles and gas vehicles, calculate the deviation rate between the real-time value of each data item in the remote OBD monitoring time series data and the detection value of the corresponding data item in the annual inspection detection data of the target heavy vehicle. Step S102: In accordance with the national standard annual inspection requirements, based on the remote OBD monitoring time series data, the annual inspection detection data, and the deviation rate, identify different abnormal annual inspection scenarios where abnormal situations occur and determine the abnormal scores. Obtain the first annual inspection abnormal score based on the abnormal scores of different abnormal annual inspection scenarios. Step S103: Based on the remote OBD monitoring time series data, the deviation rate, and the abnormal situations that occur in different abnormal annual inspection scenarios, the first annual inspection abnormal score is corrected by machine learning algorithm to obtain the second annual inspection abnormal score. Step S104: Input the remote OBD monitoring time series data into the time series anomaly detection model. The time series anomaly detection model identifies hidden anomalies in the annual inspection and outputs a third annual inspection anomaly score based on the anomaly scores of different hidden anomalies. The time series anomaly detection model includes a cascaded TCN dilated convolution module, a two-layer LSTM network, an AdditiveAttention layer, a Dropout layer, and a fully connected layer. Step S105: Merge the first annual inspection abnormality score, the second annual inspection abnormality score, and the third annual inspection abnormality score to obtain the final annual inspection abnormality score; or, when a specified abnormal annual inspection scenario is identified, directly output the preset final annual inspection abnormality score, determine the abnormality level of the annual inspection result and / or the annual inspection process based on the magnitude of the final annual inspection abnormality score, and if the final annual inspection abnormality score is zero, determine that the annual inspection result and / or the annual inspection process is normal.

[0015] In practice, the remote OBD monitoring time-series data includes: engine speed (r / min), load (%), vehicle speed (km / h), coolant temperature (°C), NOx concentration (ppm), DPF differential pressure (kPa), urea level (%), MIL status (0=off, 1=on), fault code DTC (compliant with ISO15031-6 standard), diagnostic ready items (emission-related ready items as specified in GB3847-2018), data online status (0=offline, 1=online), mileage (km), timestamp (format: YYYY-MM-DDHH:MM:SS), etc. The sampling frequency of the remote OBD monitoring time-series data is fixed at 1Hz to ensure the continuity and consistency of the time-series data.

[0016] In practice, after obtaining the remote OBD monitoring time-series data, before calculating the deviation rate between the real-time value of each data item in the remote OBD monitoring time-series data and the corresponding detection value in the annual inspection data of the target heavy vehicle, the remote OBD monitoring time-series data can be preprocessed as follows: (1) Outlier removal algorithm (3σ principle) For each data point x in the time-series data of remote OBD monitoring, the mean μ and standard deviation σ are calculated for each dimension. The specific formula is as follows: μ=(1 / N)·Σx i(i=1,2,...,N, where N is the total number of data points for each data point x in a single dimension, and i represents the i-th data point); σ=([(1 / N)·Σ(x i μ)²])0.5 (i=1,2,...,N); Satisfy x i <μ 3σ or x i Points with a value greater than μ+3σ are identified as outliers and removed directly. An outlier marking mechanism is also added to record the timestamp, value, and dimension of the outlier for subsequent anomaly tracing. For key dimensions such as rotational speed and NOx, if three or more consecutive data points are outliers, the time period is marked as an "abnormal data period" and will not be included in subsequent coupling comparison and anomaly determination.

[0017] (2) Temporal filtering denoising algorithm A 5-second sliding window mean filter is used to smooth the remote OBD monitoring time-series data. The optimized filtering logic for different dimensions of data is as follows: y i =(x i 2+x i 1+x i +x i+1 +x i+2 ) / 5 (i≥3, i≤N-2); Boundary handling: When i=1, y1=(x1+x2+x3) / 3; when i=2, y2=(x1+x2+x3+x4) / 4; When i=N-1, y N-1 =(x N-3 +x N-2 +x N-1 +x N ) / 4; when i=N, y N =(x N-2 +x N-1 +x N ) / 3; This algorithm is used to filter out instantaneous fluctuations in speed, NOx, and DPF differential pressure. The speed fluctuation threshold is set to 500 r / min (instantaneous changes > 500 r / min are considered fluctuation interference), the NOx fluctuation threshold is set to 500 ppm, and the DPF differential pressure fluctuation threshold is set to 5 kPa, ensuring the filtered data is accurate. i It retains the actual changes in operating conditions while eliminating invalid noise.

[0018] (3) Missing value handling algorithm Missing rate ≤ 5%: Linear interpolation is used for completion, with the interpolation formula being x.k =x k-1 +(x k+1 -x k-1 x × (k - (k-1)) / ((k+1) - (k-1)), where k is the index of the missing data point, x k-1 x k+1 The missing data points x are respectively k Valid data points are adjacent to each other; if the missing data point is at the beginning of the sequence (k=1), forward padding (x1=x2) is used; if it is at the end of the sequence (k=N), backward padding (x1=x2) is used. N =x N-1 ).

[0019] Missing rate > 5%: Marked as abnormal time series data of remote OBD monitoring, and not included in high-precision calculation; at the same time, the missing time period, missing dimension and missing rate are recorded as the input basis for subsequent anomaly judgment (scenario 15).

[0020] (4) Time alignment algorithm Based on the start time T0 (accurate to the second) of the annual inspection period, the remote OBD monitoring time series data window T is extracted, with the specific range as follows: T∈[T0 3600s, T0+3600s] The time alignment error is ≤30s. If the remote OBD timestamp and the annual inspection timestamp deviate by more than 30s, the time synchronization algorithm (based on NTP network time protocol) is used for calibration first. If the deviation still exceeds 30s after calibration, it is marked as "time alignment failure" and only the VIN coupling matching result is used for subsequent simplified judgment. The aligned data forms the key reliable time period of the annual inspection for anomaly identification. It is resampled at a frequency of 1Hz to ensure that the time series data corresponds accurately with the annual inspection period.

[0021] (5) VIN Coupled Matching Algorithm Using the VIN (Vehicle Identification Number, 17 characters) as the unique key, the remote OBD monitoring time-series data is matched one-to-one with the annual inspection data. The matching rules are as follows: if the VINs are completely identical, the match is considered successful. If the VINs do not match, a secondary key is used: license plate number + engine model for secondary matching. If both are completely identical, the match is considered successful. If they still do not match, the algorithm is marked as coupling failure. When coupling fails, the algorithm automatically terminates the high-precision anomaly judgment and only outputs the "coupling failure" flag and the reason (VIN mismatch / license plate number + engine model mismatch), without participating in the subsequent anomaly score calculation.

[0022] In specific implementation, before calculating the deviation rate between the real-time value of each data item in the remote OBD monitoring time-series data and the corresponding detection value in the annual inspection data of the target heavy-duty vehicle, in order to adapt to the differentiated structural requirements of diesel and gas vehicles, a method is proposed to perform a coupled comparison of data items or data dimensions between the remote OBD monitoring time-series data and the annual inspection data, distinguishing between different vehicle models. This is to further calculate the deviation rate between the real-time value and the annual inspection value for each data item or data dimension. For example: (1) Data dimensions or data items coupled to diesel vehicles: 1) Rotational speed → Test condition compliance: The average rotational speed in the remote OBD monitoring time series data is compared with the average rotational speed in the annual inspection data. If the deviation rate is ≤10%, the test condition is considered compliant. 2) DPF pressure difference → smoke opacity compliance: The average DPF (Diesel Particulate Filter) pressure difference in the remote OBD monitoring time series data is negatively correlated with the smoke opacity detection value in the annual inspection data. The smaller the DPF pressure difference, the greater the possibility that the DPF is leaking, penetrating, or removed. If the annual inspection smoke opacity is qualified in this case, the risk of abnormality is higher (because DPF failure should have caused the smoke opacity to exceed the standard, and the qualified result is suspected of being abnormal). 3) NOx concentration → Annual inspection NOx consistency: The deviation rate between the corrected NOx concentration value in the remote OBD monitoring time series data and the NOx detection value in the annual inspection data is ≤30%, which is considered consistent; 4) Urea consumption → SCR (Selective Catalytic Reduction) effectiveness: The urea consumption rate in the time series data of remote OBD monitoring is positively correlated with NOx emission reduction. If the consumption rate is too low, the SCR may fail. 5) MIL (Malfunction Indicator Lamp) / Fault Code → OBD Compliance: If there are no emission-related fault codes and the MIL is off in the remote OBD monitoring time series data, then OBD is considered compliant.

[0023] (2) Data dimensions or data items coupled to gas-powered vehicles: Gas-powered vehicles do not have SCR / DPF / NOx sensors, so the algorithm automatically skips related coupling items. The specific coupling data dimensions or data items are as follows: 1) Speed ​​→ Test condition compliance: Same as the judgment rules for diesel vehicles; 2) EGR opening → Emission effectiveness: If there are no EGR (Exhaust Gas Recirculation) related fault codes in the remote OBD monitoring time series data, and the EGR diagnostic readiness items are normal, then the EGR system is considered to be working effectively. 3) Oxygen sensor difference → TWC (Three-Way Catalyst) catalytic efficiency: If the difference between the values ​​of the front oxygen sensor and the rear oxygen sensor in the remote OBD monitoring time series data is ≥0.1V, then the TWC is considered effective; 4) MIL / Fault Code → OBD Compliance: Same as the rules for determining diesel vehicle compliance.

[0024] In practice, for diesel vehicles and natural gas vehicles, based on the aforementioned coupled data dimensions or data items, the deviation rate between the real-time values ​​and the annual inspection values ​​can be further calculated for each data item or data dimension using the following formula: Deviation rate R (%) = | Real-time value in OBD Annual inspection results | / Annual inspection results × 100% The deviation rate R for each data item or data dimension can be used for subsequent anomaly detection.

[0025] In addition, this application also proposes a method for correcting the deviation rate: when the annual inspection value is 0 (e.g., no fault code, smoke opacity is 0), if the real-time value in the OBD is >0, the deviation rate is determined to be 100%; when the real-time value in the OBD is 0 and the annual inspection value is >0, the deviation rate is also determined to be 100%; the deviation rate is retained to two decimal places to ensure calculation accuracy.

