基于PLSR自适应深度信念网络的出水总磷预测

发布时间：2018-03-21 13:47

本文选题：出水总磷　切入点：PLSR　出处：《化工学报》2017年05期 　论文类型：期刊论文

【摘要】：针对污水处理过程出水总磷预测问题存在的强非线性、大时变等特征,提出了一种基于偏最小二乘回归自适应深度信念网络(partial least square regression adaptive deep belief network,PLSR-ADBN)的出水总磷预测方法。PLSR-ADBN是基于深度学习模型DBN的一种改进型建模方法。首先,将自适应学习率引入到DBN的无监督预训练(pre-training)阶段,来提高网络收敛速度。其次,利用PLSR方法取代传统DBN中基于梯度的逐层权值精调(fine-tuning)方法,来提高网络预测精度。同时,通过构造李雅普诺夫函数证明了PLSR-ADBN学习过程的收敛性。最后,将PLSR-ADBN用于实际污水处理过程出水总磷预测中。实验结果表明所提出的PLSR-ADBN收敛速度快且预测精度高,能够满足实际污水处理过程对出水总磷监测精度和运行效率的要求。
[Abstract]:In view of the characteristics of strong nonlinearity and large time-varying in the prediction of total phosphorus in effluent from sewage treatment process, A partial least square regression adaptive deep belief network (PLSR-ADBN) based on partial least squares regression adaptive depth belief network is proposed. PLSR-ADBN is an improved modeling method based on the depth learning model DBN. The adaptive learning rate is introduced into the unsupervised pre-training stage of DBN to improve the convergence speed of the network. Secondly, the PLSR method is used to replace the gradient-based fine-tuning method of the hierarchical weights in DBN to improve the accuracy of network prediction. The convergence of PLSR-ADBN learning process is proved by constructing Lyapunov function. Finally, PLSR-ADBN is used to predict the total phosphorus of effluent from the actual sewage treatment process. The experimental results show that the proposed PLSR-ADBN converges quickly and has high prediction accuracy. It can meet the requirement of monitoring precision and running efficiency of effluent total phosphorus in actual sewage treatment process.
【作者单位】：北京工业大学信息学部;计算智能与智能系统北京市重点实验室;
【基金】：国家杰出青年科学基金项目(61225016) 国家自然科学基金项目(61533002,61603009) 中国博士后科学基金项目(2015M570910) 朝阳区博士后科研基金项目(2015ZZ-6) 北京工业大学基础研究基金项目(002000514315501)~~
【分类号】：TP18;X832

【相似文献】