A hybrid model for Forecasting Biological Oxygen Demand using CEEMDAN-LSTM

Reliable and accurate forecasting of water quality parameters is essential for water quality management. Existing methods often rely on external factors and multiple water quality parameters. In this study, we demonstrate the applicability of a hybrid approach incorporating Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Long Short-Term Memory for forecasting Biological Oxygen Demand(BOD). The approach is minimalistic that solely utilizes historical data. The CEEMDAN decomposition is applied to the original time series data to generate a set of Intrinsic Mode Functions(IMFs) with varied frequencies and a residual, thus capturing the non-linear and non-stationary characteristics of the data. LSTM is then employed to forecast the IMFs and residuals produced by CEEMDAN. Finally, all the forecasted IMFs and residuals are aggregated to generate the final forecast. To conduct a thorough and rigorous analysis, the CEEMDAN-LSTM model is used to forecast BOD levels at three monitoring stations flowing along the river Ganga in Kanpur district of the state of Uttar Pradesh, India, considering one, two, and three-hour forecasting horizons. Experimental results demonstrate that using CEEMDAN combined with LSTM can effectively detect the complex non-linear patterns in the time series data, leading to more accurate outcomes than the alternative techniques.