[0026] In practical implementation, in order to accurately identify annual inspection abnormalities according to national standard annual inspection requirements, this application proposes the concept of abnormal annual inspection scenarios, and performs abnormality identification and outputs the first annual inspection abnormality score, for example, Identify different abnormal annual inspection scenarios and determine abnormal scores, including: The identified abnormal inspection scenarios include: failure to test at the prescribed engine speed, abnormal DPF pressure difference but passing the smoke inspection, excessive NOx data deviation, urea not being consumed for a long time but passing the NOx inspection, MIL light being on or exceeding the ready level but still passing the inspection, emission system failure but passing the inspection, testing with the vehicle off, abnormal test duration, emission testing before OBD testing, uncalibrated testing equipment, insufficient smoke testing attempts, temporary repairs before the inspection, remote OBD data blocking or power failure, mismatched vehicle basic information, and missing or abnormally fluctuating remote OBD monitoring time-series data. For example, the engine speed sequence in the remote OBD monitoring time series data during the annual vehicle inspection period is read (sampling frequency 1Hz, test duration ≥60s, less than 60s is marked as abnormal test duration, associated with scenario 8); the mean μ, standard deviation σ, and fluctuation coefficient CV=σ / μ of the engine speed are calculated based on the engine speed sequence; then it is determined that: if the continuous proportion of the engine speed in the remote OBD monitoring time series data that is not within the national standard engine speed range (such as diesel vehicle free acceleration test speed: 1800±200r / min; load deceleration test speed: 2200±200r / min) is ≥60% of the annual inspection engine speed test duration, or if the continuous proportion of the engine speed with fluctuation coefficient >5% in the remote OBD monitoring time series data is ≥50% of the engine speed test duration, and the annual inspection engine speed test conclusion is qualified, then the abnormal annual inspection scenario is identified as failure to test the engine speed according to regulations (i.e., abnormal annual inspection scenario 1) and the corresponding abnormal score (such as 3 points) is output, and the abnormal engine speed period, the deviation value between the actual engine speed range and the national standard range are recorded as evidence of abnormality; If the mean Pmean of the DPF pressure difference (sampling time ≥ 30 min, less than 30 min will not be considered in this scenario) in the remote OBD monitoring time series data is < 5 kPa (indicated as DPF missing / through, further evidence of abnormality is provided by smoke opacity compliance) or > 30 kPa (indicated as DPF blockage) lasts for ≥ 30 min, and the annual smoke opacity is compliant (e.g., smoke opacity < 0.5 m)... - If the annual inspection scenario is identified as DPF pressure difference abnormal but smoke opacity annual inspection qualified (i.e., abnormal annual inspection scenario 2), the corresponding abnormal score (e.g., 3 points) will be output. In addition, if the DPF pressure difference continues to fluctuate between 5 and 30 kPa, but the fluctuation coefficient CV > 15%, it will be marked as DPF working abnormal, and will not be judged as this scenario for the time being, but will only be used as the input feature of the subsequent XGBoost model.

[0027] If the target heavy-duty vehicle is a diesel vehicle, and the NOx deviation rate R in the remote OBD monitoring time series data is greater than 30% and the NOx concentration value after environmental correction is greater than 1.5 × the NOx concentration value of the annual inspection, and the annual inspection NOx is qualified, then the abnormal annual inspection scenario is identified as excessive NOx data deviation (i.e., abnormal annual inspection scenario 3), and the corresponding abnormal score is output; for example, the correction formula for environmental correction is as follows: NOx-corr=NOx- OBD ×(25 / (T+273))×(P / 101.3) Where NOx-corr is T, which is the NOx concentration value after environmental correction, and NOx- OBD The NOx concentration value in the time series data of remote OBD monitoring is the ambient temperature (unit: °C) collected by remote OBD, and P is the ambient pressure (unit: kPa, remote OBD collection or default 101.3 kPa). If NOx-corr < 0.5 × annual inspection NOx concentration value and deviation rate R > 30%, it is marked as NOx data distortion, not considered in this scenario, and included in the abnormal data situation in scenario 15. If the target heavy-duty vehicle is a diesel vehicle, and the urea consumption rate r < 0.3% / 100km in the remote OBD monitoring time series data, the mileage within the first preset time period (e.g., 30 days) is ≥ 500km, the annual NOx inspection is qualified, and there is no urea replenishment record in the first preset time period in the remote OBD monitoring time series data (e.g., no replenishment feature of liquid level rise ≥ 10% in the remote OBD monitoring time series data, and no remote OBD data association record for the corresponding replenishment period), then the abnormal annual inspection scenario is identified as urea not being consumed for a long time but the NOx annual inspection is qualified (i.e., abnormal annual inspection scenario 4), and the corresponding abnormal score is output; where r = (initial liquid level) Current fluid level) / Mileage (Initial fluid level is the value collected remotely by OBD 30 days ago, current fluid level is the value collected 1 hour before the annual inspection, and mileage is the cumulative mileage recorded by remote OBD within 30 days). In addition, if the urea level fluctuates by ≤5% (within 30 days) and there is no record of urea replenishment (no sudden increase in level via remote OBD), it is directly determined that the urea is not consumed and there is no need to calculate the consumption rate; if there is a record of urea replenishment, but the consumption rate after replenishment is still <0.3% / 100km, and the mileage in 30 days is ≥500km and the annual inspection NOx is qualified, it is also determined to be abnormal in this scenario (i.e., abnormal annual inspection scenario 4).

[0028] If the MIL is lit in the remote OBD monitoring time series data, there are emission-related fault codes (such as fault code types that conform to emission-related codes in ISO15031-6, such as P0401, P2047, etc.), and the annual inspection conclusion is qualified; or, if there are ≥3 unready items in the remote OBD monitoring time series data (for example, for gas-powered vehicles, SCR / DPF / NOx related ready items are removed, and only EGR, TWC, oxygen sensor, etc. ready items are counted), and the annual inspection conclusion is qualified, then an abnormal annual inspection scenario is identified as MIL lit or unready items exceeding the standard but still passing the annual inspection (i.e., abnormal annual inspection scenario 5), and the corresponding abnormal score is output; the unready item statistics time is the monitoring time series data collected by remote OBD within 10 minutes before the annual inspection. If the unready items become ≤2 items during the annual inspection process, it is not judged as this scenario.

[0029] If the target heavy-duty vehicle is a diesel vehicle, and the remote OBD monitoring time series data shows SCR efficiency <80%, DPF regeneration abnormality (e.g., more than 10 regeneration attempts within 30 days, or more than 3 regeneration failures, with the remote OBD recording regeneration status as 0 = failure and 1 = success), or EGR system failure (EGR-related fault codes exist, and the EGR diagnostic readiness item is abnormal, indicating EGR system failure), then the abnormal annual inspection scenario is identified as emission system failure but the annual inspection is passed, and the corresponding abnormal score is output; if the target heavy-duty vehicle is a gas-powered vehicle, and the annual inspection is passed, and the remote OBD monitoring time series data shows EGR system failure... If the following conditions are met: (EGR-related fault codes exist and the EGR diagnostic readiness item is abnormal, indicating EGR system failure), TWC temperature difference < 50℃, or no difference between the front and rear oxygen sensors (e.g., difference < 0.05V, lasting ≥ 3min), then the abnormal annual inspection scenario is identified as emission system failure but the annual inspection is qualified (i.e., abnormal annual inspection scenario 6), and the corresponding abnormal score is output; where SCR efficiency = (NOx inlet concentration - NOx outlet concentration) / NOx inlet concentration × 100%, sampling time ≥ 10min; TWC temperature difference is the difference between the TWC inlet temperature and outlet temperature collected remotely by OBD, sampling time ≥ 5min.

[0030] If the engine is in a shutdown state (status code 0 = shutdown, 1 = running) for a duration ≥ 10s in the remote OBD monitoring time series data, and the annual inspection record for the period when the engine is in a shutdown state is in the testing phase (i.e., the test timestamp coincides with the shutdown period in the remote OBD monitoring time series data), then the abnormal annual inspection scenario is identified as the vehicle being tested while the engine is off (i.e., abnormal annual inspection scenario 7), and the corresponding abnormal score is output; if the shutdown period is ≤ 10s, and it is a normal operating condition switch (such as shifting gears, temporary parking), then it is not determined to be this scenario, and the reasonableness of the operating condition is determined by remote OBD vehicle speed and RPM data.

[0031] If the duration of the load in the remote OBD monitoring time series data that is greater than the preset value (i.e., high load, the preset value can be 70%) (e.g., the statutory test duration: free acceleration test ≥ 60s, load deceleration test ≥ 120s) deviates from the annual inspection record test duration by more than 30%, and there are no related faults (e.g., vehicle faults, temporary equipment faults, etc., if there are no related records, it is considered as no reasonable explanation), then the abnormal annual inspection scenario is identified as abnormal test duration (i.e., abnormal annual inspection scenario 8) and the corresponding abnormal score is output; if the high load period in the remote OBD monitoring time series data is less than 50% of the statutory duration, then it is directly determined as abnormal test duration, and there is no need to calculate the deviation rate.

[0032] If the emission detection start time T in the remote OBD monitoring time series data exh (i.e., the start time stamp of the emissions test in the annual inspection record) < OBD test completion time T OBD(i.e., the OBD detection completion timestamp recorded by the remote OBD), or, if there is no remote OBD detection record (the OBD detection completion time is empty), then the abnormal annual inspection scenario is identified as emission detection without OBD detection (i.e., abnormal annual inspection scenario 9) and the corresponding abnormal score is output; if OBD detection and emission detection are performed in parallel (timestamp overlap ≥80%), then it is not determined to be this scenario.

[0033] If the calibration cycle of the testing equipment is greater than 1 year (e.g., the calibration record time is greater than 365 days from the current testing time), or there is no calibration record, and at least two data points in the remote OBD monitoring time series data have a deviation rate greater than 30% (i.e., the remote OBD data deviates significantly from the annual inspection data), then the abnormal annual inspection scenario is identified as the testing equipment not being calibrated (i.e., abnormal annual inspection scenario 10), and the corresponding abnormal score is output; if the equipment calibration cycle has not expired, but the calibration record is incomplete (e.g., no calibration personnel or calibration equipment information), then it is considered as no calibration record, and is judged according to this scenario (i.e., abnormal annual inspection scenario 10).

[0034] If the number of free acceleration tests in the remote OBD monitoring time series data is less than 3 (e.g., in the remote OBD speed sequence, the number of valid tests that meet the free acceleration speed range is less than 3; valid test definition: the speed increases from idle to the specified speed and remains stable for ≥3s), or if there is no corresponding single smoke opacity test data in the remote OBD monitoring time series data (e.g., single smoke opacity value, not averaged), then the abnormal annual inspection scenario is identified as insufficient smoke opacity test counts (i.e., abnormal annual inspection scenario 11), and the corresponding abnormal score is output. If all fault codes in the remote OBD monitoring time series data are cleared within the second preset time period (e.g., fault codes go from present to absent, and no new fault codes are added), emission indicators (e.g., NOx, DPF differential pressure, urea consumption, etc.) change from continuous abnormality to normal instantly, the mileage within the second preset time period (e.g., 7 days) is <50km, and there are no maintenance records (e.g., no maintenance-related fault code clearing records or component replacement-related data in the remote OBD monitoring time series data), then the abnormal annual inspection scenario is identified as a temporary repair before the annual inspection (i.e., abnormal annual inspection scenario 12), and the corresponding abnormal score is output; if after the fault codes are cleared, the emission indicators remain stable for ≥48h, and the mileage is ≥50km, then it is not determined to be this scenario.

