Statistical Machine Learning Model for Production Simulation of Power Systems with a High Proportion of Photovoltaics

With the accelerated global transition to renewable energy, photovoltaic power generation is playing an increasingly important role in modern power systems. However, the high volatility and intermittency of PV generation present significant challenges for power system simulation. Therefore, this chapter introduces a series of statistical machine learning models designed to simulate power system production in regions with a high proportion of photovoltaic generation. These methods are organized into two time scales: long-term and short-term. On the long-term scale, style transfer and bidirectional long short-term memory networks are integrated to build a generative adversarial network to generate year-round photovoltaic scenarios. The popular adaptive neighborhood clustering method is applied to extract representative PV output scenarios. On the short-term scale, an enhanced LSTM model integrating decomposition strategies and attention mechanisms is proposed for ultra-short-term power load forecasting. Additionally, the dynamic mask technique is introduced to forecast photovoltaic generation, optimizing the model's adaptability to dynamic system conditions. The proposed methods provide support for the efficient integration of photovoltaic power into the grid, helping optimize energy management and enhance grid stability under high renewable energy penetration.