Heavy duty vehicle supervision method for remote emission management vehicle-mounted terminal data quality

Abstract

The invention relates to a heavy vehicle supervision method for remote emission management vehicle-mounted terminal data quality, and belongs to the technical field of motor vehicle emission remote monitoring and environmental protection. According to the method, multi-source heterogeneous emission related data is acquired through a vehicle-mounted terminal, multi-dimensional quality evaluation indexes such as data integrity and accuracy are fused, and abnormal behaviors such as data tampering and equipment degradation are accurately identified by adopting an unsupervised abnormal fingerprint detection and traceability algorithm. On the basis, in combination with regional environment risks, adaptive matching and dynamic feedback of data quality grading and multi-level supervision measures are realized, and the authenticity of emission data of the heavy duty vehicle and supervision response efficiency are effectively improved. According to the invention, the intelligent decision-making full link is realized through data acquisition, judgment and identification, and the method is suitable for the field of efficient, refined and intelligent emission supervision of road moving sources.

Description

Technical Field

[0001] This invention belongs to the field of remote monitoring of motor vehicle emissions and environmental protection technology, and more specifically relates to a method for monitoring the data quality of vehicle-mounted terminals for remote emission management. Background Technology

[0002] With the continuous growth of motor vehicle ownership in my country, vehicle emissions have become a significant factor affecting urban air quality and air pollution control. Heavy-duty diesel vehicles, in particular, are major contributors to nitrogen oxides (NOx) and particulate matter (PM) emissions from transportation sources due to their high emission intensity per vehicle, complex operating conditions, and difficulty in management. Traditional methods of motor vehicle emission supervision mainly rely on periodic emission testing, roadside inspections, and on-site enforcement. These methods suffer from limitations such as simplistic approaches, limited regulatory coverage, and insufficient real-time and precision, making it difficult to effectively address the realities of heavy-duty vehicles operating across regions, dynamic emission changes, and illegal modifications and cheating by some owners to evade supervision.

[0003] In recent years, thanks to the development of vehicle-mounted terminals (such as OBD and remote emission terminals), vehicle networking, and big data technologies, an increasing number of heavy-duty vehicles have been equipped with remote real-time emission monitoring equipment, enabling dynamic collection and remote uploading of key vehicle emission parameters. Theoretically, through the connection between the cloud backend and local environmental protection regulatory systems, continuous and dynamic emission data monitoring and anomaly analysis of heavy-duty vehicles across the entire region can be performed. However, many technical challenges remain in practical applications. First, vehicle emission data collection often faces complex situations such as multi-source heterogeneity (e.g., GPS speed, engine status, multiple sensor outputs), data loss and anomalies, communication delays and distortions, severely affecting data integrity and reliability. Second, data falsification, interpolation, and evasion caused by equipment aging, sensor drift, improper installation, and deliberate vehicle cheating are becoming increasingly complex, making it difficult for conventional threshold- and rule-based anomaly detection methods to accurately identify these advanced anomalies. Third, how to effectively link dynamic data quality assessment results with actual regulatory decisions to achieve a hierarchical, scenario-based, and resource-optimized intelligent regulatory strategy still lacks efficient, intelligent, and scalable system support.

[0004] Therefore, there is an urgent need to construct an intelligent data quality assessment system that supports multi-source heterogeneous data acquisition, asynchronous integrity detection, and multi-feature fusion, and to innovatively achieve unsupervised anomaly fingerprint detection, anomaly tracing, and hierarchical adaptive regulatory measures linked to regional pollution conditions. Only by establishing intelligent systems across the entire chain from the data layer and analysis layer to the decision-making layer, and achieving efficient regulation of heavy-duty vehicle emissions throughout their entire lifecycle, in all scenarios, and throughout the entire process, can we effectively improve the level of road mobile source emission control in my country and contribute to the continuous improvement of ambient air quality and the construction of ecological civilization. Therefore, in response to the above needs and challenges, this paper proposes a high-quality intelligent regulatory method for the entire process of remote emission data from heavy-duty vehicles, which has great practical value and application prospects. Summary of the Invention

[0005] This invention aims to solve the technical challenges in existing remote monitoring of heavy-duty vehicle emissions, such as multi-source heterogeneous data, incomplete data collection, weak ability to identify abnormal behavior, and crude regulatory response and poor real-time performance. Specifically, it includes how to achieve highly robust quality assessment of multi-dimensional emission-related data from vehicle terminals, automated fingerprint detection and tracing of abnormal behavior, and how to implement graded, intelligent, and adaptive regulatory strategy feedback based on data quality classification and regional environmental risks. This will improve the timeliness, accuracy, and reliability of emission data, effectively identify and trace anomalies such as cheating, tampering, and sensor degradation, and achieve efficient and intelligent management of the entire process of road mobile source emissions.

