MAP prediction method and device of engine

Abstract

The invention relates to an MAP prediction method and device of an engine. Comprising the following steps: preprocessing universal characteristic data of an engine under different working conditions, and inputting the preprocessed data into a preset neural network model to obtain a plurality of engine fuel consumption rate predicted values and engine combustion stability CIMEP predicted values; screening out at least one target CIMEP predicted value of which the CIMEP predicted value does not exceed a preset value from the plurality of engine combustion stability CIMEP predicted values, and screening out a target engine fuel consumption rate meeting a preset condition from the engine fuel consumption rates corresponding to each target CIMEP predicted value; and target VVT configuration parameters are determined according to the target engine fuel consumption rate, and the MAPs of the engine under different working conditions are predicted based on the target VVT configuration parameters. Therefore, the problem that an existing calibration method cannot accurately quantify the relation between input and engine performance is solved.

Description

Technical Field

[0001] The present application relates to the field of engine technology, and in particular to a method and device for predicting MAP of an engine. Background Art

[0002] In order to meet the increasingly stringent emission regulations and reduce vehicle fuel consumption, variable valve timing (VVT) technology is widely used in engine control systems. VVT technology is a technology used in automotive piston engines that can adjust the overlap time and timing of the engine's intake and exhaust systems to reduce fuel consumption and improve efficiency. In the calibration process of automotive engines, calibrating MAP control is a key step in optimizing engine performance. The traditional calibration method requires repeated scanning of the engine's universal characteristic map under different working conditions and then manually selecting the optimal solution. This is a time-consuming and costly task. In order to improve efficiency, computer-assisted engine calibration can be used.

[0003] In the related art, there are two main methods of computer-aided calibration: linear fitting method and virtual calibration technology.

[0004] However, the linear fitting method is only applicable to linear relationships. For complex nonlinear systems such as engines, it cannot accurately describe the relationship between input and engine performance. Virtual calibration technology can only predict engine performance in the early stages of the test and cannot make full use of the data obtained from later actual tests. Therefore, it cannot accurately quantify the relationship between input and engine performance, which needs to be solved urgently. Summary of the invention

[0005] The present application provides an engine MAP prediction method and device to solve the problem that the existing calibration method cannot accurately quantify the relationship between input and engine performance, making the engine calibration process more efficient and accurate.

[0006] To achieve the above object, the first embodiment of the present application provides a method for predicting MAP of an engine, comprising the following steps:

[0007] Obtain universal characteristic data of the engine under different working conditions;

[0008] Preprocessing the universal characteristic data, inputting the preprocessed universal characteristic data into a preset neural network model, and outputting a plurality of engine fuel consumption rate prediction values ​​and a plurality of engine combustion stability CIMEP (Cylinder Indicated Mean Effective Pressure) prediction values ​​respectively;

[0009] Filter out at least one target CIMEP predicted value whose CIMEP predicted value is less than or equal to a first preset threshold value from the plurality of engine combustion stability CIMEP predicted values, and filter out a target engine fuel consumption predicted value that meets a preset condition from the engine fuel consumption predicted values ​​corresponding to each target CIMEP predicted value;

[0010] A target VVT configuration parameter is determined according to the target engine fuel consumption rate prediction value, and a MAP of the engine under different operating conditions is predicted based on the target VVT configuration parameter.

[0011] Through the above-mentioned technical means, by using the preprocessed engine universal characteristic data and inputting it into the preset neural network model, the nonlinear relationship between the complex working characteristics of the engine and the fuel consumption rate and CIMEP can be captured. The powerful learning ability of the neural network model makes the prediction results more accurate, which helps to improve the accuracy of engine performance evaluation.

[0012] According to one embodiment of the present application, before the preprocessed universal characteristic data is input into the preset neural network model, it also includes:

[0013] Obtaining a public dataset of universal characteristics of the engine under historical operating conditions;

[0014] Preprocessing the universal characteristic public data set, and dividing the preprocessed universal characteristic public data set into a training set and a verification set according to a preset division ratio;

[0015] Based on a preset Bayesian algorithm, the preset neural network is trained using the training set to obtain an initial neural network model, and the initial neural network model is verified using the verification set until the initial neural network model meets the preset standard, and the iterative training of the preset neural network is terminated to obtain the preset neural network model; otherwise, the preset division ratio is adjusted, and iterative training is performed based on the adjusted new training set.

[0016] Through the above technical means, the Bayesian algorithm has the ability to automatically adjust model parameters. Training of neural networks based on the Bayesian algorithm reduces the dependence and uncertainty of manual parameter adjustment and improves the efficiency of model training.

[0017] According to an embodiment of the present application, the method of training a preset neural network based on a preset Bayesian algorithm using the training set to obtain an initial neural network model includes:

[0018] Determine the model training function and evaluation function;

[0019] According to the model training function and the evaluation function, based on the preset Bayesian algorithm, determining optimal model parameters within a preset parameter set;

[0020] Based on the optimal model parameters, the preset neural network is trained using the training set to obtain the initial neural network model.

[0021] Through the above technical means, the MLP neural network based on Bayesian optimization is used to predict the VVT ​​calibration map, which significantly reduces the demand for human resources; the entire prediction process can be completed on an ordinary computer without the need for large-scale computing resources.

[0022] According to one embodiment of the present application, the using the verification set to verify the initial neural network model until the initial neural network model meets a preset standard includes:

[0023] Inputting the verification set into the initial neural network model, and outputting a predicted value;

[0024] Calculating a root mean square error and a correlation coefficient of the predicted value, and obtaining a Gaussian distribution curve of the root mean square error and a scatter plot of the correlation coefficient based on the root mean square error and the correlation coefficient;

[0025] Based on the Gaussian distribution curve of the root mean square error and the correlation coefficient scatter plot, determining whether the number of points in the correlation coefficient scatter plot that are less than or equal to a second preset threshold value meets a preset condition;

[0026] When the number of points less than or equal to the second preset threshold satisfies the preset condition, it is determined that the initial neural network model satisfies the preset standard.

