Prediksi Parameter Kualitas Air Sungai Bengawan Solo Menggunakan Metode CNN-LSTM

Abstract

Water quality prediction is an essential component of sustainable river management, particularly for rivers that serve multiple functions such as domestic water supply, agriculture, and industry. The Bengawan Solo River, one of the longest and most important rivers in Indonesia, exhibits dynamic water quality conditions influenced by seasonal variation and human activities. This study aims to develop a predictive model for river water quality parameters using a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) approach based on multivariate time series data. Monthly water quality data collected from several monitoring stations along the Bengawan Solo River during the period 2021–2025 were used in this study. The observed parameters include temperature, pH, dissolved oxygen (DO), biochemical oxygen demand (BOD), chemical oxygen demand (COD), and total suspended solids (TSS). Data preprocessing involved missing value imputation, outlier handling, categorical encoding of monitoring stations, and Min–Max normalization. The CNN component was employed to extract local temporal features, while the LSTM layer captured long-term temporal dependencies. To reduce overfitting, regularization techniques such as dropout, L2 regularization, early stopping, and Bayesian hyperparameter optimization were applied. Model performance was evaluated using scale-independent metrics, namely Normalized Root Mean Squared Error (NRMSE) and Mean Absolute Scaled Error (MASE). The results demonstrate that the proposed CNN–LSTM model provides robust and reliable predictions across different data split scenarios, indicating its potential application as a decision-support tool for river water quality management.