Optimal Resource Planning Using Homer PRO with Monthly Predictions from TCN-based Bidirectional-GRU-LSTM Models

Abstract

This paper introduces an innovative approach for optimizing resource allocation in hybrid energy systems by utilizing HOMER Pro, which incorporates monthly load forecasts produced by a Temporal Convolutional Network (TCN)-based Bidirectional-GRU-LSTM (BiGRU-LSTM) model. The proposed technique seeks to overcome the shortcomings of traditional load forecasting methods by employing deep learning strategies to effectively capture both local and long-term dependencies within time-series data. The forecasted load serves as an input for HOMER simulations, facilitating a comprehensive analysis of costs and performance for renewable energy systems. Significant challenges have been identified, such as the static nature of HOMER’s load assumptions and the potential for misrepresentation in renewable energy penetration during specific periods. Our findings reveal that although the TCN-BiGRU-LSTM model greatly enhances the accuracy of load forecasting, discrepancies in HOMER’s simulations may arise when relying on monthly averaged data, particularly in scenarios with high renewable penetration. The work emphasizes the necessity for more sophisticated simulation frameworks that take into account real-time variations in load and generation data. This research offers important insights into the improvement of hybrid energy system design and optimization through the application of advanced forecasting techniques.