[0006] To achieve the above objectives, the present invention employs the following technical solution: the method comprises: Multi-source heterogeneous data acquisition and standardization integration: Construct a heterogeneous data acquisition module compatible with multiple brands, vehicle models, and terminal protocols to synchronously acquire multi-source emission-related data from CAN bus, OBD interface, and external telemetry devices; Real-time integrity and consistency detection of time-series data; forward and backward autoregressive prediction of historical and latest data; if the prediction error between the two sides jumps significantly in the sensitive range, timely capture of temporary breakpoints, distortion, and missing data integrity points of imputation abnormalities, and automatically label the integrity level. A data quality assessment that integrates multiple features can achieve highly robust self-identification of multi-dimensional quality indicators such as data validity, accuracy, and timeliness. Unsupervised dynamic detection and tracing of anomalies involves segmenting the time-series data of the vehicle terminal into different time windows, extracting feature residual vectors through a residual autoencoder, and then clustering the residual space to detect potential anomaly fingerprints. This process identifies and marks the sources of cheating devices, data tampering, and sensor degradation anomalies, and outputs alarm levels. The adaptive strategy-driven hierarchical regulatory feedback automatically matches regulatory actions based on data quality levels and adjusts regulatory priorities and measure intensity based on dynamic feedback of regional pollution status and environmental factors.

[0007] In one approach, the multi-source heterogeneous data acquisition and standardization fusion also includes a dynamic tag-driven data element integration mechanism: based on the device's self-reported ID and the time-varying sensor health code, similar parameters from different sources are automatically standardized to a unified code, thereby achieving automated heterogeneous data standardization and adaptive error correction for device anomalies.

[0008] In one approach, the real-time integrity and consistency detection of time-series data specifically includes: for standardized and integrated data streams, the system continuously maintains historical time-series sequences and integrates the latest batch of data in real time. First, based on the forward autoregressive model, the latest value is predicted using historical data. In contrast, the historical point is predicted using backward autoregression starting from the latest moment. The system captures potential sudden breaks, unnatural jumps, and spoofed interpolation in the bidirectional data. When the two prediction errors change drastically in the sensitive interval and the forward and backward prediction results are inconsistent, it determines that there is a temporal integrity defect at that point or in that segment.

[0009] In one approach, the data quality assessment that integrates multiple features includes: introducing a multi-layer nested feature quality neural network (MNFQNN), which not only receives the original signal sequence, but also integrates the output temporal integrity error features, statistical indicators, and dynamic change features of abnormal fluctuation rates. The system simultaneously incorporates the health score of the acquisition device, the self-test code of the sensor, the output of the driver's driving behavior model, and the current working scene label of the vehicle. All features are processed hierarchically through a nested feature encoder. The first layer uses multi-branch convolution and gated recurrent unit (GRU) to extract spatiotemporal features from the original sequence and statistics, and outputs the hidden state vector. The second nested layer aggregates the device health score, behavior probability and scene label context features through a fully connected transformation to obtain the comprehensive context state. Subsequently, a historical distribution reference of vehicles and environments in the same location is introduced to achieve a comparison of residuals between cross-samples and current features; Finally, the fused output is processed by a classifier and a regression head to output a hierarchical multidimensional quality score of the data. During training, a multi-task loss function is used to balance the quality level of manual annotations and the historical supervision signals of the system, thereby optimizing the overall error.

