A performance evaluation framework for deep peak shaving of the CFB boiler unit based on the DBN-LSSVM algorithm

Abstract

Under such a circumstance that the scale of renewable power connected into grids increases companied with more fluctuation, the flexibility and stability in power generation have been focus. Circulating fluidized bed (CFB) has unique merits in deep peak shaving, but its operation presents multi-influencing factors and multi-mode characteristics, which makes it very difficult to monitor the operation state. Toward this end, a novel performance evaluation framework has been proposed. The proposed framework contains two main parts: deep feature extraction conducted by deep belief networks (DBN), connecting with performance status classification by least square support vector machine (LSSVM). In this framework, massive operation data detected by sensors and reference status labels were entered into DBN for dimension reduction and feature extraction in a semi-supervised way. LSSVM finished the status classification based on these features. The final classification results are processed by DBN and LSSVM successively, which can not only make full use of the multidimensional parameters of CFB, but also avoid the influence of multimode of CFB. Besides, some comparations of the case study are conducted and analysed respectively to verify the efficiency and accuracy of the performance evaluation framework.