基于改进BP神经网络的厌氧发酵产气量预测模型  

Prediction modeling for gas production of anaerobic fermentation based on improved BP neural network

作　　者：花亚梅 赵贤林[1] 王效华[1] 滕昆辰 

机构地区：[1]南京农业大学工学院,南京210031 [2]南京绿鼎新能源有限公司,南京211800

出　　处：《环境工程学报》2016年第10期5951-5956,共6页Chinese Journal of Environmental Engineering

基　　金：江苏省自然科学基金(青年基金)资助项目(BK20130696);世界银行项目(A2-B10-CS-2009-001)

摘　　要：为准确预测大中型沼气工程日产气量,采用改进BP神经网络算法,引入动量和自适应调节学习率,根据厌氧发酵机理以及实际工程运行状况,建立以温度、TS浓度以及pH值作为输入层节点,沼气日产气量为输出层节点的预测模型。利用实验获取的150组数据作为模型的训练样本和测试样本,通过Matlab软件进行仿真,结果表明,改进的BP神经网络对沼气的日产气量具有良好的预测能力,建立的沼气日产气量预测模型不仅收敛速度快而且精度高,150组数据的平均预测准确率为84.02%。To predict the daily gas production of large- and medium-sized methane plants,a biogas prediction model was established,based on the mechanism of anaerobic fermentation and the actual operation status of the projects,with temperature,TS concentration,and pH values as the input layer nodes and methane gas production as the output layer node. In the prediction model,an improved BP neural network algorithm,momentum,and an adaptive learning rate were used. The 150 groups data were gathered from training,and the testing samples were obtained by experiment. The simulation results from the Matlab software showed that the improved BP neural network of the methane gas production model had a fast convergence speed and high accuracy. The average accuracy rate of prediction of 150 groups data was 84. 02% in the experiment.

关 键 词：沼气工程 厌氧发酵 动量 自适应学习率 产气量 

分 类 号：X705[环境科学与工程—环境工程]