[0010] In one approach, the unsupervised dynamic detection and source tracing of anomalies includes: proposing a multi-scale residual autoencoder clustering algorithm to achieve unsupervised automatic diagnosis of anomalies in heavy-duty vehicle on-board terminal data at different time scales; firstly, the time-series data that has completed quality assessment is divided into multiple time windows to obtain a multi-scale segmented dataset; for each data segment at each scale, it is input into a trained residual autoencoder structure, including an encoder and a decoder, to obtain the encoded vector and the reconstructed output, respectively, and the residual vector between the original input and the reconstructed output is calculated; Next, the residual vectors at all scales are uniformly mapped into the high-dimensional feature space, and an improved unsupervised clustering method is used to automatically identify the abnormal pattern clusters in the residual space; for residual vectors classified as low-density boundaries or isolated points, their abnormality is quantified according to the Euclidean distance or Mahalanobis distance from the cluster center. By associating the morphological characteristics and occurrence cycle of abnormal residuals with vehicle and equipment information, specific anomaly types can be further matched, including data interpolation, terminal cheating, and sensor progressive drift. Ultimately, each abnormal segment is assigned a source-tracing label and alarm level, and the results are pushed to the regulatory authorities, enabling unsupervised, dynamic, and interpretable detection and source tracing of vehicle data anomalies.

[0011] In one approach, the adaptive policy-driven hierarchical regulatory feedback includes: First, all vehicle data is assigned a multi-dimensional quality score vector and an anomaly fingerprint alarm level. An adjustable strategy activation threshold vector and priority weight are pre-defined for each regulatory action. The core of regulatory decision-making is a hierarchical mapping mechanism: First, for a single vehicle's current all-dimensional quality and anomaly status, a comprehensive evaluation score is calculated. Using a piecewise step activation function to The value range corresponds one-to-one with the available regulatory measures, i.e., step mapping, to achieve automatic hierarchical response; At the same time, the regulatory feedback strategy is not isolated. It integrates regional pollution factors in real time and combines dynamic adjustments with new dynamic thresholds to prioritize regulatory resources for key areas or periods of high environmental risk.

[0012] In one scheme, the multi-source heterogeneous data includes: NOx, CO2, particulate matter, fuel consumption, GPS trajectory, and terminal operating status.

[0013] Beneficial effects of this invention: This invention offers the following advantages: First, by introducing an automatic quality assessment and fusion mechanism for multi-source heterogeneous data, it effectively improves the completeness, accuracy, and timeliness of heavy-duty vehicle emission data, significantly reducing the risk of regulatory misjudgments caused by aging acquisition equipment, communication anomalies, or data tampering. Second, the innovatively proposed multi-scale residual autoencoder clustering algorithm enables unsupervised dynamic detection and high-precision tracing of various anomaly fingerprints, accurately identifying complex anomalies such as cheating, data tampering, and sensor degradation, providing data support for anomaly liability identification and risk classification. Third, the multi-level strategy adaptive decision-making method based on data quality and anomaly detection results can intelligently match optimal regulatory measures according to actual regulatory needs and regional environmental risks, achieving dynamic resource allocation, priority intervention for key vehicles, and closed-loop regulatory feedback, significantly improving the scientific nature, precision, and response efficiency of regulation. Overall, this invention breaks through the technical bottlenecks of traditional heavy-duty vehicle emission supervision, such as narrow coverage and crude anomaly identification. It can ensure the authenticity and reliability of emission data, and provides strong technical support and innovative solutions for combating illegal emissions, controlling air pollution and promoting smart environmental protection supervision. It has broad application prospects and significant environmental and social benefits. Attached Figure Description

[0014] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a flowchart illustrating the implementation of the method of the present invention. Detailed Implementation

[0015] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Typical embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0016] Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. To facilitate understanding, the invention will now be described more fully with reference to the accompanying drawings. Typical embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to make the disclosure of the invention more thorough and complete.

[0017] like Figure 1 and Figure 2 The aforementioned method for monitoring the data quality of onboard terminals for remote emission management of heavy-duty vehicles specifically includes: Step 1: Multi-source heterogeneous data acquisition and standardized fusion Construct a heterogeneous data acquisition module compatible with multiple brands, vehicle models, and terminal protocols to achieve synchronous acquisition of multi-source emission-related data (such as NOx, CO2, particulate matter, fuel consumption, GPS trajectory, terminal operating status, etc.) from CAN bus, OBD interface, and external telemetry devices.

[0018] In achieving multi-source heterogeneous data acquisition and standardized fusion, the first step is to build a highly adaptable data acquisition module that is compatible with different heavy-duty vehicle brands, models, and terminal protocols. This module must not only support mainstream CAN bus and OBD interfaces but also be expanded to connect to various external telemetry devices, enabling multi-channel real-time synchronous acquisition of emission-related data such as NOx, CO2, particulate matter, fuel consumption, GPS trajectory, and terminal operating status. Due to significant differences in the original data formats, data frequencies, and parameter naming among various manufacturers or devices, a dynamic tag-driven data element integration mechanism is introduced.

