Heavy diesel vehicle emission degradation prediction system and method

Abstract

The invention belongs to the technical field of vehicle exhaust emission monitoring, and discloses a heavy diesel vehicle emission degradation prediction system and method. The method comprises the following steps: vehicle type family classification and PEMS data source acquisition; pEMS data transient acquisition and steady-state window extraction are carried out; building and training a multi-mileage model; based on the accuracy verification result of the multi-mileage model, creating a digital twinborn body for each in-use vehicle in the vehicle type family; calculating a preliminary degradation coefficient; and calculating a correction degradation coefficient according to the predicted correction value of the reference model and the predicted correction value of each current mileage section model, comparing the correction degradation coefficient with a preset degradation threshold value, judging whether the correction degradation coefficient exceeds the preset degradation threshold value or not, if yes, carrying out preset degradation threshold value early warning, and carrying out in-vehicle maintenance closed loop according to early warning information. According to the invention, real-time accurate monitoring is realized, and low-cost, real-time and full-coverage emission degradation monitoring is realized; and automatic threshold judgment and precise maintenance closed loop are realized, and the supervision efficiency and compliance rate are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle tail gas emission monitoring, and in particular relates to a system and method for predicting emission degradation of heavy-duty diesel vehicles. Background Art

[0002] Heavy-duty diesel vehicles are a major source of road traffic pollutants, and their emissions of NOx and particulate matter significantly impact urban air quality. Consequently, monitoring emissions from in-use heavy-duty diesel vehicles and identifying high-emission vehicles have long been a challenge and a key focus of environmental regulation. However, traditional in-use vehicle emissions monitoring methods have numerous shortcomings in real-world road conditions, making it difficult to promptly and comprehensively identify vehicles with severely degraded emissions.

[0003] Currently, common methods for regulating in-use vehicle emissions include regular annual emissions inspections, roadside spot checks, and on-board diagnostics (OBD) fault code monitoring. However, these methods lack real-time performance, coverage, and accuracy. For example, portable emissions measurement systems (PEMS) can be used for on-road vehicle emissions testing. However, due to high equipment costs, complex and time-consuming testing, they are only used in limited situations, such as new vehicle type approvals, making it difficult to continuously monitor a large number of in-use vehicles in real time. Furthermore, while OBD systems can monitor some engine and aftertreatment faults in real time, they can only detect a limited number of pre-defined fault conditions and cannot fully assess the gradual degradation of emissions performance. When emission control component performance deteriorates but the OBD fault light has not yet been triggered, the vehicle may already be experiencing elevated emissions levels without being detected. Furthermore, studies have shown that OBD detection has a high false positive rate in identifying high-emitting vehicles, with up to approximately 50% of actual high-emitting vehicles not being accurately detected by OBD. In other words, many vehicles in use with excessive emissions may pass the OBD self-check but fail the actual emissions test, missing the opportunity for timely repair and control. This leads to the risk of underreporting in OBD-based monitoring, and simply using OBD instead of exhaust pollutant testing cannot ensure the accurate identification of all high-emitting vehicles.

[0004] At the same time, traditional emission tests (such as the simple working condition method or annual exhaust gas test) can only give a conclusion on whether the vehicle is currently meeting the standards, and cannot reflect the degree of emission degradation of the vehicle relative to its initial state when it leaves the factory. When vehicle emissions approach the critical point of regulatory limits, existing means are difficult to identify and warn in advance, and they are often not discovered until emissions exceed the standards or the fault is serious. This lagging regulatory method causes vehicles with high emission hazards to be on the road for a long time, causing excessive emissions. In summary, the existing PEMS field tests and OBD monitoring methods have problems such as poor real-time performance, narrow coverage, and high false positives and missed judgments in actual road emission supervision. It is difficult to identify and locate high-emission in-use vehicles that require maintenance and treatment in a timely and accurate manner. At present, the industry also lacks a system and method for real-time evaluation of the emission degradation trends of multiple pollutants such as NOx, THC, and CO, which cannot meet the growing demand for refined supervision of in-use vehicle emissions. Summary of the Invention

[0005] To overcome the challenges of related technologies, the present invention discloses a system and method for predicting exhaust emissions degradation of heavy-duty diesel vehicles. Specifically, the system and method are based on machine learning and digital twin technology to predict, assess, and monitor the degradation of NOx, THC, and CO emissions of heavy-duty diesel vehicles in real time.

[0006] The technical solution is as follows: A method for predicting emission degradation of heavy-duty diesel vehicles comprises the following steps:

[0007] S1, vehicle family classification and PEMS data source acquisition;

[0008] S2, based on the classification and acquisition results, performs transient acquisition of PEMS data and steady-state window extraction;

[0009] S3, builds and trains multi-mileage segment models based on the extracted data;

[0010] S4, verify the accuracy of the trained multi-mileage segment model;

[0011] S5, based on the accuracy verification results, deploy the multi-mileage segment model on the cloud digital twin platform and create a digital twin for each in-use vehicle in the model family;

[0012] S6: Based on the created digital twin, the cloud-based digital twin platform uses the average operating condition parameters of the latest steady-state window to obtain the predicted values ​​of the baseline model and the current mileage segment model, and performs preliminary degradation coefficient calculations based on the predicted values.

