Prediction of effluent from WWTPs using differential evolutionary extreme learning machines

Tools for predicting the plant process performances and forming the basis for controlling the process operation are essential for any wastwater treatment plant (WWTP). In this paper, extreme learning machine (ELM) with the ability of black-box modeling is considered. To further improve the performance of ELM, three intelligent algorithms are considered. In the proposed hybrid evolutionary ELM, intelligent algorithms (BDE, SaDE and TDE) are applied to optimize the network hidden neuron parameters and Moore-Penrose (MP) inverse is used to analytically determine the output weights. The proposed algorithms BDE-ELM SaDE-ELM and TDE-ELM are applied in the prediction of effluent from a biological WWTP. The simulation results show that hybrid evolutionary ELMs all performs better than basic ELM. And TDE is a superior method to improve the performance of ELM comparing with BDE and SaDE. Further more, TDE-ELM performs the best in the prediction accuracy among the three proposed methods.