[0027] Through the above technical means, the root mean square error and correlation coefficient of the predicted value are calculated, which provides a quantitative evaluation standard for the model performance. These two indicators together provide a basis for the comprehensive evaluation of the model performance.

[0028] According to an embodiment of the present application, the step of selecting a target engine fuel consumption rate prediction value that meets a preset condition from the engine fuel consumption rate prediction values ​​corresponding to each target CIMEP prediction value includes:

[0029] Determining an engine fuel consumption prediction value corresponding to each target CIMEP prediction value;

[0030] The minimum engine fuel consumption rate prediction value is selected from the engine fuel consumption rate prediction values ​​corresponding to each target CIMEP prediction value as the target engine fuel consumption rate prediction value.

[0031] According to the engine MAP prediction method proposed in the embodiment of the present application, by preprocessing the universal characteristic data of the engine under different working conditions, the preprocessed universal characteristic data is input into a preset neural network model (MLP), and multiple engine fuel consumption rate prediction values ​​and multiple engine combustion stability CIMEP prediction values ​​can be output respectively; at least one target CIMEP prediction value whose CIMEP prediction value is less than or equal to a preset threshold is selected from the multiple engine combustion stability CIMEP prediction values, and the target engine fuel consumption rate that meets the preset conditions is selected from the engine fuel consumption rates corresponding to each target CIMEP prediction value; the target VVT configuration parameters are determined according to the target engine fuel consumption rate, and the MAP of the engine under different working conditions is predicted based on the target VVT configuration parameters. Therefore, by adopting the neural network MLP, the optimal MAP value under different working conditions can be predicted, which solves the problem that the existing calibration method cannot accurately quantify the relationship between the input and the engine performance, making the engine calibration process more efficient and accurate.

[0032] To achieve the above-mentioned purpose, a second embodiment of the present application provides a MAP prediction device for an engine, comprising:

[0033] An acquisition module is used to obtain universal characteristic data of the engine under different working conditions;

[0034] An acquisition module, used for preprocessing the universal characteristic data, inputting the preprocessed universal characteristic data into a preset neural network model, and outputting a plurality of engine fuel consumption rate prediction values ​​and a plurality of engine combustion stability CIMEP prediction values ​​respectively;

[0035] a screening module, configured to screen out at least one target CIMEP predicted value whose CIMEP predicted value is less than or equal to a first preset threshold value from the plurality of engine combustion stability CIMEP predicted values, and to screen out a target engine fuel consumption predicted value that meets a preset condition from the engine fuel consumption predicted values ​​corresponding to each target CIMEP predicted value;

[0036] The prediction module is used to determine the target VVT configuration parameters according to the predicted value of the target engine fuel consumption rate, and predict the MAP of the engine under different working conditions based on the target VVT configuration parameters.

[0037] According to one embodiment of the present application, before inputting the preprocessed universal characteristic data into the preset neural network model, the obtaining module further includes:

[0038] An acquisition unit, used for acquiring a universal characteristic public data set of the engine under historical operating conditions;

[0039] A partitioning unit, configured to preprocess the universal characteristic public data set, and partition the preprocessed universal characteristic public data set into a training set and a validation set according to a preset partitioning ratio;

[0040] A training unit is used to train a preset neural network using the training set based on a preset Bayesian algorithm to obtain an initial neural network model, and to verify the initial neural network model using the verification set, until the initial neural network model meets a preset standard, and then terminate the iterative training of the preset neural network to obtain the preset neural network model; otherwise, adjust the preset division ratio, and perform iterative training based on the adjusted new training set.

[0041] According to one embodiment of the present application, the training unit is specifically used for:

[0042] Determine the model training function and evaluation function;

[0043] According to the model training function and the evaluation function, based on the preset Bayesian algorithm, determining optimal model parameters within a preset parameter set;

[0044] Based on the optimal model parameters, the preset neural network is trained using the training set to obtain the initial neural network model.

[0045] According to one embodiment of the present application, the training unit is specifically used for:

[0046] Inputting the verification set into the initial neural network model, and outputting a predicted value;

[0047] Calculating a root mean square error and a correlation coefficient of the predicted value, and obtaining a Gaussian distribution curve of the root mean square error and a scatter plot of the correlation coefficient based on the root mean square error and the correlation coefficient;

[0048] Based on the Gaussian distribution curve of the root mean square error and the correlation coefficient scatter plot, determining whether the number of points in the correlation coefficient scatter plot that are less than or equal to a second preset threshold value meets a preset condition;

[0049] When the number of points less than or equal to the second preset threshold satisfies the preset condition, it is determined that the initial neural network model satisfies the preset standard.

[0050] According to one embodiment of the present application, the screening module is specifically used for:

[0051] Determining an engine fuel consumption prediction value corresponding to each target CIMEP prediction value;

[0052] The minimum engine fuel consumption rate prediction value is selected from the engine fuel consumption rate prediction values ​​corresponding to each target CIMEP prediction value as the target engine fuel consumption rate prediction value.

[0053] According to the MAP prediction device of the engine proposed in the embodiment of the present application, by preprocessing the universal characteristic data of the engine under different working conditions, the preprocessed universal characteristic data is input into the preset neural network model (MLP), and multiple engine fuel consumption rate prediction values ​​and multiple engine combustion stability CIMEP prediction values ​​can be output respectively; at least one target CIMEP prediction value whose CIMEP prediction value is less than or equal to the preset threshold is selected from the multiple engine combustion stability CIMEP prediction values, and the target engine fuel consumption rate that meets the preset conditions is selected from the engine fuel consumption rates corresponding to each target CIMEP prediction value; the target VVT configuration parameters are determined according to the target engine fuel consumption rate, and the MAP of the engine under different working conditions is predicted based on the target VVT configuration parameters. Therefore, by adopting the neural network MLP, the optimal MAP value under different working conditions can be predicted, which solves the problem that the existing calibration method cannot accurately quantify the relationship between the input and the engine performance, making the engine calibration process more efficient and accurate.

