模糊方法改进的反向传输神经网络预测南美白对虾养殖的水质

发布时间：2018-12-20 22:13

【摘要】：在南美白对虾高密度、规模化围塘养殖生产中,水质参数的监测、评价及预警是至关重要的。以杭州市某南美白对虾基地日常养殖水质为研究对象,选取温度、pH值、溶解氧、氧化还原电位等4项指标作为预测参数,建立拓扑结构为40-14-4的3层前馈反向传输(back propagation,BP)神经网络模型,即以连续10个时间单位的预测变量为输入层,隐含层节点数为14个,输出层变量为温度、pH值、溶解氧和氧化还原电位。为克服传统BP神经网络存在的收敛速度慢、易陷入震荡和泛化能力不强等缺点,采用模糊方法优化了自适应变步长BP神经网络算法,缩短了BP神经网络的训练时间,提高了网络收敛效率与稳定度。结果表明,以模糊方法改进的BP神经网络(FABPM)方法具有收敛速度快、预测精度高、稳定度好等特点,对选取的4项水质指标都给出了较好的训练与预测结果,预测的平均相对误差小于2.5%,从而为水产养殖水质预测与评价提供了一种新方法。
[Abstract]:The monitoring, evaluation and early warning of water quality parameters are very important in the production of high density and large scale ponds of Penaeus vannamei (Penaeus vannamei). The daily water quality of Penaeus vannamei (Penaeus vannamei) base in Hangzhou was studied. Four indexes, such as temperature, pH value, dissolved oxygen, redox potential and so on, were selected as prediction parameters. A three-layer feedforward and backward transmission (back propagation,BP) neural network model with topological structure of 40-14-4 is established. The input layer is a prediction variable of 10 consecutive time units, the number of hidden layer nodes is 14, the output layer variable is temperature, and the pH value is the output layer variable. Dissolved oxygen and redox potential. In order to overcome the disadvantages of the traditional BP neural network, such as slow convergence speed, easy to fall into turbulence and weak generalization ability, the adaptive variable step size BP neural network algorithm is optimized by fuzzy method, which shortens the training time of BP neural network. The convergence efficiency and stability of the network are improved. The results show that the improved BP neural network (FABPM) method based on fuzzy method has the advantages of fast convergence, high prediction accuracy and good stability. The four selected water quality indexes are well trained and predicted. The average relative error of prediction is less than 2.5, which provides a new method for the prediction and evaluation of aquaculture water quality.
【作者单位】：浙江大学城市学院信息与电气工程学院;
【基金】：浙江省重大科技专项(2012C03009-2) 杭州市科技发展计划项目(20140432B08)
【分类号】：S959;TP183

【参考文献】