关键词: 动力装置, 分形, 支持向量回归, 时间序列, 预测

Abstract:

The operational condition characteristics and the forecasting requirements for power devices were analyzed. A condition forecasting model based on the fractal theory and support vector regression method was presented. As a condition feature, vibration intensity was adopted to construct a time series. Minimum embedding dimension of the phase space was calculated by using phase space restructure technique, which was used to decide the input nodes of the support vector regression model.The forecasting ability and the validity of the model were studied and validated by means of application case and contrastive experiments.The research results demonstrate that relative mean error (RME) of the forecasting model is only 1.7881% for one step forecasting and 3.3983% for thirty steps forecasting; and the model can better meet the requirements of the condition forecasting for power devices.

Key words: power device, fractal, support vector regression, time series, forecasting