Engine inflation efficiency virtual sensing method based on physical information neural network

Abstract

The invention discloses an engine inflation efficiency virtual sensing method based on a physical information neural network. The method comprises the following steps: step 1, constructing a physical model based on a simplified physics first law; step 2, constructing a feedforward neural network model FNN for estimating the inflation efficiency of the engine; 3, training a physical model, fixing parameters of the physical model, and further training a feedforward neural network model FNN; 4, forming a physical information neural network PDM in combination with the physical model and the feed-forward neural network model FNN; step 5, training parameters of the feed-forward neural network model FNN by using the comprehensive loss function of the physical loss and the data loss, realizing fusion of the physical model and the feed-forward neural network model FNN, and obtaining a trained physical information neural network PDM; and step 6, applying the trained physical information neural network PDM to predict the inflation efficiency of the engine under different operation conditions. The method can effectively improve the prediction precision of the inflation efficiency of the engine.

Description

Technical Field

[0001] The present invention relates to the field of mechanical engineering and data science technology, and in particular to a virtual perception method of engine charging efficiency based on physical information neural network. Background Art

[0002] In the modern automobile industry, engine performance, thermal efficiency and environmental adaptability are key indicators of its technological advancement. Precise regulation of charging efficiency can not only improve the thermal efficiency of the engine and reduce fuel consumption, but also effectively control emissions to meet increasingly stringent environmental regulations. In addition, effective management of charging efficiency is also crucial to improving the response speed and driving performance of the engine. In a turbocharger system, accurate estimation and control of charging efficiency is directly related to the matching effect of the supercharger and the dynamic response characteristics of the entire system. Accurate control and optimization of charging efficiency is the basis for achieving efficient and environmentally friendly engine design.

[0003] In the past decade, researchers have achieved many important results in engine charging efficiency modeling. Overall, these methods can be divided into three categories: physical model-based modeling methods, data-based modeling methods, and physical data hybrid-driven modeling methods. Physical models are based on rigorous physical theories and formulas to estimate volumetric efficiency by capturing the basic physical laws of the intake process. Kocher et al. (Kocher L, Koeberlein E, Van Alstine DG, et al. Physically based volumetric efficiency model for diesel engines utilizing variable intake valve actuation [J]. International Journal of Engine Research, 2012, 13 (2): 169-184.) developed a physical model that estimates volumetric efficiency through the law of conservation of energy and has been experimentally verified to be applicable to a variety of engine configurations. Menzel et al. (Menzel G, Och SH, Mariani VC, et al. Multi-objective optimization of the volumetric and thermal efficiencies applied to a multi-cylinder internal combustion engine [J]. Energy Conversion and Management, 2020, 216: 112930.) developed a physical model for optimizing the volumetric efficiency of a multi-cylinder internal combustion engine. The model uses multi-objective optimization methods, including non-dominated sorting genetic algorithm II (NSGA-II) and multi-objective differential evolution (MODE), to find the optimal valve timing setting. A one-dimensional non-steady compressible gas flow model and a zero-dimensional combustion model were used in the study, and they were solved and optimized by numerical methods. Although these methods have achieved certain success in the past, they usually have limitations, especially in terms of cost, time efficiency and adaptability to dynamic operating conditions.

