SCR outlet temperature judgment method based on machine learning model
By using machine learning models to determine the SCR outlet temperature, the problem of dependence on dedicated test data and difficulty in fault condition detection in existing technologies has been solved, achieving low-cost and efficient temperature determination and fault detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-25
- Publication Date
- 2026-04-03
AI Technical Summary
Existing methods for determining SCR outlet temperature rely on virtual sensors based on physical models, require a large amount of dedicated experimental data, and are limited in their ability to detect fault conditions and application scenarios, resulting in high costs.
A machine learning model is adopted. Specific index data is read from the engine ECM, standardized, and then input into the machine learning model for training. A multilayer perceptron model is used to determine the SCR outlet temperature, reducing reliance on dedicated test data and maintaining stable performance under abnormal conditions.
It enables accurate determination of SCR outlet temperature without increasing costs, overcomes the limitations of physical model virtual sensors, adapts to different application scenarios, reduces development costs, and improves fault detection capabilities.
Smart Images
Figure CN121782011A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of temperature measurement technology for selective catalytic reduction (SCR) devices, and in particular to a method for determining the SCR outlet temperature based on a machine learning model. Background Technology
[0002] Currently, the method for determining the outlet temperature of a diesel engine's aftertreatment SCR (Selective Catalytic Reduction) device is to install a physical temperature sensor at the SCR outlet. The drawback is that a single sensor costs 50-120 yuan, resulting in high equipment costs. Virtual SCR outlet temperature sensors also exist, but these are primarily implemented using physical models. Virtual sensors based on physical models have certain limitations: 1. Complex physical models require extensive dedicated experimental data, necessitating significant calibration work by calibration engineers; 2. Fault conditions cannot be detected; for example, existing virtual SCR outlet temperature sensors based on physical models cannot detect high temperatures when there is significant hydrocarbon escape; 3. Application scenarios are limited; special operating conditions require recalibration, such as virtual carbon loading sensors requiring recalibration at high altitudes. This paper aims to overcome the pain points of virtual sensors developed from physical models while ensuring basic sensor performance. It seeks to find potential correlations between data points from large datasets without requiring extensive dedicated experimental data, thus reducing development costs. Furthermore, it addresses how to maintain stable performance and adapt to different environments when faced with data experiencing interference or anomalies.
[0003] Therefore, it is necessary to invent a method for determining the SCR outlet temperature based on a machine learning model, thereby forming a virtual sensor for SCR outlet temperature detection. This would reduce the need for physical SCR outlet temperature sensors and thus lower costs, while overcoming the aforementioned limitations of virtual sensors implemented using physical models. Summary of the Invention
[0004] The purpose of this invention is to provide a method for determining the SCR outlet temperature based on a machine learning model, addressing the aforementioned problems. This method solves the technical issues of traditional SCR outlet temperature determination methods, which rely on physical models and virtual sensors that require extensive dedicated experimental data, cannot detect fault conditions, and have limited application scenarios.
[0005] To achieve the above objectives, the technical solution adopted by this invention is: a method for determining SCR outlet temperature based on a machine learning model, comprising the following steps:
[0006] Step S1: Read specific indicator data from the engine ECM;
[0007] Step S2: Standardize the above specific indicator data;
[0008] Step S3: Input the standardized data into the machine learning model for training;
[0009] Step S4: The SCR outlet temperature can be obtained by denormalizing the output of the machine learning model.
[0010] Furthermore, the specific index data in step S1 are SCR airflow rate, DOC In temperature, DOC Out temperature, and DPF Out temperature.
[0011] Furthermore, the specific indicator data in step S1 also includes the SCR OUT temperature at the previous moment.
[0012] Furthermore, in step S3, the machine learning model training method is as follows:
[0013] Step S3.1: Select the data with post-processing and valid test results from the test data of the development bench NRTC and NRSC;
[0014] Step S3.2: Continue to extract multiple sets of specific index data that are strongly correlated with the SCR outlet temperature from the valid experimental data; the multiple sets of specific index data are divided into training set, validation set and test set.