[0054] To achieve the above objectives, the third aspect of the present application proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the engine MAP prediction method as described in the above embodiment.

[0055] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the MAP prediction method of the engine as described in the above embodiments.

[0056] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0058] Figure 1 A flowchart of a MAP prediction method for an engine provided according to an embodiment of the present application;

[0059] Figure 2 is a schematic diagram of a hardware system according to an embodiment of the present application;

[0060] Figure 3 is a schematic diagram of the extraction result and the screening result of the input data according to one embodiment of the present application;

[0061] Figure 4 is a schematic diagram of a prediction progress bar according to an embodiment of the present application;

[0062] Figure 5 A schematic diagram of a Bayesian optimization result according to an embodiment of the present application;

[0063] Figure 6 is a schematic diagram of a Gaussian distribution curve of a root mean square error and a scatter plot of a correlation coefficient according to an embodiment of the present application;

[0064] Figure 7 A schematic diagram of calculating a prediction value credibility result using Euclidean distance according to an embodiment of the present application;

[0065] Figure 8 A flowchart of another engine MAP prediction method provided according to an embodiment of the present application;

[0066] Fig. 9 A schematic diagram showing a comparison between a predicted VVT value and an actual value under corresponding speed loads according to an embodiment of the present application;

[0067] Fig.10 A schematic diagram of a UI interface according to an embodiment of the present application;

[0068] Fig.11 MAP diagram of intake and exhaust VVT according to one embodiment of the present application;

[0069] Fig.12 It is a block diagram of a MAP prediction device for an engine provided according to an embodiment of the present application;

[0070] Fig.13 It is a schematic diagram of the structure of an electronic device provided according to an embodiment of the present application.

[0071] Explanation of reference numerals: 10 - MAP prediction device of engine, 100 - acquisition module, 200 - acquisition module, 300 - screening module, 400 - prediction module; 1301 - memory, 1302 - processor, 1303 - communication interface. DETAILED DESCRIPTION

[0072] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0073] The following describes the engine MAP prediction method and device proposed according to the embodiments of the present application with reference to the accompanying drawings.

[0074] Figure 1 It is a flowchart of a method for predicting MAP of an engine according to an embodiment of the present application.

[0075] Before introducing the engine MAP prediction method proposed in the embodiment of the present application, the hardware system and software parts involved in the engine MAP prediction method are briefly introduced. Figure 2 As shown in the figure, the hardware part includes: engine, engine ECU (Engine Control Unit), INCA (Integrated Calibration and Application Tool, measurement, calibration and diagnostic software), test bench and host computer. Among them, the engine is the core component, which is precisely controlled by the engine ECU to maintain the stability of the working condition. The engine ECU is connected to the INCA system and is responsible for collecting and recording the universal characteristic data of the engine under various working conditions. In the hardware setting, the test bench is responsible for controlling the dynamometer, recording all test data, and controlling the speed and throttle opening through the electrical interface to ensure the same working conditions. These recorded data will be used in the subsequent host computer software to train the neural network. The host computer is installed with the required software environment and tools for training the neural network. By training the collected data, an accurate prediction can be obtained, providing a powerful tool for vehicle performance optimization and fuel economy analysis.

[0076] The software uses Python 3.12 and the following library functions: tkinter (for creating user interfaces), subprocess and os (for executing external commands in multiple threads), pandas (for processing and analyzing CSV (Comma Separated Values) data), numpy (for data calculations), scikit-learn (for machine learning), joblib (for importing and exporting models, and parallel computing), tqdm (for creating progress bars), matplotlib (for drawing), and scipy (for interpolation).

[0077] Next, the MAP prediction method of the engine of the embodiment of the present application is introduced in detail based on the software processing part.

[0078] For example, Figure 1 As shown, the MAP prediction method of the engine includes the following steps:

[0079] In step S101, universal characteristic data of the engine under different working conditions are obtained.

[0080] Specifically, experimental equipment or simulation software (such as the INCA system) can be used to collect universal characteristic data of the engine under different operating conditions. These data usually include speed, load, intake and exhaust VVT, and fuel consumption rate.

[0081] In step S102, the universal characteristic data is preprocessed, and the preprocessed universal characteristic data is input into a preset neural network model to output a plurality of engine fuel consumption rate prediction values ​​and a plurality of engine combustion stability CIMEP prediction values ​​respectively.

[0082] Among them, in the embodiment of the present application, the preset neural network model refers to a trained MLP (Multilayer Perceptron) prediction model, which can better adapt to nonlinear relationships, accurately describe the complex relationship between input and engine performance, and can make full use of all measured data, fully mine and utilize historical data, and improve the use value of data.

[0083] Specifically, after obtaining the universal characteristic data, the universal characteristic data can be preprocessed, that is, data cleaning (removing outliers, missing values ​​or unreasonable data points) and formatting are performed to ensure data quality, and then the preprocessed universal characteristic data is input into a preset neural network model (i.e., MLP) to predict the engine fuel consumption rate and CIMEP (characterizing the combustion stability of the engine) under different working conditions, and output multiple engine fuel consumption rate prediction values ​​and multiple CIMEP prediction values ​​respectively, wherein each engine fuel consumption rate prediction value corresponds one-to-one to each CIMEP prediction value.