[0013] S7, using the average residual to correct the obtained prediction value, respectively obtaining a corrected baseline model prediction correction value and a corrected prediction correction value of each current mileage segment model; calculating a correction degradation coefficient based on the prediction correction value and outputting it;

[0014] S8, comparing the output corrected degradation coefficient with the preset degradation threshold to determine whether the corrected degradation coefficient exceeds the preset degradation threshold. If so, a preset degradation threshold warning is issued, and closed-loop maintenance of the in-use vehicle is performed based on the warning information.

[0015] In step S1, the vehicle type family classification includes: classifying the heavy-duty diesel vehicles in use into several vehicle type families according to the vehicle emission limit category, engine model and after-treatment device configuration elements;

[0016] PEMS data sources include: PEMS data used by regulatory authorities for compliance supervision and inspection and PEMS data used by manufacturers for compliance self-inspection.

[0017] In step S2, transient acquisition includes: the PEMS device collects transient data of the vehicle under real road conditions at a frequency of ≥1 Hz, including engine speed, load, SCR temperature, exhaust flow, fuel injection amount, and transient concentrations of NOx / THC / CO emissions;

[0018] Steady-state window extraction involves binning the raw transient data by vehicle-specific power, applying interquartile range (IQR) filtering to the intake, exhaust, and injection system parameters, and filtering for continuous steady-state data windows within each power interval.

[0019] In step S3, the multi-mileage segment model includes a baseline model and a segmented model;

[0020] Multi-mileage segment model construction, including:

[0021] Build a benchmark model and a PEMS window dataset for new vehicles in factory condition, which is the first mileage segment model;

[0022] Build segmented models: Construct window data sets of 0–100,000, 100,000–200,000, 200,000–300,000, 300,000–400,000, 400,000–500,000, 500,000–600,000, and 600,000–700,000 kilometers for the 2nd to 8th mileage segment models;

[0023] For each vehicle family, the random forest RF algorithm is used to train the baseline model and the segmented model using steady-state window data, so that the performance standards of the baseline model and the segmented model include: the corrected determination coefficient R 2 ≥0.90; mean absolute error MAE ≤ 5%; root mean square error RMSE ≤ 10%.

[0024] In step S4, the accuracy of the trained multi-mileage segment model is verified, including independent test set evaluation, prediction-measurement comparison, and degradation coefficient error verification.

[0025] In step S5, a digital twin is created for each in-use vehicle in the vehicle family, including real-time uploading of transient operating condition data such as engine speed, load, SCR temperature, exhaust flow, fuel injection amount, and transient concentration of NOx / THC / CO emissions via remote OBD. The digital twin aggregates and displays the real-time data and uses it as input for the baseline model and segmented model.

[0026] In step S6, the cloud digital twin platform uses the average operating condition parameters of the latest steady-state window to obtain the predicted values ​​of the benchmark model and the current mileage segment model, and performs preliminary degradation coefficient calculation based on the predicted values, including: inputting the average operating condition parameters extracted from the latest steady-state window into the benchmark model and the current mileage segment model at the same time to obtain P base and P current , calculate the preliminary degradation coefficient, the expression is:

[0027]

[0028] Where K0 is the initial degradation coefficient, P base is the pollutant emission value predicted by the benchmark model, P base The pollutant emission values ​​predicted by the current model for each mileage segment.

[0029] In step S7, the obtained prediction value is corrected using the average residual to obtain the corrected prediction value of the baseline model and the corrected prediction value of each current mileage segment model, including:

[0030] Periodic calibration: randomly select multiple vehicles from each mileage segment every month to conduct PEMS emergency tests to obtain the measured emission value P real , calculate the residual Δ base and Δ current :

[0031] Δ base =P real -P base

[0032] Δ current =P real -P current

[0033] Bias correction, which corrects the predicted values ​​by the mean residuals:

[0034]

[0035] Then calculate the correction degradation coefficient K1:

[0036]

[0037] Where, P real is the actual emission value of the vehicle under the same working conditions measured by PEMS, Δ base is the residual between the predicted value of the benchmark model and the measured value, Δ current is the residual between the current model prediction value and the measured value, N is the number of test vehicles involved in the calibration, and P' base is the predicted value of the calibrated benchmark model, P' current is the predicted value of the current model after correction.

[0038] Furthermore, in the process of calculating the correction degradation coefficient K1, the average residual is continuously optimized dynamically to obtain the average residual;

[0039] If the average residual after continuous periodic correction and bias correction exceeds the tolerance, the new data will be included in the training set and the multi-mileage segment model will be updated using incremental learning or retraining.

[0040] In step S8, the preset degradation thresholds include: NOx, 1.20; THC, 1.10; CO, 1.15;

[0041] When the correction degradation coefficient K1 continues to exceed the threshold, the system automatically triggers an early warning and notifies the operator and regulatory authorities via SMS, email, and APP.

