Development of Water Level Prediction Improvement Method Using Multivariate Time Series Data by GRU Model

The methods for improving the accuracy of water level prediction were proposed in this study by selecting Gated Recurrent Unit (GRU) model, which is effective multivariate learning at Paldang Bridge station Han River, South Korea, where fluctuates seasonally. hydrological data (i.e., and flow rate) entered into GRU model; provided Water Resources Management Information System (WAMIS), meteorological Seoul Meteorological Observatory Yangpyeong through Korea Administration. Correlation analysis was used to select training data. Important input affecting daily (DWL) rate (DFR), vapor pressure (DVP), dew point temperature (DDPT), 1 h max precipitation (1HP), as prediction. However, DWL did not improve even if from a single observatory far Therefore, study, predictive that effectively utilize each presented. First, it method arithmetically averaging two observatories using model. Second, use weighted distances point. improved results obtained with some correlation between located used.