Ensemble forecasting of ERCOT efficiency and residual price variability using GFDL-CM4 climate projections

This paper investigates the climate-induced reliability issues within the Texas ERCOT grid by developing a multi-phase ensemble learning forecasting infrastructure. Meteorological variable forecasts from the CMIP6 GFDL-CM4 climate model were synthesized to project operational efficiency and resulting electricity price fluctuations from 2025 to 2050. Random forest regression and a hybrid SARIMA-XGBoost structure were used to estimate power plant generation and corresponding fuel consumption across seven major energy sources, achieving tolerance accuracies up to 93.7%. Renewables were projected with the most severe efficiency deficits, with wind and solar averaging 8.14% and 12.44% per month, respectively, over the forecasting range. A supporting Gaussian process model directly trained on the projected efficiencies and estimated long-term losses of $29.8 billion for residential consumers and $31.4 billion for retail energy providers under moderate conditions. Risk analysis identified key vulnerable climate zones and dominant predictive weather variables, namely heat-related metrics. Mitigation recommendations were developed based on model outputs, including the deployment of resilient technologies (e.g., anti-icing systems, perovskite PVs, etc.), integrated energy storage, and time-of-use pricing mechanisms. This paper is the first of a multi-part series, with future work extending current data for outage modeling and nonlinear optimization of resilience investment.