[0084] For example, if Figure 3 As shown, use the library os (for file path operations) and pandas (for data processing) to read all excel files and all sheets in the "original data" folder and export them to the process folder (such as Figure 3 (a)). After using pandas methods such as dropna() and replace() to remove missing values ​​and unnecessary text data, select the data name and range according to the requirements of the data to be filtered to remove outliers (such as Figure 3(b)). In this embodiment, the setting standards are FUELCOSP>0&FUELCOSP<400, SPEED>0&SPEED<5500, RL_MESS>0&RL_MESS<150, CIMEP0>0&CIMEP0<30. The names to be selected can be adjusted at will according to different data sources, and the range can be changed. At this time, the exported data structure is that the last row is the output value, and all the previous data columns are the input values ​​during prediction. Therefore, it can be quickly input during the later data import, and there is no need to change the input quantity name when facing data update. Finally, a complete file directory is constructed to facilitate the search of process files and prevent subsequent program errors.

[0085] Since the MLP model is a multi-input single-output model, it needs to be retrained for different output values ​​(i.e., engine fuel consumption rate prediction value and CIMEP prediction value), and the corresponding data name and data screening range need to be changed when adjusting the training content. After the data is imported, it is normalized and the data obtained by automatic parameter selection is filled into the reserved interface. When the automatic parameter selection does not meet the requirements, the reserved interface can be used to manually adjust the parameters according to the MLP model parameter definition. The training time can be estimated according to the progress bar prompt, such as Figure 4 To speed up the process, when predicting the engine fuel consumption rate and CIMEP, these two sets of training can be carried out simultaneously. The two preset neural network models, model1 and model2, can be exported through the joblib library for easy calling at any time to avoid repeated training in subsequent processes.

[0086] In step S103, at least one target CIMEP prediction value whose CIMEP prediction value is less than or equal to a first preset threshold is selected from a plurality of engine combustion stability CIMEP prediction values, and a target engine fuel consumption prediction value that meets preset conditions is selected from the engine fuel consumption prediction values ​​corresponding to each target CIMEP prediction value.

[0087] The first preset threshold value may be a value pre-set by researchers in the field, a value obtained through a limited number of experiments, or a value obtained through a limited number of computer simulations, which is not specifically limited here. Preferably, in the embodiment of the present application, the first preset threshold value is set to 3.

[0088] Specifically, after obtaining multiple engine fuel consumption rate prediction values ​​and multiple CIMEP prediction values, the CIMEP prediction value is used as a restriction condition to screen out at least one target CIMEP prediction value whose CIMEP prediction value is less than or equal to a first preset threshold value (such as 3) from the multiple CIMEP prediction values. For each target CIMEP prediction value, an engine fuel consumption rate prediction value that meets the preset condition (i.e., the minimum fuel consumption rate) can be screened out from the corresponding engine fuel consumption rate prediction values ​​as the target engine fuel consumption rate prediction value.

[0089] It can be understood that introducing parameters such as CIMEP prediction value as constraints can prevent VVT selection from falling into the local optimal solution.

[0090] The following describes in detail how to filter out the target engine fuel consumption rate prediction value that meets the preset conditions.

[0091] As a possible implementation method, in some embodiments, a target engine fuel consumption rate prediction value that meets preset conditions is screened out from the engine fuel consumption rate prediction values ​​corresponding to each target CIMEP prediction value, including: determining the engine fuel consumption rate prediction value corresponding to each target CIMEP prediction value; selecting the minimum engine fuel consumption rate prediction value from the engine fuel consumption rate prediction values ​​corresponding to each target CIMEP prediction value as the target engine fuel consumption rate prediction value.

[0092] Specifically, based on the correspondence between the predicted value of each engine fuel consumption rate and each CIMEP predicted value, the engine fuel consumption rate predicted value corresponding to each target CIMEP predicted value is found and determined, and the engine fuel consumption rate predicted value corresponding to each target CIMEP predicted value is arranged in ascending or descending order, and the smallest engine fuel consumption rate predicted value is selected as the target engine fuel consumption rate predicted value.

[0093] In step S104, target VVT configuration parameters are determined according to the target engine fuel consumption rate prediction value, and the MAP of the engine under different operating conditions is predicted based on the target VVT configuration parameters.

[0094] It can be understood that since each engine fuel consumption rate corresponds to a VVT configuration parameter, after determining the target engine fuel consumption rate, the corresponding target VVT configuration parameters (i.e., optimal VVT) can be determined based on the target engine fuel consumption rate through a lookup table, interpolation method, etc. Based on the target VVT configuration parameters, a neural network model or other prediction methods are used to predict the MAP of the engine under different operating conditions, analyze the prediction results, evaluate the impact of the target VVT configuration parameters on the engine performance, and make adjustments and optimizations as needed.

[0095] It should be noted that after the target VVT configuration parameters are determined, the target VVT configuration parameters may be verified to verify whether the configuration parameters can achieve the expected fuel economy and performance on the actual engine.

[0096] For ease of understanding, the following describes in detail how to obtain the preset neural network model.

[0097] Furthermore, in some embodiments, before the preprocessed universal characteristic data is input into the preset neural network model, it also includes: obtaining a universal characteristic public data set of the engine under historical operating conditions; preprocessing the universal characteristic public data set, and dividing the preprocessed universal characteristic public data set into a training set and a verification set according to a preset division ratio; based on a preset Bayesian algorithm, using the training set to train the preset neural network to obtain an initial neural network model, and using the verification set to verify the initial neural network model until the initial neural network model meets the preset standard, ending the iterative training of the preset neural network to obtain the preset neural network model, otherwise, adjusting the preset division ratio, and performing iterative training based on the adjusted new training set.

