Exploration and Prediction of Lake Water Levels Based on the MOACO Algorithm and CNN-LSTM Model

This paper presents a study of lake level detection and prediction based on the MOACO algorithm and CNN-LSTM model. The study begins by considering each water level as a node and comprehensively accounting for temporal and regional factors to calculate the directed influences between water levels as weights, thereby establishing a directed influence network. Subsequently, the study utilizes the Multi-Objective Ant Colony Optimization (MOACO) algorithm to analyze the influence network. It computes the degree centrality of water levels in the network and further evaluates the influencing factors of water levels using the PageRank-corrected eigenvector centrality. It is observed that precipitation and evaporation have the most significant impact, and the power-law distribution indicates that a minority of water levels significantly influence the majority. Furthermore, the study meticulously constructs a CNN-LSTM model to analyze the time series data of the water levels of five major lakes for precise control. The analysis reveals strong correlations between certain lakes, suggesting shared hydrological characteristics. The training process involves handling missing data, normalization, and setting targets for LSTM to predict the best monthly water levels based on 100 training iterations. The results demonstrate the high predictive accuracy of the algorithm designed in this study.