摘要
针对发动机特征参数预测中的参数选择及预测模型建立的问题,提出一种基于粒子群优化核极限学习机(PSO-KELM)的发动机特征参数预测方法。根据能够反映发动机性能及技术状况变化趋势、不解体检测和抗干扰的原侧,选择排气温度作为表征发动机运行状况的特征参数,以发动机水温、油温、尺杆位移和转速作为关联特征量。通过核极限学习机(KELM)建立预测模型,采用粒子群算法对KELM的惩罚系数和核参数进行优化,以减少人为因素的影响。车辆在怠速和行驶工况下的特征参数预测结果表明,与基于粒子群优化的最小二乘支持向量机预测模型(PSO-LS-SVM)和基于网格搜索优化的核极限学习机预测模型(NS-KELM)相比,PSO-KELM的预测精度更高,为发动机特征参数预测提供了一条有效途径。
To achieve appropriate characteristic parameters and build optimized prediction model in the process of engine characteristic parameters prediction,an algorithm named Particle Swarm Optimization Kernel Extreme Learning Machine(PSO-KELM) is proposed.According to the principle of being able to reflect the transformation of the engine performance,un-disassembly examination and anti-jamming,exhaust gas temperature is chosen as the characteristic parameter representing the engine operating status.The water temperature,oil temperature,gear-lever shift and rotate speed are regard as the related variable.To reduce human factor influence,the particle swarm optimization algorithm is used to choose punishment coefficient and kernel coefficient in KELM model.Experiment results show that PSO-KELM reflects higher prediction accuracy than PSO-LS-SVM and NS-KELM.A effective approach is build for engine characteristic parameters prediction by PSO-KELM.
出处
《控制工程》
CSCD
北大核心
2014年第S1期28-32,共5页
Control Engineering of China
基金
军内科研项目
关键词
发动机
特征参数
核极限学习机
粒子群算法
engine
characteristic parameters
kernel extreme learning machine
particle swarm optimization
