GMM聚类和ICEEMDAN-IBWO-BiLSTM短期 光伏发电功率预测

打开文本图片集

中图分类号：TB9;TM615 文献标志码：A 文章编号：1674-5124（2026）02-0175-10

Abstract： Accurate PV power prediction can effctively help the power scheduling department to formulate effective scheduling plans forthe power system. Aiming at the problems of poor adaptability and insufficient accuracy of a single prediction model， this paper proposes a short-term PV power prediction model based on the combination of Gaussian mixture clustering （GMM） and ICEEMDAN-IBWO-BiLSTM.First， the input features are screened by Pearson correlation analysis，and the historical PV power data are segmented using GMM to select similar days for testing. Secondly，the historical data of PV power are decomposed by using the improved complete ensemble empirical mode decomposition with adaptive noise （ICEEMDAN） and the modal components are reconstructed by using the arrangement entropy. And then the IBWO-BiLSTM model is used to predict the reconstructed subsequence，and the subsequence prediction results are superimposed to obtain the PV power prediction value.Finaly，the model is validated with the real data of a PV power plant as an example， and the results show that under sunny，cloudy and rainy weather，compared with other comparative models， the RMSE of the proposed model is reduced on average by 56.30% 45.40% and 37.95% ，the MAE is reduced on average by 57.52% 45.62% and 31.99% ，and the R2 is improved on average by 1.55% ， 4.72% and 5.64% ， and AIC was reduced on average by 36.39% ， 21.42% and 22.89% ，which verified the validity and superiority of the model.

Keywords： short-term photovoltaic power prediction; Gaussian mixture model; ICEEMDAN; beluga whale optimization; BiLSTM

0 引言

随着全球能源短缺和环境污染等问题日益突出，可再生能源已成为全球能源结构中越来越重要的一部分。（剩余13104字）