[0098] Specifically, we can obtain the universal characteristic public data set of the engine under historical operating conditions (including speed, load, intake and exhaust VVT and fuel consumption rate, etc.) from reliable sources, remove outliers, missing values ​​or unreasonable data points in the universal characteristic public data set, and standardize the data so that the neural network model can better process it. According to experience or experimental requirements, the preprocessed universal characteristic public data set is divided into a training set and a validation set according to a preset division ratio (such as 80% for the training set and 20% for the validation set). The Bayesian algorithm (such as Bayesian hyperparameter optimization) is used to find the optimal neural network hyperparameters (such as learning rate, regularization coefficient, etc.) to improve the model training efficiency and performance. The training set is used to iteratively train the preset neural network (MLP), and the validation set is used to evaluate the model performance after each iteration. The performance evaluation results on the validation set (such as accuracy, loss value, etc.) are used to determine whether the model meets the preset standards. If not, the preset division ratio can be adjusted (or the neural network structure or hyperparameters can be adjusted). A new round of iterative training is performed based on the adjusted new training set until the performance of the model on the validation set meets the preset standards. The iterative training is terminated to obtain the final preset neural network model.

[0099] Therefore, by inputting the sorted data into the MLP model for training, the complex relationship between input features and fuel consumption is learned. When predicting fuel consumption, the speed, load, and VVT are input. The preset neural network model obtained after training can output the corresponding fuel consumption prediction value. If you want to filter out the VVT ​​value under specific fuel consumption conditions, you can also use fuel consumption as a criterion, speed and load as input, and perform reverse prediction. The preset neural network model obtained after training can quickly and accurately predict the optimal MAP value under different working conditions, lowering the threshold for engine calibration, providing VVT prediction values ​​in a short time, speeding up the calibration process, and making the engine calibration process more efficient and accurate.

[0100] In some embodiments, based on a preset Bayesian algorithm, a preset neural network is trained using a training set to obtain an initial neural network model, including: determining a model training function and an evaluation function; according to the model training function and the evaluation function, based on a preset Bayesian algorithm, determining optimal model parameters within a preset parameter set; based on the optimal model parameters, training the preset neural network using a training set to obtain an initial neural network model.

[0101] Among them, the model training function is a function used to train the neural network. It can accept the neural network model, training data and hyperparameters as input, and update the weights and biases of the neural network through the back propagation algorithm to minimize the loss function; the evaluation function is mainly used to calculate the model's accuracy, recall rate, F1 score, loss value and other indicators.

[0102] The Bayesian optimization algorithm is mainly used for parameter adjustment in the field of machine learning and for optimizing functions with high evaluation costs. For traditional machine learning parameter optimization problems, you can choose a gradient descent (SGD) method to directly access the derivative of the function. Given some cost functions to be minimized, the algorithm iteratively takes the step with the maximum downward slope, and theoretically reaches the minimum after a sufficient number of iterations; or in some optimization scenarios, the evaluation cost of the function is low, and simple brute force search methods can be used, such as manual parameter adjustment, grid search, random grid search, etc., to traverse possible parameter combinations as much as possible. The Bayesian optimization algorithm is mainly used in scenarios where the function cost evaluation is complex (long training time, high parameter adjustment cost), model black box (unable to directly observe the target function), and the optimization purpose is global optimization.

[0103] In Bayesian optimization, we can first assume that the objective function follows a prior distribution (usually a Gaussian process model), and then in each iteration, we select a point that is most likely to improve performance based on the current objective function probability model for evaluation. After the evaluation is completed, the new observations are added to the model and the probability model is updated. This process is repeated until the preset number of iterations is reached or other stopping conditions are met.

[0104] like Figure 5 As shown in the figure, the Bayesian method is used to adjust the parameters multiple times, and the parameter adjustment efficiency is improved by 70% compared with the grid search, and the optimal training effect is automatically achieved without occupying human resources. Compared with grid search and random search, Bayesian optimization has the following advantages: Intelligent selection of evaluation points: Bayesian optimization can intelligently select the next evaluation point based on historical observations, so as to find the parameter configuration close to the optimal solution in a small number of iterations; Processing of complex objective functions: Bayesian optimization can handle complex objective functions such as multi-peak and non-convex; Adaptive adjustment of search strategy: Bayesian optimization can adaptively adjust the search strategy according to the nature of the objective function to better balance the relationship between exploration and utilization.

[0105] Specifically, in the Bayesian optimization process of the embodiment of the present application, a hyperparameter search space can be preset (i.e., a parameter set, and the range of the space can be manually adjusted to ensure that the optimal model parameters are basically within the space). The Bayesian optimization algorithm constructs a probability model (Gaussian process) of the objective function, and uses the probability model to select the next optimal parameter point that hopes to find the optimal solution for evaluation, and updates the probability model, thereby achieving the parameter value that maximizes the objective function (the evaluation function value on the validation set) in the preset parameter set as the optimal model parameter, and the optimal model parameters are exported to a CSV file for easy viewing. After obtaining the optimal model parameters, these parameters are used as the initial parameters of the neural network, and the neural network is trained using the training set to obtain the initial neural network model.

[0106] Among them, in some embodiments, the initial neural network model is verified using a verification set until the initial neural network model meets the preset standards, including: inputting the verification set into the initial neural network model and outputting a predicted value; calculating the root mean square error and correlation coefficient of the predicted value, and obtaining a Gaussian distribution curve of the root mean square error and a scatter plot of the correlation coefficient based on the root mean square error and the correlation coefficient; based on the Gaussian distribution curve of the root mean square error and the scatter plot of the correlation coefficient, determining whether the number of points in the correlation coefficient scatter plot that are less than or equal to a second preset threshold value meets the preset conditions; when the number of points that are less than or equal to the second preset threshold value meets the preset conditions, determining that the initial neural network model meets the preset standards.

[0107] Among them, the root mean square error is a common indicator to measure the difference between the model prediction value and the true value. It calculates the square root of the average of the squares of the difference between the predicted value and the true value. The smaller the root mean square error, the more accurate the model prediction; the correlation coefficient (such as the Pearson correlation coefficient) refers to the correlation coefficient between the predicted value and the true value, which can measure the strength and direction of the linear relationship between the two. The closer the absolute value of the correlation coefficient is to 1, the stronger the linear relationship between the predicted value and the true value.

