Optimization of TCN-BiLSTM for dissolved oxygen prediction based on improved sparrow search algorithm.

Dissolved oxygen (DO) is a crucial indicator of water quality in river ecosystems, and its accurate prediction plays a vital role in the protection and sustainable utilization of these ecosystems. However, current DO prediction models often struggle with issues such as noise in the water quality data and insufficient feature extraction. To address these challenges, this paper proposes a dissolved oxygen prediction method based on an improved sparrow search algorithm optimized TCN- BiLSTM (SMI-TCN BiLSTM). Initially, the Savitzky-Golay (SG) filter is employed to denoise the water quality data, producing smoother and more consistent datasets. Next, the Maximum Information Coefficient (MIC) is applied to quantify the correlation between input features, enabling the identification and selection of key influencing factors. In addition, the traditional Temporal Convolutional Network (TCN) often fails to capture the dynamic fluctuations present in DO data, resulting in suboptimal prediction performance. To overcome this limitation, a Bi-directional Long Short-Term Memory (BiLSTM) network is integrated into the TCN framework, forming a TCN-BiLSTM prediction module. This module effectively captures both forward and backward temporal dependencies, improving the model's ability to track the dynamic trends in the data and enhancing its prediction accuracy. Finally, to address the stochastic nature of hyperparameter optimization in the TCN-BiLSTM module, we introduce an improved Sparrow Search Algorithm (ISSA). The ISSA is applied to optimize the hyperparameters of the TCN-BiLSTM model, thereby improving the overall prediction performance. To validate the proposed model, experiments are conducted on real datasets and compared with other water quality prediction models. The experimental results demonstrate that our method achieves the best prediction results.