Real-time water quality detection based on fluctuation feature analysis with the LSTM model

Abstract Signal analysis and anomaly detection for water pollution early warning systems are important necessary. In view of the nonlinear volatile characteristics quality time series, this paper proposes a new method based on fluctuation feature analysis. The has two steps. First, series data used to calculate residuals between observed value predicted with long short-term memory (LSTM) network. Second, dynamic features extracted by sliding window described Approximate Entropy (ApEn) which input model Isolation Forest. Compared traditional methods, results obtained proposed show better performance in distinguishing anomalies. can be applied real-time warning.