DEVELOPMENT OF A LONG SHORT-TERM MEMORY PREDICTIVE MODEL FOR SMALL HYDRO POWER TURBINE LOCATION

Abstract

Among the many advantages of water, is its use for power generation and since a dam is expensive to construct, the run-of-river storage is considered for predicting suitable location of Small Hydro Power turbines installation. There is also the challenge of non-availability of hydrological data for some of the rivers in Nigeria for national planning and statistics. This research work applied one of the deep learning algorithms; Long Short-Term Memory (LSTM) networks to develop a predictive model from the historical data of the Benin-Owena River Basin Development Authority. The predictive model was implemented as a web-based application with Hypertext Processor and Laravel as the framework with Hypertext Markup Language, Cascading Style Sheet, Bootstrap, JQuery, and JavaScript for the front end side while Python 3 with Tensorflow as the backend engine alongside Keras packages. The LSTM predictive model was trained with seventeen river datasets and the results indicated appropriate turbines that can be installed in each river from the prediction and power generation estimate. The LSTM predictive model used a head of 16.95m, discharge of 22.10 m3/s from for Ethiope River and resulted in power output of 32.92kw.Also, the Root Mean Square Error used for performance evaluation gave a value of 17.29 for Ethiope River water height and 3.61 for discharge. The overall energy delivered was high and the recommended turbines were crossflow, propeller or kaplan. Therefore, the results showed that LSTM model resulted in a higher power output generation.