A small sample machine learning reinforcement method for engine performance prediction and optimization
By optimizing the dataset using Latin hypercube sampling and genetic algorithms, and combining it with machine learning models, the problem of insufficient accuracy in engine performance prediction under small sample datasets was solved, achieving efficient engine performance prediction and optimization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JIANGSU UNIV
- Filing Date
- 2024-12-30
- Publication Date
- 2026-04-10
AI Technical Summary
The lack of publicly available engine performance datasets in existing technologies limits the application of machine learning in engine performance prediction and optimization. Furthermore, the limited size of self-built small sample datasets results in poor model prediction accuracy, low dataset generation efficiency, and reduced effectiveness.
An initial dataset containing 100-1000 samples was generated using the Latin hypercube sampling method. Output parameters were calculated through numerical simulation. The dataset was then optimized using machine learning models and genetic algorithms. Classification and regression models were used for operating condition identification and performance prediction. Combustion parameters with high determination coefficients were selected as input features, and a small-sample machine learning enhancement method was constructed.
It significantly improves the prediction accuracy and generalization ability of machine learning models on small sample data, reduces the time cost of engine optimization design, and realizes multi-objective optimization and rapid evaluation of engine performance.
Smart Images
Figure CN119903729B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application relates to an engine performance prediction and optimization small-sample machine learning reinforcement method and belongs to the technical field of engine performance optimization. BACKGROUND
[0002] With the rapid development of artificial intelligence, machine learning is increasingly applied in the energy field, such as engine performance prediction and optimization. However, since there is no public engine performance dataset at present, the application of machine learning in engine performance prediction is greatly limited, and model training can only be performed through self-built datasets. The limitation of self-built datasets is that the dataset size is usually no more than 1000, because a large size will result in high computing cost. Under such a small sample dataset, the prediction accuracy of the currently popular machine learning models (such as random forest, support vector machine, neural network, etc.) is poor, so it is difficult to be practically applied to engine performance prediction and operation strategy optimization.
[0003] In addition, the efficiency of the random sampling method for generating datasets is low, and a large number of abnormal combustion conditions will be obtained, resulting in a decrease in the effectiveness of the dataset for performance prediction. The method of self-built dataset needs to be further improved. SUMMARY
[0004] The application provides an engine performance prediction and optimization small-sample machine learning reinforcement method to solve the problems mentioned in the background.
[0005] Technical scheme: An engine performance prediction and optimization small-sample machine learning reinforcement method comprises the following steps:
[0006] S1, dataset generation: the dataset contains input features and output parameters; Latin hypercube sampling method is used to randomly generate input features containing 100-1000 samples, and numerical simulation method is used to calculate the output parameters of the engine under the input features, so as to construct an initial dataset;
[0007] S2, data preprocessing: the input parameters in the dataset are normalized, and for one-dimensional parameters in the output parameters, whether the in-cylinder average temperature exceeds 1100 K and whether the acoustic intensity RI is higher than 5 MW / m 2 are divided into three working conditions of normal combustion, misfire and knock;
[0008] The label value of normal combustion condition is set to 0, misfire condition is 1, and knock condition is 2. Finally, the data set 1 is arranged for classification model training, the specific engine output parameters of the normal combustion condition are saved, and the data set 2 is arranged for regression model prediction. For the two-dimensional parameters in the data set, uniform grid interpolation needs to be performed according to the crank angle;
[0009] S3, condition recognition: the data set obtained in step S2 contains different conditions, and a classification model in machine learning is used for training to realize recognition and prediction of different conditions;
[0010] S4, feature construction: the label value of the data set composed of the normal combustion condition obtained in step S2 contains engine performance parameters, combustion parameters and emission parameters. The combustion parameters are related to the performance parameters, but have no direct influence on the optimization of engine performance. First, all combustion parameters are trained and predicted by using a machine learning model, and parameters with a determination coefficient R 2 greater than 0.95 are screened out;
[0011] S5: one-dimensional parameter model training: the training set and the test set are the same as those in step S4, and the combustion parameters screened out in step S4 are used as new input features together with engine operating parameters for secondary training to predict engine performance parameters and emission parameters;
[0012] S6: two-dimensional parameter model training: consistent with step S5, the combustion parameters screened out in step S4, engine operating parameters and crank angle values are used as new input features of the machine learning model for training to predict engine in-cylinder pressure, heat release rate and temperature parameters;
[0013] S7: engine performance prediction: for given engine operating parameters, first, the classification model trained in step S3 is used for recognition, if it is a normal combustion condition, the prediction models trained in steps S5 and S6 are used for specific performance parameter prediction; if it is a non-normal combustion condition, the engine performance parameters are not specifically predicted.
