The Research Of Prediction Algorithm Of Oil Spectral Data In The Running-in Diesel Engine

The running-in quality of diesel engine is directly related to the operation state and follow-up service life, so it must proceed to the running-in evaluation before put into operation.Predicting the running-in process through the data of oil spectral analysis, is one of the effective methods. The oil spectral data of diesel running-in process are often non-linear changes which the commonly used predictive models cannot adapt to, such as curve fitting, time series, Kalman filtering, the gray prediction, fuzzy models, resulting in low prediction accuracy and limiting the use of diesel engine running-in process. So the research of predict algorithm and model of diesel engine oil spectral data, not only has a certain theoretical significance, but also for the assessment of diesel engine running-in has important application value.According to the theory of BP neural network, RBF neural network and GA-RBF neural network, this paper set up the diesel engine running-in prediction model and provide a new method of running-in evaluation.In order to train and test the neural network with enough of the training sample and test sample, this paper apply with the spline fitting to fit the oil spectrum data. The predicted results of test sample shows that the established models of BP neural network, RBF neural network and GA-RBF neural network have high prediction accuracy.This paper applied the established neural network to predict the oil spectrum data of diesel engine running-in process, the results shows that the BP neural network and RBF neural network can be used in the oil spectrum data prediction during the diesel engines running-in process, but both have certain deficiencies:BP neural network algorithm have low accuracy and slow convergence speed; RBF neural network algorithm have fast convergence speed and low accuracy.In order to overcome the low accuracy of RBF algorithm due to the random selection of initialized weights and retain its advantages of fast convergence speed, this paper optimizes the RBF algorithm using the genetic algorithm, and establishes a GA-RBF neural network. The results show that the prediction accuracy of GA-RBF neural network can be attained to 85% above, and can predict short-term trends of oil spectrum data in diesel engine running-in process.The results of this paper provided a new method of the diesel engine running-in assessment, and established the foundation on the studying of oil spectrum data for long-term prediction model, have a certain theoretical significance and application value.