Enhanced Cascade Schemes for Advancing Machine Learning-Based Prediction of Heating and Cooling Loads in Residential Buildings

The contemporary imperative of addressing the challenge of energy-efficient building design underscores the significance of employing advanced methodologies. Leveraging machine learning (ML) tools has emerged as a viable strategy to streamline the modeling phase, a pivotal step in the design process for such structures. Despite the potential benefits, the precision of ML methods may be compromised due to various factors. This research introduces two modifications to the cascade correction schemes applied to regression models, aiming to enhance the accuracy of the selected ML method. These enhancements are grounded in the concept of establishing a cascade of two ML algorithms utilizing rational fractions (direct approach). The modification involves an additional modeling step that focuses on elucidating the relationships among independent inputs for the second ML method. This is achieved through nonlinear expansion using the Wiener polynomial. Such an approach contributes to an augmented accuracy of the ML algorithm, thereby elevating the overall precision of the cascade. The modeling of these modifications was conducted using a real-world dataset, specifically addressing the prediction of heating and cooling loads in residential buildings. Optimal parameters for the proposed modifications were identified, showcasing a noteworthy enhancement in accuracy compared to existing methods.