[0035] If the data interruption period of the remote OBD monitoring time series data is ≥ the third preset duration (e.g., 24h), the overlap between the data interruption period and the annual inspection time (overlap = the intersection duration of the interruption period and the annual inspection time / the duration of the annual inspection period) is > 0.8, the data jump value before and after the data interruption is > 3σ, where σ is the standard deviation, and there are no fault records related to the data interruption (e.g., vehicle repair, power outage, signal interruption, etc., if there are no relevant records, it is considered as no reasonable cause), then the abnormal annual inspection scenario is identified as remote OBD data shielding or power outage (i.e., abnormal annual inspection scenario 13) and the corresponding abnormal score is output; where (jump value = |first data value after interruption -last data value before interruption|, compared with the 3σ threshold of the data within 30 days of this dimension); if the data interruption is a phased interruption (cumulative interruption duration ≥ 24h, single interruption < 1h), and the overlap with the annual inspection time is > 0.8, it is also determined to be this scenario (i.e., abnormal annual inspection scenario 13).

[0036] If the vehicle's basic information (such as VIN, fuel type (diesel / gas), emission standard) in the remote OBD monitoring time series data is inconsistent with the vehicle registration information for the annual inspection, and there are no records related to changes in vehicle information (such as vehicle modification, untimely updates of registration information, etc.), then the abnormal annual inspection scenario is identified as a mismatch of vehicle basic information (i.e., abnormal annual inspection scenario 14) and the corresponding abnormal score is output; if the fuel type and emission standard are consistent, only the last 4 digits of the VIN are inconsistent, and there are records of changes to the registered vehicle information, then it is not determined to be this scenario.

[0037] If the data loss rate of the remote OBD monitoring time series data is >5%, the data is continuously interrupted for ≥24 hours, or the instantaneous change in the running data does not conform to the working conditions of the corresponding time period (i.e., the change value is >3σ, and the vehicle speed and engine status do not change significantly during the corresponding time period, such as the engine speed suddenly increasing from 1000r / min to 3000r / min, while the vehicle speed remains at 0), then the abnormal annual inspection scenario is identified as a missing or abnormal fluctuation in the remote OBD monitoring time series data (i.e., abnormal annual inspection scenario 15), and the corresponding abnormal score is output. The missing rate is calculated as: missing data point / total number of data points × 100%. If the abnormal fluctuation is a normal working condition change (such as rapid acceleration or deceleration), then the remote OBD vehicle speed and load data are used to assist in the judgment, and reasonable working condition fluctuations are excluded.

[0038] In practice, the first annual inspection anomaly score is corrected using a machine learning algorithm to obtain the second annual inspection anomaly score, including: The machine learning algorithm dynamically calculates weights and anomaly confidence based on the severity of the identified abnormal annual inspection scenarios, and corrects the first annual inspection anomaly score based on the weights and the anomaly confidence to obtain the second annual inspection anomaly score. The machine learning algorithm is a hybrid XGBoost model that includes binary classification and regression algorithms, and the hybrid XGBoost model integrates multiple decision trees.

[0039] For example, the formula for correcting the first annual inspection abnormality score is as follows: ML Score =Rule Score ×W×P Among them, ML Score For the second year's abnormal score, Rule Score The score represents the first annual inspection anomaly score, where W is the weight and P is the anomaly confidence level. In addition, if the rule Score =0 (no irregular anomaly), but P≥0.8, then MLScore=1.0 (basic score for minor anomalies), which triggers the subsequent TCN-LSTM-Attention model (instantaneous anomaly detection model) to further verify and output the third annual anomaly score, avoiding missed detection of hidden anomalies.

[0040] In practice, the anomaly confidence level P∈[0,1] is as follows: P≥0.8 is judged as high confidence anomaly, 0.5≤P<0.8 is medium confidence, and P<0.5 is low confidence (low confidence anomalies do not participate in the final level determination and are only used as a reference); Adaptive weight W: dynamically adjusted according to the anomaly type. For severe anomalies (identified anomaly inspection scenarios 2, 3, 4, 6, 12, 13), W∈[1.5, 2.0]; for general anomalies (identified anomaly inspection scenarios 5, 7, 9, 10, 11, 14), W∈[1.0, 1.5]; and for minor anomalies (identified anomaly inspection scenarios 1, 8, 15), W∈[0.8, 1.0].

[0041] In practice, during the training of the hybrid XGBoost model, 36-dimensional data characteristics, including remote OBD monitoring time-series data, deviation rates, and anomalies occurring in different abnormal annual inspection scenarios, can be input into the hybrid XGBoost model for training. For example... Input features (36 dimensions, all from remote OBD, with explicit definition for each feature dimension). 1) Vehicle Basics: 6 dimensions (Emission Standard: China V=5, China VI=6; Fuel Type: Diesel=0, Gas=1; Vehicle Age: Current Year - Vehicle Manufacturing Year; Gross Vehicle Weight: Total vehicle weight in relevant static information, unit: t; Displacement: Relevant static information, unit: L; Cumulative Mileage: Total mileage recorded by remote OBD, unit: km). 2) Remote OBD timing: 12 dimensions (average speed, standard deviation of speed, average load, standard deviation of load, average NOx, standard deviation of NOx, average DPF pressure difference, standard deviation of DPF pressure difference, average urea level, urea level change rate, average coolant temperature, average exhaust temperature); 3) Remote OBD Faults: 6 dimensions (MIL illumination duration percentage, number of emission fault codes, number of unready items, number of fault code clearing times, number of EGR faults, number of DPF faults); Supplementary note: The number of EGR faults is based only on statistics of EGR-related fault codes read by remote OBD and does not involve EGR opening monitoring data; 4) Coupling comparison: 5 dimensions (speed deviation rate, NOx deviation rate, DPF pressure difference deviation rate, urea consumption deviation rate, oxygen sensor difference deviation rate); 5) Rule output: 7 dimensions (number of abnormal scenarios, number of serious abnormal scenarios, number of general abnormal scenarios, number of minor abnormal scenarios, rule score, total duration of abnormal periods, percentage of abnormal data points).

[0042] In practice, the hybrid XGBoost model uses an ensemble of multiple decision trees, as shown in the following formula:

[0043] in, The output of the hybrid XGBoost (Extreme Gradient Boosting) model is given by M, where M is the number of decision trees. This represents the output of the m-th decision tree; the loss function L combines classification and regression losses, and includes a regularization term to prevent overfitting. The specific formula is: L=L class +α·L reg +β·L regu Among them, L class For binary cross-entropy loss (used for anomaly / normal classification), L reg For mean squared error loss (used for score regression), L regu α is the L2 regularization term (used for weight penalty); α and β are hyperparameters, with α taking the value of 0.8 and β taking the value of 0.1, determined through grid search optimization.

[0044] Parameters of the hybrid XGBoost model: Number of decision trees M=100, maximum tree depth=6, learning rate=0.1, minimum number of splits=20, minimum number of leaf nodes=10, regularization coefficient λ=0.01; Training strategy: 1) Offline pre-training: Pre-training was performed using remote OBD data from the annual inspection of heavy vehicles over the past 3 years and anomaly labeled data (100,000+ samples), with a training set to test set ratio of 8:2; 2) Online incremental learning: New remote OBD and annual inspection data are accessed daily and updated incrementally at a rate of 1000 samples / batch, with an update frequency of 24 hours / time; Early stopping mechanism: If the test set loss does not decrease for 10 consecutive rounds, stop training to prevent overfitting.

[0045] In practical implementation, to further accurately identify hidden anomalies, a time-series anomaly detection model is proposed for identification. For example, the time-series anomaly detection model identifies hidden anomalies in annual inspections and outputs a third annual inspection anomaly score, including: The TCN dilated convolution module extracts short-term mutation and local anomaly features (such as mutations, spikes, steps, etc.) from the remote OBD monitoring time-series data. These short-term mutation and local anomaly features are then input into the two-layer LSTM network. The two-layer LSTM network generates and outputs a hidden state sequence H={h1, h2, ... h...} for each time step (i.e., a preset duration) by arranging the short-term mutation and local anomaly features in chronological order. t …,h T}, h t Let T be the t-th short-term mutation and local anomaly feature, and T be the total number of short-term mutations and local anomalies in each time step. The hidden state sequence is input into the AdditiveAttention layer, which automatically strengthens the weights of relevant features in the hidden state sequence within the window 1 hour before the annual inspection. Based on the weights, the relevant features in the hidden state sequence are weighted and fused to obtain a context vector. The context vector is then regularized by the Dropout layer and input into a fully connected layer. The fully connected layer identifies covert cheating behavior based on the regularized context vector and generates a temporal anomaly score corresponding to the covert cheating behavior identification (i.e., the fully connected layer maps the context vector into three dimensions of data through linear mapping and Sigmoid / Softmax activation functions: third annual inspection anomaly score, anomaly type (such as temporary repair / OBD blocking / soft cheating / segmented cheating / composite type), and anomaly confidence). The temporal anomaly score is output as the third annual inspection anomaly score, and anomaly confidence can also be output.

[0046] In practice, this also includes: During the process of identifying covert cheating behavior based on the regularized context vector in the fully connected layer, if the changes in emission indicators (such as NOx, DPF pressure difference, etc.) show a V-shaped reversal (the decrease is greater than 50%, and the stabilization time after the reversal is less than 24 hours) or there are traces of fault code clearing within the preset period of the annual inspection (such as the number of fault codes changing from ≥3 to 0 in the past 7 days), and the time series anomaly score S≥0.7, then the type of covert cheating behavior output by the fully connected layer is temporary repair; if the time series anomaly score S<0.7, it can be marked only as a suspicious case of covert cheating behavior with temporary repair, and there is no need to output the type of covert cheating behavior; In the process of identifying covert cheating behavior based on the regularized context vector in the fully connected layer, if the data interruption pattern is fixed (such as interruption at a fixed time every day), the overlap between the data interruption period and the annual inspection time is >0.8, and the data jump before and after the data interruption is >3σ, and the time series anomaly score S≥0.8, then the type of covert cheating behavior output by the fully connected layer is OBD masking; if the time series anomaly score S<0.7, it can be marked only as a suspicious case of covert cheating behavior with OBD masking, and there is no need to output the covert cheating behavior type. During the process of identifying covert cheating behavior based on the regularized context vector in the fully connected layer, if the urea level fluctuation does not match the NOx concentration decrease (e.g., spurious urea level change, level fluctuation ≤3%, but NOx concentration decrease >30%), the SCR efficiency does not match the urea consumption rate (e.g., efficiency >95%, but urea consumption rate <0.1% / 100km), or NOx intermittently exceeds the standard (e.g., the excess duration <10%, and concentrated in non-test periods), and the time-series anomaly score S ≥ 0.75, then the type of covert cheating behavior output by the fully connected layer is soft cheating; if the time-series anomaly score S < 0.7, it can be marked only as a suspicious case of covert cheating behavior with soft cheating, without needing to output the covert cheating behavior type; In the process of identifying covert cheating behavior based on the regularized context vector in the fully connected layer, if the emission index data meets the standard during the test period (e.g., deviation rate ≤ 10%), but exceeds the standard continuously during the non-test period (e.g., deviation rate > 30%, lasting ≥ 2 hours), the temporal feature similarity between the test period and the non-test period is < 0.3, and the temporal anomaly score S ≥ 0.8, then the type of covert cheating behavior output by the fully connected layer is segmented cheating; if the temporal anomaly score S < 0.7, it can be marked only as a suspicious case of covert cheating behavior with segmented cheating, and there is no need to output the covert cheating behavior type.

