Kernel Based Method For Online Gas Flow Prediction Intervals In Metallurgical Enterprise

In the field of time series analysis and prediction, prediction interval model is getting more and more attention because it can better reflect the dynamic characteristics of the changes in the data compared to the ordinary single-valued prediction model. And in time series forecasting, kernel-based learning method is becoming a hot field in machine learning and pattern recognition since its model is easy to be determined and the algorithm is simple and fast, and have been successfully applied in the gas energy data regression prediction by the scholars.As the traditional interval prediction methods took too long time and got low accuracy when applied to some industrial problems, a double Gaussian kernel based online method is proposed in this paper. The method can describe the dynamics of nonlinear systems well by the collaboration of the two kernels; meanwhile, it can largely simplify the calculation of the Jacobian matrix into a class of kernel computation so as to reduce the computing cost for the practical demand. To determine the hyper-parameters of the proposed model, a conjugate gradient algorithm that makes the prediction error close to the effective noise of the sample data is then derived for high prediction accuracy.To verify the effectiveness of the proposed method, the real gas flow data coming from the energy center of a metallurgical enterprise is employed. In the experiments, four indicators: predicted range coverage, the average interval size, comprehensive evaluation index and interval time-consuming are adopted to evaluate the quality of the prediction interval compared to some other algorithms. The results indicate that the proposed method exhibits high accuracy and reliability with a low computing cost, meets the requirements of real-time and accuracy on the gas flow online forecasting, and the accuracy of the prediction model is greatly improved by the optimization of the hyper parameters.