Quantum LSTM-Based Deep Learning and Hybrid Hydrodynamic Modeling of Mahanadi River Network

Real-time data-driven daily hydrodynamic variables such as rain gauge, water level, and water discharge prediction are vitally important to manage flood situations. However, it is difficult to predict these time-series data precisely due to intrinsically dynamic behavior and complex meteorological factors of weather. Developing hybrid models, ANN, and quantum LSTM-based deep learning models for forecasting have been gaining traction in recent years, yet conventional models fall short in assimilating the information from features. Therefore, to improve the accuracy, the random shock and signal decomposition algorithm is paired with the individual model to construct “hybrid” model. For solving this major problem, a hybrid ARIMA-ARNN model is developed through the training datasets of quantitative gauge and water level with prediction testing datasets of the four gauging stations, namely Bamnidhi, Basantpur, Kesinga, and Kantamal under study. These four gauging stations are selected using principal component analysis (PCA) and different methods of cluster analysis, namely K-means, Partitioning Around Medoids (PAM), and Hierarchical Clustering (HC) Partitioning. Using ANN model, the impact of rain gauge and water level on water discharge of these rain gauge (RG) stations are studied. The quantum LSTM-based deep learning model is developed for out-of-sample forecasting for 30 days ahead of three hydrodynamic datasets, namely gauge, water level, and discharge four the above four gauging stations with higher forecast accuracy and better constancy in the results. It is observed that the proposed models outperform other high-tech models with high accuracy in terms of all metrics on all four datasets, respectively. The hybrid ARIMA-ARNN model that is presented to forecast gauge and water level, ANN model to predict discharge in terms of gauge and water level and advanced quantum LSTM-based deep learning model for out-of-sample forecasting of gauge, water level, and discharge provide attractive value for daily hydrodynamic variable prediction of the flood management of Mahanadi River network in Odisha.