[0014] Preferably, step S1 specifically comprises:
[0015] On the basis of the initial data set, a new data set is generated by using a genetic algorithm, until the determination coefficient R 2 of the prediction result of the machine learning model on the test set is greater than 0.95;
[0016] The input features are engine operating parameters, including intake temperature, intake pressure, fuel injection amount, injection timing and injection pressure. The output parameters are one-dimensional parameters and two-dimensional parameters. The one-dimensional parameters are engine performance parameters, combustion parameters and emission parameters. The two-dimensional parameters are engine combustion parameters varying with the crank angle.
[0017] Preferably, step S2 is specifically:
[0018] In step S2, the formula expression of the normalization processing is: , wherein X is the normalized value, represents the original value, is the average value, is the maximum value, is the minimum value;
[0019] In step S2, the engine performance parameters are specifically fuel consumption, combustion noise, efficiency, mean indicated pressure IMEP, and total output power; the combustion parameters are specifically in-cylinder maximum explosion pressure, pressure rise rate, crank angle CA10 corresponding to 10% burned mass fraction, crank angle CA50 corresponding to 50% burned mass fraction, and crank angle CA90 corresponding to 90% burned mass fraction; and the emission parameters are specifically nitrogen oxide NOx emission, soot Soot emission, unburned hydrocarbon HC emission, and carbon monoxide CO emission.
[0020] The two-dimensional engine parameters specifically refer to the changes of in-cylinder pressure, heat release rate, and temperature parameters with crank angle.
[0021] The uniform grid interpolation specifically refers to interpolation calculation of in-cylinder pressure, heat release rate, and temperature at the same crank angle interval, and the interpolation method is linear interpolation, and the formula is:
[0022]
[0023] wherein, and are the abscissas of two data points of the original data, y and are the ordinates corresponding to the two data points, is the abscissa of the interpolation point, is the ordinate.
[0024] Preferably, step S4 is specifically:
[0025] In step S4, the determination coefficient R 2 , and the formula expression is:
[0026]
[0027] The formula expression of MSE is:
[0028]
[0029] wherein, the value of R is between 0 and 1, and the closer to 1, the better the model prediction performance.MSE, which is between 0 and positive infinity, the closer to 0, the better the model prediction performance, represents the number of data points, represents the predicted value, represents the true value, represents the mean of the true value.
[0030] Preferably, in steps S3 and S4, the machine learning model is selected as a regression model, including but not limited to support vector machine, random forest, extreme gradient boosting, neural network and Gaussian process regression.
[0031] Preferably, in steps S3 and S4, the machine learning model is selected as an automatic machine learning model, including but not limited to Autogluon, Auto-sklearn, TPOT and H2O model.
[0032] The machine learning model is selected as a support vector machine SVM for modeling, and the optimization objective of SVM is to find a function , so that the error between the predicted value and the true value is within a certain range, and the expression of the loss function is:
[0033]
[0034] , wherein, is a kernel function, commonly used kernel functions include linear kernel, radial basis function kernel and polynomial kernel, is a bias term, is a true value, is a predicted value, is an insensitive bandwidth, used to control the tolerance of the model to small errors;
[0035] In addition to finding the regression function with the smallest loss function, the goal of SVM is to keep the model simple, that is, the norm of the weight is small, which can be achieved by minimizing the following function:
[0036]
[0037] , wherein, is a model complexity term, is a regularization parameter, used to control the complexity of the model, and are slack variables.