[0004] With the rapid development of big data and high-performance computing technology, data-driven methods have opened up new ways to optimize engine performance. By analyzing massive amounts of engine operating data, machine learning algorithms can reveal complex data patterns and potential performance optimization opportunities. The literature (Uzun A. Air mass flow estimation of diesel engines using neural network [J]. Fuel, 2014, 117: 833-838.) uses a recursive neural network to predict air mass flow, considering crankshaft angle, load and engine speed as inputs. The results show high accuracy, but are too dependent on the quantity and quality of data. The literature (Luján JM, Climent H, García-Cuevas LM, et al. Volumetric efficiency modelling of internal combustion engines based on anovel adaptive learning algorithm of artificial neural networks[J]. Applied Thermal Engineering, 2017, 123: 625-634.) proposed a new adaptive learning algorithm, which improved the limitations of the standard gradient descent method by increasing the update speed of the hidden layer weights and improved the learning speed of the network by applying different learning rates between network layers. However, this method has only been verified within a limited range of operating conditions, and the generalization ability of the method is not strong enough. The paper (Li J, Zhou Q, Williams H, et al. Fuzzy-tree-constructed data-efficient modelling methodology for volumetric efficiency of dedicated hybrid engines[J]. Applied Energy, 2022, 310: 118534.) proposed a method based on a hierarchical fuzzy inference tree. The article simplifies model construction by combining several low-dimensional fuzzy inference systems, and develops a Gaussian distribution-based resampling technique to maintain sample diversity by screening a small number of diverse samples.The literature (Gao J, Liu X, Sun B, et al. In-cylinder air mass flow estimation of gasoline engines based on map self-learning [J]. Control Engineering Practice, 2023, 140: 105674.) proposed an estimation method based on MAP self-learning, which eliminates the periodic fluctuations of the signal by designing a suitable filtering method, and uses an adaptive observer to estimate the system state and volumetric efficiency. The above-mentioned data-based optimization strategies can not only provide real-time performance monitoring and prediction, but also reveal performance improvement points that are difficult to discover through traditional methods. However, although data-driven methods have significant advantages in processing large-scale data sets and discovering nonlinear relationships, they often ignore the physical principles behind them. Sometimes they may ignore the basic laws of engine operation due to over-reliance on data, thereby limiting the versatility and accuracy of the prediction model.

[0005] In summary, there is an urgent need for a new method that combines the engine physical model with advanced data-driven technology to achieve high-precision virtual perception of the engine charging efficiency and provide support for high-precision control of the engine. Therefore, the present invention proposes an innovative solution that subtly combines the rigor of the physical model with the flexibility of the data-driven model (feedforward neural network model FNN). The present invention constructs a physical model based on a simplified first law of physics and combines it with an FNN model constructed using advanced machine learning technology. It takes advantage of the physical model based on physical information neural network and the FNN model to achieve high-precision prediction of engine charging efficiency. Summary of the invention

[0006] The purpose of the present invention is to provide a virtual perception method of engine charging efficiency based on physical information neural network in order to address the technical defects existing in the prior art.

[0007] The technical solution adopted to achieve the purpose of the present invention is:

[0008] A virtual perception method of engine charging efficiency based on physical information neural network includes the following steps:

[0009] Step 1, constructing a physical model based on a simplified first law of physics, wherein the physical model estimates the engine charging efficiency according to the engine speed, intake manifold pressure, exhaust pressure and intake manifold temperature;

[0010] Step 2, constructing a feedforward neural network model FNN for estimating engine charging efficiency to fully capture the nonlinear characteristics in the engine operation data, wherein the input of the feedforward neural network model FNN is the engine speed, intake manifold pressure, exhaust pressure and intake manifold temperature, and the output is the engine charging efficiency;

[0011] Step 3, first train the physical model of step 1, fix its parameters, and then further train the feedforward neural network model FNN of step 2;

[0012] Step 4, combining the physical model of step 1 and the feedforward neural network model FNN of step 3 to form a physical information neural network PDM;

[0013] Step 5, using the comprehensive loss function of physical loss and data loss to train the parameters of the feedforward neural network model FNN, realizing the fusion of the physical model and the feedforward neural network model FNN, and obtaining the trained physical information neural network PDM;

[0014] Step 6, using the trained physical information neural network PDM to predict the charging efficiency of the engine under different operating conditions.

[0015] In the above technical solution, the physical model in step 1 is:

[0016]

[0017] Where n is the engine speed, P im is the intake manifold pressure, P em is the exhaust pressure, T im is the intake manifold temperature, θ1-θ 13 is the physical model parameter, and η is the estimated value of the engine charging efficiency.

[0018] In the above technical solution, the structure of the feedforward neural network model FNN in step 2 includes an input layer, multiple hidden layers and an output layer, the hidden layer is composed of multiple linear layers and nonlinear activation functions, the linear part of each hidden layer is implemented by nn.Linear, and the LeakyReLU activation function is applied to the output of each linear part.