[0019] A dynamic tag-driven data element integration mechanism is adopted: based on the device's self-reported ID and the time-varying sensor health code, similar parameters from different sources are automatically standardized into a unified code, realizing automated heterogeneous data standardization and adaptive error correction for device anomalies.

[0020] This mechanism uses each device's self-reported unique ID as an index, combined with the embedded time-varying sensor health code as a dynamic reference for data validity. First, all collected parameters are tagged: each data stream is automatically matched to a unified mapping table before being entered into the database. Algorithms automatically identify and correct the names, units, precision, and value ranges of the same parameters under different protocols, unifying them under the platform's standard data coding system. Based on this, for duplicate or abnormal data, the sensor self-check status and historical stability recorded in the device's health code are prioritized for detection. Cross-comparison of the same type of parameters from different sources and time points enables redundancy verification and anomaly correction for individual data elements. When short-term information loss or abnormal jumps are detected, automatic interpolation or correction can be performed based on historical data patterns and the internal logical relationships of data within the same group, improving the completeness and consistency of the final data quality. The entire process is driven by dynamic tags, maintaining high robustness in heterogeneous data integration and standardization under complex operating conditions such as device model changes, mixed use of new and old sensors, and protocol iteration upgrades, laying a high-quality data foundation for subsequent intelligent data quality assessment and refined supervision.

[0021] Step 2: Real-time integrity and consistency detection of time-series data Design a bidirectional autoregressive sequence consistency algorithm: simultaneously perform forward and backward autoregressive predictions on the collected historical data and the latest data. If the prediction error between the two sides changes significantly in the sensitive range, it will promptly capture data integrity gaps such as temporary breakpoints, distortions, and imputation anomalies, and automatically label the integrity level.

[0022] In the process of real-time integrity and consistency detection of time-series data, an innovative bidirectional autoregressive time-series comparison algorithm was adopted to efficiently identify integrity anomalies and time-series distortions in heavy-duty vehicle on-board terminal data during remote emission management. Specifically, for standardized and integrated data streams, historical time-series sequences are continuously maintained. And access the latest batch of data in real time. First, based on a forward autoregressive model (such as first-order or multi-order autoregressive AR, with parameters adaptively selected), the latest value is predicted using historical data: in These are the regression coefficients for dynamic self-learning. In contrast, backward autoregression is used to predict historical points starting from the latest timeframe: at this time These are the regression parameters learned from backward samples. The error is calculated by comparing the latest true values ​​with the forward predicted values. And the error between historical values ​​and backward predictions. This algorithm can bidirectionally capture potential sudden breaks, unnatural jumps, and falsified interpolation in the data. When these two prediction errors change drastically within a sensitive range (such as a dynamic threshold based on statistical distribution) and the forward and backward prediction results are inconsistent, it can be determined that there is a temporal integrity defect at that point or in that segment. Furthermore, based on the distribution characteristics and magnitude of the error, an integrity level (such as high, moderate, or low) is assigned to each data record and annotated on the data element in real time, facilitating subsequent intelligent supervision and key investigations. This algorithm can handle the characteristics of time-varying data and has the ability to identify complex problems such as abnormal disturbances and falsified interpolation, effectively supporting high-reliability quality assurance of key data streams in remote emission supervision of heavy-duty vehicles.

[0023] Step 3: Data quality assessment by fusing multiple features Construct a multi-level nested feature quality neural network (MNFQNN): The input includes multi-dimensional features such as original signal, statistical features, abnormal morphology, data fluctuation rate, and health score of acquisition equipment. It embeds driver behavior model and work scene label, and performs cross-sample comparison based on the distribution of the same vehicle and environment in history, so as to achieve highly robust self-identification of multi-dimensional quality indicators such as data validity, accuracy, and timeliness.

