Comparison of machine learning techniques for reservoir outflow forecasting

Abstract. Reservoirs play a key role in many human societies due to their capability manage water resources. In addition supply and hydropower production, ability retain control the flow makes them valuable asset for flood mitigation. This is function, since extreme events have increased last few decades as result of climate change, therefore, application mechanisms capable mitigating damage will be coming decades. Having good estimation outflow reservoir can an advantage management or early warning systems. When historical data are available, data-driven models been proven useful tool different hydrological applications. this sense, study analyzes efficiency machine learning techniques predict outflow, namely multivariate linear regression (MLR) three artificial neural networks: multilayer perceptron (MLP), nonlinear autoregressive exogenous (NARX) long short-term memory (LSTM). These were applied forecast eight reservoirs characteristics located Miño River (northwest Spain). general, results obtained showed that proposed provided reservoirs, improving with classical approaches such consider equal previous day. Among analyzed, NARX approach was option best estimations on average.