基于RBF神经网络和小波分析的水质异常检测方法

快速准确地检测出自然或人为事件下的水质异常,对保护水环境和保障公众健康具有重要意义。针对水质背景数据波动大时异常检测性能不理想的问题,提出一种基于RBF神经网络和小波分析的水质异常检测算法。通过引入RBF神经网络预测水质,对预测值和真实值相比较得到的残差序列进行滑动窗口小波去噪,对各时刻偏离原点的距离与特定阈值比较后判定水质是否异常。以某水源水库在线监测氨氮值为研究对象验证算法,实验结果表明:与时间序列增量方法比较,算法具有更高的异常检测率和较低的误报率。

Water quality anomaly detection method based on RBF neural network and wavelet analysis

Fast and accurate detection of nature or artificial water contamination events plays a vital role in protecting the water environment and public health.To solve the problem that conventional water quality anomaly detection algorithms have bad performance under high background noise,an algorithm based on RBF neural network and wavelet analysis is proposed.By introducing RBF neural network to predict future water quality values using previous observations of the water quality,wavelet de-noising residual time series is obtained by comparing prediction value with real value,then using wavelet de-noising in sliding windows,if the new residual at one time is greater than specific threshold/baseline,water quality is claimed abnormal.Using the online monitoring ammonia nitrogen value as the research object,it proves that the proposed algorithm has lower false alarm rate(FAR)and higher probability of detection(PD)comparing with time series increments.