Study on photovoltaic power forecasting based on meteorological information and optimization of active/reactive power support capabilities for photovoltaic-storage distribution transformer areas

Against the backdrop of rapidly growing global photovoltaic (PV) power generation, its inherent intermittency and volatility challenge power grid stability. This paper presents a comprehensive model for PV power forecasting and optimizing active/reactive support capabilities of distributed PV-storage stations. First, a hybrid EMD-KPCA-LSTM model is developed to address PV output variability. It employs Empirical Mode Decomposition (EMD) for multi-scale feature extraction from meteorological data, Kernel Principal Component Analysis (KPCA) for dimensionality reduction, and Long Short-Term Memory (LSTM) networks for high-precision forecasting. Experimental results show significantly improved accuracy and robustness compared to single LSTM or EMD-LSTM models, with effective quantification of prediction uncertainties. Second, a nonlinear programming model based on robust optimization and Sequential Quadratic Programming (SQP) is proposed to optimize active/reactive support. By refining energy storage charging/discharging strategies, the model maximizes grid support while mitigating PV uncertainty via robust optimization and solving efficiently with SQP. Case studies validate that optimized PV-storage stations effectively dampen PV fluctuations, providing stable and substantial active/reactive support to enhance grid flexibility, stability, and renewable energy integration. This work offers critical theoretical and practical insights for developing smarter, greener, and more reliable modern power systems.