CN-121656505-B - Heavy vehicle NO based on multistage dynamic modelingxOn-line monitoring data correction method

CN121656505BCN 121656505 BCN121656505 BCN 121656505BCN-121656505-B

Abstract

The invention provides a heavy vehicle NO x on-line monitoring data correction method based on multistage dynamic modeling, which relates to the field of automobile emission monitoring and intelligent data calibration, and comprises the steps of acquiring heavy vehicle remote on-line monitoring data in real time through vehicle-mounted terminal equipment; PEMS comparison test data of the heavy vehicle are obtained through PEMS test equipment, and the vehicle running process is divided into five stages according to an SCR device starting temperature threshold, an NO x sensor activation temperature threshold, an NO x sensor starting temperature threshold and an engine heat engine stable temperature threshold; for each stage, based on the online monitoring data and PEMS test data of the stage, a corresponding NO x monitoring data correction model is constructed, and NO x concentration data in the online monitoring data is corrected through the model. The invention overcomes the defects of poor working condition adaptation, low completion precision, weak interpretation, insufficient linkage and the like in the prior art, and ensures stable correction precision under different working conditions.

Inventors

YANG YANYAN
SHEN XIUE
FENG QIAN
WANG PENGRUI
LU YANG
ZHOU YANQING
SUN SHIDA

Assignees

北京市生态环境监测中心

Dates

Publication Date: 20260505
Application Date: 20260205

Claims (8)