[0108] The second preset threshold can be a value pre-set by researchers in this field, or a value obtained through a limited number of experiments, or a value obtained through a limited number of computer simulations. It is not specifically limited here and is used to determine whether the linear relationship between the predicted value and the true value is strong enough.

[0109] Specifically, the prepared validation set is input into the trained initial neural network model. The model can generate corresponding prediction values ​​according to the input data. By calculating the root mean square error and correlation coefficient between the predicted value and the true value, a Gaussian distribution curve of the root mean square error can be drawn (such as Figure 6 (b)) and the correlation coefficient scatter plot (as shown in Figure 6 (a) shows), the distribution of the root mean square error can be visualized through the Gaussian distribution curve, and the distribution of the correlation coefficient can be intuitively observed through the correlation coefficient scatter plot. The training effect of the model can be judged by image analysis, that is, based on the correlation coefficient scatter plot, traverse each point in the correlation coefficient scatter plot, calculate the number of points in the figure that are less than or equal to the second preset threshold, and after setting the second preset threshold, Figure 6 The blue area in (a) will identify the data points that meet the threshold condition, while the red area represents the data points that do not meet the condition. It is determined whether the number of points less than or equal to the second preset threshold meets the preset condition (which can be a specific value). If the number of points less than or equal to the second preset threshold meets the preset condition, then it is considered that the initial neural network model performs well and meets the preset standard.

[0110] In addition, when judging that the initial neural network model meets the preset standard, the Euclidean distance can also be used to judge the error size. Taking N = 1500 and RL = 60 as an example to predict the engine fuel consumption rate, the credibility is calculated by the Euclidean distance, as shown in Figure 7 shown.

[0111] In order to facilitate those skilled in the art to further understand the engine MAP prediction method proposed in the embodiment of the present application, it is further elaborated below in conjunction with specific embodiments.

[0112] like Figure 8 As shown, the MAP prediction method of the engine may further include the following steps:

[0113] Step S801, inputting the gantry data.

[0114] Step S802: data screening and extraction.

[0115] Specifically, useful information is extracted from the data collected by the hardware system and filtered.

[0116] Step S803: automatic parameter selection for training.

[0117] Specifically, Bayesian optimization is used to find the optimal parameter combination in the parameter space to improve training accuracy. The parameter space can be ranged based on manual parameter adjustment to ensure that the optimal parameter combination is basically within the parameter space, and the optimal parameter combination is exported to a CSV file for later viewing.

[0118] Step S804, perform model training respectively, and output engine fuel consumption rate and engine combustion stability CIMEP.

[0119] It is understandable that since the MLP model is a multi-input single-output model, it needs to be retrained for different output values. In order to speed up the progress, two sets of training can be carried out at the same time to obtain two models, model1 and model2. After the model is exported, it can be called at any time to avoid repeated training in subsequent processes.

[0120] Step S805: error analysis.

[0121] Step S806, determine whether the analysis result meets the accuracy requirement. If yes, execute step S810, otherwise, execute step S807.

[0122] Specifically, since neural network regression analysis mainly focuses on the root mean square error and correlation coefficient, after the required data is input into the model in batches, the predicted value can be obtained. Subsequently, the root mean square error and correlation coefficient of the predicted value are calculated and the corresponding images are drawn. In the image, the scatter plot of the correlation coefficient and the Gaussian distribution curve of the predicted value error will be presented. Through the analysis of these images, the training effect of the model can be evaluated.

[0123] Step S807, training manual parameter selection.

[0124] Step S808, perform model training, and output engine fuel consumption rate and engine combustion stability CIMEP.

[0125] Step S809: error analysis.

[0126] Step S810, predicting the engine fuel consumption rate and the engine combustion stability CIMEP.

[0127] Step S811, predicting VVT and MAP under a single operating condition.

[0128] With N=1500 and RL=60 as parameters, VVT prediction is performed and a comparison chart between VVT and actual value under corresponding speed and load is generated. Fig. 9 . Design user interfaces such as Fig.10 As shown, the main function is automatically executed to facilitate user operation, and the MAP diagram of intake and exhaust VVT is drawn according to the predicted VVT value, such as Fig.11 As shown, Fig.11(a) is the MAP diagram of intake VVT, Fig.11 (b) is the MAP diagram of exhaust VVT.

[0129] The prediction process of the MLP model is automated, without the need for real-time human intervention. It can be started automatically and even run at night, which significantly improves work efficiency. Compared with manual operations, machine learning models have shown higher efficiency in processing large amounts of data and complex relationships. By applying the Bayesian optimization method, the parameter adjustment of the MLP model is also automated. In addition, the Map prediction operation using the MLP model is intuitive and easy. Thanks to the free installation of Python and the convenient library installation method of the pip command, setting up the environment becomes a breeze. When using it, users only need to click buttons in the integrated graphical user interface to complete the prediction task without in-depth understanding of complex professional knowledge, which greatly reduces the technical threshold.

[0130] It should be noted that the embodiments of the present application can also use traditional methods such as linear fitting to achieve similar effects; in addition to MLP, other neural network models such as CNN (Convolutional Neural Network), RNN (Recurrent Neural Network) can also be used to achieve similar effects; in addition, Auto ML (Automated Machine Learning) tools such as TPOT can also be used to achieve similar effects. Therefore, it is necessary to protect the application scenario of this technology in engine bench calibration.

[0131] According to the engine MAP prediction method proposed in the embodiment of the present application, by preprocessing the universal characteristic data of the engine under different working conditions, the preprocessed universal characteristic data is input into a preset neural network model (MLP), and multiple engine fuel consumption rate prediction values ​​and multiple engine combustion stability CIMEP prediction values ​​can be output respectively; at least one target CIMEP prediction value whose CIMEP prediction value is less than or equal to a preset threshold is selected from the multiple engine combustion stability CIMEP prediction values, and the target engine fuel consumption rate that meets the preset conditions is selected from the engine fuel consumption rates corresponding to each target CIMEP prediction value; the target VVT configuration parameters are determined according to the target engine fuel consumption rate, and the MAP of the engine under different working conditions is predicted based on the target VVT configuration parameters. Therefore, by adopting the neural network MLP, the optimal MAP value under different working conditions can be predicted, which solves the problem that the existing calibration method cannot accurately quantify the relationship between the input and the engine performance, making the engine calibration process more efficient and accurate.

