基于模糊信息粒化和长短期记忆网络的短期风速预测

针对风速点预测无法对预测结果进行风险评估、区间预测难以满足电网精细化要求,以及现有静态预测方法难以描述风速序列长期相关性的现象,提出一种基于模糊信息粒化(Fuzzy Information Granulation,FIG)和长短期记忆(Long Short-Term Memory,LSTM)网络的动态预测模型。该方法先对风速序列进行模糊信息粒化,提取出粒化后数据的最大值(区间上界)、最小值(区间下界)和平均值。其次采用ADAM算法优化的LSTM网络对各粒化数据进行动态建模,得到能描述风速波动性的区间预测结果和点预测结果。算列表明,所提动态模型的预测效果比其它基本模型的预测效果更好。

Short-term wind speed prediction based on fuzzy information#$NLgranulation and LSTM

In view of the fact that the deterministic wind speed forecast can not assess the risk of the prediction results and the long-term correlation of wind speed sequences can not be described by the existing static prediction methods, a prediction model based on fuzzy information granulation (FIG) and long short-term memory (LSTM) network is proposed. In this method, the wind speed sequence is firstly granulated by fuzzy information, and the maximum value (interval upper bound), average value and minimum value (interval lower bound) of the data after granulation are extracted. Secondly, using the LSTM optimized by ADAM algorithm, the granulated data are respectively dynamically modeled to obtain interval forecasting results and point forecasting results which can describe the wind speed volatility. Calculations show that the proposed dynamic model predicts better than other basic models.