Short-term photovoltaic power prediction based on volatility analysis and uncertainty measurement

Accurately predicting photovoltaic output is an effective way to address the impact of uncertainty and volatility in photovoltaic power generation. Therefore, this article proposes a new photovoltaic power combination prediction model based on hybrid complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), permutation entropy optimization (PEO) (briefly as CPO), and improved whale optimization algorithm (IWOA) with bidirectional long short-term memory (BiLSTM) (namely, CPO-IWOA-BiLSTM), which explores the fluctuation characteristics, to achieve accurate prediction of short-term photovoltaic power. Using CEEMDAN technology to decompose photovoltaic power into different modes, and then using PEO to group and reconstruct the data of each mode, significantly characterizing the fluctuation process of photovoltaic power. The IWOA utilizes adaptive weights to improve convergence speed, optimizing the parameters of the BiLSTM model, and thereby improving prediction accuracy. Using kernel density estimation (KDE) to accurately predict the fluctuation range of photovoltaic output power and overcome the uncertainty of photovoltaic output power. The experimental results show that the proposed model can effectively achieve short-term photovoltaic power generation prediction and guide power grid scheduling.