[0047] In practice, the output and threshold settings of the time-series anomaly detection model are as follows: 1) Temporal anomaly score S∈[0,1]: S≥0.7 is judged as a hidden anomaly (i.e., there is a hidden cheating behavior), 0.5≤S<0.7 is a suspicious anomaly, and S<0.5 is normal; 2) Abnormal Period (second level): Outputs the start and end timestamps of the abnormal period with an accuracy of ±1s, and marks the remote OBD data segment corresponding to the abnormal period. 3) Anomaly Type: Output the specific type of one of the four types of covert cheating behaviors. If multiple types are met simultaneously, mark it as a "composite anomaly". 4) Anomaly confidence: It is linked to the confidence score output by the XGBoost model. If the mean of the two is ≥0.7, it is confirmed as an anomaly; otherwise, it is considered as a suspicious anomaly and requires manual review.

[0048] In practical implementation, for the time-series anomaly detection model, among which... 1) TCN (i.e., TCN dilated convolution module): It adopts 3 layers of dilated convolution with dilation coefficients of 1, 2 and 4 respectively, a kernel size of 3, and the activation function is ReLU. Residual connections are added to each layer to extract short-term abrupt changes and local anomalies such as mutations, spikes and steps in remote OBD time series data. 2) LSTM: A 2-layer LSTM network is used with 64 hidden layer units and a dropout rate of 0.3. The model uses a single heavy-duty vehicle as an independent analysis unit, and the input is the remote OBD effective runtime sequence data of the vehicle in the past 30 days. The sampling frequency is 1Hz (1 time / second), and data is only collected when the engine is running. No data is generated when the engine is off. Calculated based on an average of 4 hours of effective operation per day for heavy-duty vehicles, the total effective time sequence length for 30 days is: 30×4×3600=43200 time steps. It is used to learn the long-term changing trends, progressive failure characteristics, and periodic abnormal patterns of urea level, DPF pressure difference, NOx concentration, MIL fault status, and OBD online status to identify hidden cheating behaviors such as long-term failure of emission systems, temporary repairs before annual inspection, and OBD data masking. 3) AdditiveAttention layer: The additive attention mechanism is adopted, and the dimension of the attention weight is consistent with the dimension of the LSTM output (64 dimensions). It automatically strengthens the weight of relevant features in the window of 1 hour before the annual inspection (3600 data points), and the weight coefficient is increased by 3 to 5 times to suppress the interference of relevant features or data in irrelevant time periods. 4) Dropout layer: Dropout rate of 0.3 to prevent model overfitting; 5) Fully connected layer: The output dimension is 3 (temporal anomaly score, anomaly type, anomaly confidence).

[0049] In practice, the input sequence of the time-series anomaly detection model includes: Remote OBD monitoring time-series data includes engine speed (0~4000r / min), load (0~100%), NOx (0~3000ppm, diesel vehicle), DPF differential pressure (0~30kPa, diesel vehicle), urea level (0~100%, diesel vehicle), MIL status (0 / 1), online status (0 / 1), and fault code status (0=no fault, 1=fault). The input sequence must be preprocessed by an algorithm (outlier removal, filtering, and missing value handling) before being input into the time-series anomaly detection model to ensure data validity. For gas-powered vehicles, NOx, DPF differential pressure, and urea level are automatically removed, and the input sequence is 5-dimensional.

[0050] In practice, after obtaining the abnormal scores for the first, second, and third annual inspections, they can be merged to obtain the final abnormal score. For example, a weighted sum can be performed on the abnormal scores for the first, second, and third annual inspections. Final Score =0.4×Rule Score +0.3×ML Score +0.3×DL Score Among them, Rule Score : First annual inspection abnormality score (diesel vehicles 0~25 points, natural gas vehicles 0~14 points); ML Score : Second year inspection abnormality score (with Rule) Score Within the same interval, adjust using linear normalization: ML Score =ML Score / ML Score_max ×Rule Score_max , of which ML Score_max For the maximum machine learning score, Rule Score_max (The maximum score according to the corresponding vehicle model rules); DL Score Third-year abnormal score (normalized to 0 ~ Rule) Score_max The interval is normalized using the same method as ML_Score.

[0051] In practical implementation, to achieve strong constraint as a fallback, when a specified abnormal annual inspection scenario is identified, a preset final annual inspection abnormality score is directly output, and the vehicle is judged to be severely abnormal. For example, when any of the following specified abnormal annual inspection scenarios are identified, a preset final annual inspection abnormality score is directly output: a. Detect before OBD (i.e., scenario 9, anomaly marker = 1); b. DPF missing / through but smoke opacity acceptable (i.e., scenario 2, P_mean < 5kPa and smoke opacity acceptable, but because DPF missing / through should have caused smoke opacity to exceed the standard, the acceptable result is suspected of being abnormal). c. Remote OBD shielding / power off (i.e., scenario 13, anomaly marker = 1 and confidence level P ≥ 0.8); d. Obvious temporary repair before the exam (i.e., scenario 12, anomaly marker = 1 and time-series anomaly score S ≥ 0.8); e. MIL is always on and there are serious emission fault codes such as P0401 and P2047, but the annual inspection is still qualified (i.e., scenario 5, MIL is on and there are serious emission fault codes such as P0401 and P2047, but the annual inspection is qualified). If any of the above conditions are met, it will be directly judged as a serious abnormality. The final annual inspection abnormality score, Final_Score, will be calculated based on the lowest score for serious abnormalities of the corresponding vehicle type (14 points for diesel vehicles and 10 points for gas vehicles), without needing to participate in the fusion formula calculation.

[0052] In practice, if the specified abnormal annual inspection scenario is not identified, the abnormality level of the annual inspection result and / or the annual inspection process is determined based on the magnitude of the final annual inspection abnormality score. For example... Diesel vehicles: Final annual inspection abnormality score ≥ 14 points: Serious abnormality (record all serious abnormality scenarios and scores); Final annual inspection abnormality score 7~13 points: General abnormality (record the main abnormality scenarios and scores); Final annual inspection abnormality score 1~6 points: Minor abnormality (record the abnormality scenarios and scores); Final annual inspection abnormality score 0 points: Normal; Gas-powered vehicles: Final annual inspection abnormality score ≥ 10 points: serious abnormality; final annual inspection abnormality score 5~9 points: general abnormality; final annual inspection abnormality score 1~4 points: minor abnormality; final annual inspection abnormality score 0 points: normal.

[0053] In practice, the final annual inspection score (FinalScore) is kept to one decimal place. If the score is at the level threshold (e.g., 13.5 points for a diesel vehicle), the level is determined by rounding up (14 points, severe abnormality). After the abnormality level is determined, a level traceability report is output to clarify the score contribution of each algorithm module and key abnormal evidence.

[0054] The following example, taken from a recent annual vehicle inspection in April 2026, illustrates the accuracy and effectiveness of the aforementioned method for identifying abnormalities in heavy-duty vehicle inspections based on remote OBD monitoring. A single National VI emission standard heavy-duty diesel semi-trailer truck was selected as the implementation subject. The entire implementation process, parameter settings, algorithm execution, and output results of the aforementioned method for identifying abnormalities in heavy-duty vehicle inspections based on remote OBD monitoring are presented, ensuring that each step is reproducible and each parameter is verifiable, thus meeting the actual needs of current annual vehicle inspection supervision.

[0055] 1. Basic information of the target vehicle (recent actual vehicle parameters) Vehicle type: China VI emission standard heavy-duty diesel semi-trailer tractor; Vehicle Identification Number (VIN): Confidential for now; Year of manufacture: August 2020; vehicle age: 5.8 years. Gross vehicle weight: 42t, displacement: 11.8L, mileage: 386,520km; Emission standard: China VI (corresponding characteristic value 6); Fuel type: Diesel (corresponding characteristic value 0); Annual inspection time: April 16, 2026, 09:15:00 (T0), annual inspection line number: 3, inspection personnel number: JC003; Testing equipment: Idle gas analyzer (calibration date: March 20, 2026, calibration certificate number: JZ20260320001), exhaust gas analyzer (calibration date: March 18, 2026, calibration certificate number: JZ20260318005).

[0056] 2. Preliminary preparations (recent actual data collection) (1) Data collection scope Collect remote OBD 1Hz time-series data for the vehicle over the past 30 days (March 17, 2026 - April 16, 2026), and high-frequency data for one hour before and after the annual inspection (08:15:00-10:15:00 on April 16, 2026), ensuring a 100% sampling rate and no large-scale data loss.

[0057] (2) Data collection dimensions The system includes 13 dimensions: engine speed, load, vehicle speed, coolant temperature, NOx concentration, DPF differential pressure, urea level, MIL status, fault codes, diagnostic ready items, online data status, mileage, and timestamp, which conform to the remote OBD raw data definition of this invention.

[0058] (3) Annual inspection data collection Simultaneously collect on-site data from the vehicle's annual inspection, including test conditions (free acceleration) and smoke opacity detection values ​​(0.3m). - ¹) NOx detection value (180ppm), test duration (120s), number of tests (3 times), OBD test record (no fault code, MIL off), annual inspection conclusion (qualified).

[0059] 3. Specific implementation process: (1) Step 1: Data preprocessing 1) Outlier Removal: Calculate the mean μ and standard deviation σ of 30-day OBD data dimension by dimension. Taking NOx concentration as an example, μ=850ppm, σ=120ppm, remove x outliers. i <850-3×120=490ppm or x iOutliers of >850+3×120=1210ppm were removed, totaling 12 outlier data points, and the outlier timestamps were marked (all from 14:20 to 14:21 on April 10, 2026). Three consecutive outliers in the rotational speed dimension (3800r / min, 3900r / min, 4050r / min, μ=1800r / min, σ=500r / min) were marked as data outlier period (09:35 to 09:36 on April 15, 2026) and were not included in subsequent coupling comparisons.

[0060] 2) Time-series filtering and noise reduction: A 5-second sliding window mean filter is used. Taking the speed data as an example, when i=5, y5=(x3+x4+x5+x6+x7) / 5. Boundary points are processed according to the rules of this invention to filter instantaneous speed jumps (such as at 10:10 on April 12, 2026, the speed jumps from 1500r / min to 2100r / min, the jump variable 600r / min>500r / min, which is judged as a jump interference, and is corrected to 1780r / min after filtering). NOx jumps and DPF pressure difference jumps are filtered according to the corresponding thresholds to ensure that the data is true and valid.

