| The development of information technology in coal-fired power plants provides a convenient platform for the research on the data-driven modeling technique. The operation data have the characteristics of large volume and high degree, in addition, there are high correlation and coupling among the variables. The real thermal process also exhibits heavy nonlinear feature, besides, there are outliers existing in the original data because of the sensor failure and electromagnetic disturbance. All the characteristics mentioned above bring hard difficulties and challenges to construct an accurate model. Artificial intelligent techniques provide a new theory foundation for solving such problems. Based on the theory of partial least square (PLS) and least square support vector machine (LSSVM), modeling method using the operation data in power plant is studied in this paper. The main research work of this dissertation is summarized on the following aspects:The characteristics of operation data in power plant are analyzed, and three stages of data-driven modeling are summarized, i.e., data preparation, model construction and model validation. Meanwhile, data process methods such as dimension reduction, outliers detection, data normalization and variable selection are elaborated.PLS-LSSVM method is proposed to tackle the data multicollinearity, and process nonlinearity. PLS is used to extract the feature variables and eliminate the correlation between variables. Then the feature variables are taken as the inputs of the LSSVM model to describe the nonlinear relation and improve the prediction accuracy.Considered the reheat steam temperature system in thermal process, the influence of relevant variables are analyzed. Then the data process theory and PLS-LSSVM aforementioned are applied to predict the steam temperature. Comparisons between other different models are also made, and the results reveal that the proposed method has better approximation performance. |
PDF Full Text Request