基于自组织递归模糊神经网络的BOD软测量  

BOD soft-sensing based on self-organizing recurrent fuzzy neural network

作　　者：丁海旭 李文静 叶旭东 乔俊飞 DING Haixu;LI Wenjing;YE Xudong;Qiao Junfei(Faculty of Information Technology,Beijing University of Technology,Beijing 100124,Beijing,China;Beijing Key Laboratory of Computational Intelligence and Intelligent System,Beijing 100124,Beijing,China;Huludao Electric Power Bureau,State Grid Liaoning Electric Power Supply Company,Huludao 125000,Liaoning,China)

机构地区：[1]北京工业大学信息学部,北京市100124 [2]计算智能与智能系统北京市重点实验室,北京市100124 [3]国网辽宁省电力有限公司葫芦岛供电公司,辽宁省葫芦岛市125000

出　　处：《计算机与应用化学》2019年第4期331-336,共6页Computers and Applied Chemistry

基　　金：国家自然科学基金资助项目(61533002,61603009);北京市自然科学基金(4182007);北京工业大学日新人计划(2017-RX(1)-04)

摘　　要：生化需氧量是污水处理过程中评价水质的重要指标之一,神经网络软测量是解决其在线测量困难的主要方法。污水处理是一个动态的过程,而前馈神经网络由于缺乏动态性而难以保证对其的测量精度。本文提出了一种自组织递归模糊神经网络,建立了内部的反馈连接以增强网络动态性能,通过评估神经元的互信息关系和激活强度以增长或修剪规则层神经元,采用梯度下降学习算法进行参数更新,并结合自适应学习率以提高收敛精度。通过对实际污水厂数据的实验结果表明,本文提出的模型结构更紧凑,对出水生化需氧量的预测精度更高。Biochemical Oxygen Demand(BOD) is one of the most important indices for evaluating water quality in wastewater treatment, and soft-sensing based on neural networks is a primary method to solve the problem of on-line measurement. Wastewater treatment is a dynamic process, while feedforward neural network is difficult to guarantee measurement accuracy due to its lack of dynamics. This paper presents a Self-organizing Recurrent Fuzzy Neural Network, which establishes the internal feedback connection to enhance the dynamic performance. It adds or prunes neurons in the regular layer based on the mutual information and activation intensity of the neurons. Finally, it uses gradient descent learning algorithm to update parameters and combine adaptive learning rate to improve convergence accuracy. The experimental results of sewage treatment plant samples show that the model proposed in this paper is compact and can achieve high accuracy for the prediction of BOD.

关 键 词：生化需氧量 自组织递归模糊神经网络 互信息 自适应学习率 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]