[0038] Beneficial effects: the present application provides a small sample machine learning enhancement method for engine parameter prediction, which can be used to enhance the prediction accuracy of the machine learning model for one-dimensional performance parameters and two-dimensional parameters changing with time of the engine, improve the generalization ability of the machine learning model, and make the small sample machine learning model have R 2The performance of the mean square error (MSE) is greatly improved. The method can be applied to multi-objective optimization of engine performance and rapid evaluation of engine cycle variation, and can significantly reduce the time cost of engine optimization design. BRIEF DESCRIPTION OF DRAWINGS
[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only a part of the embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of the provided drawings.
[0040] Figure 1 The flowchart of the present application.
[0041] Figure 2 The flowchart of data preprocessing of the present application.
[0042] Figure 3 The principle diagram of the feature construction method of the present application.
[0043] Figure 4 The MSE comparison diagram of the pre-cylinder pressure prediction value and the CFD calculation value of the present application before and after feature construction. DETAILED DESCRIPTION
[0044] The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.
[0045] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore it cannot be understood as a limitation of the present application.
[0046] In the present application, unless otherwise explicitly specified and limited, the first feature is "on" or "under" the second feature can include that the first and second features are in direct contact, or that the first and second features are not in direct contact but are in contact through another feature between them. Moreover, the first feature "on", "above" and "over" the second feature includes that the first feature is directly above and obliquely above the second feature, or only indicates that the first feature is higher in horizontal height than the second feature. The first feature "under", "below" and "under" the second feature includes that the first feature is directly below and obliquely below the second feature, or only indicates that the first feature is lower in horizontal height than the second feature.
[0047] Taking the performance prediction and optimization of a methanol-hydrogen blended engine as an example, Figure 1 is a flowchart of a small sample machine learning reinforcement method for engine performance prediction and optimization proposed by the present application, and the steps are as follows:
[0048] S1: Data set generation: using Latin hypercube random sampling method to randomly generate an initial data set containing 200 samples in the specified input feature range, and then using genetic algorithm to generate a new generation of data set containing 40 samples through a series of processes such as screening and sorting, crossing and mutation, and repeating iteration for 20 times, finally generating a data set containing 1000 samples. The range of input features is shown in Table 1:
[0049]
[0050] S2: Data preprocessing: normalizing the input features in the data set. The output features are divided into one-dimensional parameters and two-dimensional parameters, wherein the one-dimensional parameters are divided into whether the engine misfires according to whether the maximum average temperature in the engine cylinder is greater than 1100 K, and whether the combustion noise RI exceeds 5 MW / m 2 determine whether knock occurs, the label value of misfire is 1, knock is 2, and normal combustion condition is 0. Secondly, the cylinder pressure in the data set is uniformly grid interpolated, and the crank angle is uniformly grid interpolated according to an interval of 1 degree from -143 to 130°CA;
[0051] S3: Condition identification: the data set obtained in step S2 contains label values of different conditions. First, randomly divide the data set, and the proportion of the training set and the test set is 4:1. Then, use the LightGBM model in machine learning to train and realize the identification and prediction of different conditions. Since there is a sample imbalance problem in the data set, i.e. the number of samples of misfire, knock and normal combustion condition is unbalanced, the synthetic minority over-sampling technique (SMOTE) is used to process the data set, and the final classification model is saved. The prediction results of the classification model are shown in Table 2:
[0052]
[0053] In which, the results on the diagonal line of the table are all correct classifications. If no classification model is used for preliminary screening, the results of step S6 will be very different from the numerical simulation results.
