基于粒子群的模糊神经网络养殖池塘溶氧预测研究  

Particle swarm based fuzzy neural network for studying the prediction of dissolved oxygen in aquaculture ponds

作　　者：赵景波 薛秉鑫 ZHAO Jingbo;XUE Bingxin(School of Information and Control Engineering,Qingdao University of Technology, Qingdao 266520,Shandong,China)

机构地区：[1]青岛理工大学信息与控制工程学院,山东青岛266520

出　　处：《渔业现代化》2018年第4期8-14,20,共8页Fishery Modernization

基　　金：国家自然科学基金项目(51475251);山东省自然科学基金(ZR2013FM014);山东省高等学校科技计划项目(J12LN37);青岛市科技计划项目(15-9-2-109-nsh)

摘　　要：溶氧是水产养殖中的一项重要指标,与水产品生长有着十分密切的关系。为准确预测养殖池塘的溶氧量,降低水产养殖风险,提出基于小波包分析和粒子群算法优化模糊神经网络的组合预测模型。首先使用小波包变换对采集的原始信号进行消噪处理,然后将处理后的逼近信号分为训练数据和测试数据,利用训练数据对模糊神经网络进行训练,并使用粒子群算法对网络参数进行优化,最后利用测试数据进行溶氧预测并检验预测模型的性能。通过对比试验,分别证明了粒子群算法和小波包变换的有效性:预测溶氧值时,基于小波包变换,粒子群算法与BP算法相比,误差指标均方根误差(RMSE)、平均相对误差均值(MAPE)和平均绝对误差(MAE)分别降低了22.75、3.97和22.86个百分点;基于粒子群算法,有小波包变换和无小波包变换相比,3项指标分别降低了16.82、3.36和16.65个百分点。研究表明:小波包分析和粒子群算法可提高预测精度,该组合模型可对溶氧进行有效预测。As an important indicator in aquaculture, dissolved oxygen has a very close relationship with the growth of aquatic products. In order to accurately predict dissolved oxygen content in aquaculture ponds and reduce aquaculture risk,a combined fuzzy neural network model for prediction based on wavelet packet analysis and particle swarm optimization is proposed. First, wavelet packet transform is used to the collected original signals. Then the processed approximation signal is divided into training data and test data. Training data is used to train fuzzy neural network, and particle swarm optimization is used to optimize network parameters. Finally,test data is used to predict the dissolved oxygen and test the performance of the prediction model. Through comparison, the effectiveness of particle swarai optimization and wavelet packet transform is proved respectively. When predicting dissolved oxygen,based on wavelet packet transform,compared with BP algorithm,the error indexes such as RMSE,MAPE and MAE of particle swarm optimization are decreased by 22. 75,3. 97 and 22. 86 percentage points respectively ； based on particle swarai optimization, compared with absence of wavelet packet transform，the three indexes are decreased by 16. 82,3. 36 and 16. 65 percentage points respectively for existence of wavelet packet transform. The study shows that wavelet packet analysis and particle swarai optimization can improve the prediction accuracy,and the combined model can effectively predict dissolved oxygen.

关 键 词：溶氧预测 模糊神经网络 粒子群算法 小波包分析 

分 类 号：S967.4[农业科学—水产养殖] TP183[农业科学—水产科学]