CAREER: An Integrated Hybrid Forecasting Framework for Increased Wind Power Penetration

该研究的目标是开发下一代算法，以实现短期风预测的显着改善。我们无法准确地捕捉高维空间中风的非高斯不确定性，这意味着低可预测性、高风险和对昂贵的平衡电力资源的需求。因此，它是一个主要目标，拟议的研究是显着提高风预测提前48小时，从而提高调度，调度和单位承诺在日前的电力market.Intellectual Merit操作：拟议的研究将开发一个综合框架的随机算法可扩展的非线性不确定性传播。算法输出将与贝叶斯融合意义上的现场测量数据相结合，从而形成混合预测结构。主要的技术挑战是：（i）由于多个时间和空间尺度、湍流和地形效应而导致的复杂的风动力学;（ii）非高斯风的不确定性;（iii）由于高维性而需要可扩展的算法;以及（iv）需要融合来自多种算法和异构测量源的信息。建议的框架将有以下主要特点，以满足这些挑战：（1）制定风状态作为一个随机混合过程，由多个降阶微尺度和中尺度模式;（2）一种新的随机粒子不确定性传播方法的基础上的方法的特征，马尔可夫链蒙特-卡罗和Karhunen-Lo`eve扩展。开发的算法的实际有效性将从两个风电场的数据进行测量：科洛顿风电场/纽约州和罗斯科风电场/TX。更广泛的影响：拟议的预测框架将导致增加风电的渗透，减少目前与它相关的风险，并使我们能够实现我们的全球目标，减少对化石燃料为基础的电力的依赖。教育计划包括培训可持续能源多学科领域的高中教师，这些教师将反过来影响数千名学生。

The goal of the proposed research is to develop the next generation of algorithms to achieve significant improvement in short-term wind forecasting. Our inability to accurately capture non-Gaussian uncertainty of wind in high dimensional spaces translates to low predictability; high risk and the need for expensive balancing power resources. It is thus a primary objective of the proposed research is to significantly improve wind forecasts upto 48 hours in advance, leading to enhanced dispatch, scheduling and unit commitment operations in the day-ahead electricity market.Intellectual Merit: The proposed research will develop an integrated framework of randomized algorithms for scalable nonlinear uncertainty propagation. Algorithm output will be combined with measured on-site data in the sense of Bayesian fusion, leading to a hybrid forecasting structure. The main technical challenges are: (i) complex wind dynamics due to multiple temporal and spatial scales, turbulence and orographic effects; (ii) non-Gaussian wind uncertainty; (iii) need for scalable algorithms due to high dimensionality; and (iv) need for fusion of information arriving from multiple algorithms and heterogeneous measurement sources. The proposed framework will have the following key features to meet these challenges: (1) formulation of the wind state as a stochastic hybrid process, governed by multiple reduced ordermicro and mesoscale models; (2) a novel randomized particle uncertainty propagation approach based on the method of characteristics, Markov chain Monte-Carlo and the Karhunen-Lo`eve expansion. Practical effectiveness of developed algorithms will be measured against data from two wind-farms: the Cohocton Wind Farm/NY and the Roscoe Wind Farm/TX.Broader Impact: The proposed forecasting framework will lead to increased penetration of wind power by reducing the risks currently associated with it; and enable us to achieve our global targets of reducing dependence on fossil-fuel based electricity. The education plan includes training high school teachers in the multidisciplinary area of sustainable energy who will in turn reach thousands of students.