A modified online sequential extreme learning machine for building circulation fluidized bed boiler's NOx emission model

Autor: Peifeng Niu, Yunpeng Ma, Shanshan Yan, Guoqiang Li
Rok vydání: 2018
Předmět:
Zdroj: Applied Mathematics and Computation. 334:214-226
ISSN: 0096-3003
Popis: In the last decade, the online sequential extreme learning machine (OS-ELM) has become an effective online modeling tool for the regression problem and time series prediction areas. However, the random initialization input-weights remain unchanged when the new large testing data arrive, which maybe reduce its training accuracy and generalization ability gradually. In this paper, based on the conventional OS-ELM, a kind of input data sample increment online sequential extreme learning machine is proposed, namely SIOS-ELM. As its name suggests, the sample increment is the actual error value between the present training input data sample and the new arriving input data sample. In SIOS-ELM, the parameters of hidden layer nodes (the input-weights and threshold values of hidden layer) are calculated in real time based on the sample increment by twice least square method when the new input data arrives one by one or chunk by chunk. In addition, the output weights are adjusted online as the OS-ELM. Compared with OS-ELM and its variants on benchmark problems, the proposed SIOS-ELM possesses better model accuracy and generalization ability. Additionally, the SIOS-ELM is applied to build the NOx emissions model of one 330 MW circulating fluidized bed boiler. The experiment result reveals that the SIOS-ELM is an effective online machine learning tool.
Databáze: OpenAIRE