[0019] In the above technical solution, in step 3, the nonlinear least squares method is used to adjust the parameters θ1-θ 13 .

[0020] In the above technical solution, in step 3, when training the feedforward neural network model FNN, the mean square error MSE is used as the loss function, the loss function calculates the difference between the predicted value and the actual measured value of the feedforward neural network model FNN, and the optimization algorithm adopts the first Adam optimizer.

[0021] In the above technical solution, in step 3, after the physical model parameters are optimized, the freezing technology is used to lock the values ​​of the physical parameters, and then the non-physical parameters of the feedforward neural network model FNN are trained, and the performance of the feedforward neural network model FNN is further optimized by a data-driven method:

[0022] if θ∈Θ physics ,thenθ.require_grad=False

[0023] Among them, Θ physics are the parameters of all physical models θ1-θ 13 A collection of .

[0024] In the above technical solution, the first Adam optimizer is:

[0025]

[0026] Among them, θ next represents the parameters of the optimized feedforward neural network model FNN, θ represents the parameters of the feedforward neural network model FNN, l r is the learning rate, and Adam(·) represents the update function of the first Adam optimizer.

[0027] In the above technical solution, in step 5, the comprehensive loss function is:

[0028] L total =(1-λ physics )·L data +λ physics ·L physics

[0029]

[0030] Where: N is the number of samples, y physics,i is the predicted value of the physical model for the i-th sample, y pred,i is the prediction of the feedforward neural network model FNN for the i-th sample, y true,i is the measured value of the inflation efficiency of the i-th sample, L data Represents data loss, which is the mean square error between the predicted value of the feedforward neural network model FNN and the actual measured value, L physics represents the physical loss, which is the mean square error between the predicted value of the physical model and the actual measured value, λ physics is the weight coefficient used to balance the two losses, and its value is between 0 and 1.

[0031] In the above technical solution, in step 5, a second Adam optimizer is used to perform fusion loss training to minimize the overall error between the predicted value and the actual measured value.

[0032] In the above technical solution, the second Adam optimizer is Among them, θ next represents the parameters of the optimized feedforward neural network model FNN, θ represents the parameters of the feedforward neural network model FNN, l r is the learning rate, and Adam(·) represents the update function of the second Adam optimizer.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] This invention innovatively combines the accuracy of physical models and the flexibility of data-driven models (feedforward neural network models FNN), and aims to achieve high-precision prediction of engine charging efficiency by integrating the advantages of different models, while reducing computing costs and improving the adaptability of models in complex environments. Through this method, engine charging efficiency can be perceived virtually in real time at a lower cost and higher accuracy, thus providing strong technical support for the optimization and control of engine performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 PDM architecture diagram;

[0036] Figure 2 Comparison chart of the prediction results of the physical model, FNN model and PDM with the measured data;

[0037] Figure 3 Schematic diagram of the relative errors between the physical model, FNN model, PDM and measured data;

[0038] Figure 4 Schematic diagram of the correlation between the physical model, FNN model and PDM and the measured data;

[0039] Figure 5 Distribution diagram of relative errors between the physical model, FNN model, PDM and measured data. DETAILED DESCRIPTION

[0040] The present invention is further described in detail below in conjunction with specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0041] like Figure 1 As shown, a virtual perception method of engine charging efficiency based on physical information neural network includes the following steps:

[0042] Step 1, constructing a physical model based on the simplified first law of physics according to domain knowledge, wherein the physical model estimates the engine charging efficiency according to the engine speed, intake manifold pressure, exhaust pressure and intake manifold temperature;

[0043] The physical model is shown in formula (1):

[0044]

[0045] Where n is the engine speed, P im is the intake manifold pressure, P em is the exhaust pressure, T im is the intake manifold temperature, θ1-θ 13 is the physical model parameter, η is the estimated value of the engine charging efficiency;

[0046] Step 2: In order to fully capture the nonlinear characteristics in the engine operation data and establish an accurate mapping relationship between the input parameters and the charging efficiency, a feedforward neural network (FNN) model for charging efficiency estimation is constructed;