[0042] Another object of the present invention is to provide a heavy-duty diesel vehicle emission degradation prediction system, which implements the heavy-duty diesel vehicle emission degradation prediction method, and the system includes:

[0043] The data transient acquisition and steady-state window extraction module is used for vehicle family classification and PEMS data source acquisition. Based on the classification and acquisition results, PEMS data transient acquisition and steady-state window extraction are performed.

[0044] The multi-mileage segment model construction and training module is used to construct and train the multi-mileage segment model based on the extracted data; and to verify the accuracy of the trained multi-mileage segment model;

[0045] A preliminary degradation coefficient calculation module is used to deploy multi-mileage segment models on the cloud-based digital twin platform based on accuracy verification results and create a digital twin for each in-use vehicle in the model family. Based on the created digital twin, the cloud-based digital twin platform uses the average operating condition parameters extracted from the latest steady-state window to obtain predicted values ​​for the baseline model and the current mileage segment model, and performs preliminary degradation coefficient calculations based on the predicted values.

[0046] The correction degradation coefficient calculation and warning module is used to correct the obtained prediction value using the average residual to obtain the corrected baseline model prediction correction value and the prediction correction value of the current mileage segment model respectively; calculate the correction degradation coefficient based on the prediction correction value and output it; compare the output correction degradation coefficient with the preset degradation threshold to determine whether the correction degradation coefficient exceeds the preset degradation threshold. If it exceeds, a preset degradation threshold warning is issued, and a closed-loop maintenance of the in-use vehicle is performed based on the warning information.

[0047] In combination with all the above technical solutions, the beneficial effects of the present invention are as follows:

[0048] First, the purpose of the present invention is to provide a "dual-model comparative drive" emission degradation assessment and supervision system, which constructs multi-mileage prediction models for different vehicle model families, utilizes existing PEMS data from regulatory authorities' compliance supervision and manufacturers' compliance self-inspection, and connects digital twins with remote OBD in real time to achieve full life cycle segmented degradation monitoring and intelligent early warning maintenance of NOx, THC, and CO emission performance at low cost.

[0049] Second, the present invention fully utilizes existing data: relying on regulatory and enterprise self-inspection PEMS data, it eliminates the need for extensive new testing;

[0050] Vehicle family-specific models: Develop an independent model for each vehicle family to enhance prediction targeting;

[0051] Full life cycle coverage: 8 mileage segment models cover the entire process from 0 to 700,000 kilometers;

[0052] Real-time and precise monitoring: Second-level transient data + steady-state window extraction + dual-model comparison + residual correction, achieving low-cost, real-time, and full-coverage emission degradation monitoring;

[0053] Intelligent early warning maintenance closed loop: Automatic threshold judgment and precise maintenance closed loop improve supervision efficiency and compliance rate.

[0054] Third, the present invention is expected to significantly reduce the manpower and equipment costs of emission monitoring and reduce resource consumption for vehicle maintenance and supervision by constructing a refined emission degradation prediction model and real-time monitoring on a digital twin platform; commercially, it can form a cloud-based intelligent monitoring and maintenance service platform, generate stable commercial service income, and has broad market prospects.

[0055] Existing emissions monitoring primarily relies on periodic spot checks or static alarms, lacking a real-time, accurate emission degradation prediction and closed-loop maintenance management mechanism for large-scale in-use vehicle fleets. This invention, for the first time, proposes an online monitoring approach that combines a multi-mileage segment model with a digital twin and enables real-time correction. This fills a technological gap in the field of real-time, dynamic emission degradation monitoring for in-use diesel vehicles, both domestically and internationally.

[0056] Fourth, this invention addresses the challenges of traditional emissions monitoring methods, which are unable to track emissions degradation trends in vehicles in real time and prevent timely proactive intervention. By leveraging digital twins for real-time prediction and residual correction, it overcomes bottlenecks such as expensive PEMS equipment, time-consuming monitoring, and insufficient prediction accuracy.

[0057] Fifth, the present invention overcomes the industry's prejudice that emission degradation can only be monitored with high precision by relying on high-cost PEMS or bench test equipment. It proposes an innovative technical route based on the combination of digital twins, real-time data prediction and periodic small-scale PEMS residual correction. It can achieve low-cost, high-precision online emission degradation prediction and maintenance warning, and is a new type of emission control technology idea and method. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure;

[0059] Figure 1 This is a schematic diagram of a method for predicting emission degradation of heavy-duty diesel vehicles provided by an embodiment of the present invention;

[0060] Figure 2 This is a flow chart of a method for predicting emission degradation of heavy-duty diesel vehicles provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0061] To make the above-mentioned objects, features, and advantages of the present invention more readily apparent, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. The following description sets forth numerous specific details to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art may make similar modifications without departing from the scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0062] The innovation of this invention lies in: the invention proposes a heavy-duty diesel vehicle emission degradation prediction method that integrates a multi-mileage random forest model with digital twin technology. By combining real-time OBD data prediction and periodic PEMS data correction, it achieves accurate quantification and active early warning maintenance of emission degradation of in-use vehicles, breaking through the limitations of traditional methods that are difficult to conduct real-time and accurate supervision.

