基于BP神经网络的汽车起重机油耗预测研究  

A Study on Fuel Consuming Pre-Estimation in Truck Cranes Based on BP Neural Network

作　　者：周道良[1] 蔡祖戈 刘彦辉[1] 吴斌[1] 杨国秀[1] 

机构地区：[1]徐工集团江苏徐州工程机械研究院

出　　处：《工程机械》2015年第5期23-27,7,共5页Construction Machinery and Equipment

摘　　要：以油门踏板操作(行程百分比)、卷扬手柄操作(开度百分比)及起重量作为输入参数,以燃油消耗量作为输出结果,基于BP神经网络建立针对用户操作的汽车起重机燃油消耗量预测模型;采集汽车起重机各工况下燃油消耗量试验数据,对比定发动机转速试验样本与不定发动机转速试验样本对BP神经网络模型预测精度的影响,并在神经网络训练样本前处理过程中引入滑动平均法;727个预测值与目标值的对比显示,模型相对误差平方和为0.000 31,线性回归值为0.999 49;仿真结果表明,定发动机转速试验样本及滑动平均法能够显著提高BP神经网络的泛化能力及仿真精度,从而提高汽车起重机燃油消耗量预测的精确度。In a study of presumed fuel consumption in auto-cranes a prediction model was constructed with given data input as throttle pedal operation(stroke ratio in%),winch handle operation(opening ratio in%),and craning capacity,and with fuel consumption as the output result,which was based on a BP neural network in view of fuel consumption in autocranes as operated by the owner itself.Resultant test data were collected under various working conditions to identify influence of test samples of specified RPM in comparison with those un-specified RPM on the predictive precision of the BP neural network model.Besides,a sliding average method was introduced in the pre-process of the neural network for exercise.Finally,727 pieces of predicted value in comparison with the target value showed that the model's relative error sum of squares was 0.000 31,and the linear average regressed value was 0.999 49.the simulation results show that test samples of engine RPM and the sliding average method can obviously increase generic capability of the BP neural network and the simulating precision,thus increasing a predictive precision for fuel consumption in auto-cranes.

关 键 词：BP神经网络 燃油消耗量 汽车起重机 泛化能力 

分 类 号：TH213.6[机械工程—机械制造及自动化] TP183[自动化与计算机技术—控制理论与控制工程]