[0054] S4: Feature construction: the combustion parameters of the normal combustion condition obtained in step S2 are first trained by a support vector machine model, and the training set and the test set are randomly divided in a ratio of 4:1. The Bayesian optimization method is used for hyperparameter optimization, and the number of trials is 200. The R 2 and MSE of different combustion parameters on the test set are calculated, as shown in Table 3:
[0055]
[0056] S5: One-dimensional parameter model training: during model training, the training set and the test set are consistent with those in step S4, and Pmax, PPRR, IMEP, CA50 selected in step S4 are used as new input features together with engine operating parameters to predict engine fuel consumption, combustion noise, emission and other parameters, as shown in Figure 2 The R 2 and MSE of different combustion parameters on the test set are calculated, as shown in Table 4. In the table, the prediction results of the AutoGluon automatic machine learning model without processing and the AutoGluon model after using the method of the application are compared, and it can be obviously seen that the prediction performance of the method of the application on the small sample data set is significantly improved in R 2 and MSE.
[0057]
[0058] S6: Two-dimensional parameter model training: the division of the training set and the test set is consistent with that in step S5, and the combustion parameters selected in step S4, the engine operating parameters and the crank angle value are used as new input features to train the machine learning model to predict engine in-cylinder pressure, heat release rate, temperature and other parameters. Finally, the results of all crank angles are spliced into the final results, so as to realize the prediction of the cylinder pressure curve. The MSE distribution of the model on the test set is shown in Figure 3 It can be found that the feature construction method proposed in this paper has obvious improvement effect on the prediction of engine cylinder pressure curve.
[0059] S7: Engine performance prediction: Based on the classification model and regression model established in step S3 and S5. For a specified engine operating condition, first use the classification model to identify the condition, if it is a misfire or knock example, set its value to a large number, such as 999. If it is a normal combustion condition, continue to call the regression model to predict the engine performance parameters.
[0060] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts between the various embodiments can be referred to each other. For the device disclosed by the embodiments, since it corresponds to the method disclosed by the embodiments, the description is relatively simple, and the relevant part can be referred to the method part.
[0061] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A small sample machine learning reinforcement method for engine performance prediction and optimization, characterized in that, Comprise the following steps: S1, data set generation: the input features and output parameters are contained in the data set; the input features containing 100-1000 samples are randomly generated by using Latin hypercube sampling method, and the output parameters of the engine under the input features are calculated by using numerical simulation method, so as to construct an initial data set; S2, data preprocessing: the input parameters in the data set are normalized, and for one-dimensional parameters in the output parameters, whether the in-cylinder average temperature exceeds 1100 K and whether the acoustic intensity RI is higher than 5 MW / m 2 are divided into three working conditions: normal combustion, misfire and knock. The label value of normal combustion condition is set to 0, misfire condition is 1, and knock condition is 2, finally the data set 1 is arranged for classification model training, the specific engine output parameters of normal combustion condition are saved, and the data set 2 is arranged for regression model prediction, for the two-dimensional parameters in the data set, it is necessary to carry out uniform grid interpolation according to the crank angle; S3, condition identification: the data set obtained in step S2 contains different conditions, which is trained by using classification model in machine learning to realize identification and prediction of different conditions; S4, Feature construction: the label value in the data set composed of normal combustion conditions obtained in step S2 contains engine performance parameters, combustion parameters and emission parameters, wherein the combustion parameters are related to the performance parameters, but have no direct effect on the optimization of engine performance; first, all combustion parameters are trained and predicted by a machine learning model, and parameters with a determination coefficient R 2 greater than 0.95 are screened out. S5: one-dimensional parameter model training: the training set and test set are the same as step S4, the combustion parameters screened out in step S4 are used as new input features together with engine operating parameters for secondary training to predict engine performance parameters and emission parameters; S6: two-dimensional parameter model training: consistent with step S5, on the same training set and test set, the combustion parameters screened out in step S4, engine operating parameters and crank angle value are used as new input features for machine learning model training to predict engine in-cylinder pressure, heat release rate and temperature parameters; S7: engine performance prediction: for given engine operating parameters, first use the classification model trained in step S3 for identification, if it is normal combustion condition, use the prediction models trained in steps S5 and S6 to predict specific performance parameters; If it is not normal combustion condition, the engine performance parameters are not specifically predicted.
