【作者】 黄新宇； 陈静； 王尤刚； 江灏； 缪希仁；

【Author】 Xinyu Huang;Jing Chen;Yougang Wang;Hao Jiang;Xiren Miao;College of Electrical Engineering and Automation,Fuzhou University;Power System and Device Industry Research Institute,Fuzhou University;

【机构】 福州大学电气工程与自动化学院； 福州大学电力系统与装置产业研究院；

【摘要】 泛在电力物联网的建设带来了海量的电力用户侧数据,如何利用这些数据进行准确高效的负荷预测,是新一轮电力体制改革下电网公司经济稳定运营的基础。本文以泛在电力物联网建设为背景,研究了区域日最大负荷预测精度提升与算法耗时减小的问题,对此提出一种基于离散小波变换和改进粒子群优化的极限学习机相结合的多步预测模型。利用离散小波变换将负荷序列分解为3层细节分量和1层趋势分量,采用改进的粒子群算法对极限学习机的参数进行优化,融合重构的负荷序列和日期特征对优化后的模型进行学习,得到日最大负荷多步预测模型。依托某省泛在建设平台,对其A、B两地实际数据的预测结果表明,所提方法在RMSE、MAPE精度指标上比DWT-ELM、IWPSO-ELM、ELM和WNN均有明显改善,在训练耗时上比WNN缩短51.78s,验证了所提算法在提高预测精度的情况下大大降低模型的训练时间。更多还原

【Abstract】 The construction of ubiquitous power Internet of things has brought a large number of power user data. How to use these data to accurately and efficiently predict the load is the basis for the stable operation of power grid companies under the new round of power system reform. In this paper, based on the construction of ubiquitous power Internet of things, the problem of improving the prediction accuracy and reducing the algorithm time is studied, a multi-step prediction model based on discrete wavelet transform and improved particle swarm optimization extreme learning machine is proposed. The discrete wavelet transform is used to decompose the load sequence into three layers of detail components and one layer of trend components, the improved particle swarm optimization was used to optimize the parameters of the extreme learning machine, and the reconstructed load sequence and date characteristics were combined to train the optimized model, then the multi-step prediction model of daily maximum load is obtained. Based on the ubiquitous construction platform in A province, the prediction results of the actual data from A and B show that the proposed method is significantly better than DWT-ELM, IWPSO-ELM, ELM and WNN in RMSE and MAPE precision indexes, and the training time is 51.78 s shorter than WNN, which verifies that the proposed algorithm can significantly reduce the training time of the model when the prediction accuracy is improved.更多还原