[0132] Next, the MAP prediction device of the engine proposed in accordance with the embodiment of the present application will be described with reference to the accompanying drawings.

[0133] Fig.12 It is a block diagram of a MAP prediction device for an engine according to an embodiment of the present application.

[0134] like Fig.12 As shown, the MAP prediction device 10 of the engine includes: an acquisition module 100 , an acquisition module 200 , a screening module 300 and a prediction module 400 .

[0135] Wherein, the acquisition module 100 is used to acquire universal characteristic data of the engine under different working conditions;

[0136] The acquisition module 200 is used to pre-process the universal characteristic data, input the pre-processed universal characteristic data into a preset neural network model, and output a plurality of engine fuel consumption rate prediction values ​​and a plurality of engine combustion stability CIMEP prediction values ​​respectively;

[0137] A screening module 300 is used to screen out at least one target CIMEP predicted value whose CIMEP predicted value is less than or equal to a first preset threshold value from a plurality of engine combustion stability CIMEP predicted values, and screen out a target engine fuel consumption predicted value that meets a preset condition from the engine fuel consumption predicted values ​​corresponding to each target CIMEP predicted value;

[0138] The prediction module 400 is used to determine the target VVT configuration parameters according to the predicted value of the target engine fuel consumption rate, and predict the MAP of the engine under different working conditions based on the target VVT configuration parameters.

[0139] Furthermore, in some embodiments, before the preprocessed universal characteristic data is input into the preset neural network model, the obtaining module 200 further includes:

[0140] An acquisition unit, used to acquire a public dataset of universal characteristics of the engine under historical operating conditions;

[0141] A partitioning unit, used for preprocessing the universal characteristic public data set, and dividing the preprocessed universal characteristic public data set into a training set and a verification set according to a preset partitioning ratio;

[0142] The training unit is used to train the preset neural network using the training set based on the preset Bayesian algorithm to obtain an initial neural network model, and to verify the initial neural network model using the verification set, until the initial neural network model meets the preset standard, and then terminate the iterative training of the preset neural network to obtain the preset neural network model; otherwise, adjust the preset division ratio, and perform iterative training based on the adjusted new training set.

[0143] Furthermore, in some embodiments, the training unit is specifically configured to:

[0144] Determine the model training function and evaluation function;

[0145] According to the model training function and the evaluation function, based on the preset Bayesian algorithm, the optimal model parameters are determined within the preset parameter set;

[0146] Based on the optimal model parameters, the preset neural network is trained using the training set to obtain the initial neural network model.

[0147] Furthermore, in some embodiments, the training unit is specifically configured to:

[0148] Input the validation set into the initial neural network model and output the predicted value;

[0149] Calculate the root mean square error and correlation coefficient of the predicted value, and obtain the Gaussian distribution curve of the root mean square error and the scatter plot of the correlation coefficient based on the root mean square error and the correlation coefficient;

[0150] Based on the Gaussian distribution curve of the root mean square error and the correlation coefficient scatter plot, determining whether the number of points in the correlation coefficient scatter plot that is less than or equal to the second preset threshold value meets a preset condition;

[0151] When the number of points less than or equal to the second preset threshold satisfies the preset condition, it is determined that the initial neural network model meets the preset standard.

[0152] Furthermore, in some embodiments, the screening module 300 is specifically configured to:

[0153] Determine the engine fuel consumption prediction value corresponding to each target CIMEP prediction value;

[0154] The minimum engine fuel consumption rate prediction value is selected from the engine fuel consumption rate prediction values ​​corresponding to each target CIMEP prediction value as the target engine fuel consumption rate prediction value.

[0155] It should be noted that the above explanation of the embodiment of the engine MAP prediction method is also applicable to the engine MAP prediction device of this embodiment, which will not be repeated here.

[0156] According to the MAP prediction device of the engine proposed in the embodiment of the present application, by preprocessing the universal characteristic data of the engine under different working conditions, the preprocessed universal characteristic data is input into the preset neural network model (MLP), and multiple engine fuel consumption rate prediction values ​​and multiple engine combustion stability CIMEP prediction values ​​can be output respectively; at least one target CIMEP prediction value whose CIMEP prediction value is less than or equal to the preset threshold is selected from the multiple engine combustion stability CIMEP prediction values, and the target engine fuel consumption rate that meets the preset conditions is selected from the engine fuel consumption rates corresponding to each target CIMEP prediction value; the target VVT configuration parameters are determined according to the target engine fuel consumption rate, and the MAP of the engine under different working conditions is predicted based on the target VVT configuration parameters. Therefore, by adopting the neural network MLP, the optimal MAP value under different working conditions can be predicted, which solves the problem that the existing calibration method cannot accurately quantify the relationship between the input and the engine performance, making the engine calibration process more efficient and accurate.

[0157] Fig.13 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include:

[0158] A memory 1301 , a processor 1302 , and a computer program stored in the memory 1301 and executable on the processor 1302 .

[0159] When the processor 1302 executes the program, the engine MAP prediction method provided in the above embodiment is implemented.

[0160] Furthermore, the electronic device further comprises:

[0161] The communication interface 1303 is used for communication between the memory 1301 and the processor 1302 .

[0162] The memory 1301 is used to store computer programs that can be executed on the processor 1302 .

[0163] The memory 1301 may include a high-speed RAM (Random Access Memory) memory, and may also include a non-volatile memory, such as at least one disk memory.