1. A heavy vehicle NO x on-line monitoring data correction method based on multistage dynamic modeling is characterized by comprising the following steps: the method comprises the steps of acquiring remote on-line monitoring data of a heavy vehicle in real time through vehicle-mounted terminal equipment of the heavy vehicle, wherein the remote on-line monitoring data of the heavy vehicle at least comprise vehicle speed, engine parameters, NO x concentration data, exhaust flow and exhaust temperature; PEMS comparison test data of the heavy vehicle are obtained through PEMS test equipment, wherein the PEMS comparison test data comprise NO x concentration data and exhaust flow measured by the PEMS; Dividing the vehicle operation process into five stages according to a starting temperature threshold of the SCR device of the heavy vehicle, an activation temperature threshold of the NO x sensor, a starting temperature threshold of the NO x sensor and a stabilizing temperature threshold of the engine heat engine; for each stage, based on the remote online monitoring data of the heavy vehicle and the PEMS comparative test data of the stage, respectively constructing a corresponding NO x monitoring data correction model, and correcting NO x concentration data in the remote online monitoring data of the heavy vehicle through the correction model; dynamically updating parameters of the correction model at each stage based on the real-time PEMS comparison test data to realize continuous calibration of the remote on-line monitoring data of the heavy truck; Wherein the five stages include: The first stage, the engine is started until the exhaust temperature reaches 200 ℃; the second stage, wherein the exhaust temperature reaches 200 ℃ to the cooling liquid temperature reaches 70 ℃; The third stage, the temperature of the cooling liquid reaches 70 ℃ to the start of the operation of the NO x sensor; The fourth stage, the NO x sensor starts to work until the temperature of the cooling liquid reaches 85-90 ℃; A fifth stage, wherein the temperature of the cooling liquid reaches 85-90 ℃ until the test is finished; In the first stage, adopting a multiple linear regression model and combining a piecewise linear interpolation method to correct and complement data; in the second stage, adopting a multiple stepwise regression model and a weighted moving average smoothing method to correct data; In the third stage, a partial least square regression model is combined with a Kalman filtering method to conduct data prediction and completion; In the fourth stage, a polynomial regression model is adopted to combine with a deviation correction formula with temperature compensation for data calibration; And in the fifth stage, adopting a transient emission prediction model based on a long-short-term memory network to carry out data correction.
2. The method for correcting NO x on-line monitoring data of a heavy truck according to claim 1, further comprising the step of performing data complementation on NO x concentration data in the remote on-line monitoring data of the heavy truck by adopting a corresponding data complementation method aiming at three types of data missing scenes including single data missing, continuous period data missing and data missing when a sensor is not in operation.
3. The method for correcting the online monitoring data of the heavy vehicle NO x according to claim 2, wherein the data complement method corresponding to the three types of data missing scenes includes: For single data deletion, according to the working condition consistency of the data at the front and rear moments, adopting a mean value supplementing method or a weighted mean value supplementing method to complete the data; for the data missing of the continuous time period, carrying out data complementation based on the accumulated mileage, average vehicle speed and historical emission rate association in the time period; And for the data missing when the sensor does not work, carrying out data complementation based on the correction model of the corresponding stage.
4. The method for correcting online monitoring data of NO x of a heavy truck according to claim 1, further comprising performing an anomaly cleaning on the remote online monitoring data of the heavy truck before the correction model is constructed, the anomaly cleaning comprising: Judging whether NO x concentration data in the remote on-line monitoring data of the heavy truck exceeds a preset reasonable range or not; Judging whether each monitoring parameter in the remote on-line monitoring data of the heavy truck accords with a value range specified by a standard; Performing time sequence consistency check on the remote on-line monitoring data of the heavy truck, and identifying and eliminating mutation abnormal data; And carrying out logic relevance verification on the remote on-line monitoring data of the heavy truck, and identifying and removing the data combination which does not accord with the physical logic.
5. The method for correcting the online monitoring data of the NO x of the heavy truck according to claim 1, wherein the implementation manner of data prediction and completion in the third stage includes: Taking a set main parameter affecting the emission of the diesel engine NO x in the third stage as an independent variable matrix, taking NO x concentration data measured by PEMS as a dependent variable vector, extracting a main component with a cumulative variance contribution rate of more than or equal to 95%, and constructing a partial least square regression model; Taking the extracted principal component as a state vector, and constructing a Kalman filtering state space model; Taking the predicted value of the partial least square regression model on the concentration of NO x as a priori estimated value of Kalman filtering; and using the PEMS contrast test data which are effective at adjacent moments as an observation value, and performing iterative updating through Kalman filtering to realize dynamic optimal estimation and data completion of the concentration of NO x in the non-working stage of the NO x sensor.
6. The method for correcting the on-line monitoring data of the heavy truck NO x according to claim 1, wherein the deviation correction formula of the fourth stage is: y 3_corrected = y 3_pred +Δ×(1 + k×(x 10 - T ref )) Wherein y 3_corrected is the corrected concentration of NO x , y 3_pred is a polynomial regression model predicted value, delta is the average deviation of NO x sensor data and PEMS data, k is a temperature compensation coefficient, x 10 is the SCR inlet temperature, and T ref is the SCR efficient working reference temperature.
7. The method for correcting online monitoring data of NO x of heavy duty vehicle according to claim 1, wherein the transient emission prediction model structure based on long-short-term memory network in the fifth stage includes: an input layer for receiving key parameters including barometric pressure, net output torque, engine speed, fuel flow, SCR inlet temperature, exhaust flow, and historical NO x concentration; At least one hidden layer using a ReLU activation function; Dropout layer for preventing overfitting; An output layer for outputting a predicted NO x concentration value; The model training is carried out by adopting a dynamic learning rate adjustment strategy and a data enhancement method.
8. The method for correcting the on-line monitoring data of the NO x of the heavy truck according to claim 1, wherein dynamically updating parameters of the correction model of each stage based on the PEMS comparison test data in real time includes: periodically updating coefficients or weights of the correction model at each stage by taking the real-time PEMS comparison test data as a reference; And compensating and correcting the atmospheric pressure measured value by combining the environmental temperature and humidity data so as to reduce environmental interference.

