Photovoltaic power data preprocessing and law analysis based on sky state stratification and time delay correction

In response to the common problems of non-steady, sample heterogeneity and time misalignment in photovoltaic power prediction, this paper takes a photovoltaic power station in Hubei Province as the research object and builds a systematic data preprocessing and analysis framework that integrates physical constraints and statistical consistency. The framework includes data integrity verification, scattering score feature extraction, sky state stratification, time delay scanning and time alignment, and multi-scale feature construction, aiming to improve the physical rationality and statistical robustness of power prediction. The research results show that the system response under sunny conditions has significant linear characteristics, and the power output is stable and the fluctuation is small; under mixed cloud-day conditions, affected by the rapid evolution of the cloud field, the output power shows heteroscedastic and nonlinear characteristics, and the correlation peak decreases and the time delay is wide. The overall correlation analysis results show that the short-term response of the photovoltaic system is mainly controlled by meteorological factors such as cloud movement and local scattering enhancement, rather than acquisition link delay or equipment inertia. Research shows that the preprocessing strategy based on sky state stratification and time delay correction can significantly reduce the data structural deviation, and provide a unified data basis and theoretical support for the construction of a photovoltaic power prediction model with physical consistency, timing robustness and interpretable.