[0063] Example 1, as Figure 1 As shown in FIG, the principle diagram of the method for predicting emission degradation of heavy-duty diesel vehicles provided by an embodiment of the present invention; Figure 2 As shown, the heavy-duty diesel vehicle emission degradation prediction method provided by the embodiment of the present invention includes the following steps:

[0064] S1, vehicle family classification and PEMS data source acquisition;

[0065] In-use heavy-duty diesel vehicles are divided into several "model families" based on factors such as vehicle emission limit category, engine model, and after-treatment device configuration. Relevant PEMS data comes from PEMS data generated by regulatory authorities during compliance inspections and by manufacturers through self-inspections of in-use PEMS.

[0066] Regulatory authorities use random PEMS sampling data for compliance monitoring;

[0067] PEMS self-inspection data for in-use compliance self-inspection by manufacturers;

[0068] The two extract steady-state windows according to unified standards to form the original data set for training and calibration.

[0069] S2, based on the classification and acquisition results, performs transient acquisition of PEMS data and steady-state window extraction;

[0070] Transient acquisition: The PEMS device collects the vehicle's transient data under real road conditions at a frequency of ≥1Hz, including engine speed, load, SCR temperature, exhaust flow, fuel injection amount, and transient concentrations of NOx / THC / CO emissions.

[0071] Steady-state window extraction: The raw transient data is binned by vehicle-specific power. Interquartile range (IQR) filtering is applied to key parameters of the intake, exhaust, and fuel injection systems. A continuous 20–30 second window of steady-state data is selected within each power range. These windows serve as training and validation samples for the multi-mileage model.

[0072] S3, builds and trains multi-mileage segment models based on the extracted data;

[0073] The multi-mileage segment model includes a benchmark model and a segmented model;

[0074] The multi-mileage segment model construction includes:

[0075] Build a benchmark model and construct a PEMS window dataset for new vehicles in factory condition (Test 1, mileage segment 1), representing the ideal, undegraded emission level.

[0076] Build segmented models: Build window data sets for 0–100,000, 100,000–200,000, 200,000–300,000, 300,000–400,000, 400,000–500,000, 500,000–600,000, and 600,000–700,000 kilometers (Test2–Test8, i.e., the second to eighth mileage segments), and obtain models for each mileage segment; illustratively, models for the second to eighth mileage segments are included;

[0077] For each vehicle family, the random forest (RF) algorithm is used to train the baseline model and the segmented model using steady-state window data.

[0078] The random forest (RF) algorithm optimizes hyperparameters through five-fold cross validation and GridSearchCV. Specifically, the samples are randomly divided into five equal parts on the preprocessed training set, and four parts of the data are used to train the model and the remaining part is used to evaluate the performance. The parameters such as "number of decision trees" (n_estimators), "maximum tree depth" (max_depth), "minimum number of split samples" (min_samples_split), "minimum number of leaf node samples" (min_samples_leaf) and "maximum number of features" (max_features) are traversed one by one within the preset range. GridSearchCV will perform a complete five-fold validation under each hyperparameter combination and calculate the average determination coefficient (R 2 ), mean absolute error (MAE) and root mean square error (RMSE) to determine the optimal configuration. Ensure that the performance of the baseline model and segmented model meets the following standards: Corrected determination coefficient R 2 ≥0.90 (e.g., baseline model R 2 ≥0.95); mean absolute error MAE ≤ 5%; root mean square error RMSE ≤ 10% (relative value).

[0079] S4, verify the accuracy of the trained multi-mileage segment model;

[0080] Independent test set evaluation: 20% of the original window data is set aside for each mileage segment as a test set to evaluate the multi-mileage segment model R 2 , MAE, RMSE;

[0081] Prediction-measurement comparison: Regularly use PEMS data outside the test set to compare the model predictions and measured values ​​for multiple mileage segments, and draw scatter plots and residual distribution diagrams;

[0082] Deterioration coefficient error verification: Evaluate the maximum relative error between the degradation coefficient predicted by the multi-mileage segment model and the degradation coefficient calculated by the steady-state operating method to be ≤10%.

[0083] S5, based on the accuracy verification results, deploy the multi-mileage segment model on the cloud digital twin platform and create a digital twin for each in-use vehicle in the model family;

[0084] Deploy the baseline model and segmented model of each vehicle family in step S3 on the cloud digital twin platform and provide RESTful API services. The baseline model includes the PEMS window data of the new vehicle in the factory state in the first mileage segment, and the segmented model includes the window data of the second to eighth mileage segments, that is, the models of each mileage segment.

[0085] Create a digital twin for each in-use vehicle and upload transient operating data such as engine speed, load, SCR temperature, exhaust flow, fuel injection amount, and transient concentrations of NOx / THC / CO emissions in real time via remote OBD (CAN+MQTT);

[0086] The digital twin aggregates and displays real-time data and uses it as input for baseline and segmented models.

[0087] S6: Based on the created digital twin, the cloud-based digital twin platform uses the average operating condition parameters of the latest steady-state window to obtain the predicted values ​​of the baseline model and the current mileage segment model, and performs preliminary degradation coefficient calculations based on the predicted values.