[0061] 3) Missing value handling: The missing rate for each dimension was calculated. The missing rate for urea level was 3.2% (≤5%). Linear interpolation was used to complete the missing values. For example, for a missing point k=1200, x... 1199 =28%, x 1201 =27%, interpolation calculation x 1200 =28%+(27%-28%)×(1200-1199) / (1201-1199)=27.5%; Dimensions with no missing data rate > 5% are not marked as outliers.

[0062] 4) Time alignment: Based on the annual inspection start time T0 = April 16, 2026 09:15:00, the data window T∈[08:15:00,10:15:00] is extracted. The remote OBD timestamp and the annual inspection timestamp have a deviation of 12s (≤30s), which does not require calibration. After alignment, the data is resampled at 1Hz to obtain 7200 time steps of time-series data, which accurately corresponds to the annual inspection period.

[0063] 5) VIN Coupling Matching: Using the VIN as a unique key, the remote OBD data and annual inspection data are matched. If the two are completely consistent, the match is successful and proceeds to the subsequent high-precision judgment.

[0064] (2) Step 2: Remote OBD-Annual Inspection Data Coupling and Comparison 1) Comparison of coupling dimensions (i.e., each data item): a. Speed ​​comparison: The average speed of remote OBD is 1750 r / min, and the average speed of the annual inspection test is 1800 r / min. The deviation rate R = |1750-1800| / 1800×100%≈2.78%≤10%, which means the operating condition is compliant. b. DPF differential pressure comparison: The average remote OBD differential pressure is 3.2 kPa (relatively low), and the annual smoke opacity is 0.3 m. - ¹(Qualified), the two are negatively correlated, indicating a high risk of abnormality; c. NOx concentration comparison: The average NOx value after remote OBD correction is 850 ppm, and the annual inspection NOx detection value is 180 ppm. The deviation rate R=|850-180| / 180×100%≈372.22%>30%, indicating abnormal NOx consistency. d. Urea consumption comparison: The urea level in the remote OBD system decreased from 28% to 27.8% within 30 days, which is an extremely low consumption rate. It is not positively correlated with NOx emission reduction (theoretical emission reduction rate of 30%), indicating that the SCR effectiveness is abnormal. e. MIL / Fault Code Comparison: The remote OBD detected fault code P20EE (low SCR efficiency), the MIL light is constantly on, the annual inspection OBD test record shows no fault, and the OBD is determined to be non-compliant.

[0065] 2) Deviation rate correction: In the case of no annual inspection value or OBD value of 0, the deviation rate is retained to 2 decimal places, consistent with the above calculation.

[0066] (3) Step 3: Rule baseline engine anomaly determination. After determination, the vehicle triggers the following 6 types of abnormal annual inspection scenarios: Scenario 2: DPF pressure difference is abnormal (average 3.2 kPa), but smoke opacity is acceptable (0.3 m). - ¹); Scenario 3: NOx data deviation is too large (deviation rate 372.22%); Scenario 4: Urea is not consumed for a long period (30-day consumption rate of 0.2%), but NOx levels are within acceptable limits; Scenario 5: MIL is on, emission fault code (P20EE) is present, but the system is still qualified; Scenario 6: Emission system failure (low SCR efficiency) but passes annual inspection; Scenario 12: Temporary repair before annual inspection (April 15, 2026, 20:00, fault codes cleared, NOx concentration dropped from 1020ppm to 450ppm, showing a V-shaped reversal). Rule Scoring Score (i.e., the first annual inspection abnormal score) = Scenario 2 (4 points) + Scenario 3 (4 points) + Scenario 4 (4 points) + Scenario 5 (3 points) + Scenario 6 (4 points) + Scenario 12 (4 points) = 23 points (diesel vehicles, 0~25 points).

[0067] (4) Step 4: Execution of XGBoost model adaptive scoring algorithm 1) Feature Extraction: Extract 36-dimensional remote OBD features. Some key features are as follows: a. Vehicle basic 6 dimensions: Emission standard = 6, Fuel type = 0, Vehicle age = 5.8, Gross vehicle weight = 42, Engine displacement = 11.8, Cumulative mileage = 386520; b. 12-dimensional time series: mean rotation speed 1750, standard deviation of rotation speed 320, mean NOx 850, standard deviation of NOx 120, mean DPF pressure difference 3.2, urea level change rate -0.0067; c. Six-dimensional fault analysis: MIL illumination time percentage 85%, number of emission fault codes 1, number of fault code clearing times 1; d. Coupling comparison 5 dimensions: rotational speed deviation rate 2.78%, NOx deviation rate 372.22%, DPF pressure difference deviation rate 45.00%; rule output 7 dimensions: number of abnormal scenarios 6, number of severe abnormal scenarios 6, rule score 23.

[0068] 2) Model Invocation: Invoke the pre-trained XGBoost model (number of decision trees M=100, maximum tree depth=6, learning rate=0.1, λ=0.01), input 36-dimensional features, and perform inference.

[0069] 3) Output and Correction: a. Anomaly confidence P = 0.92 (high confidence anomaly, ≥ 0.8); adaptive weight W = 1.8 (all 6 specified scenarios are severe anomalies, W ∈ [1.5, 2.0]); b. Corrected Score ML Score =Rule Score ×W×P=23×1.8×0.92≈37.99, after normalization (Diesel vehicle Range) max =25), second annual inspection abnormality score ML Score =37.99 / 37.99×25=25 points.

[0070] (5) Step 5: Algorithm execution of the TCN-LSTM-Attention temporal anomaly detection model 1) Input sequence: Extract the OBD time series data of the vehicle after 30 days of preprocessing, a total of 43,200 time steps (30×4×3600), with 8 input dimensions (diesel vehicle), covering speed, load, NOx, DPF differential pressure, urea level, MIL status, online status, and fault code status.

[0071] 2) Model call: Call the TCN-LSTM-Attention model (3 layers of TCN, dilation coefficient 1 / 2 / 4, convolution kernel 3; 2 layers of LSTM, hidden units 64, dropout=0.3; 4-head Attention, weights are increased by 4 times in the hour before the annual inspection) to perform the extraction of short-term mutations and local abnormal features and the detection of anomalies in covert cheating behavior.

[0072] 3) Identification of concealed abnormal behavior: a. Temporary repair: NOx concentration shows a V-shaped reversal (|780-920| / 920≈15.22%<50%, but combined with the fault code clearing (Fault_count(t2)-Fault_count(t1)=0-1=-1>-3, not fully meeting the characteristics, but combined with the time series trend), the time series abnormality score S=0.82≥0.7, and the temporary repair is judged to be abnormal; b. OBD masking: No data interruption (Data_missing_duration=0s), OBD masking is excluded; - Soft cheating: urea level standard deviation Std(urea_level) = 0.025 < 0.03, NOx_reduction = 30% > 0.3, time series anomaly score S = 0.78 ≥ 0.75, judged as soft cheating anomaly; c. Segmented cheating: NOx deviation rate during the test period is 2.78%≤10%, deviation rate during the non-test period is 372.22%>30%, duration is ≥2h, feature similarity is 0.25<0.3, and time series anomaly score S=0.85≥0.8, which is considered segmented cheating anomaly; 4) Output results: The time-series anomaly score S = 0.82 (normalized DLScore = 0.82 / 1 × 25 = 20.5 points), the anomaly type is a composite anomaly of "temporary fix + soft cheating + segmented cheating", the anomaly period is from 20:00 on April 15, 2026 to 09:15 on April 16, 2026, the anomaly confidence is 0.91, and the XGBoost confidence mean = (0.92 + 0.91) / 2 = 0.915 ≥ 0.7, confirming the anomaly.

[0073] (6) Step 6: Weighted fusion decision and anomaly level quantification 1) Fusion computing: Final Score =0.4×Rule Score + 0.3×ML Score + 0.3×DL Score =0.4×23 +0.3×25 + 0.3×20.5=9.2 + 7.5 + 6.15=22.85 points, rounded to one decimal place, is 22.9 points. 2) Strong constraint fallback check: The vehicle triggers strong constraint condition d (obvious temporary repair before the exam, scenario 12 marker = 1 and S = 0.82 ≥ 0.8), directly judged as a serious anomaly, Final Score The score is calculated based on the minimum score of 14 for serious abnormalities in diesel vehicles (the actual score of 22.9 points is greater than 14 points, so it is still judged as serious abnormality).

[0074] 3) Abnormality level determination: Diesel vehicles, 22.9 points ≥ 14 points, are determined to be seriously abnormal.

[0075] (7) Step 7: Final output result Output a complete anomaly assessment report, including: Exception flag: 1 (Exception); Final Score 22.9 points; Anomaly level: Severe anomaly; Abnormal types: DPF pressure difference, NOx data distortion, urea not consumed, MIL still qualified, emission system failure, temporary repair, soft cheating, segmented cheating (8 types of abnormalities); Abnormal period: 20:00 on April 15, 2026 to 09:15 on April 16, 2026 (with second-level precision); Anomaly confidence level: 0.915; Key evidence chain: remote OBD time-series data (NOx concentration V-shaped reversal, urea level pseudo-fluctuation data), coupling deviation data (NOx deviation rate 372.22%), abnormal scene records (6 types of regular anomalies, 3 types of hidden anomalies), fault code records (P20EE), annual inspection data (smoke opacity qualified, NOx qualified).

[0076] It is evident that, compared with the prior art, the present invention possesses outstanding substantive features and significant progress: Existing technologies often rely on simple comparisons between local OBD data and annual inspection results, or depend solely on fixed threshold rules for anomaly detection, which cannot handle complex operating conditions, data noise, and covert cheating behaviors. This invention, for the first time, constructs a three-layer heterogeneous fusion algorithm based on remote OBD high-resolution time-series data, combining a rule engine, XGBoost machine learning, and TCN-LSTM-Attention deep learning. It supplements key details such as data preprocessing, model training, and threshold setting, enabling joint identification of explicit and covert anomalies with non-obviousness.

[0077] Existing technologies lack in-depth temporal analysis for annual vehicle inspection scenarios, and cannot identify hidden anomalies such as temporary repairs before the inspection, OBD data masking, soft cheating, and segmented cheating. This invention extracts local mutation features through TCN, learns long-term trends through LSTM, and focuses attention on key periods of the annual inspection. It clarifies the model parameters and hidden anomaly judgment thresholds, and achieves second-level localization and type prediction of hidden anomalies, achieving significant technological progress.

[0078] Existing technologies do not have algorithm branches designed for the structural differences between heavy-duty diesel vehicles and gas-powered vehicles, which can easily lead to misjudgments and invalid decisions. This invention achieves fuel type adaptation, allowing gas-powered vehicles to automatically skip the SCR, DPF, and NOx-related decision logic. It also supplements the decision rules and deviation rate correction mechanism for coupled dimensions, improving versatility and accuracy.

