摘要
利用人工神经网络对实验室中短程硝化过程进行仿真模拟,采用误差反向传播算法,并结合自适应学习率,在MATLAB语言环境下建立了进水NH4+-N﹑DO﹑温度以及外加碳源与出水NH4+-N和NO2--N之间的非线性映射函数关系,确立了相关的动态模型.结合最优化网络模型运行参数,对样本进行仿真学习,仿真输出值与实际值的拟合程度相当高,最大误差仅有13.8955%.通过权重分析,探究了各输入因素与输出结果之间的价值贡献关系,进水NH4+-N和温度对短程硝化过程表现出较大的影响.
The feasibility of dynamic simulation of shortcut nitrification process based on artificial neural network (ANN) was studied. With Back-Propagation algorithm and the adaptive study rate, a dynamic simulation model was established by MATLAB software, which could reflect the nonlinear function relationship between NH4^+-N, DO, temperature, external carbon source of influent and NH4^+-N and NO2-N of effluent. The numerical outputs and the experimental values matched well, with a highest error of 13.8955%. The value contribution relationships between each input factor and output results were investigated by weighted average analysis, which indicated that NH4+-N and temperature had tremendous influence on the shortcut nitrification process. The results suggested that the ANN could reflect the nonlinear function between influent and effluent parameters, and was suitable for the dynamic monitoring of the shortcut nitrification bio-process for wastewater.
出处
《中国环境科学》
EI
CAS
CSCD
北大核心
2008年第8期694-698,共5页
China Environmental Science
基金
国家“973”项目(2005CB724203)
国家自然科学基金资助项目(50478053)
关键词
人工神经网络
误差反向传播算法
自适应学习率
短程硝化
artificial neural network: back propagation algorithm
the adaptive study rate
shortcut nitrification