Intelligent Prediction of Photovoltaic Power Generation Capacity Based on Uncertainty and Mesoscale Characteristics

太阳能辐照存在可变性、气象因素复杂性等不确定性因素，光伏发电功率数据信息冗杂等问题，是当前光伏发电行业急需关注并解决的问题。基于此，本文提出了一种基于EWT分解、粒子群优化算法优化长短期记忆算法的混合预测模型，即EWT-PSOBOA-LSTM模型。借助粒子群优化算法(PSOBOA)对LSTM进行优化，运用自适应信号分解经验小波变换(EWT)将具有波动性的电力数据分解为不同的模态，通过介尺度特征分析数据特征，并利用信号分析中的频域分析进行分类预测。本文以某公司电力数据为例，通过与GRNN、ARIMA、BiLSTM等模型的对比，验证了所提出的混合模型具有良好的提升预测精度的效果。The uncertainty factors, such as the variability of solar irradiation and the complexity of meteorological factors, as well as the redundancy of photovoltaic power generation capacity data information, are the issues that the current photovoltaic power generation industry urgently needs to pay attention to and solve. Based on this, this paper proposes a hybrid prediction model based on EWT decomposition and particle swarm optimization algorithm to optimize the long short-term memory algorithm, namely the EWT-PSOBOA-LSTM model. The LSTM is optimized by means of the Particle Swarm Optimization (PSOBOA) algorithm. The adaptive signal decomposition Empirical Wavelet Transform (EWT) is used to decompose the fluctuating power data into different modes. The data features are analyzed through mesoscale characteristics, and the frequency domain analysis in signal analysis is utilized for classification and prediction. This paper takes the power data of a certain company as an example and, through comparison with models such as GRNN, ARIMA, and BiLSTM, verifies that the proposed hybrid model has a good effect on improving the prediction accuracy.