Selection of Most Relevant Input Parameters Using WEKA for Artificial Neural Network Based Oar-Based Water Turbine Harvester Hydrokinetic Efficiency Prediction Models

Abstract

Accurate hydrokinetic efficiency prediction plays a crucial role in renewable energy research, particularly in assessing the effectiveness of small water turbine harvesters. Various models, including both conventional and Artificial Neural Network (ANN) approaches are being utilized for hydrokinetic efficiency prediction. External factors and geographical variables significantly influences these predictions, underscoring the importance of identifying relevant variables for precise forecasts. To address this, the Waikato Environment for Knowledge Analysis (WEKA) software analyzed 25 diverse water sources with distinct environmental factors to pinpoint influential input parameters for ANN-based hydrokinetic efficiency prediction. Key parameters identified included turbine hydraulic power, shaft mechanical power, turbine efficiency, the ratio of blade length to blade width, blade axle length, and blade inclination angle. Three ANN models (ANN-1, ANN-2, and ANN-3) were developed, with maximum Mean Absolute Percentage Errors (MAPE) of 19.36%, 11.29%, and 8.31%, respectively. Impressively, the ANN-3 model, incorporating specific input variables, demonstrated a 12.67% enhanced prediction accuracy compared to ANN-1 and ANN-2. WEKA identified blade width, blade axle length, blade inclination angle, and turbine efficiency as the most relevant input variables, culminating in a robust hydrokinetic efficiency prediction of 29.6%.