[0047] The structure of the feedforward neural network model FNN includes an input layer, multiple hidden layers, and an output layer, which is specially designed to handle multi-dimensional engine operating parameters. The input parameters of the model include engine speed, intake manifold pressure, exhaust pressure, and intake manifold temperature (obtained by automatic feature engineering contrast learning). The hidden layer of the model consists of multiple linear layers and nonlinear activation functions. The linear part of each hidden layer is implemented by nn.Linear, which is responsible for linear transformation in the feature space. Subsequently, the LeakyReLU activation function is applied to the output of each linear part to introduce nonlinear characteristics, which helps the network learn more complex function mappings. The design of LeakyReLU avoids the problem of gradient disappearance during training, allowing the network to learn more effectively during optimization. The output parameter of the feedforward neural network model FNN is the engine charging efficiency.

[0048] Step 3, when training the physical model of step 1, according to the engine operating parameters or historical data, specifically including inputting the intake manifold pressure P im , intake manifold temperature T im , engine speed n and exhaust pressure P em Data information, using nonlinear least squares method to adjust the parameters θ1-θ of the physical model 13 so that it can accurately output an estimate of the charging efficiency;

[0049] In the physical model, the parameters θ1-θ 13After the determination, the feedforward neural network model FNN is trained in one step, and the mean square error (MSE) is used as the loss function, which calculates the difference between the predicted value of the feedforward neural network model FNN and the actual measured value (data sample generated by the bench test). The optimization algorithm uses the first Adam optimizer, which is an optimization method with adaptive learning rate. It can automatically adjust the learning rate during the training process, thereby accelerating the convergence speed and improving the prediction accuracy of the feedforward neural network model FNN;

[0050] Step 4: The physical model of step 1 is combined with the feedforward neural network model FNN of step 2 to form a physical information neural network model (PDM). The physical model of step 1 is used to make a basic estimate of the inflation efficiency. The feedforward neural network model FNN estimates the parameters θ1-θ 13 Adjustments and optimizations are performed to learn the underlying patterns and nonlinear relationships in the data. This hybrid-driven modeling approach not only retains the interpretability of the physical model, but also enhances the model's ability to capture complex phenomena through data-driven adjustments.

[0051] Step 5, use the comprehensive loss function of physical loss and data loss to train the parameters of the feedforward neural network model FNN, realize the fusion of the physical model and the feedforward neural network model FNN, and obtain the trained physical information neural network model PDM. Use the second Adam optimizer to perform fusion loss training to minimize the overall error between the predicted value and the actual measured value, while considering data loss and physical loss. By adjusting the relative weights between physical parameters and data-driven parameters, the accuracy of the physical model is retained while enhancing the model's ability to capture complex data patterns.

[0052] The comprehensive loss function is:

[0053] L total =(1-λ physics )·L data +λ physics ·L physics (2)

[0054]

[0055] Where: N is the number of samples, y physics,i is the predicted value of the physical model for the i-th sample, y pred,i is the prediction of the feedforward neural network model FNN for the i-th sample, y true,i is the measured value of the inflation efficiency of the i-th sample, L datarepresents data loss, which is the mean square error between the predicted value of the feedforward neural network model FNN and the actual measured value. physics Represents the physical loss, which is the mean square error between the predicted value of the physical model and the actual measured value. physics is the weight coefficient used to balance the two losses, and its value is between 0 and 1.

[0056] In step 5, the second Adam optimizer is shown in formula (5), which ensures that the feedforward neural network model FNN can be accurately adjusted in different stages of training, thereby improving the accuracy and robustness of the overall prediction model.

[0057]

[0058] Among them, θ next represents the parameters of the optimized feedforward neural network model FNN, θ represents the parameters of the feedforward neural network model FNN, l r is the learning rate, and Adam(·) represents the update function of the second Adam optimizer.

[0059] Step 6: Use the trained PDM to predict the charging efficiency of the engine under different operating conditions.

[0060] Example 2

[0061] This embodiment is further optimized based on embodiment 1.

[0062] In step 3, after optimizing the physical model parameters, the freezing technique is used to lock the values ​​of the physical parameters, expressed as formula (6), and then the non-physical parameters of the feedforward neural network model FNN are trained to ensure that the physical knowledge is effectively utilized and maintained consistent, while further optimizing the model performance through data-driven methods.

