基于相空间重构和遗传优化SVR的机械设备状态趋势预测

                        Aiming at the nonlinear and non-stationary characteristics of vibration signal sequence of mechanical equipments, a condition trend prediction method for mechanical equipments is proposed based on phase space reconstruction and genetic optimization support vector regression (SVR). First of all, a one-dimensional time series of vibration signals are transformed into a matrix by use of phase space reconstruction technique, and its features are selected adaptively. The phase points are imported to SVR model as input features and the SVR predictor is trained. Then, adaptive genetic algorithm is applied to optimize the penalty factor C, non-sensitive factor and Gaussian kernel width synchronously. The best model parameters are obtained automatically. Finally, the SVR prediction model is constructed and is applied to vibration signal prediction of a machine unit. The experimental results show that whether for single-step or 24-step prediction, the prediction accuracy of the proposed genetic optimization SVR is higher than that of the conventional SVR, indicating that the proposed method has a good ability for prediction of the condition trend of the mechanical equipments.