[0088] For example, the cloud-based digital twin platform (the vehicle is connected to the digital twin platform) automatically matches the current baseline model based on the vehicle's accumulated mileage, and automatically matches each mileage segment model in the segmented model;

[0089] The average operating parameters extracted from the latest steady-state window in step S2 are simultaneously input into the benchmark model and the current mileage segment model, P base and P current ;

[0090] Calculate the preliminary degradation coefficient, the expression is:

[0091]

[0092] Where K0 is the initial degradation coefficient, P base is the pollutant emission value predicted by the baseline model, in g / kWh; P base The pollutant emission values ​​predicted by the current model for each mileage segment are in g / kWh.

[0093] S7, using the average residual to correct the obtained prediction value, respectively obtaining a corrected baseline model prediction correction value and a corrected prediction correction value of each current mileage segment model; calculating a correction degradation coefficient based on the prediction correction value and outputting it;

[0094] Periodic calibration: Every month, 20-30 vehicles are randomly selected from each mileage segment for PEMS emergency testing to obtain the measured emission values ​​P real , calculate the residual Δ base and Δ current :

[0095] Δ base =P real -P base ,Δ current =P real -P current

[0096] Bias correction: Correct the predicted values ​​by the mean residual:

[0097]

[0098] Then calculate the correction degradation coefficient K1:

[0099]

[0100] Where, P real is the actual emission value of the vehicle under the same working conditions obtained by PEMS measurement, in g / kWh; Δ base is the residual between the baseline model prediction value and the measured value, in g / kWh; Δ current is the residual between the current model prediction value and the measured value, in g / kWh; N is the number of test vehicles involved in the calibration, P' base is the predicted value of the benchmark model after correction, in g / kWh; P' current It is the current model prediction value after correction, in g / kWh.

[0101] For example, during the calculation of the correction degradation coefficient, the average residual is continuously and dynamically optimized to obtain an accurate average residual;

[0102] If the average residual after continuous periodic correction and bias correction exceeds the tolerance, the new data will be included in the training set and the multi-mileage segment model will be updated using incremental learning or retraining until the average residual is accurate;

[0103] S8, comparing the output corrected degradation coefficient with a preset degradation threshold to determine whether the corrected degradation coefficient exceeds the preset degradation threshold. If so, a preset degradation threshold warning is issued, and closed-loop maintenance of the in-use vehicle is performed based on the warning information;

[0104] Preset degradation thresholds (e.g. NOx: 1.20, THC: 1.10, CO: 1.15);

[0105] When K1 exceeds the threshold continuously (e.g., cumulatively for 30 minutes), the system automatically triggers an early warning and notifies the operator and regulatory authorities via SMS, email, APP, etc., including the vehicle ID, pollutant type, and degree of degradation;

[0106] The operator arranges for the vehicle to be inspected at a designated maintenance station within a specified time (e.g., 72 hours). Based on the degradation report, technicians will accurately check and perform maintenance on key components such as the SCR catalyst, DOC, DPF, and fuel injection system.

[0107] After the maintenance is completed, the vehicle is run again and data is uploaded. The platform recalculates K1. If it returns to within the threshold, the warning is automatically lifted and the complete maintenance closed-loop record is archived; otherwise, maintenance continues until the requirements are met.

[0108] It can be seen from the above embodiments that the present invention relies on the PEMS data of in-use compliance supervision by regulatory authorities and in-use self-inspection by manufacturers, trains multiple models according to vehicle model families and mileage segments, calculates degradation coefficients through steady-state window extraction of transient data, dual-model comparison, and residual correction, and automatically triggers high emission warnings and maintenance closed loops in combination with thresholds. It can evaluate vehicle emission degradation trends in a low-cost, real-time, and full-coverage manner, significantly improving regulatory efficiency and reliability.

[0109] Example 2, as another possible implementation of the present invention, the method for predicting emission degradation of heavy-duty diesel vehicles provided in this embodiment of the present invention includes:

[0110] Step 1: Vehicle access and data collection.

[0111] Each heavy-duty diesel vehicle in use is equipped with an OBD+MQTT terminal to collect engine speed, load, SCR temperature, exhaust flow, fuel injection amount and PEMS transient emission data in real time through the CAN bus.

[0112] The data is uploaded to the cloud-based digital twin platform at a frequency of 1 Hz.

[0113] Step 2: Steady-state window extraction.

[0114] The cloud-based digital twin platform bins the uploaded transient data according to VSP (Vehicle Specific Power);

[0115] Apply interquartile range filtering to parameters such as intake manifold absolute pressure, exhaust temperature, and fuel injection quantity to eliminate anomalies;

[0116] In each power interval, a continuous 20–30 s stable operating condition segment is selected, and the average value of each parameter is calculated as the steady-state window, which is used as the model input feature.

[0117] Step 3: Model training and verification.