[0024] In the intelligent assessment of multi-feature data quality, a multi-layer nested feature quality neural network (MNFQNN) was introduced for the first time to achieve highly robust automatic assessment of the multi-dimensional quality of heavy-duty vehicle emission data. This model not only receives the original signal sequence... It also incorporates the temporal integrity error feature E output from step 2, and statistical indicators such as the mean. ,variance Anomaly score A, and abnormal fluctuation rate Dynamic change characteristics, etc. In addition, a health score of the data collection device will be introduced simultaneously. Sensor self-check codes and driver behavior model outputs (such as probability distributions for rapid acceleration and prolonged idling) ), and the vehicle's current working scenario label (e.g., nested categories such as mountainous areas, cities, and nighttime). All features are processed hierarchically through a nested feature encoder. The first layer uses multi-branch convolution and gated recurrent units (GRUs) to extract spatiotemporal features from the original sequence and statistics, outputting a latent state vector. The second nested layer aggregates contextual features such as device health score, behavior probability, and scene label through fully connected transformation to obtain a comprehensive contextual state. Subsequently, the multi-layer fusion module uses a gating fusion unit to... and The data is then stitched together, and a historical distribution reference for vehicles in the same vehicle and environment is introduced (by retrieving high-quality historical samples of vehicles with the same VIN or on the same route to obtain the distribution baseline). This allows for residual comparison between cross-samples and the current feature: Finally, the fused output is processed by a classifier and a regression head to produce a hierarchical multidimensional quality score of the data. During training, a multi-task loss function is used, taking into account the quality level of manual annotation. Optimize overall error by comparing with historical monitoring signals. This multi-layered nested neural network possesses the comprehensive ability to identify and attribute quality issues to complex factors such as original anomalies, equipment heterogeneity, and environmental changes. It enables adaptive, multi-dimensional, and dynamic evaluation of indicators such as data validity, accuracy, and timeliness in remote emission monitoring operations, providing strong support for data anomaly detection, key tracking, and intelligent strategy grading.

[0025] Step 4: Unsupervised dynamic detection and source tracing of anomalies A multi-scale residual autoencoder clustering (MSRAC) algorithm is proposed: Time-series data from vehicle terminals is segmented into different time windows, feature residual vectors are extracted using a residual autoencoder, and then the residual space is clustered to detect potential anomaly fingerprints. This enables the identification and tracing of anomalies such as cheating devices, data tampering, and sensor degradation, and outputs alarm levels.

[0026] In the unsupervised dynamic detection and source tracing of anomaly fingerprints, a multi-scale residual autoencoder clustering (MSRAC) algorithm is proposed to achieve unsupervised automatic diagnosis of anomalies in heavy-duty vehicle on-board terminal data at different time scales. First, time-series data that has undergone quality assessment is... By sliding the segmentation according to multiple time windows (e.g., 1 minute, 10 minutes, 1 hour, etc.), a multi-scale segmented dataset is obtained. , where m represents the time scale. For each data segment at each scale The input is fed into the trained residual autoencoder structure, including the encoder. With decoder , respectively obtain the encoding vector and reconstructed output Calculate the residual vector between the original input and the reconstructed output. This residual vector can characterize the reconstruction bias patterns of time-series data at different scales due to anomalies, tampering, sensor degradation, etc. Next, the residual vectors at all scales are uniformly mapped into a high-dimensional feature space, and an improved unsupervised clustering method (such as adaptive density peak clustering (ADC) or spectral clustering) is used to automatically identify anomalous pattern clusters in the residual space. For residual vectors classified as low-density boundaries or isolated points, their anomalousness is determined based on the Euclidean or Mahalanobis distance from the cluster center. Quantify it. For example... in Let be the center and covariance matrix of the normal cluster. When the threshold of the multi-scale residual distribution is exceeded, the segment is identified as a potential anomaly fingerprint. By associating the morphological features and occurrence cycle of the anomaly residuals with vehicle equipment information, specific anomaly types are further matched, such as data imputation, terminal fraud (e.g., sudden zeroing, forged continuity), and sensor progressive drift. Finally, the algorithm assigns a source-tracing label and alarm level (e.g., high-risk equipment tampering, moderate sensor aging, mild short-term anomaly) to each anomaly segment and pushes the results to the regulatory end, achieving unsupervised, dynamic, and interpretable detection and source tracing of vehicle data anomalies.

[0027] Step 5: Adaptive Policy-Driven Hierarchical Regulatory Feedback Based on the multidimensional results of data quality assessment, a hierarchical policy-driven adaptation algorithm is adopted: it automatically matches regulatory actions (such as remote review, on-site spot checks, terminal calibration, vehicle restrictions, non-intervention and other multi-level measures) according to the data quality level, and dynamically adjusts the regulatory priority and the intensity of measures in combination with environmental factors such as regional pollution status.

