Assessing Methodological Validity and Variable Considerations for Forecasting Renewable Energy Load Requirements: A Review

This paper presents a systematic literature review conducted to investigate the validity of the adopted methodologies and variable considerations of forecasting renewable energy load requirements. A total of 213 studies from renowned scientific databases, including IEEE, Springer, Elsevier, ACM, and Taylor & Francis, were initially considered, resulting in a final selection of 23 primary studies using the inclusion and exclusion criterion. The review focused on analyzing the internal and external validity threats associated with the methods employed in these studies. Findings revealed several critical gaps in the methodology of the reviewed studies. Studies failed to address the bias-variance tradeoff, which is crucial for ensuring accurate predictions. A significant number of studies ignored parameter optimization in deep learning models, potentially impacting the reliability and precision of the load requirement forecasts. Studies did not conduct tests for multicollinearity of predictive attributes, which may lead to misleading results and inflated model performance. Few studies incorporated environmental and socioeconomic attributes in their datasets, which could have significant implications for energy consumption. Observations from this study highlight the importance of considering these gaps in renewable energy load requirement forecasting to enhance the accuracy and applicability of the resulting models. It emphasizes the significance of properly accounting for bias-variance tradeoffs, optimizing model parameters, and conducting tests for multicollinearity. The inclusion of environmental and socioeconomic variables in training sets is encouraged to capture the intricate relationships between these factors and energy consumption accurately. The study contributes to the improvement of renewable energy load requirement forecasting methodologies by promoting robust practices and considering comprehensive sets of variables.