[0118] The random forest (RF) algorithm model is trained separately on the steady-state window data of Test1–Test8 according to the vehicle family, and the hyperparameters such as the number of trees and depth are optimized using GridSearchCV;

[0119] Divide the dataset into 80% training set and 20% test set, and evaluate the model R. 2 , MAE, RMSE;

[0120] Test1 benchmark model example: R 2=0.96, MAE=2.8%, RMSE=4.5%; Test4 (200,000–300,000 km) model: R 2 =0.92, MAE=3.9%, RMSE=4.8%.

[0121] Step 4: Verify model accuracy.

[0122] Using additional PEMS data as a validation set under all operating conditions, a scatter plot of predicted values ​​versus measured values ​​was drawn. The scatter plots were close to the 45° line, and the residuals were concentrated within ±10%.

[0123] The difference between the degradation coefficient prediction and the calculation based on the steady-state operating method is less than 10%, proving the feasibility and stability of the prediction.

[0124] Step 5: Real-time prediction and degradation calculation.

[0125] The cloud-based digital twin platform automatically selects the baseline model and the corresponding mileage models based on the vehicle's accumulated mileage;

[0126] Extract the current steady-state window parameters in real time and call the model to predict P in parallel base and P current ;

[0127] Calculate the preliminary degradation coefficient K0 and store the trend.

[0128] Step 6: Residual correction and model update.

[0129] Every month, 20-30 vehicles with different mileage ranges are randomly tested for PEMS road test to obtain the actual emission P real ;

[0130] Calculate Δ base =P real -P base ,Δ current =P real -P current , take the average

[0131] Correction Then calculate the correction degradation coefficient K1;

[0132] when If the deviation exceeds the limit continuously, incrementally update the model or fully retrain it and replace the old model in the cloud.

[0133] Step 7: Threshold warning and closed-loop maintenance.

[0134] Assume NOx: 1.20, THC: 1.10, CO: 1.15 as degradation thresholds;

[0135] When K1 exceeds the threshold for ≥30 minutes, the system automatically sends a text message / email / App push warning, including the license plate, pollutant type, and degradation degree;

[0136] The operator will take the vehicle to an authorized maintenance point, where technicians will review the deterioration report, focusing on key components such as the SCR, DOC, DPF, and fuel injection system.

[0137] After the repair is completed, the vehicle is driven again and data is uploaded in real time, and the platform recalculates K1;

[0138] If K1 ≤ threshold, the system cancels the warning and archives the maintenance record; otherwise, maintenance continues until it is qualified, forming a complete closed-loop management.

[0139] Through the above implementation methods, the system can realize real-time and accurate emission degradation monitoring and intelligent maintenance closed-loop management of a large number of heavy-duty diesel vehicles in use, significantly improving supervision efficiency and environmental benefits.

[0140] Example 3, as another possible implementation of the present invention, the method for predicting emission degradation of heavy-duty diesel vehicles provided in this embodiment of the present invention includes:

[0141] a. Divide in-use heavy-duty diesel vehicles into several vehicle families based on emission limits, engine models, and after-treatment configurations;

[0142] b. For each vehicle family, collect PEMS transient data for the factory-fresh state and eight mileage segments (0–100,000 km, 100,000–200,000 km, and 600,000–700,000 km) and extract the steady-state window.

[0143] The PEMS transient data acquisition frequency is ≥1 Hz, and the steady-state window is a continuous 20-30 second data segment that meets the Vehicle Specific Power binning and IQR filtering conditions of intake, exhaust, and injection parameters.

[0144] c. Use the random forest algorithm to train eight mileage prediction models, with the new car model serving as the baseline model.

[0145] d. Use remote OBD to collect vehicle transient conditions in real time and extract steady-state window parameters, and call the benchmark model and the corresponding mileage segment model to predict the emission value P in parallel. base and P current ;

[0146] e. Calculate the preliminary degradation coefficient And the bias correction is performed in combination with the periodic PEMS measured residual to obtain K1;

[0147] f. When K1 continuously exceeds the preset threshold, a high emission warning is automatically triggered and maintenance is instructed. After maintenance, if K1 is retested and returns to within the threshold, the warning is lifted and closed-loop recording is completed.

[0148] The PEMS transient data acquisition frequency is ≥1 Hz, and the steady-state window is a continuous 20-30 second data segment that meets the Vehicle Specific Power binning and IQR filtering conditions of intake, exhaust, and injection parameters.

[0149] To further illustrate the effects of the embodiments of the present invention, the following experiments were conducted.

[0150] (1) Experimental purpose:

[0151] Verify the accuracy of the multi-mileage model in predicting vehicle emission degradation in different mileage segments; examine the improvement effect of digital twin real-time prediction and periodic PEMS correction on model performance; and evaluate the practical value of early warning triggering and subsequent maintenance closed loop on emission recovery.

[0152] (2) Experimental subjects and groups:

[0153] Vehicle samples: A total of 48 in-use heavy-duty diesel vehicles of the same model family were selected and divided into 8 groups (6 vehicles in each group) according to mileage range: Test 1 (new vehicle factory status), Test 2 (0-100,000 kilometers), Test 3 (100,000-200,000 kilometers), Test 4 (200,000-300,000 kilometers), Test 5 (300,000-400,000 kilometers), Test 6 (400,000-500,000 kilometers), Test 7 (500,000-600,000 kilometers) and Test 8 (600,000-700,000 kilometers).