[0079] Existing technologies cannot quantify or explain abnormal results, making it difficult to support law enforcement evidence. This invention supplements details such as score normalization, level tracing, and evidence chain recording by using weighted fusion scoring, anomaly level quantification, and feature importance output. This makes anomalies quantifiable, traceable, and explainable, meeting the needs of supervision and law enforcement.

[0080] In this embodiment, a computer device is provided, such as... Figure 2 As shown, it includes a memory 201, a processor 202, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-mentioned methods for identifying abnormalities in the annual inspection of heavy vehicles based on remote OBD monitoring.

[0081] Specifically, the computer device can be a computer terminal, a server, or a similar computing device.

[0082] In this embodiment, a computer-readable storage medium is provided, which stores a computer program that executes any of the above-described methods for identifying abnormalities in the annual inspection of heavy vehicles based on remote OBD monitoring.

[0083] Specifically, computer-readable storage media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer-readable storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable storage media does not include transient media, such as modulated data signals and carrier waves.

[0084] Based on the same inventive concept, this invention also provides a device for identifying abnormalities in the annual inspection of heavy-duty vehicles based on remote OBD monitoring, as described in the following embodiments. Since the principle of the device for identifying abnormalities in the annual inspection of heavy-duty vehicles based on remote OBD monitoring is similar to that of the method for identifying abnormalities in the annual inspection of heavy-duty vehicles based on remote OBD monitoring, the implementation of the device for identifying abnormalities in the annual inspection of heavy-duty vehicles based on remote OBD monitoring can refer to the implementation of the method for identifying abnormalities in the annual inspection of heavy-duty vehicles based on remote OBD monitoring, and will not be repeated. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0085] Figure 3 This is a structural block diagram of a heavy vehicle annual inspection anomaly identification device based on remote OBD monitoring according to an embodiment of the present invention, such as... Figure 3 As shown, it includes: The deviation rate calculation module 301 is used to acquire the remote OBD monitoring time series data of the target heavy vehicle during the annual inspection period, and calculate the deviation rate between the real-time value of each data item in the remote OBD monitoring time series data and the detection value of the corresponding data item in the annual inspection detection data of the target heavy vehicle based on the structural differences between diesel vehicles and gas vehicles. The first identification module 302 is used to identify different abnormal annual inspection scenarios and determine the abnormal score according to the remote OBD monitoring time series data, the annual inspection detection data and the deviation rate, in accordance with the national standard annual inspection requirements, and to obtain the first annual inspection abnormal score based on the abnormal score of the different abnormal annual inspection scenarios. The second identification module 303 is used to correct the first annual inspection abnormality score based on the remote OBD monitoring time series data, the deviation rate, and the abnormal situations that occur in different abnormal annual inspection scenarios, and obtain the second annual inspection abnormality score by using a machine learning algorithm. The third identification module 304 is used to input the remote OBD monitoring time series data into the time series anomaly detection model. The time series anomaly detection model identifies hidden anomalies in the annual inspection and outputs a third annual inspection anomaly score. The time series anomaly detection model includes a cascaded TCN dilated convolution module, a two-layer LSTM network, an AdditiveAttention layer, a Dropout layer, and a fully connected layer. The anomaly identification module 305 is used to fuse the first annual inspection anomaly score, the second annual inspection anomaly score, and the third annual inspection anomaly score to obtain a final annual inspection anomaly score, or, when a specified abnormal annual inspection scenario is identified, directly output a preset final annual inspection anomaly score, and determine the anomaly level of the annual inspection result and / or the annual inspection process based on the magnitude of the final annual inspection anomaly score. If the final annual inspection anomaly score is zero, then the annual inspection result and / or the annual inspection process is determined to be normal.

[0086] In one embodiment, the first identification module is used to identify abnormal annual inspection scenarios including: failure to test at the prescribed engine speed, abnormal DPF pressure difference but passing the annual smoke inspection, excessive NOx data deviation, long-term non-consumption of urea but passing the annual NOx inspection, passing the annual inspection despite MIL being lit or not ready, emission system failure but passing the annual inspection, testing after the vehicle is turned off, abnormal test duration, emission testing before OBD testing, uncalibrated testing equipment, insufficient number of smoke tests, temporary repairs before the annual inspection, remote OBD data shielding or power failure, mismatched vehicle basic information, and missing or abnormal fluctuations in remote OBD monitoring time-series data. If the percentage of speeds outside the national standard speed range in the remote OBD monitoring time series data is greater than or equal to 60% of the annual inspection speed test duration, or if the percentage of speeds with a fluctuation coefficient greater than 5% in the remote OBD monitoring time series data is greater than or equal to 50% of the speed test duration, and the annual inspection speed test conclusion is qualified, then the abnormal annual inspection scenario is identified as failure to test the speed according to regulations and the corresponding abnormal score is output. If the duration of the mean value Pmean of the DPF pressure difference in the remote OBD monitoring time series data is ≥30 min and the annual inspection smoke opacity is qualified, then the abnormal annual inspection scenario is identified as abnormal DPF pressure difference but qualified smoke opacity and the corresponding abnormal score is output. If the target heavy vehicle is a diesel vehicle, and the deviation rate R of NOx in the remote OBD monitoring time series data is greater than 30% and the NOx concentration value after environmental correction is greater than 1.5 × the NOx concentration value of the annual inspection, and the annual inspection NOx is qualified, then the abnormal annual inspection scenario is identified as excessive NOx data deviation and the corresponding abnormal score is output. If the target heavy vehicle is a diesel vehicle, and the urea consumption rate r < 0.3% / 100km in the remote OBD monitoring time series data, the driving mileage within the first preset time period is ≥ 500km, the annual inspection NOx is qualified, and there is no urea replenishment record in the first preset time period in the remote OBD monitoring time series data, then the abnormal annual inspection scenario is identified as urea not being consumed for a long time but the NOx annual inspection is qualified and the corresponding abnormal score is output. If MIL= is lit in the remote OBD monitoring time series data, there is an emission-related fault code, and the annual inspection conclusion is qualified; or if there are ≥3 unready items in the remote OBD monitoring time series data, and the annual inspection conclusion is qualified, then an abnormal annual inspection scenario is identified as MIL being lit or unready items exceeding the standard but the annual inspection is still qualified, and the corresponding abnormal score is output. If the target heavy-duty vehicle is a diesel vehicle, and the remote OBD monitoring time series data shows SCR efficiency <80%, DPF regeneration abnormality, or EGR system failure, then the abnormal annual inspection scenario is identified as emission system failure but the annual inspection is passed, and the corresponding abnormal score is output. If the target heavy-duty vehicle is a gas-powered vehicle, and the annual inspection is passed, and the remote OBD monitoring time series data shows EGR system failure, TWC temperature difference <50℃, or no difference between the front and rear oxygen sensors, then the abnormal annual inspection scenario is identified as emission system failure but the annual inspection is passed, and the corresponding abnormal score is output. If the annual inspection record for the period when the engine is in a shutdown state for a duration of ≥10s and the engine is in a shutdown state is during testing, then the abnormal annual inspection scenario is identified as the vehicle is being tested while the engine is off and the corresponding abnormal score is output. If the deviation rate between the load exceeding the preset value in the remote OBD monitoring time series data and the test duration in the annual inspection record is >30%, and there are no related faults, the annual inspection is qualified. Then, the abnormal annual inspection scenario is identified as abnormal test duration and the corresponding abnormal score is output. If the emission detection start time T in the remote OBD monitoring time series data exh <OBD detection completion time T OBD Alternatively, if there is no remote OBD detection record but the annual inspection is qualified, the abnormal annual inspection scenario is identified as emissions testing without OBD testing and the corresponding abnormal score is output. If the calibration cycle of the testing equipment is greater than 1 year, and at least two data points in the remote OBD monitoring time series data have a deviation rate greater than 30%, and the annual inspection is qualified, then the abnormal annual inspection scenario is identified as the testing equipment not being calibrated and the corresponding abnormal score is output. If the number of free acceleration method tests in the remote OBD monitoring time series data is less than 3, or if there is no corresponding single smoke opacity test data in the remote OBD monitoring time series data, and the annual inspection is qualified, then the abnormal annual inspection scenario is identified as insufficient smoke opacity test count and the corresponding abnormal score is output. If all fault codes in the remote OBD monitoring time series data are cleared within the second preset time period, the emission index changes from continuous abnormal to normal instantly, the mileage within the second preset time period is <50km, there is no maintenance record, and the annual inspection is qualified, then the abnormal annual inspection scenario is identified as temporary repair before the annual inspection and the corresponding abnormal score is output. If the data interruption period of the remote OBD monitoring time series data is ≥ the third preset duration, the overlap between the data interruption period and the annual inspection time is > 0.8, the data jump value before and after the data interruption is > 3σ, and there are no fault records related to the data interruption, and the annual inspection OBD detection record shows normal or the annual inspection conclusion is qualified, then the abnormal annual inspection scenario is identified as remote OBD data shielding or power failure and the corresponding abnormal score is output. If the vehicle basic information in the remote OBD monitoring time series data is inconsistent with the vehicle registration information for the annual inspection, and there are no records related to changes in vehicle information, and the annual inspection is qualified, then the abnormal annual inspection scenario is identified as a mismatch of vehicle basic information and the corresponding abnormal score is output. If the data loss rate of the remote OBD monitoring time series data is greater than 5%, the data is continuously interrupted for ≥24 hours, or the instantaneous change in the running data does not conform to the working conditions of the corresponding time period, resulting in the inability to verify the annual inspection results or the annual inspection conclusion being qualified, then the abnormal annual inspection scenario is identified as missing or abnormal fluctuations in the remote OBD monitoring time series data and the corresponding abnormal score is output.

[0087] In one implementation, the second identification module is used to dynamically calculate weights and anomaly confidence based on the severity of the anomalies identified in the abnormal annual inspection scenario using the machine learning algorithm, and to correct the first annual inspection anomaly score based on the weights and the anomaly confidence to obtain the second annual inspection anomaly score. The machine learning algorithm is a hybrid XGBoost model that includes a binary classification algorithm and a regression algorithm, and the hybrid XGBoost model integrates multiple decision trees.

[0088] In one implementation, the third identification module is used to extract short-term mutation and local anomaly features from the remote OBD monitoring time-series data through the TCN dilated convolution module, input the short-term mutation and local anomaly features into the dual-layer LSTM network, the dual-layer LSTM network generates and outputs a hidden state sequence by arranging the short-term mutation and local anomaly features at each time step in chronological order; the hidden state sequence is input into the AdditiveAttention layer, the AdditiveAttention layer automatically strengthens the weights of relevant features in the hidden state sequence within the window one hour before the annual inspection, and the relevant features in the hidden state sequence are weighted and fused based on the weights to obtain a context vector; the context vector is regularized by the Dropout layer and then input into a fully connected layer, the fully connected layer identifies covert cheating behavior based on the regularized context vector and generates a time-series anomaly score corresponding to the covert cheating behavior identification.