[0015] Step S3.3: Create a multilayer perceptron model with 100 hidden units and 1 output corresponding to the number of 4 specific indicator data.
[0016] Step S3.4: The maximum number of iterations for training the model is 2000. If the R² value of the validation set does not improve in the regression analysis over 100 consecutive rounds, then the training of the model is stopped.
[0017] In step S3.5, if R² is greater than 76% on the test set, then the model is considered successfully trained and can be used to determine the output of SCR outlet temperature.
[0018] Furthermore, in step S3.2, the original data is divided into a training set, a validation set, and a test set in a ratio of 6:2.5:1.5.
[0019] Furthermore, in step S3.4, the initial learning rate used to train the model is 0.001, and the learning rate is dynamically adjusted using an adaptive learning rate during the training process.
[0020] Furthermore, in step S2, the standardization method is Z-Score standardization, and the formula for the standardized value is:
[0021]
[0022] in It is the mean calculated from the original experimental data. It is the standard deviation calculated from the original experimental data.
[0023] By adopting the above technical solution, the present invention has the following beneficial effects:
[0024] This invention, based on a machine learning model, develops an SCR outlet temperature sensor that overcomes the limitations of virtual sensors developed using physical models while maintaining basic sensor performance. It can identify potential correlations between data points from large datasets without requiring extensive dedicated experimental data, thus reducing development costs. The machine learning model maintains stable performance even under interference or abnormal conditions. It is easily scalable, allowing for rapid adaptation to new application areas through model transfer. This invention addresses the technical problems of traditional SCR outlet temperature determination using physical model virtual sensors, which require large amounts of dedicated experimental data, cannot detect fault conditions, and have limited application scenarios. Attached Figure Description
[0025] Figure 1 This is a flowchart of the steps of the present invention.
[0026] Figure 2 This describes the performance of the machine learning model of this invention on a training set with four specific parameters.
[0027] Figure 3 This describes the performance of the machine learning model of this invention on a test set with four specific parameters.
[0028] Figure 4 This describes the performance of the machine learning model of this invention on a validation set with four specific parameters.
[0029] Figure 5 This is a comparison chart of the performance curves of the machine learning model of this invention on the training set with five specific parameters, the performance curves on the training set with four specific parameters, and the actual value curves.
[0030] Figure 6 This is a comparison chart of the performance curves of the machine learning model of this invention on a test set with five specific parameters, the performance curves on a training set with four specific parameters, and the actual value curves.
[0031] Figure 7 This is a comparison chart of the performance curves of the machine learning model of this invention on the validation set with five specific parameters, the training set with four specific parameters, and the true value curve. Detailed Implementation
[0032] The specific implementation of the invention will be further described below with reference to the accompanying drawings.
[0033] like Figure 1 As shown, a method for determining the SCR outlet temperature based on a machine learning model includes the following steps:
[0034] Step S1: Read specific indicator data from the engine ECM;
[0035] Step S2: Standardize the above specific indicator data;
[0036] In this step, the standardization method is Z-Score standardization, and the formula for the standardized value is:
[0037]
[0038] in It is the mean calculated from the original experimental data. It is the standard deviation calculated from the original experimental data.
[0039] Step S3: Input the standardized data into the machine learning model for training;
[0040] In this step, the machine learning model training method is as follows:
[0041] Step S3.1: Select valid test data with aftertreatment from the NRTC and NRSC test data of the development bench. Specifically, select valid test data with aftertreatment from the NRTC and NRSC test data of the internal development bench over 23-24 years, extracting a total of 14,000 test data points. The specific index data mentioned in this step are SCR air velocity, DOC In temperature, DOC Out temperature, and DPF Out temperature. Specifically, DOC In temperature refers to the inlet temperature of the diesel oxidation catalyst in the aftertreatment, DOC Out temperature refers to the outlet temperature of the diesel oxidation catalyst in the aftertreatment, and DPF Out temperature refers to the outlet temperature of the particulate filter in the aftertreatment.
[0042] Step S3.2: Continue to extract multiple sets of specific index data that are strongly correlated with the SCR outlet temperature from the valid experimental data; these multiple sets of specific index data are divided into training set, validation set, and test set. Specifically, in this step, the original data is divided into training set, validation set, and test set in a ratio of 6:2.5:1.5.