[0154] Control group: 3 vehicles were randomly selected from each mileage group as the "calibration group" (periodic PEMS calibration and early warning maintenance), and the remaining 3 vehicles were used as the "non-calibration group".

[0155] (3) Data collection and preprocessing:

[0156] OBD+PEMS transient data collection: Install the OBD+MQTT terminal to synchronously collect: engine speed, load, SCR temperature, exhaust flow, fuel injection amount, NOx / THC / CO concentration, frequency ≥ 1HZ.

[0157] Steady-state window extraction: Based on the vehicle-specific power binning, an IQR filter is applied to screen the 20–30 s continuous steady-state window within each bin and calculate the average operating parameters.

[0158] (4) Model training and deployment:

[0159] Baseline model and segmented model training: For the steady-state window data of Test1–Test8, the random forest algorithm (RF) was used, and five-fold cross validation + GridSearchCV hyperparameter optimization were performed in parallel. Performance standard: R2 ≥0.90, MAE≤5%, RMSE≤10%.

[0160] Cloud deployment: Deploy the eight trained models on the digital twin platform and enable RESTful API services.

[0161] (5) Prediction, correction and early warning process:

[0162] Real-time prediction: The platform automatically identifies the current vehicle condition steady-state window periodically (e.g., every hour), and calls the corresponding mileage segment model and the benchmark model to predict P in parallel. base and P current , calculate the preliminary degradation coefficient K0.

[0163] Periodic calibration (only for the “calibration group”): Randomly select 3 vehicles from each mileage group every month and conduct on-site PEMS emergency test to obtain the measured value P real Calculate the residuals, take the average residual of each group, correct the predicted value according to formula (2), and then calculate the corrected degradation coefficient K1.

[0164] Warning and maintenance: If K1 exceeds the threshold for 30 consecutive minutes, an SMS / email / APP push warning will be automatically sent; the operator must complete vehicle inspection and maintenance within 72 hours; after maintenance, the platform will retest K1. If it returns to within the threshold, the warning will be lifted and archived.

[0165] (6) Performance evaluation indicators:

[0166] Evaluate model accuracy and calculate R 2 , MAE, RMSE; evaluate the effectiveness of warning, warning trigger rate, false alarm rate, missed alarm rate, K1 recovery rate after maintenance and recovery time distribution; check the consistency between the predicted value and the measured value, draw a scatter plot comparison diagram and a residual distribution diagram; evaluate the long-term stability of the method of the present invention, and analyze the trend change of vehicle K1 over time and mileage.

[0167] The above description is only a preferred specific implementation method of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for predicting emission degradation of heavy-duty diesel vehicles, characterized in that: The method comprises the following steps: S1, vehicle family classification and PEMS data source acquisition; S2, based on the classification and acquisition results, performs transient acquisition of PEMS data and steady-state window extraction; S3, builds and trains multi-mileage segment models based on the extracted data; S4, verify the accuracy of the trained multi-mileage segment model; S5, based on the accuracy verification results, deploy the multi-mileage segment model on the cloud digital twin platform and create a digital twin for each in-use vehicle in the model family; S6: Based on the created digital twin, the cloud-based digital twin platform uses the average operating condition parameters of the latest steady-state window to obtain the predicted values ​​of the baseline model and the current mileage segment model, and performs preliminary degradation coefficient calculations based on the predicted values. S7, using the average residual to correct the obtained prediction value, respectively obtaining a corrected baseline model prediction correction value and a corrected prediction correction value of each current mileage segment model; calculating a correction degradation coefficient based on the prediction correction value and outputting it; S8, comparing the output corrected degradation coefficient with the preset degradation threshold to determine whether the corrected degradation coefficient exceeds the preset degradation threshold. If so, a preset degradation threshold warning is issued, and closed-loop maintenance of the in-use vehicle is performed based on the warning information.

2. The method for predicting emission degradation of heavy-duty diesel vehicles according to claim 1, characterized in that: In step S1, the vehicle type family classification includes: classifying the heavy-duty diesel vehicles in use into several vehicle type families according to the vehicle emission limit category, engine model and after-treatment device configuration elements; PEMS data sources include: PEMS data used by regulatory authorities for compliance supervision and inspection and PEMS data used by manufacturers for compliance self-inspection.

3. The method for predicting emission degradation of heavy-duty diesel vehicles according to claim 1, characterized in that: In step S2, transient acquisition includes: the PEMS device collects transient data of the vehicle under real road conditions at a frequency of ≥1 Hz, including engine speed, load, SCR temperature, exhaust flow, fuel injection amount, and transient concentrations of NOx / THC / CO emissions; Steady-state window extraction involves binning the raw transient data by vehicle-specific power, applying interquartile range (IQR) filtering to the intake, exhaust, and injection system parameters, and filtering for continuous steady-state data windows within each power interval.