[0089] In one embodiment, the third identification module is further configured to, during the process of identifying covert cheating behavior based on the regularized context vector, if the emission index data changes show a V-shaped reversal or there are traces of fault code clearing within a preset period of the annual inspection, and the time-series anomaly score S≥0.7, then the type of covert cheating behavior output by the fully connected layer is temporary repair; during the process of identifying covert cheating behavior based on the regularized context vector, if the data interruption pattern is fixed, the overlap between the data interruption period and the annual inspection time is >0.8, and the data jump before and after the data interruption is >3σ, and the time-series anomaly score S≥0.8, then the type of covert cheating behavior output by the fully connected layer is OBD masking; In the process of identifying covert cheating behavior based on the regularized context vector, if the urea level fluctuation does not match the NOx concentration decrease, the SCR efficiency does not match the urea consumption rate, and NOx intermittently exceeds the standard, and the time-series anomaly score S ≥ 0.75, then the type of covert cheating behavior output by the fully connected layer is soft cheating; In the process of identifying covert cheating behavior based on the regularized context vector, if the emission index data meets the standard during the test period but continuously exceeds the standard during the non-test period, and the time-series feature similarity between the test period and the non-test period is < 0.3, and the time-series anomaly score S ≥ 0.8, then the type of covert cheating behavior output by the fully connected layer is segmented cheating.

[0090] In one embodiment, the TCN dilated convolution module is a 3-layer dilated convolution with residual connections added to each layer to extract the short-term mutation and local anomaly features.

[0091] The embodiments of this invention achieve the following technical effects: using remote OBD real-time / historical time-series data as the sole basic data source, and through a three-layer heterogeneous fusion algorithm combining abnormal annual inspection scenario identification according to national standard annual inspection requirements, machine learning algorithms, and time-series anomaly detection models, high-precision identification of explicit and implicit annual inspection anomalies is achieved. This adapts to the differentiated structural needs of diesel and gas vehicles, strictly adheres to relevant national standard annual inspection requirements such as GB3847-2018, and outputs quantifiable, traceable, and directly data-based evidence for law enforcement. This fills the gap in existing technology, helps improve the intelligence and precision of heavy vehicle annual inspection supervision, and is conducive to ensuring the authenticity and fairness of annual inspection results.

[0092] Obviously, those skilled in the art should understand that the modules or steps of the above-described embodiments of the present invention can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the embodiments of the present invention are not limited to any particular hardware and software combination.

[0093] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying abnormalities in the annual inspection of heavy-duty vehicles based on remote OBD monitoring, characterized in that, include: Acquire remote OBD monitoring time-series data of the target heavy vehicle during the annual inspection period and within 30 days prior to the annual inspection period. Based on the structural differences between diesel and gas vehicles, calculate the deviation rate between the real-time value of each data item in the remote OBD monitoring time-series data and the detection value of the corresponding data item in the annual inspection data of the target heavy vehicle during the annual inspection period. According to the national standard annual inspection requirements, based on the remote OBD monitoring time series data, the annual inspection data and the deviation rate, different abnormal annual inspection scenarios with abnormal situations are identified and abnormal scores are determined. The first annual inspection abnormal score is obtained based on the abnormal scores of different abnormal annual inspection scenarios. Based on the remote OBD monitoring time series data, the deviation rate, and the abnormal situations that occur in different abnormal annual inspection scenarios, the first annual inspection abnormal score is corrected by machine learning algorithm to obtain the second annual inspection abnormal score. The remote OBD monitoring time series data is input into the time series anomaly detection model. The time series anomaly detection model identifies hidden anomalies in the annual inspection and outputs a third annual inspection anomaly score. The time series anomaly detection model includes a cascaded TCN dilated convolution module, a two-layer LSTM network, an AdditiveAttention layer, a Dropout layer, and a fully connected layer. The first annual inspection abnormality score, the second annual inspection abnormality score, and the third annual inspection abnormality score are fused to obtain the final annual inspection abnormality score. Alternatively, when a specified abnormal annual inspection scenario is identified, a preset final annual inspection abnormality score is directly output. The abnormality level of the annual inspection result and / or the annual inspection process is determined based on the magnitude of the final annual inspection abnormality score. If the final annual inspection abnormality score is zero, the annual inspection result and / or the annual inspection process is determined to be normal.

2. The method as described in claim 1, characterized in that, Identify different abnormal annual inspection scenarios and determine abnormal scores, including: The identified abnormal inspection scenarios include: failure to test at the prescribed engine speed, abnormal DPF pressure difference but passing the smoke inspection, excessive NOx data deviation, urea not being consumed for a long time but passing the NOx inspection, MIL light being on or exceeding the ready level but still passing the inspection, emission system failure but passing the inspection, testing with the vehicle off, abnormal test duration, emission testing before OBD testing, uncalibrated testing equipment, insufficient smoke testing attempts, temporary repairs before the inspection, remote OBD data blocking or power failure, mismatched vehicle basic information, and missing or abnormally fluctuating remote OBD monitoring time-series data. If the percentage of speeds outside the national standard speed range in the remote OBD monitoring time series data is greater than or equal to 60% of the annual inspection speed test duration, or if the percentage of speeds with a fluctuation coefficient greater than 5% in the remote OBD monitoring time series data is greater than or equal to 50% of the speed test duration, and the annual inspection speed test conclusion is qualified, then the abnormal annual inspection scenario is identified as failure to test the speed according to regulations and the corresponding abnormal score is output. If the duration of the mean value Pmean of the DPF pressure difference in the remote OBD monitoring time series data is ≥30 min and the annual inspection smoke opacity is qualified, then the abnormal annual inspection scenario is identified as abnormal DPF pressure difference but qualified smoke opacity and the corresponding abnormal score is output. If the target heavy vehicle is a diesel vehicle, and the deviation rate R of NOx in the remote OBD monitoring time series data is greater than 30% and the NOx concentration value after environmental correction is greater than 1.5 × the NOx concentration value of the annual inspection, and the annual inspection NOx is qualified, then the abnormal annual inspection scenario is identified as excessive NOx data deviation and the corresponding abnormal score is output. If the target heavy vehicle is a diesel vehicle, and the urea consumption rate r < 0.3% / 100km in the remote OBD monitoring time series data, the driving mileage within the first preset time period is ≥ 500km, the annual inspection NOx is qualified, and there is no urea replenishment record in the first preset time period in the remote OBD monitoring time series data, then the abnormal annual inspection scenario is identified as urea not being consumed for a long time but the NOx annual inspection is qualified and the corresponding abnormal score is output. If MIL= is lit in the remote OBD monitoring time series data, there is an emission-related fault code, and the annual inspection conclusion is qualified; or if there are ≥3 unready items in the remote OBD monitoring time series data, and the annual inspection conclusion is qualified, then an abnormal annual inspection scenario is identified as MIL being lit or unready items exceeding the standard but the annual inspection is still qualified, and the corresponding abnormal score is output. If the target heavy-duty vehicle is a diesel vehicle, and the remote OBD monitoring time series data shows SCR efficiency <80%, DPF regeneration abnormality, or EGR system failure, then the abnormal annual inspection scenario is identified as emission system failure but the annual inspection is passed, and the corresponding abnormal score is output. If the target heavy-duty vehicle is a gas-powered vehicle, and the annual inspection is passed, and the remote OBD monitoring time series data shows EGR system failure, TWC temperature difference <50℃, or no difference between the front and rear oxygen sensors, then the abnormal annual inspection scenario is identified as emission system failure but the annual inspection is passed, and the corresponding abnormal score is output. If the annual inspection record for the period when the engine is in a shutdown state for a duration of ≥10s and the engine is in a shutdown state is during testing, then the abnormal annual inspection scenario is identified as the vehicle is being tested while the engine is off and the corresponding abnormal score is output. If the deviation rate between the load exceeding the preset value in the remote OBD monitoring time series data and the test duration in the annual inspection record is >30%, and there are no related faults, the annual inspection is qualified. Then, the abnormal annual inspection scenario is identified as abnormal test duration and the corresponding abnormal score is output. If the emission detection start time T in the remote OBD monitoring time series data exh <OBD detection completion time T OBD Alternatively, if there is no remote OBD detection record but the annual inspection is qualified, the abnormal annual inspection scenario is identified as emissions testing without OBD testing and the corresponding abnormal score is output. If the calibration cycle of the testing equipment is greater than 1 year, and at least two data points in the remote OBD monitoring time series data have a deviation rate greater than 30%, and the annual inspection is qualified, then the abnormal annual inspection scenario is identified as the testing equipment not being calibrated and the corresponding abnormal score is output. If the number of free acceleration method tests in the remote OBD monitoring time series data is less than 3, or if there is no corresponding single smoke opacity test data in the remote OBD monitoring time series data, and the annual inspection is qualified, then the abnormal annual inspection scenario is identified as insufficient smoke opacity test count and the corresponding abnormal score is output. If all fault codes in the remote OBD monitoring time series data are cleared within the second preset time period, the emission index changes from continuous abnormal to normal instantly, the mileage within the second preset time period is <50km, there is no maintenance record, and the annual inspection is qualified, then the abnormal annual inspection scenario is identified as temporary repair before the annual inspection and the corresponding abnormal score is output. If the data interruption period of the remote OBD monitoring time series data is ≥ the third preset duration, the overlap between the data interruption period and the annual inspection time is > 0.8, the data jump value before and after the data interruption is > 3σ, where σ is the standard deviation, and there are no fault records related to the data interruption, and the annual inspection OBD detection record shows normal or the annual inspection conclusion is qualified, then the abnormal annual inspection scenario is identified as remote OBD data shielding or power failure and the corresponding abnormal score is output. If the vehicle basic information in the remote OBD monitoring time series data is inconsistent with the vehicle registration information for the annual inspection, and there are no records related to changes in vehicle information, and the annual inspection is qualified, then the abnormal annual inspection scenario is identified as a mismatch of vehicle basic information and the corresponding abnormal score is output. If the data loss rate of the remote OBD monitoring time series data is greater than 5%, the data is continuously interrupted for ≥24 hours, or the instantaneous change in the running data does not conform to the working conditions of the corresponding time period, resulting in the inability to verify the annual inspection results or the annual inspection conclusion being qualified, then the abnormal annual inspection scenario is identified as missing or abnormal fluctuations in the remote OBD monitoring time series data and the corresponding abnormal score is output.

3. The method as described in claim 1, characterized in that, The first annual inspection anomaly score is corrected using a machine learning algorithm to obtain the second annual inspection anomaly score, including: The machine learning algorithm dynamically calculates weights and anomaly confidence based on the severity of the identified abnormal annual inspection scenarios, and corrects the first annual inspection anomaly score based on the weights and the anomaly confidence to obtain the second annual inspection anomaly score. The machine learning algorithm is a hybrid XGBoost model that includes binary classification and regression algorithms, and the hybrid XGBoost model integrates multiple decision trees.