[0043] Step S3.3: Create a multilayer perceptron model with 4 inputs, 100 hidden units, and 1 output;
[0044] Step S3.4: The maximum number of iterations for training the model is 2000. If the R² value of the validation set does not improve in 100 consecutive rounds of regression analysis, then the model training is stopped. In this step, the initial learning rate used for training the model is 0.001, and the learning rate is dynamically adjusted using an adaptive learning rate during the training process.
[0045] In step S3.5, if R² is greater than 76% on the test set, then the model is considered successfully trained and can be used to determine the output of SCR outlet temperature.
[0046] Step S4: The SCR outlet temperature can be obtained by denormalizing the output of the machine learning model.
[0047] Please see Figures 2 to 4 This involves testing the model's performance using different training, testing, and validation sets. The performance graph shows the actual values (blue line) and the model's output (orange line). If the blue and orange lines largely overlap, it indicates that the model performs well.
[0048] In a further embodiment, the specific index data in step S1 also includes the SCR OUT temperature at the previous moment; see [link to previous embodiment]. Figures 5 to 7 Compared to specific indicator data that only includes the predicted curves of the model tested on the training, testing, and validation sets of four parameters—SCR airflow velocity, DOC In temperature, DOC Out temperature, and DPF Out temperature—specific indicator data also includes the predicted curve of the SCR OUT temperature at the previous moment, which is closer to the true value curve.
[0049] The above description is a detailed description of the preferred embodiments of the present invention. However, the embodiments are not intended to limit the scope of the patent application of the present invention. All equivalent changes or modifications made under the technical spirit of the present invention should fall within the patent scope covered by the present invention.
Claims
1. A method for determining the outlet temperature of an SCR based on a machine learning model, characterized in that: Includes the following steps: Step S1: Read specific indicator data from the engine ECM; Step S2: Standardize the above specific indicator data; Step S3: Input the standardized data into the machine learning model for training; Step S4: The SCR outlet temperature can be obtained by denormalizing the output of the machine learning model.
2. The method for determining SCR outlet temperature based on a machine learning model according to claim 1, characterized in that: The specific index data in step S1 are SCR airflow rate, DOC In temperature, DOC Out temperature, and DPF Out temperature.
3. The method for determining SCR outlet temperature based on a machine learning model according to claim 2, characterized in that: The specific index data in step S1 also includes the SCR OUT temperature at the previous moment.
4. The method for determining SCR outlet temperature based on a machine learning model according to any one of claims 2 to 3, characterized in that: In step S3, the machine learning model training method is as follows: Step S3.1: Select the data with post-processing and valid test results from the test data of the development bench NRTC and NRSC; Step S3.2: Continue to extract multiple sets of specific index data that are strongly correlated with the SCR outlet temperature from the valid experimental data; the multiple sets of specific index data are divided into training set, validation set and test set. Step S3.3: Create a multilayer perceptron model with inputs corresponding to a specific number of indicator data, 100 hidden units, and 1 output; Step S3.4: The maximum number of iterations for training the model is 2000. If the R² value of the validation set does not improve in the regression analysis over 100 consecutive rounds, then the training of the model is stopped. In step S3.5, if R² is greater than 76% on the test set, then the model is considered successfully trained and can be used to determine the output of SCR outlet temperature.
5. The method for determining SCR outlet temperature based on a machine learning model according to claim 4, characterized in that: In step S3.2, the original data is divided into training set, validation set and test set in a ratio of 6:2.5:1.
5.
6. The method for determining SCR outlet temperature based on a machine learning model according to claim 4, characterized in that: In step S3.4, the initial learning rate used to train the model is 0.001, and the learning rate is dynamically adjusted using an adaptive learning rate during the training process.
7. The method for determining SCR outlet temperature based on a machine learning model according to claim 1, characterized in that: In step S2, the standardization method is Z-Score standardization, and the formula for the standardized value is: in It is the mean calculated from the original experimental data. It is the standard deviation calculated from the original experimental data.