基于风电功率在线概率预测的风储联合系统随机预测控制

The limited dispatchability of wind energy poses a challenge to its increased penetration. One technically feasible solution to this challenge is to integrate an energy storage system (ESS) with a wind farm, i.e., to form a wind-ESS hybrid generation system. Currently, all the proposed control schemes for wind-ESS system are based on the deterministic predictive control, which only considers the offline point wind power forecasts. However, these schemes cannot handle two major features of wind energy. First, due to the stochastic nature of wind, there are significant inherent uncertainties in wind power, but the point forecasts cannot reflect these uncertainties. Second, the characteristic of wind generation is far from being deterministic, but the offline forecasting model cannot follow this time-varying characteristics. As a result, the performance of existing wind-ESS control schemes cannot be guaranteed and fairly low on average. To address these issues, this proposal proposes a stochastic predictive control scheme for the wind-ESS system. This scheme is based on the online probabilistic wind power forecasts, thus taking into account the censored non-Gaussian uncertainties in wind power, and tracking the changes in wind generation characteristics. In the view of forecasting model and control algorithm, these uncertainties and time-varying characteristics pose a series of challenges, i.e., how to model and predict the censored non-Gaussian distributions, how to update the forecasting model recursively, and how to solve a stochastic optimization problem involving censored non-Gaussian disturbance. Focusing on these challenges, this project will establish an online probabilistic wind power forecasting model, which provides the censored non-Gaussian predictive distributions using Bayesian inference and warping function. Based on these predictive distributions, stochastic predictive control algorithms for ESS and wind turbine will be developed, which incorporate the quantified wind power uncertainties using the warping chance constraints. This project will lay the theoretical foundation and develop practical methods for improving the wind power dispatchability.

风储联合发电系统为提升风电可调度性提供了物理支撑。然而，现有的风储联合系统控制方法多基于风电功率点预测模型和确定性预测控制策略，无法有效应对风电的两个本质特性，即出力呈截尾非高斯不确定性和出力特性呈时变性。对此，本项目从如何处理风电的这两个本质特性出发，探索建立一种能给出风电功率预测分布的在线概率预测模型，以及基于此的风储联合系统随机预测控制方法，从而系统级提升风储联合系统控制水平。对预测模型和控制算法来说，上述不确定性和时变性的引入，将带来截尾非高斯分布建模和预测、预测模型在线更新、含截尾非高斯扰动的随机系统优化等一系列挑战性科学问题。围绕这些难题，本项目将分层次地研究基于在线贝叶斯推理和翘曲映射的风电功率概率预测模型、基于翘曲机会约束的储能装置和风电机组随机预测控制策略，以及相应的随机规划问题求解算法。本项工作将为提升风电可调度性提供理论支持，并为相关实际应用提供一定的技术铺垫。

将储能装置和变速恒频风电机组整合配置，构成风储联合发电系统，可为提升风电可调度性提供强有力的物理支撑。对于风储联合发电系统而言，提升风电可调度性的前提条件之一是对风电功率进行有效的预测。针对此问题，本项目基于在线模型选择和翘曲高斯过程(WGP)，提出了一种概率风功率预测的集合模型。该模型提供了非高斯预测分布，量化了与风功率相关的非高斯不确定性。为了适应风力发电的时变特性，建立了时间依赖的多个基本预测模型和在线模型选择策略，从而自适应地选择每种预测的最有可能的基本模型。此外，设计了一种状态转换策略来动态地改变输入特征集，从而增强模型的适应性。在在线学习框架中，基本模型也应该是时间自适应的。为了实现这个目标，介绍了一种递归算法从而允许WGP基本模型进行在线更新。提出了一种随机协调预测控制方案。该控制方案具有双层结构。基于风电功率非高斯预测分布，上层随机预测控制器协调风电机组和储能装置的运行。计算出的功率参考值传递给下层风电机组和储能装置各自的本地控制器来具体执行。由此，则风储联合发电系统的总功率输出可被调节至期望的调度水平。该控制策略的显著特点是基于风电功率的非高斯预测分布来优化控制指令，故而可以有效处理风电功率的非高斯不确定性。