Description

Heavy vehicle NO x on-line monitoring data correction method based on multistage dynamic modeling Technical Field The invention relates to the technical field of automobile emission monitoring and intelligent data calibration, in particular to a heavy truck NO x on-line monitoring data correction method based on multistage dynamic modeling. Background The current heavy vehicle NO x on-line monitoring data correction technology faces three core pain points of poor working condition suitability, low data complement precision and weak model generalization capability, and is specifically characterized in that: 1. the working condition-correction strategy is disjoint and the whole period coverage is insufficient. In the prior art, heavy vehicle operation is mostly regarded as a cold start stage and a heat engine stage, dynamic working characteristics of key equipment (SCR and NO x sensor) are not considered, for example, the SCR needs to start catalytic reaction at the exhaust temperature of more than or equal to 200 ℃, the NO x sensor needs to stably output at the cooling liquid temperature of more than or equal to 70 ℃, and the existing single correction model cannot adapt to the emission rule of transition stages such as 'the exhaust temperature does not reach the SCR start threshold value', 'the sensor is not activated', and the like, so that data deviation in the early stage of cold start (0-5 minutes) is up to 30% -50%, and the accuracy requirement of +/-15% specified in GB17691-2018 is far exceeded. 2. The data missing processing is rough, and the scene suitability is weak. Aiming at NO x data loss, the existing method mostly adopts 'fixed interpolation' (such as linear interpolation) or 'global average value' supplementation, and does not distinguish the difference between 'single loss' and 'continuous period loss', namely, for single loss, simple interpolation does not utilize the working condition correlation at adjacent moments, and for continuous loss (such as 5-10 minutes of signal interruption of a vehicle-mounted terminal), only depends on historical average value, and does not combine the dynamic correlation of 'accumulated mileage-average vehicle speed-emission rate', so that the continuous loss data complementation error exceeds 25%, and the emission total amount statistical requirement cannot be met. 3. The multi-parameter collinearity is not solved, and the model precision and the interpretability are unbalanced. In the online monitoring data, the exhaust flow (y 1) and the engine fuel flow (x 7), the net output torque (x 4) and the rotating speed (x 6) have strong collinearity (correlation coefficient > 0.8), the prior art either adopts single linear regression (neglecting collinearity to cause model distortion) or adopts pure LSTM deep learning (high precision but poor interpretability and difficult error source tracing), and cannot consider both the "calibration precision" and the "supervision traceability requirement", for example, although a certain existing LSTM scheme reduces the RMSE to 60ppm, the prior art cannot explain why the increase of the SCR inlet temperature by 10 ℃ can cause the change of the correction value by 5ppm, and the traceability requirement of environmental protection supervision is not met. 4. PEMS is not dynamically linked with on-line monitoring data. The existing correction model takes off-line PEMS data as a static reference, and a dynamic calibration mechanism of 'real-time PEMS data-on-line monitoring data' is not established, so that when the vehicle working condition deviates from an off-line test working condition (such as actual road gradient change), the correction model fails and the data deviation increases suddenly. In summary, the prior art cannot meet the requirement of accurately correcting the NO x on-line monitoring data under the full working condition and the full scene of the heavy vehicle, and an innovative scheme of 'working condition segmentation and accuracy, method fusion coordination and scene adaptation dynamics' is needed. Disclosure of Invention Therefore, the embodiment of the application provides a multi-stage diesel vehicle NOx monitoring data correction method based on models such as an LSTM network and the like, so as to overcome the defects of poor working condition adaptation, low completion precision, poor interpretation, insufficient linkage and the like in the prior art. The embodiment of the application provides a heavy vehicle NO x on-line monitoring data correction method based on multistage dynamic modeling, which comprises the following steps: the method comprises the steps of acquiring remote on-line monitoring data of a heavy vehicle in real time through vehicle-mounted terminal equipment of the heavy vehicle, wherein the remote on-line monitoring data of the heavy vehicle at least comprise vehicle speed, engine parameters, NO x concentration data, exhaust flow and exhaust temperature; PEMS comparison test data of th