[0063] if θ∈Θ physics ,thenθ.require_grad=False (6)

[0064] Among them, Θ physics are the parameters of all physical models θ1-θ 13 A collection of;

[0065] The first Adam optimizer in step 3 is

[0066]

[0067] Among them, θ next represents the parameters of the optimized feedforward neural network model FNN, θ represents the parameters of the feedforward neural network model FNN, l r is the learning rate, and Adam(·) represents the update function of the Adam optimizer.

[0068] Example 3

[0069] The physical model, the feedforward neural network model FNN and the physical information neural network model PDM are verified. Figure 2-Figure 5 .

[0070] Figure 2 To show the comparison between the estimated values ​​of the physical model, FNN model and PDM model and the measured inflation efficiency values, the blue dots in the figure represent the measured inflation efficiency values, the green line represents the estimated values ​​of the physical model, the purple line represents the estimated values ​​of the FNN model, and the red line represents the estimated values ​​of the PDM model. It can be clearly seen from the figure that the prediction curve of the PDM model is closest to the actual measured value, indicating that it has the highest prediction accuracy.

[0071] Figure 3 Indicates the relative errors between the estimated values ​​of the physical model, FNN model and PDM model and the measured charging efficiency. The green line in the figure indicates the relative error between the estimated value of the physical model and the measured charging efficiency, the purple line in the figure indicates the relative error between the estimated value of the FNN model and the measured charging efficiency, and the red line in the figure indicates the relative error between the estimated value of the PDM model and the measured charging efficiency. It can be seen from the figure that the relative error of the PDM model is lower than that of the physical model and the FNN model as a whole, showing a smaller fluctuation range, which means that the PDM model is more accurate in prediction at each measurement point.

[0072] Figure 4 The figure shows the correlation diagram between the physical model, FNN model and PDM model and the measured inflation efficiency. The black line in the figure is the measured inflation efficiency, the green point is the estimated value of the physical model, the purple point is the estimated value of the FNN model, and the red point is the estimated value of the PDM model. It can be seen from the figure that the correlation diagram of the PDM model (red points) shows the tightest distribution, and all points are closely around the diagonal line, which indicates that the predicted value of the PDM model has the strongest linear relationship with the actual measured value.

[0073] Figure 5 The frequency distribution of relative errors of the physical model, neural network model and PDM model is shown. The green bar graph represents the frequency distribution of relative errors between the physical model estimate and the measured charging efficiency value, the purple bar graph represents the frequency distribution of relative errors between the FNN model estimate and the measured charging efficiency value, and the red bar graph represents the frequency distribution of relative errors between the PDM model estimate and the measured charging efficiency value.

[0074] The relative error of the physical model estimate relative to the measured charging efficiency value accounts for 41.03% of the total estimates within the range of ±1%, and 71.79% of the total estimates within the range of ±2%. The relative error of the FNN model estimate relative to the measured charging efficiency value accounts for 78.21% of the total estimates within the range of ±1%, and 87.82% of the total estimates within the range of ±2%. The relative error of the PDM model estimate relative to the measured charging efficiency value accounts for 80.13% of the total estimates within the range of ±1%, and 94.23% of the total estimates within the range of ±2%. In summary, the relative error distribution of the PDM model is the most concentrated and obviously biased towards the zero error line, which shows that the PDM model has a significant advantage in minimizing the prediction error. Compared with the physical model and the FNN model, the PDM model not only performs better in error size, but also in prediction stability. This high generalization ability means that the PDM model can adapt to different operating conditions and provide accurate real-time prediction of engine charging efficiency.

