基于改进RBF径向基神经网络的车载油耗软测量

Soft Measurement of Vehicle Fuel Consumption Based on Improved RBF Neural Network

针对现有的燃油消耗量测量方法存在成本较高、结构复杂且难以实现车载实时测量油耗的问题,提出基于改进RBF径向基神经网络的汽车油耗软测量方法。依据油耗产生机理,选取影响油耗且容易直接测量的参数作为测量模型的输入部分,采用油耗试验实测数据作为测量模型的训练样本和测试样本,并结合主元分析方法改进径向基神经网络的训练速度,自学习得出油耗软测量模型。利用Matlab仿真并将计算机仿真结果与油耗仪测量数据对比分析,平均误差在5.0%以内,验证了上述油耗测量方法的有效性,降低了车载实时油耗测量的成本和复杂程度。

The soft measurement method of vehicle fuel consumption,based on the radial basis neural networks,was put forward in view of the existing fuel consumption measurement's high cost,complicated structure,and difficulty to realize the on-board and real-time measurements of fuel consumption. This method based on the combustion mechanisms of engine,avoided the fuel consumption information difficult to be directly measured,and selected some parameters which affected fuel consumption and was easy to be directly measured as the input part of the measurement model. The measured data from fuel consumption test was treated as training samples and test samples of measurement model. Principal component analysis was used to improve the training speed and the measurement model can learn by itself. The computer simulation results in Matlab,compared with the measured data from fuel consumption test,showed that the measurement errors were less than 5. 0%,this soft measurement method verified valid and reduced the cost and complexity of the on-board and real-time fuel consumption measurement.