摘要：Aiming at the problem of difficult measurement in wastewater treatment operation, a soft-sensor modeling method for wastewater treatment process in a chemical plant using improved deep neural network was proposed. Firstly, a deep neural network multi-output adaptive soft-sensing model based on stack auto-encoder (SAE) was proposed, which was used only for synchronous online prediction of multiple target variables in the wastewater treatment process. Secondly, two algorithms of time difference modeling and variable importance projection are put in place to deal with the performance degradation problem and select auxiliary variables. Finally, the proposed model is checked by an actual case. Experimental results show that the proposed soft-sensor model not only performs well in single target prediction, but also has good performance in multi-output prediction.
关键词：Environmental pollution;Sewage from chemical plants;Deep neural network;Soft measurement;More output;Soft measurement;The time difference modeling