4. The method according to any one of claims 1 to 3, characterized in that, The time-series anomaly detection model identifies hidden anomalies in annual inspections and outputs a third annual inspection anomaly score, including: The TCN dilated convolution module extracts short-term mutation and local anomaly features from the remote OBD monitoring time-series data. These features are then input into a two-layer LSTM network. The LSTM network generates and outputs a hidden state sequence by arranging the short-term mutation and local anomaly features at each time step in chronological order. The hidden state sequence is then input into an AdditiveAttention layer, which automatically strengthens the weights of relevant features within the window one hour before the annual inspection in the hidden state sequence. Based on these weights, the relevant features in the hidden state sequence are weighted and fused to obtain a context vector. The context vector is then regularized by a Dropout layer and input into a fully connected layer. The fully connected layer identifies covert cheating behavior based on the regularized context vector and generates a time-series anomaly score corresponding to the identified covert cheating behavior. This time-series anomaly score is then output as the third annual inspection anomaly score.

5. The method as described in claim 4, characterized in that, Also includes: In the process of identifying covert cheating behavior based on the regularized context vector, if the change of emission index data shows a V-shaped reversal or there are traces of fault code clearing within the preset period of the annual inspection, and the time-series anomaly score S≥0.7, then the type of covert cheating behavior output by the fully connected layer is temporary repair. In the process of identifying covert cheating behavior based on the regularized context vector, if the data interruption pattern is fixed, the overlap between the data interruption period and the annual inspection time is >0.8, and the data jump before and after the data interruption is >3σ, and the time-series anomaly score S≥0.8, then the type of covert cheating behavior output by the fully connected layer is OBD masking. In the process of identifying covert cheating behavior based on the regularized context vector, if the urea level fluctuation does not match the decrease in NOx concentration, the SCR efficiency does not match the urea consumption rate, and NOx intermittently exceeds the standard, and the time-series anomaly score S ≥ 0.75, then the type of covert cheating behavior output by the fully connected layer is soft cheating. In the process of identifying covert cheating behavior based on the regularized context vector, if the emission index data meets the standard during the test period but exceeds the standard during the non-test period, the temporal feature similarity between the test period and the non-test period is <0.3, and the temporal anomaly score S≥0.8, then the type of covert cheating behavior output by the fully connected layer is segmented cheating.

6. The method as described in claim 4, characterized in that, The TCN dilated convolution module consists of three layers of dilated convolution, with residual connections added to each layer to extract the short-term mutation and local anomaly features.

7. A device for identifying abnormalities in the annual inspection of heavy-duty vehicles based on remote OBD monitoring, characterized in that, include: The deviation rate calculation module is used to acquire remote OBD monitoring time series data of the target heavy vehicle during the annual inspection period and within 30 days before the annual inspection period. Based on the structural differences between diesel vehicles and gas vehicles, it calculates the deviation rate between the real-time value of each data item in the remote OBD monitoring time series data and the detection value of the corresponding data item in the annual inspection data of the target heavy vehicle during the annual inspection period. The first identification module is used to identify different abnormal annual inspection scenarios and determine the abnormal score according to the remote OBD monitoring time series data, the annual inspection data and the deviation rate, in accordance with the national standard annual inspection requirements, and to obtain the first annual inspection abnormal score based on the abnormal score of different abnormal annual inspection scenarios. The second identification module is used to correct the first annual inspection abnormality score based on the remote OBD monitoring time series data, the deviation rate, and the abnormal situations that occur in different abnormal annual inspection scenarios, and to obtain the second annual inspection abnormality score by using a machine learning algorithm. The third identification module is used to input the remote OBD monitoring time series data into the time series anomaly detection model. The time series anomaly detection model identifies hidden anomalies in the annual inspection and outputs a third annual inspection anomaly score. The time series anomaly detection model includes a cascaded TCN dilated convolution module, a two-layer LSTM network, an AdditiveAttention layer, a Dropout layer, and a fully connected layer. An anomaly identification module is used to fuse the first annual inspection anomaly score, the second annual inspection anomaly score, and the third annual inspection anomaly score to obtain a final annual inspection anomaly score. Alternatively, when a specified abnormal annual inspection scenario is identified, a preset final annual inspection anomaly score is directly output. The anomaly level of the annual inspection result and / or the annual inspection process is determined based on the magnitude of the final annual inspection anomaly score. If the final annual inspection anomaly score is zero, the annual inspection result and / or the annual inspection process is determined to be normal.

8. The apparatus as claimed in claim 7, characterized in that, The first identification module is used to identify abnormal annual inspection scenarios including: failure to test at the prescribed engine speed, abnormal DPF pressure difference but passing the annual smoke inspection, excessive NOx data deviation, long-term non-consumption of urea but passing the annual NOx inspection, passing the annual inspection despite MIL being lit or not ready, emission system failure but passing the annual inspection, testing after the vehicle is turned off, abnormal test duration, emission testing before OBD testing, uncalibrated testing equipment, insufficient number of smoke tests, temporary repairs before the annual inspection, remote OBD data shielding or power failure, mismatched vehicle basic information, and missing or abnormally fluctuating remote OBD monitoring time series data. If the percentage of speeds outside the national standard speed range in the remote OBD monitoring time series data is greater than or equal to 60% of the annual inspection speed test duration, or if the percentage of speeds with a fluctuation coefficient greater than 5% in the remote OBD monitoring time series data is greater than or equal to 50% of the speed test duration, and the annual inspection speed test conclusion is qualified, then the abnormal annual inspection scenario is identified as failure to test the speed according to regulations and the corresponding abnormal score is output. If the duration of the mean value Pmean of the DPF pressure difference in the remote OBD monitoring time series data is ≥30 min and the annual inspection smoke opacity is qualified, then the abnormal annual inspection scenario is identified as abnormal DPF pressure difference but qualified smoke opacity and the corresponding abnormal score is output. If the target heavy vehicle is a diesel vehicle, and the deviation rate R of NOx in the remote OBD monitoring time series data is greater than 30% and the NOx concentration value after environmental correction is greater than 1.5 × the NOx concentration value of the annual inspection, and the annual inspection NOx is qualified, then the abnormal annual inspection scenario is identified as excessive NOx data deviation and the corresponding abnormal score is output. If the target heavy vehicle is a diesel vehicle, and the urea consumption rate r < 0.3% / 100km in the remote OBD monitoring time series data, the driving mileage within the first preset time period is ≥ 500km, the annual inspection NOx is qualified, and there is no urea replenishment record in the first preset time period in the remote OBD monitoring time series data, then the abnormal annual inspection scenario is identified as urea not being consumed for a long time but the NOx annual inspection is qualified and the corresponding abnormal score is output. If MIL= is lit in the remote OBD monitoring time series data, there is an emission-related fault code, and the annual inspection conclusion is qualified; or if there are ≥3 unready items in the remote OBD monitoring time series data, and the annual inspection conclusion is qualified, then an abnormal annual inspection scenario is identified as MIL being lit or unready items exceeding the standard but the annual inspection is still qualified, and the corresponding abnormal score is output. If the target heavy-duty vehicle is a diesel vehicle, and the remote OBD monitoring time series data shows SCR efficiency <80%, DPF regeneration abnormality, or EGR system failure, then the abnormal annual inspection scenario is identified as emission system failure but the annual inspection is passed, and the corresponding abnormal score is output. If the target heavy-duty vehicle is a gas-powered vehicle, and the annual inspection is passed, and the remote OBD monitoring time series data shows EGR system failure, TWC temperature difference <50℃, or no difference between the front and rear oxygen sensors, then the abnormal annual inspection scenario is identified as emission system failure but the annual inspection is passed, and the corresponding abnormal score is output. If the annual inspection record for the period when the engine is in a shutdown state for a duration of ≥10s and the engine is in a shutdown state is during testing, then the abnormal annual inspection scenario is identified as the vehicle is being tested while the engine is off and the corresponding abnormal score is output. If the deviation rate between the load exceeding the preset value in the remote OBD monitoring time series data and the test duration in the annual inspection record is >30%, and there are no related faults, the annual inspection is qualified. Then, the abnormal annual inspection scenario is identified as abnormal test duration and the corresponding abnormal score is output. If the emission detection start time T in the remote OBD monitoring time series data exh <OBD detection completion time T OBD Alternatively, if there is no remote OBD detection record but the annual inspection is qualified, the abnormal annual inspection scenario is identified as emissions testing without OBD testing and the corresponding abnormal score is output. If the calibration cycle of the testing equipment is greater than 1 year, and at least two data points in the remote OBD monitoring time series data have a deviation rate greater than 30%, and the annual inspection is qualified, then the abnormal annual inspection scenario is identified as the testing equipment not being calibrated and the corresponding abnormal score is output. If the number of free acceleration method tests in the remote OBD monitoring time series data is less than 3, or if there is no corresponding single smoke opacity test data in the remote OBD monitoring time series data, and the annual inspection is qualified, then the abnormal annual inspection scenario is identified as insufficient number of smoke opacity tests and the corresponding abnormal score is output. If all fault codes in the remote OBD monitoring time series data are cleared within the second preset time period, the emission index changes from continuous abnormal to normal instantly, the mileage within the second preset time period is <50km, there is no maintenance record, and the annual inspection is qualified, then the abnormal annual inspection scenario is identified as temporary repair before the annual inspection and the corresponding abnormal score is output. If the data interruption period of the remote OBD monitoring time series data is ≥ the third preset duration, the overlap between the data interruption period and the annual inspection time is > 0.8, the data jump before and after the data interruption is > 3σ, where σ is the standard deviation, and there are no fault records related to the data interruption, and the annual inspection OBD detection record shows normal or the annual inspection conclusion is qualified, then the abnormal annual inspection scenario is identified as remote OBD data shielding or power failure and the corresponding abnormal score is output. If the vehicle basic information in the remote OBD monitoring time series data is inconsistent with the vehicle registration information for the annual inspection, and there are no records related to changes in vehicle information, and the annual inspection is qualified, then the abnormal annual inspection scenario is identified as a mismatch of vehicle basic information and the corresponding abnormal score is output. If the data loss rate of the remote OBD monitoring time series data is greater than 5%, the data is continuously interrupted for ≥24 hours, or the instantaneous change in the running data does not conform to the working conditions of the corresponding time period, resulting in the inability to verify the annual inspection results or the annual inspection conclusion being qualified, then the abnormal annual inspection scenario is identified as missing or abnormal fluctuations in the remote OBD monitoring time series data and the corresponding abnormal score is output.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method for identifying abnormalities in the annual inspection of heavy vehicles based on remote OBD monitoring, as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that executes the method for identifying abnormalities in the annual inspection of heavy vehicles based on remote OBD monitoring, as described in any one of claims 1 to 6.

Citation Information

Patent Citations

  • Oil consumption and NOx emission evaluation method based on actual working condition of Internet-of-vehicles heavy duty vehicle

    CN113743715A

  • Vehicle detection data monitoring method and system for vehicle detection station

    CN114659804A