Enhancing typical Meteorological Year generation for diverse energy systems: A hybrid Sandia-machine learning approach

Accurate performance assessment of energy systems heavily relies on Typical Meteorological Year (TMY) data. The Sandia method, commonly used for TMY generation, is limited by default weighting criteria for meteorological parameters, restricting its suitability for diverse energy system analyses. In response, this study presents a novel framework for generating TMY files customized to various energy systems. Utilizing three tree-based algorithms (CART, RF, XGBoost) and interpretable machine learning techniques, the framework quantifies and personalizes weighting schemes. Validation of the method's applicability is conducted using long-term historical weather data from Beijing and Lhasa, encompassing three distinct energy systems (a full air conditioning building system and two renewable energy systems). Results indicate that the new TMY generation method excels over the original Sandia method for building and photovoltaic systems but encounters limitations with wind power system. Additionally, incorporating meteorological parameters highly relevant to specific energy systems and comprehensively considering their seasonality will contribute to the development of more representative TMY data. The proposed method facilitates precise foundational climate data acquisition, enabling more accurate energy performance analysis and decision-making.