A Kind of Parameters Self-adjusting Extreme Learning Machine

Abstract

Extreme learning machine (ELM) is a kind of feed-forward single hidden layer neural network, whose input weights and thresholds of hidden layers are generated randomly. Because the output-weights of ELM are calculated by the least-square method, the ELM presents a high speed on training and testing. However, the random input-weights and thresholds of hidden layers are not the best parameters, which can not pledge the training goals of the ELM to achieve the global minimum. In order to obtain the optimal input-weights and bias of hidden layer, this paper proposes the self-adjusting extreme learning machine, called SA-ELM. Based on the idea of the ameliorated teaching learning based optimization, the input-weights and the bias of hidden layer of extreme learning machine are adjusted with "teaching phase" and "learning phase" to minimize the objective function values. The SA-ELM is applied to the eight benchmark functions to test its validity and feasibility. Compared with ELM and fast learning network, the SA-ELM owns good regression accuracy and generalization performance. Besides, the SA-ELM is applied to build the thermal efficiency model of a 300 MW pulverized coal furnace. The experiment results reveal that the proposed algorithm owns engineering practical application value.