Abstract:
In order to improve prediction accuracy of gas concentration with noise, a back-propagation artificial neural network(BP-ANN) prediction model based on independent component analysis(ICA) and k-nearest neighbor(k-NN) was proposed. Firstly, training samples are got by use of sliding time window algorithm, ICA is used to estimate independent component(IC) in the training samples, and training set is reconstructed with the IC which does not contain noise. Then, k-NN is used to reduce size of the training set and mixed distance measure function is introduced to reduce computation complexity of the training. The experimental results show that the prediction model effectively reduces prediction error and training time than traditional BP-ANN model.