人工智能技术在新能源功率预测的应用及展望

Application and Prospect of Artificial Intelligence Technology in Renewable Energy Forecasting

构建新型电力系统是实现我国“双碳”战略目标的主要举措之一,风力发电和光伏发电作为两种最具代表性的新能源,其波动性和随机性给电网安全和新能源消纳带来了重大挑战,新能源功率预测是降低其随机性影响的核心关键技术。近年来,随着大数据技术和以深度学习、强化学习为代表的新一代人工智能(artificial intelligence,AI)技术在诸多领域的成功应用,其在新能源功率预测方面的应用仍有方兴未艾之势。首先该文论述AI技术在新能源功率预测应用的理论基础,并对AI技术在风电和光伏功率预测方面的应用进行系统总结,包括数据增强和特征构建等多种数据处理技术的应用,传统机器学习算法、深度学习算法以及组合算法在模型构建方面的应用,以及进化算法、群智能优化算法、强化学习等多种智能优化算法在模型训练和超参数优化方面的应用。然后,对当前相关文献进行统计分析,并基于新能源预测大赛结果和实际预测系统调研情况,对当前学术界研究热点和趋势、产业界模型应用情况进行对比和分析。最后,对当前新能源功率预测在场景自适应、小样本学习、数值天气预报系统(numerical weather prediction,NWP)数据时空分辨率、分布式新能源预测等方面存在的一些问题进行剖析,并对采用强化学习、元学习、图神经网络(graph neural network,GNN)等多种AI技术解决相关问题的前景进行展望。

Building a new power system is one of the main measures to achieve the"carbon peak and carbon neutrality"strategic goal.Wind power and photovoltaic power generation are the most representative renewable energy,and their volatility and randomness have brought major challenges to grid security and renewable energy consumption;while renewable energy power prediction is the key technology to reduce its random impact.In recent years,with the successful application of big data technology and the latest artificial intelligence(AI)technology represented by deep learning and reinforcement learning in many fields,its application in renewable energy power forecasting is still in the ascendant.This paper first briefly discusses the theoretical basis of AI technology in the application of renewable energy power forecasting,and systematically summarizes the application of AI technology in wind power and photovoltaic power forecasting,including the application of various data processing technologies such as data enhancement and feature construction,application of traditional machine learning,deep learning,combined algorithms in model construction,the application of evolutionary algorithm,swarm intelligence optimization,reinforcement learning,and other intelligent optimization algorithms in model training and hyperparameter optimization.Then,the current related literature is statistically analyzed.Based on the results of the renewable energy forecasting contest as well as the actual forecasting system research situation,the current academic research hotspots and trends,and the application of industrial models are compared and analyzed.Finally,some problems existing in the current renewable energy power forecasting in scene adaptation,few-shot learning,numerical weather prediction(NWP)data spatio-temporal resolution,distributed renewable energy prediction,etc.are analyzed,and the prospects of a variety of AI technologies such as reinforcement learning,meta-learning,graph neural network(GNN),to solve related problems are prospected.