基于BP神经网络的柴油发动机特性建模

建立连续的发动机燃油特性和调速特性数学模型作为液压机械无级变速器虚拟试验平台的动力源。根据虚拟平台对不同特性区域的精度需求对柴油发动机不同特性区域的试验数据进行不同的密度选取、乱序和归一化处理,采用单隐层BP神经网络对试验数据进行训练,对比不同隐层节点数网络的训练误差和测试误差,选取误差最小的网络,求解出网络的数学表达式。通过该方法以ISLe310柴油发动机为例建立燃油特性和调速特性的连续数学模型,这两个简单的数学表达式准确反映了发动机万有特性和外特性,连续模型避免了虚拟试验中出现信号的突变和奇异点。通过和经典的最小二乘法拟合得到的最优特性模型进行对比,其具有更小的误差、更强的泛化能力,能够更好地反映柴油发动机的相关特性。

Modeling of Diesel Engine Characteristics Based on BP Neural Network

The of continuous mathematical model of engine fuel characteristics and speed regulation characteristics was established as the power source for the virtual test platform of hydro-mechanical continuously variable transmission.According to the accuracy requirements of different characteristic regions of the virtual platform,the test data of different characteristic regions of the diesel engine were selected with different densities,and the disordered and normalized processing was performed.The test data was trained by a single hidden layer BP neural network.Compare training error and test error of different hidden layer node number networks,and the network with the smallest error was selected to solve the mathematical expression of the network.This method used the ISLe310 diesel engine as an example to establish a continuous mathematical model of fuel characteristics and speed regulation characteristics.These two simple mathematical expressions accurately reflect the engine's universal and external characteristics,and the continuous model avoids the sudden change of signal and singularities in the virtual test.By comparing with the optimal characteristic model obtained by the classical least squares fitting method,it has smaller error and stronger generalization ability,and can better reflect the relevant characteristics of the diesel engine.