Assessing Distributed Solar Power Generation Potential under Multi-GCMs: A Factorial-Analysis-Based Random Forest Method

Abstract

The development of renewable energy is important for climate change mitigation and socioeconomic sustainability, and the prediction of renewable energy potential (e.g., solar) under the consideration of climate change impact is challenged. In this study, a factorial-analysis-based random forest (FARF) method is developed for the distributed solar power generation (DSPG) predication under multiple global climate models (GCMs). FARF has advantages in (i) downscaling large-scale climate variables to local scales, (ii) avoiding the problem of overfitting in traditional models; and (iii) reflecting the main and interactive effects of climate variables on solar radiation intensity (SRI). Then, the FARF method is applied to the Jing-Jin-Ji region of China to predict the DSPG potential under three GCMs and two emission scenarios (RCP4.5 and 8.5). Multiple validation coefficients prove that the FARF method is effective and feasible. Major findings are as follows: (i) during 2021−2100, the regional SRI would increase under all GCMs and RCPs, and the southern region is obviously higher than the northern region; (ii) the main impact factors are temperature (contribution >51%) and humidity (contribution >28%), and the interactive effects of multiple factors are insignificant; (iii) the regional DSGP would continuously rise and its contribution to electricity consumption would continue to increase; and (iv) under all GCMs, SRI and DSPG under RCP8.5 would be higher than those under RCP4.5. The findings can help decision makers to use the desired strategies for promoting renewable energy utilization and energy system sustainable development.

Development of a random-forest-copula-factorial analysis (RFCFA) method for predicting propagation between meteorological and hydrological drought

“…RF is an ensemble technique that utilizes multiple decision trees trained through bootstrap aggregating [51]. RF offers the advantage of generating reasonable predictions without requiring hyper-parameter tuning and mitigating overfitting issues commonly observed in decision trees [52][53][54]. To ensure the universality of RF models, the historical datasets, including observed streamflow datasets of global runoff data center (GRDC), GCMs datasets from Coupled Model Intercomparison Project Phase 6 (CMIP6) and, Inter-Sectoral Impact Model Intercomparison Project (ISIMIP) are divided into two subsets by random sampling: The subset containing 70% of the data is used to calibrate the model, and the subset containing the remaining 30% data is used for validation.…”