[0028] In the adaptive policy-driven hierarchical regulatory feedback mechanism, a multi-level policy-driven adaptive decision-making algorithm is innovatively constructed by fully utilizing the aforementioned multi-dimensional data quality assessment and anomaly fingerprint tracing results. First, the current data of all vehicles is assigned a multi-dimensional quality score vector. and abnormal fingerprint alarm levels (e.g., high, medium, low, no alarm). For each regulatory action (e.g., remote review) On-site spot checks Terminal calibration Vehicle restrictions No intervention (etc.) Pre-set adjustable strategy activation threshold vector and priority weight The core of regulatory decision-making is a hierarchical mapping mechanism: First, for the current overall quality and abnormal state of a single vehicle, a comprehensive evaluation score is calculated. in The strategy weights for each quality dimension, This represents the alarm weight coefficient for abnormal fingerprints. A piecewise step activation function is used to... The range of values ​​corresponds one-to-one with the available regulatory measures, i.e., a step mapping. The system enables automated tiered response: high-risk vehicles are prioritized for on-site inspections or traffic restrictions; suspected abnormal equipment is remotely verified; low- to mid-level objections trigger terminal calibration; and vehicles with good quality enter a no-intervention state. Simultaneously, the regulatory feedback strategy is not isolated, but integrates regional pollution factors in real time. (Regional environmental heavy pollution early warning, road section sensitivity level, traffic saturation, etc.), through dynamic adjustment With the new dynamic threshold In conjunction with this approach, regulatory resources are prioritized for key areas or periods of high environmental risk. Furthermore, based on historical regulatory results and strategy implementation effectiveness, the weighting is dynamically adjusted after each regulatory feedback session, leveraging a learning-based approach. Abnormal weights The mechanism establishes various activation thresholds to improve overall handling efficiency and the ability to rationally allocate resources. It achieves a hierarchical, fine-grained, strategy-adaptive, and environmentally-linked closed-loop system for remote heavy-duty vehicle emission monitoring, ensuring precise and efficient supervision and real-time risk control.

[0029] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0030] It should be understood that the above detailed description of the technical solutions of the present invention with reference to preferred embodiments is illustrative and not restrictive. Those skilled in the art can modify the technical solutions described in the embodiments or make equivalent substitutions for some of the technical features based on reading this specification; however, these modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring the data quality of heavy-duty vehicles using on-board terminals for remote emission management, characterized in that: The method includes: Multi-source heterogeneous data acquisition and standardization integration: Construct a heterogeneous data acquisition module compatible with multiple brands, vehicle models, and terminal protocols to synchronously acquire multi-source emission-related data from CAN bus, OBD interface, and external telemetry devices; Real-time integrity and consistency detection of time-series data; forward and backward autoregressive prediction of historical and latest data; if the prediction error between the two sides jumps significantly in the sensitive range, timely capture of temporary breakpoints, distortion, and missing data integrity points of imputation abnormalities, and automatically label the integrity level. A data quality assessment that integrates multiple features can achieve highly robust self-identification of multi-dimensional quality indicators such as data validity, accuracy, and timeliness. Unsupervised dynamic detection and tracing of anomalies involves segmenting the time-series data of the vehicle terminal into different time windows, extracting feature residual vectors through a residual autoencoder, and then clustering the residual space to detect potential anomaly fingerprints. This process identifies and marks the sources of cheating devices, data tampering, and sensor degradation anomalies, and outputs alarm levels. The adaptive strategy-driven hierarchical regulatory feedback automatically matches regulatory actions based on data quality levels and adjusts regulatory priorities and measure intensity based on dynamic feedback of regional pollution status and environmental factors.

2. The method for monitoring the data quality of vehicle-mounted terminals for remote emission management according to claim 1, characterized in that: The multi-source heterogeneous data acquisition and standardization fusion also includes a data element integration mechanism driven by dynamic tags: based on the device's self-reported ID and the time-varying sensor health code, similar parameters from different sources are automatically standardized into a unified code, realizing automated heterogeneous data standardization and adaptive error correction for device anomalies.

