基于GA-BP神经网络的池塘养殖水温短期预测系统

为解决传统的水温小样本非实时预测方法预测精度低、鲁棒性差等问题,基于物联网实时数据,提出了遗传算法(GA)优化BP神经网络的池塘养殖水温短期预测方法,并在此基础上设计开发了池塘养殖水温预测系统,首先采用主成分分析法筛选出影响池塘水温的关键影响因子,减少输入元素;然后使用遗传算法对初始权重和阈值进行优化,获取最优参数并构建了基于BP神经网络的水温预测模型;最后采用Java语言开发了基于B/S体系结构的预测系统。该系统在江苏省宜兴市河蟹养殖池塘进行了预测验证。结果表明:该系统在短期的水温预测中具有准确的预测效果,与传统的BP神经网络算法相比,研究内容评价指标平均绝对误差(MAE)、平均绝对百分误差(MAPE)和误差均方根(MSE)分别为0.196 8、0.007 9和0.059 2,均优于单一BP神经网络预测,可满足实际的养殖池塘水温管理需要。

Short-term Prediction System of Water Temperature in Pond Aquaculture Based on GA-BP Neural Network

The pond water temperature is one of the most important parameters which directly affect the feeding, growth, livability and reproduction of aquaculture animals. Thus it is significant to grasp the pond water temperature change for the healthy aquaculture. In order to solve the problems of low precision and poor robustness of traditional forecasting methods, a short-term prediction model of water temperature in aquaculture pond was proposed based on BP neural network optimized by genetic algorithm, and pond aquaculture water temperature prediction system was designed and developed. Firstly, the principal component analysis (PCA) was used to ensure the factors that influenced the water temperature in aquaculture pond. Secondly, the genetic algorithm and BP neural network were integrated to optimize initial weights and threshold. The method not only can get optimal parameter, but also can reduce the errors generated by random initialization. Thirdly, the short-term prediction system was developed by using Java language based on B/S architecture. Finally, the system was applied in Yixing City, Jiangsu Province. Results showed that the mean absolute error (MAE), mean absolute percentage error (MAPE) and root mean square error (RMSE) from GA-BP neural network method were 0.1968, 0.0079 and 0.0592, respectively. It was clear that GA-BP neural network was better than BP neural network algorithm. The research result met the practical needs of the pond water temperature management.