Daily flow forecast using artificial neural network (ANN) and Wavelet neural network (WNN) model, in high Andean basins of Peru

Yaranga Lázaro D.; Espinoza Varillas K.; Zapana Arpasi J. · Proceedings of the IAHR World Congress (2022) · DOI: 10.3850/iahr-39wc252171192022372

Resumen

The occurrence of hydrological phenomena is a constant every year in Peru and can cause serious flooding with economic losses. In this sense, hydrological forecasting is a tool that can provide substantial benefits for damage control. Therefore, the purpose of this research is to analyze the forecast of daily flows using artificial neural networks (ANN) and a hybrid model that combines wavelet and multiresolution ANN called wavelet neural network model (WNN). This research has as study area the Chicama and Huancane river basin. The methodology was to compare the flow forecasts generated by ANN and WRN. Likewise, several simulations were carried out with different scenarios, in order to identify the optimal structuring of the neural network, in an analysis based on efficiency indicators (RMSE, MAE, MARE y NSE) in the training and validation phase. The time series of daily flows used for the Chicama basin was from January 1, 1975 to December 6, 2016 and for the Huancane basin it was from December 1, 1975 to August 31, 2018. Subsequently, flow forecasts were made for different horizons. The results achieved showed a better performance in the WNN model compared to the ANN; determining that in a maximum forecast horizon of 15 days, the Huancane basin obtained an RMSE = 10.8 m3/s and NSE = 0.77, while in the Chicama basin an RMSE = 20.1 m3/s and NSE = 0.62.

Instituciones: Universidad Nacional Agraria La Molina