基于深度学习的湖库藻类水华预测研究  

Research on algal bloom prediction based on deep learning

作　　者：姚俊杨 许继平[1] 王小艺[1] 黄振芳 

机构地区：[1]北京工商大学计算机与信息工程学院,北京100048 [2]北京市水文总站,北京100038

出　　处：《计算机与应用化学》2015年第10期1265-1268,共4页Computers and Applied Chemistry

基　　金：国家自然科学基金(51179002);北京市教育委员会市属高校科技创新能力提升计划项目(PXM2014-014213-000033);北京市属高等学校高层次人才引进与培养计划项目(CIT&TCD201404031);北京工商大学研究生科研创新基金项目

摘　　要：随着经济社会发展,水体富营养化程度日趋严重,藻类水华频发,水体生态安全受到严重威胁,直接影响城市发展建设和居民的正常生活。深入研究藻类水华形成机理,对藻类水华暴发这一非常规突发事件进行有效防治,对促进水环境保护具有重要意义。本文在对藻类水华形成机理深入分析的基础上,考虑常规藻类水华预测方法在样本选择、模型参数设置、预测精度等方面存在的问题,提出基于深度置信网络(Deep Belief Nets,DBNs)的藻类水华预测方法,通过网络结构与参数设计,实现了算法学习与训练。结果表明,该方法对数据信息利用率充分,藻类水华预测精度高,相对其他方法具有明显的优势,为湖库藻类水华预测提供一种新思路。The development of industry leads to an increasing degree of eutrophication and a rising frequency of algal bloom occurrence, which threats water ecological security seriously and affects the city development and the public lives directly. Therefore, an intensive research of algal bloom formation mechanism has great significance for promoting the water environment protection and technology progress. Considering the shortcomings in sample selection, parameter setting and prediction accuracy of conventional mathematical statistics based algal bloom prediction method, this paper puts forward an algal bloom prediction method based on Deep Believe Network （DBN） after deep analysis of algal bloom formation mechanism. This paper also realized algorithm training by network architecture design. The results show that the method makes full use of data information with high precision and distinct advantages compared with other methods, which provides a new idea for algal bloom prediction.

关 键 词：深度学习 水华预测 人工神经网络 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]