[0164] If the memory 1301, the processor 1302 and the communication interface 1303 are implemented independently, the communication interface 1303, the memory 1301 and the processor 1302 can be connected to each other through a bus and communicate with each other. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.13 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0165] Optionally, in a specific implementation, if the memory 1301, the processor 1302 and the communication interface 1303 are integrated on a chip, the memory 1301, the processor 1302 and the communication interface 1303 can communicate with each other through an internal interface.

[0166] The processor 1302 may be a CPU (Central Processing Unit), or an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present application.

[0167] The embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the above-mentioned engine MAP prediction method when executed by a processor.

[0168] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0169] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms are not necessarily directed to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0170] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for predicting MAP of an engine, characterized in that: The following steps are involved: Obtain universal characteristic data of the engine under different working conditions; Preprocessing the universal characteristic data, inputting the preprocessed universal characteristic data into a preset neural network model, and outputting a plurality of engine fuel consumption rate prediction values ​​and a plurality of engine combustion stability CIMEP prediction values ​​respectively; Filter out at least one target CIMEP predicted value whose CIMEP predicted value is less than or equal to a first preset threshold value from the plurality of engine combustion stability CIMEP predicted values, and filter out a target engine fuel consumption predicted value that meets a preset condition from the engine fuel consumption predicted values ​​corresponding to each target CIMEP predicted value; A target VVT configuration parameter is determined according to the target engine fuel consumption rate prediction value, and a MAP of the engine under different operating conditions is predicted based on the target VVT configuration parameter.

2. The method according to claim 1, characterized in that Before inputting the preprocessed universal characteristic data into the preset neural network model, the method further includes: Obtaining a public dataset of universal characteristics of the engine under historical operating conditions; Preprocessing the universal characteristic public data set, and dividing the preprocessed universal characteristic public data set into a training set and a verification set according to a preset division ratio; Based on a preset Bayesian algorithm, the preset neural network is trained using the training set to obtain an initial neural network model, and the initial neural network model is verified using the verification set until the initial neural network model meets the preset standard, and the iterative training of the preset neural network is terminated to obtain the preset neural network model; otherwise, the preset division ratio is adjusted, and iterative training is performed based on the adjusted new training set.

3. The method according to claim 2, characterized in that The method of training a preset neural network based on a preset Bayesian algorithm using the training set to obtain an initial neural network model includes: Determine the model training function and evaluation function; According to the model training function and the evaluation function, based on the preset Bayesian algorithm, determining optimal model parameters within a preset parameter set; Based on the optimal model parameters, the preset neural network is trained using the training set to obtain the initial neural network model.

4. The method according to claim 2, characterized in that: The using the verification set to verify the initial neural network model until the initial neural network model meets a preset standard includes: Inputting the verification set into the initial neural network model, and outputting a predicted value; Calculating a root mean square error and a correlation coefficient of the predicted value, and obtaining a Gaussian distribution curve of the root mean square error and a scatter plot of the correlation coefficient based on the root mean square error and the correlation coefficient; Based on the Gaussian distribution curve of the root mean square error and the correlation coefficient scatter plot, determining whether the number of points in the correlation coefficient scatter plot that are less than or equal to a second preset threshold value meets a preset condition; When the number of points less than or equal to the second preset threshold satisfies the preset condition, it is determined that the initial neural network model satisfies the preset standard.

5. The method according to claim 1, characterized in that The step of selecting a target engine fuel consumption rate prediction value that meets a preset condition from the engine fuel consumption rate prediction values ​​corresponding to each target CIMEP prediction value includes: Determining an engine fuel consumption prediction value corresponding to each target CIMEP prediction value; The minimum engine fuel consumption rate prediction value is selected from the engine fuel consumption rate prediction values ​​corresponding to each target CIMEP prediction value as the target engine fuel consumption rate prediction value.

6. A MAP prediction device for an engine, characterized in that: include: An acquisition module is used to obtain universal characteristic data of the engine under different working conditions; An acquisition module, used for preprocessing the universal characteristic data, inputting the preprocessed universal characteristic data into a preset neural network model, and outputting a plurality of engine fuel consumption rate prediction values ​​and a plurality of engine combustion stability CIMEP prediction values ​​respectively; a screening module, configured to screen out at least one target CIMEP predicted value whose CIMEP predicted value is less than or equal to a first preset threshold value from the plurality of engine combustion stability CIMEP predicted values, and to screen out a target engine fuel consumption predicted value that meets a preset condition from the engine fuel consumption predicted values ​​corresponding to each target CIMEP predicted value; The prediction module is used to determine the target VVT configuration parameters according to the predicted value of the target engine fuel consumption rate, and predict the MAP of the engine under different working conditions based on the target VVT configuration parameters.

7. The device according to claim 6, characterized in that Before inputting the preprocessed universal characteristic data into the preset neural network model, the obtaining module further includes: An acquisition unit, used for acquiring a universal characteristic public data set of the engine under historical operating conditions; A partitioning unit, configured to preprocess the universal characteristic public data set, and partition the preprocessed universal characteristic public data set into a training set and a validation set according to a preset partitioning ratio; A training unit is used to train a preset neural network using the training set based on a preset Bayesian algorithm to obtain an initial neural network model, and to verify the initial neural network model using the verification set, until the initial neural network model meets a preset standard, and then terminate the iterative training of the preset neural network to obtain the preset neural network model; otherwise, adjust the preset division ratio, and perform iterative training based on the adjusted new training set.

8. The device according to claim 7, characterized in that The training unit is specifically used for: Determine the model training function and evaluation function; According to the model training function and the evaluation function, based on the preset Bayesian algorithm, determining optimal model parameters within a preset parameter set; Based on the optimal model parameters, the preset neural network is trained using the training set to obtain the initial neural network model.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the MAP prediction method for the engine according to any one of claims 1 to 5.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the MAP prediction method of the engine as claimed in any one of claims 1 to 5.

Images