Effective Short-Term Continuous Data Prediction Using LSTM-NHITS

Abstract

Accurate short-term continuous data prediction is crucial for timely water quality assessment and pollution prevention. However, the nonlinear and temporally dependent nature of water quality data presents significant challenges for traditional forecasting models. To address these challenges, we propose an effective short-term continuous prediction model, LSTM-NHITS, which combines Long Short-Term Memory (LSTM) networks with Neural Hierarchical Interpolation for Time Series (NHITS). This model effectively captures multi-scale features and complex temporal dependencies, improving prediction accuracy. Experimental results from datasets collected at multiple monitoring stations in Zhanjiang City show that LSTM-NHITS outperforms traditional models across different short-term forecasting horizons (4-hour, 12-hour, and 1-day). By accurately modeling both long- and short-term dependencies, this approach ensures precise continuous water quality prediction, demonstrating its potential for real-time environmental monitoring and management.