Abstract:In order to solve the serious non-linearity and time-varying problems of industrial process affected by its own structural characteristics and external factors, a real-time learning Gaussian process soft sensor modeling method based on the comprehensive similarity index of input and output is proposed. In this method, the sample data are normalized. Firstly, the similarity between input and output variables of samples is calculated by using the traditional similarity index based on distance and angle, and then the similarity is synthesized. Finally, the final relevant sample set is selected and the soft sensor model of Gaussian process regression is established. A real-time learning Gaussian process soft sensor model based on input-output similarity index is proposed to predict urban daily electricity consumption data. The results show that the proposed soft sensing modeling method can achieve high accuracy prediction and the prediction results have small errors. Therefore, it can be shown that this method has certain application reliability in electricity forecasting and can be widely used in power market forecasting and analysis.