[0075] The above is only a preferred embodiment of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A virtual perception method of engine charging efficiency based on physical information neural network, characterized in that: The following steps are involved: Step 1, constructing a physical model based on a simplified first law of physics, wherein the physical model estimates the engine charging efficiency according to the engine speed, intake manifold pressure, exhaust pressure and intake manifold temperature; Step 2, constructing a feedforward neural network model FNN for estimating engine charging efficiency to fully capture the nonlinear characteristics in the engine operation data, wherein the input of the feedforward neural network model FNN is the engine speed, intake manifold pressure, exhaust pressure and intake manifold temperature, and the output is the engine charging efficiency; Step 3, first train the physical model of step 1, fix its parameters, and then further train the feedforward neural network model FNN of step 2; Step 4, combining the physical model of step 1 and the feedforward neural network model FNN of step 3 to form a physical information neural network PDM; Step 5, using the comprehensive loss function of physical loss and data loss to train the parameters of the feedforward neural network model FNN, realizing the fusion of the physical model and the feedforward neural network model FNN, and obtaining the trained physical information neural network PDM; Step 6, using the trained physical information neural network PDM to predict the charging efficiency of the engine under different operating conditions.

2. The virtual perception method of engine charging efficiency based on physical information neural network according to claim 1, characterized in that: The physical model in step 1 is: Where n is the engine speed, P im is the intake manifold pressure, P em is the exhaust pressure, T im is the intake manifold temperature, θ1-θ 13 is the physical model parameter, and η is the estimated value of the engine charging efficiency.

3. The virtual perception method of engine charging efficiency based on physical information neural network according to claim 1, characterized in that: The structure of the feedforward neural network model FNN in step 2 includes an input layer, multiple hidden layers and an output layer, wherein the hidden layer is composed of multiple linear layers and nonlinear activation functions, the linear part of each hidden layer is implemented by nn.Linear, and the LeakyReLU activation function is applied to the output of each linear part.

4. The virtual perception method of engine charging efficiency based on physical information neural network according to claim 1, characterized in that: In step 3, the nonlinear least squares method is used to adjust the parameters θ1-θ of the physical model. 13 .

5. The virtual perception method of engine charging efficiency based on physical information neural network according to claim 4, characterized in that: In the step 3, when training the feedforward neural network model FNN, the mean square error MSE is used as the loss function, the loss function calculates the difference between the predicted value and the actual measured value of the feedforward neural network model FNN, and the optimization algorithm uses the first Adam optimizer.

6. The virtual perception method of engine charging efficiency based on physical information neural network according to claim 5, characterized in that: In step 3, after the physical model parameters are optimized, the freezing technology is used to lock the values ​​of the physical parameters, and then the non-physical parameters of the feedforward neural network model FNN are trained, and the performance of the feedforward neural network model FNN is further optimized by a data-driven method: if θ∈Θ physics ,then θ.require_grad=False Among them, Θ physics are the parameters of all physical models θ1-θ 13 A collection of .

7. The virtual perception method of engine charging efficiency based on physical information neural network according to claim 5, characterized in that: The first Adam optimizer is: Among them, θ next represents the parameters of the optimized feedforward neural network model FNN, θ represents the parameters of the feedforward neural network model FNN, l r is the learning rate, and Adam(·) represents the update function of the first Adam optimizer.

8. The virtual perception method of engine charging efficiency based on physical information neural network according to claim 1, characterized in that: In step 5, the comprehensive loss function is: L total =(1-λ physics )·L data +λ physics ·L physics Where: N is the number of samples, y physics,i is the predicted value of the physical model for the i-th sample, y pred，i is the prediction of the feedforward neural network model FNN for the i-th sample, y true，i is the measured value of the inflation efficiency of the i-th sample, L data Represents data loss, which is the mean square error between the predicted value of the feedforward neural network model FNN and the actual measured value, L physics represents the physical loss, which is the mean square error between the predicted value of the physical model and the actual measured value, λ physics is the weight coefficient used to balance the two losses, and its value is between 0 and 1.

9. The virtual perception method of engine charging efficiency based on physical information neural network according to claim 1, characterized in that: In step 5, the second Adam optimizer is used to perform fusion loss training to minimize the overall error between the predicted value and the actual measured value.

10. The virtual perception method of engine charging efficiency based on physical information neural network according to claim 9, characterized in that: The second Adam optimizer is Among them, θ next represents the parameters of the optimized feedforward neural network model FNN, θ represents the parameters of the feedforward neural network model FNN, l r is the learning rate, and Adam(·) represents the update function of the second Adam optimizer.

Images