为了提高光伏发电功率预测精度，建立了基于ICEEMDANDTW和ISMAWLSSVM的光伏发电功率超短期组合预测模型。首先，根据Pearson相关性分析，确定光辐照度、环境温度以及湿度为光伏发电功率的关键气象影响因素，继而使用改进的自适应白噪声完备集成经验模态分解(Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise，ICEEMDAN)对历史光伏功率和气象因素进行分解，降低其复杂度和随机波动性，并利用动态时间弯曲(Dynamic Time Warping，DTW)算法确定每个光伏功率子序列的输入特征向量。其次，对最小二乘支持向量机(Least Squares Support Vector Machine，LSSVM)在建模过程中的误差进行权重分配，得到加权最小二乘支持向量机(Weighted Least Squares Support Vector Machine，WLSSVM)，其解决了LSSVM模型鲁棒性低的缺陷。最后，通过改进黏菌算法(Improve Slime Mould Algorithm，ISMA)对WLSSVM进行参数优化，搭建ISMAWLSSVM预测模型，并在多种不同天气类型下进行光伏发电功率预测仿真实验。实验证明：相比EOSSAELM预测模型，该模型的RMSE在晴天、多云和雨天分别降低了57.4%、57.5%和52.5%。
In order to improve the prediction accuracy of photovoltaic power generation, an ultrashortterm combined prediction model of photovoltaic power generation based on ICEEMDANDTW and ISMAWLSSVM was proposed. Firstly, according to Pearson correlation analysis, light irradiance, ambient temperature and humidity were determined to be the key meteorological factors affecting photovoltaic power generation, then, an improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) was used to decompose the historical PV power and meteorological factors and reduce their complexity and random volatility, and the dynamic time warping (DTW) algorithm was used to determine the input eigenvectors of each PV power subsequence. Secondly, the weighted least squares support vector machine (WLSSVM) was obtained by assigning weights to the errors in the modeling process of least squares support vector machine (LSSVM), which solved the defect of low robustness of LSSVM model. Finally, the parameters of WLSSVM were optimized through the improved slime mould algorithm (ISMA), the ISMAWLSSVM prediction model was built, and the photovoltaic power generation prediction simulation experiment was conducted under various weather types. Experimental results show that compared with the EOSSAELM prediction model, this model reduces RMSE by 57.4%, 57.5% and 525% on sunny, cloudy and rainy days, respectively.