基于BKA-BiLSTM-KD模型的超短期光伏发电功率预测OA
Ultra-short-term Photovoltaic Power Prediction Based on BKA-BiLSTM-KD Model
为提高超短期光伏发电功率的预测精度,缩短预测时间,提出了一种基于 BKA-BiLSTM-KD的超短期光伏发电功率预测方法.首先,采用 K-近邻算法(KNN)清洗异常数据以提升样本质量,并基于随机森林(RF)进行特征选择与组合增强;其次,提出黑翅鸢优化算法(BKA)优化双向长短期记忆网络(BiL-STM)预测模型,并对该预测模型进行知识蒸馏(KD),得到计算时间更短的学生模型;最后,以第 17 届"中国电机工程学会杯"比赛数据集和华北某光伏场站实际功率为算例样本,分别采用 BiLSTM、BKA-BiL-STM、WOA-BiLSTM 和 BKA-BiLSTM-KD 共 4 种方法 5 种模型进行预测.预测试结果表明,与 BKA-BiLSTM 模型相比,所提方法的决定系数R2 提升了 2.7%以上,预测耗时缩短超过了 67.8%,能够为光伏场站生产规划和电网调度决策提供高效、精准的数据支撑.
To improve the accuracy and shorten the time consumption of ultra-short-term photovol-taic(PV)power forecasting,a method based on BKA-BiLSTM-KD is proposed.First,the K-Nearest Neighbors(KNN)algorithm is employed to clean abnormal data and enhance sample quality,and the ran-dom forest(RF)is utilized for feature selection and combination augmentation.Second,the black-winged kite algorithm(BKA)is used to optimize the prediction model of Bidirectional Long Short-Term Memory(BiLSTM),and the knowledge distillation(KD)is applied to this model to obtain a more computationally efficient student model.Finally,taking the data set from the 17th"The Electrician Mathematical Contest in Modeling"competition and actual power data of a PV station in North China as case samples,four methods with five models,namely BiLSTM,BKA-BiLSTM,WOA-BiLSTM and BKA-BiLSTM-KD,are used for prediction respectively.The prediction results demonstrate that,compared with the BKA-BiLSTM model,the proposed method improves the coefficient of determination(R2)by more than 2.7%and re-duces the prediction time consumption by over 67.8%.It can provide efficient and accurate data support for PV station production planning and power grid dispatching decisions.
侯鹏飞;母宇宸;王月璠;常奇峰;李俏菡
华北水利水电大学电气信息工程学院,河南 郑州 450045华北水利水电大学电气信息工程学院,河南 郑州 450045华北水利水电大学电气信息工程学院,河南 郑州 450045许继电气股份有限公司,河南 许昌 461000华北水利水电大学电气信息工程学院,河南 郑州 450045
信息技术与安全科学
功率预测随机森林双向长短期记忆网络黑翅鸢优化算法知识蒸馏
power forecastingrandom forestbidirectional long short-term memoryblack-winged kite algorithmknowledge distillation
《机械与电子》 2026 (8)
103-109,116,8
河南省自然科学基金青年基金项目(252300420471)
评论