Optimizing solar power plant site selection in Java-Bali using a hybrid FCM-SARIMA-NARNET framework

This study presents an integrated framework for optimizing solar power plant (SPP) site selection in the Java-Bali region, combining Fuzzy C Means (FCM) clustering with a hybrid Seasonal ARIMA and Nonlinear Autoregressive Neural Network (SARIMA-NARNET) forecasting model. The methodology addresses solar irradiance variability and grid constraints by grouping geographical zones and generating high accuracy solar forecasts. The hybrid SARIMA-NARNET model achieved superior predictive performance, enabling precise estimation of operational reserves through a Dynamic Confidence Level (DCL) approach. Results identify Cimanggis as the optimal location with a net system value of $23 per MWh, compared to $14.5/MWh for Godean and -$2.1/MWh for Purwosari. Implementing DCL at the P-94 percentile yielded annual energy savings of 191,978 MWh (17.22%), equivalent to approximately $172 million in system cost reductions. Solar PV demonstrated strong economic viability, with a levelized cost ranging from $59–68.3 per MWh—approximately 21% below the marginal cost of natural gas generation at optimal sites—and grid integration payback periods of 0.3–2.1 years. The study concludes that integrating spatial clustering, advanced forecasting, and dynamic reserve optimization significantly enhances the efficiency and reliability of solar PV integration. This framework provides a replicable decision-support tool for renewable energy planning in regions with similar grid and climatic characteristics.