Abstract:To solve the problem of improving the regression accuracy and model stability of the extreme learning machine(ELM), a new approach based on an improved M-estimation optimized double-parallel extreme learning machine is proposed in this study, namely robust double-parallel extreme learning machine(RDELM). Firstly, RD-ELM is constructed with a double parallel forward structure, thus the information can be received from both hidden layer neurons and input layer neurons. Secondly, we use an improved M-estimation to calculate output weights of neural network by iteratively reweighted Least-Squares Estimation(LSE), with weights assigned by the least absolute residual estimation of the samples. Finally, we establish a regression prediction model utilized to test the goodness of fit in a SinC function and verify the regression ability in eight benchmark regression problems. Then the proposed method is applied to an actual operational condition of a power plant. Experimental results show that the proposed method can efficiently process the influence of outliers and noise with strong anti-jamming ability. Compared with other methods, RD-ELM has superior performance that is stronger robustness and better generalization performance in many benchmark data and practical experiments.

A robust double-parallel extreme learning machine based on an improved M-estimation algorithm