养殖水体水质的神经网络预测模型研究

Study on neural network model for aquaculture water quality prediction

以池塘养殖水体常规水质指标作为训练样本,在分析传统水质预测模型的基础上,构建神经网络水质预测模型。运用改进的BP算法对在线监测的水质指标进行分析、分类和预测,确定水质指标与其影响因子间的非线性关系,研究养殖水体水质指数变化梯度和分布规律,同时对水质状况进行模糊判别,为养殖生产提供预警控制,并对不同情况下的输出结果做出了比较。结果表明:该网络具有较好的泛化能力,预测平均误差在3%以内,实现了水质指标的准确预测和判别,收敛速度快,具有较好的实用性和较高的预测精度,基本满足环境管理的需要。

A novel neural network model for water quality prediction is constructed with aquaculture water quality indexes in ponds as the training samples for the network after analyzing classical water quality prediction models. The water quality indexes monitored online are analyzed, classified and predicted with improved BP algorithm so as to confirm the nonlinear relationship between the indexes and their impact factors. The movement gradient and distribution regulation of water quality index in aquaculture water is studied while the fuzzy differentiation for the water condition is done in the meantime in order to provide pre-warning for next step in aquaculture. The results in different conditions are compared, which shows that the network has great generalization capability and high convergent speed with accurate prediction and differentiation of water quality indexes, the average forecast error was less than 3 percent. Simulation results prove that the proposed approach has high precision, good practicability and extensive applicability for engineering application. The arisen problems and solutions in the process of realizing the algorithm are also discussed.