Time Series Prediction Method Based on E-CRBM

To solve the problems of delayed prediction results and large errors in one-dimensional time series prediction, a method based on Error-Continuous Restricted Boltzmann Machines (E-CRBM) is proposed this paper. This constructs deep conversion framework, which composed two E-CRBMs neural network (NN). Firstly, E-CRBM models original input sequence target are trained, respectively, to extract features sequences. Then NN model used connect transform features. Secondly, feature H1 extracted from test data through E-CRBM1, as obtain transformation H2. Finally, obtained by reverse reconstruction H2 E-CRBM2. The paper introduces residual hidden layer CRBM, increases robustness CRBM improves overall accuracy. classical (sunspot series) actual operation reciprocating compressor selected experiment. Compared with traditional method, verify effectiveness single-step multi-step prediction.