2. The small sample machine learning reinforcement method for engine performance prediction and optimization of claim 1, wherein, Step S1 is specifically: On the basis of the initial data set, the genetic algorithm is used to automatically generate a new data set until the determination coefficient R of the prediction result of the machine learning model on the test set is greater than 0.95 2 greater than 0.95; Wherein, the input features are engine operating parameters, including intake temperature, intake pressure, fuel injection quantity, injection timing and injection pressure; the output parameters are divided into one-dimensional parameters and two-dimensional parameters, the one-dimensional parameters are engine performance parameters, combustion parameters and emission parameters; the two-dimensional parameters are engine combustion parameters varying with crank angle.
3. The small sample machine learning reinforcement method for engine performance prediction and optimization of claim 2, wherein, Step S2 is specifically: In step S2, the formula expression of the normalization processing is: wherein X is the normalized value, represents the original value, is the average value, is the maximum value, is the minimum value; In step S2, the engine performance parameters are specifically fuel consumption, combustion noise, efficiency, mean indicated pressure IMEP and total output power; the combustion parameters are specifically in-cylinder maximum explosion pressure, pressure rise rate, 10% burned mass fraction corresponding to crank angle CA10, 50% burned mass fraction corresponding to crank angle CA50 and 90% burned mass fraction corresponding to crank angle CA90; the emission parameters are specifically nitrogen oxide NOx emission, soot Soot emission, unburned carbon hydrogen HC and carbon monoxide CO emission; The engine two-dimensional parameters are specifically in-cylinder pressure, heat release rate and temperature parameters varying with crank angle; The uniform grid interpolation is specifically that the in-cylinder pressure, heat release rate and temperature are calculated by interpolation under the same crank angle interval, and the interpolation method is linear interpolation, and the formula is: ； wherein, and are the abscissas of the two data points of the original data respectively, y and are the ordinates of the two data points respectively, is the abscissa of the interpolation point, is the ordinate.
4. The small sample machine learning reinforcement method for engine performance prediction and optimization of claim 3, wherein, Step S4 is specifically: In step S4, the coefficient R is determined 2 , the formula expression is: ； The public expression of MSE is: ； wherein, the value of R2 is between 0 and 1, the closer to 1 the better the model's predictive performance, the value of RMSE is between 0 and positive infinity, the closer to 0 the better the model's predictive performance, denotes the number of data points, denotes the predicted value, denotes the true value, denotes the mean of the true values.
5. The small sample machine learning reinforcement method for engine performance prediction and optimization of claim 4, wherein, In the step S3 and the step S4, the machine learning model is selected as a regression model, including but not limited to a support vector machine, a random forest, an extreme gradient boosting, a neural network and a Gaussian process regression.
6. The small sample machine learning reinforcement method for engine performance prediction and optimization of claim 4, wherein, In the step S3 and the step S4, the machine learning model is selected as an automatic machine learning model, including but not limited to Autogluon, Auto-sklearn, TPOT and H2O model.
7. The small sample machine learning reinforcement method for engine performance prediction and optimization of claim 5, wherein: The machine learning model is selected to be modeled as a support vector machine (SVM), and an optimization objective of the SVM is to find a function that makes the error between the predicted value and the true value within a set range, and an expression of the loss function is: ； wherein, is a kernel function, commonly used kernel functions are linear kernel, radial basis function kernel and polynomial kernel, is a bias term, is a true value, is a predicted value, is an insensitive bandwidth, used to control the tolerance of the model to small errors; In addition to finding a regression function that minimizes the loss function, the goal of SVM is to keep the model simple, i.e. the norm of the weights is small, which can be achieved by minimizing the following function: ； wherein, is a complexity term for the model, is a regularization parameter for controlling the complexity of the model, and are slack variables.