3. The method for monitoring the data quality of vehicle-mounted terminals for remote emission management according to claim 1, characterized in that: The aforementioned real-time integrity and consistency detection of time-series data specifically includes: for standardized and integrated data streams, the system continuously maintains historical time-series sequences and integrates the latest batch of data in real time. First, based on the forward autoregressive model, the latest value is predicted using historical data. In contrast, the historical point is predicted using backward autoregression starting from the latest moment. The system captures sudden breaks, unnatural jumps, and spoofing in the bidirectional data. When these two prediction errors change drastically in the sensitive interval and the forward and backward prediction results are inconsistent, it is determined that there is a temporal integrity defect at that point or in that segment.

4. The method for monitoring the data quality of vehicle-mounted terminals for remote emission management according to claim 1, characterized in that: The data quality assessment that integrates multiple features includes the introduction of a multi-layer nested feature quality neural network (MNFQNN), which not only receives the original signal sequence, but also integrates the output temporal integrity error features, statistical indicators, and dynamic change features of abnormal fluctuation rates. The system simultaneously incorporates the health score of the acquisition device, the self-test code of the sensor, the output of the driver's driving behavior model, and the current working scene label of the vehicle. All features are processed hierarchically through a nested feature encoder. The first layer uses multi-branch convolution and gated recurrent unit (GRU) to extract spatiotemporal features from the original sequence and statistics, and outputs the hidden state vector. The second nested layer aggregates the device health score, behavior probability and scene label context features through a fully connected transformation to obtain the comprehensive context state. Subsequently, a historical distribution reference of vehicles and environments in the same location is introduced to achieve a comparison of residuals between cross-samples and current features; Finally, the fused output is processed by a classifier and a regression head to output a hierarchical multidimensional quality score of the data. During training, a multi-task loss function is used to balance the quality level of manual annotations and the historical supervision signals of the system, thereby optimizing the overall error.

5. A method for monitoring the data quality of a remote emission management vehicle terminal, as described in claim 1, characterized in that: The unsupervised dynamic detection and source tracing of anomalies includes: proposing a multi-scale residual autoencoder clustering algorithm to achieve unsupervised automatic diagnosis of anomalies in heavy-duty vehicle on-board terminal data at different time scales; firstly, the time-series data that has completed quality assessment is divided into multiple time windows to obtain a multi-scale segmented dataset; for each data segment at each scale, it is input into a trained residual autoencoder structure, including an encoder and a decoder, to obtain the encoded vector and the reconstructed output, respectively, and the residual vector of the original input and the reconstructed output is calculated; Next, the residual vectors at all scales are uniformly mapped into the high-dimensional feature space, and an improved unsupervised clustering method is used to automatically identify the abnormal pattern clusters in the residual space; for residual vectors classified as low-density boundaries or isolated points, their abnormality is quantified according to the Euclidean distance or Mahalanobis distance from the cluster center. By associating the morphological characteristics and occurrence cycle of abnormal residuals with vehicle and equipment information, specific anomaly types can be further matched, including data interpolation, terminal cheating, and sensor progressive drift. Ultimately, each abnormal segment is assigned a source-tracing label and alarm level, and the results are pushed to the regulatory authorities, enabling unsupervised, dynamic, and interpretable detection and source tracing of vehicle data anomalies.

6. A method for monitoring the data quality of a remote emission management vehicle terminal, as described in claim 1, characterized in that: The adaptive policy-driven hierarchical regulatory feedback includes: First, all vehicle data is assigned a multi-dimensional quality score vector and an anomaly fingerprint alarm level. An adjustable strategy activation threshold vector and priority weight are pre-defined for each regulatory action. The core of regulatory decision-making is a hierarchical mapping mechanism: First, for a single vehicle's current all-dimensional quality and anomaly status, a comprehensive evaluation score is calculated. Using a piecewise step activation function to The value range corresponds one-to-one with the available regulatory measures, i.e., step mapping, to achieve automatic hierarchical response; At the same time, the regulatory feedback strategy is not isolated. It integrates regional pollution factors in real time and combines dynamic adjustments with new dynamic thresholds to prioritize regulatory resources for key areas or periods of high environmental risk.

7. A method for monitoring the data quality of a remote emission management vehicle terminal as described in claim 1, characterized in that: The aforementioned multi-source heterogeneous data includes: NOx, CO2, particulate matter, fuel consumption, GPS trajectory, and terminal operating status.

Images