4. The method for predicting emission degradation of heavy-duty diesel vehicles according to claim 1, characterized in that: In step S3, the multi-mileage segment model includes a baseline model and a segmented model; Multi-mileage segment model construction, including: Build a benchmark model and a PEMS window dataset for new vehicles in factory condition, which is the first mileage segment model; Build segmented models: Construct window data sets of 0–100,000, 100,000–200,000, 200,000–300,000, 300,000–400,000, 400,000–500,000, 500,000–600,000, and 600,000–700,000 kilometers for the 2nd to 8th mileage segment models; For each vehicle family, the random forest RF algorithm is used to train the baseline model and the segmented model using steady-state window data, so that the performance standards of the baseline model and the segmented model include: the corrected determination coefficient R 2 ≥0.90; mean absolute error MAE ≤ 5%; root mean square error RMSE ≤ 10%.

5. The method for predicting emission degradation of heavy-duty diesel vehicles according to claim 1, characterized in that: In step S4, the accuracy of the trained multi-mileage segment model is verified, including independent test set evaluation, prediction-measurement comparison, and degradation coefficient error verification. In step S5, a digital twin is created for each in-use vehicle in the vehicle family, including real-time uploading of transient operating condition data such as engine speed, load, SCR temperature, exhaust flow, fuel injection amount, and transient concentration of NOx / THC / CO emissions via remote OBD. The digital twin aggregates and displays the real-time data and uses it as input for the baseline model and segmented model.

6. The method for predicting emission degradation of heavy-duty diesel vehicles according to claim 1, characterized in that: In step S6, the cloud digital twin platform uses the average operating condition parameters of the latest steady-state window to obtain the predicted values ​​of the benchmark model and the current mileage segment model, and performs preliminary degradation coefficient calculation based on the predicted values, including: inputting the average operating condition parameters extracted from the latest steady-state window into the benchmark model and the current mileage segment model at the same time to obtain P base and P current , calculate the preliminary degradation coefficient, the expression is: Where K0 is the initial degradation coefficient, P base is the pollutant emission value predicted by the benchmark model, P base The pollutant emission values ​​predicted by the current model for each mileage segment.

7. The method for predicting emission degradation of heavy-duty diesel vehicles according to claim 6, characterized in that: In step S7, the obtained prediction value is corrected using the average residual to obtain the corrected prediction value of the baseline model and the corrected prediction value of each current mileage segment model, including: Periodic calibration: randomly select multiple vehicles from each mileage segment every month to conduct PEMS emergency tests to obtain the measured emission value P real , calculate the residual Δ base and Δ current : D base =P real -P base D current =P real -P current Bias correction, which corrects the predicted values ​​by the mean residuals: Then calculate the correction degradation coefficient K1: Where, P real is the actual emission value of the vehicle under the same working conditions measured by PEMS, Δ base is the residual between the predicted value of the benchmark model and the measured value, Δ current is the residual between the current model prediction value and the measured value, N is the number of test vehicles involved in the calibration, and P' base is the predicted value of the calibrated benchmark model, P' current is the predicted value of the current model after correction.

8. The method for predicting emission degradation of heavy-duty diesel vehicles according to claim 7, characterized in that: In the process of calculating the correction degradation coefficient K1, the average residual is continuously optimized dynamically to obtain the average residual; If the average residual after continuous periodic correction and bias correction exceeds the tolerance, the new data will be included in the training set and the multi-mileage segment model will be updated using incremental learning or retraining.

9. The method for predicting emission degradation of heavy-duty diesel vehicles according to claim 6, characterized in that: In step S8, the preset degradation thresholds include: NOx, 1.20; THC, 1.10; CO, 1.15; When the correction degradation coefficient K1 continues to exceed the threshold, the system automatically triggers an early warning and notifies the operator and regulatory authorities via SMS, email, and APP.

10. A heavy-duty diesel vehicle emission degradation prediction system, characterized in that: The system implements the heavy-duty diesel vehicle emission degradation prediction method according to any one of claims 1 to 9, and the system comprises: The data transient acquisition and steady-state window extraction module is used for vehicle family classification and PEMS data source acquisition. Based on the classification and acquisition results, PEMS data transient acquisition and steady-state window extraction are performed. The multi-mileage segment model construction and training module is used to construct and train the multi-mileage segment model based on the extracted data; and to verify the accuracy of the trained multi-mileage segment model; A preliminary degradation coefficient calculation module is used to deploy multi-mileage segment models on the cloud-based digital twin platform based on accuracy verification results and create a digital twin for each in-use vehicle in the model family. Based on the created digital twin, the cloud-based digital twin platform uses the average operating condition parameters extracted from the latest steady-state window to obtain predicted values ​​for the baseline model and the current mileage segment model, and performs preliminary degradation coefficient calculations based on the predicted values. The correction degradation coefficient calculation and warning module is used to correct the obtained prediction value using the average residual to obtain the corrected baseline model prediction correction value and the prediction correction value of the current mileage segment model respectively; calculate the correction degradation coefficient based on the prediction correction value and output it; compare the output correction degradation coefficient with the preset degradation threshold to determine whether the correction degradation coefficient exceeds the preset degradation threshold. If it exceeds, a preset degradation threshold warning is issued, and a closed-loop maintenance of the in-use vehicle is performed based on the warning information.

Images