神经网络在无人机电控活塞发动机试验中应用

Application of neural networks in the test for electronic-controlled gasoline engine of unmanned aerial vehicle

分析了无人机用电控活塞发动机试验特点以及试验中存在的难点,针对电控发动机高海拔标定试验中进气歧管压力(manifold air pressure,简称MAP)传感器数据的传统线性插值方法不能完全表述电控发动机非线性特性的缺陷,提出采用BP(back propagation)神经网络模型的解决方案.为避免目前应用神经网络方法中所存在的不足,通过采用原始数据分组方法进行网络训练误差的实时反馈和控制,较好地解决了神经网络训练过程中容易陷入"局部最优"和"过拟合"状态,并对BP神经网络预测结果给予了详细研究,训练误差和预测误差分析结果表明了该方法的可行性和计算结果的可信性.

The characters and difficulties in the test for electronic-controlled gasoline en gine of unmanned aerial vehicle （UAV） were analyzed in detail. For mending the traditional liner interpolation methods, which are not completely in conformity with nonlinear charac- teristics of electronic-controlled gasoline engine, a solution was applied in the predication for the manifold air pressure （MAP） of high-altitude calibration of electronic-controlled gasoline engine, by adopting back propagation （BP） neural networks. To avoid the limitation of neural networks in application at present stage, a grouping method of raw data was put forward to control the feedback of the training error in real-time; this method provided a good solution for the problem that the neural network training result easily leads to the situation of ＂local optimum＂ and ＂over-fitting＂. The prediction results based on BP neural networks have been studied thoroughly; and the results of the training error and the prediction error reveal that the method is feasible and the result is effective.