计及不确定性的主动配电网自适应鲁棒优化运行研究

Future active distribution network with high penetration of distributed generations (DGs) will present high uncertainty and high complexity and variability of operation conditions, due to the uncertainty and fluctuation of renewable energy generations. The most existing distribution network optimization and control strategies are established based on deterministic optimal power flow (OPF) model and real-time control under multi-time scale framework, which weaken adverse impact of grid-connected DGs by ‘stepwise refining’ and implement regulations passively after distribution network security constraints violations and can’t realize active control and real-time optimization. In this proposal, the theory and method of adaptive robust optimization will be introduced into the establishment of active distribution network optimal operation strategy, which integrates the OPF and real-time control to accomplish true active control and real-time optimization of active distribution network. (1) Interval prediction model and prediction model on-line self-correction algorithm for the output power of renewable energy generations will be developed based on Ensemble Learning and Incremental Learning technology; (2) High precise linear power flow equations will be built for unbalanced distribution network. Based on the developed linear power flow equations, the robust feasible region estimation method is studied by successive linear approximation technology. (3) Multiple adaptive robust optimization models based on different adaptive functions (linear/non-linear) will be established for different operation demands (reactive power optimization/ active-reactive power coordinated optimization) of active distribution network and the corresponding solvers will be designed. (4) Optimal partition model of the distribution network will be built based on the sparsity of the network topology and asynchronous iterative distributed optimization algorithm for the adaptive robust optimization model will be developed according to the partition.

分布式能源高渗透率的主动配电网将呈现高不确定性、运行工况复杂多变等特征。基于确定性优化模型和实时自治控制的配电网多时间尺度优化调度策略，通过“逐级细化”的方式削弱分布式电源出力不确定性的影响，在安全约束越限后被动地调节。本项目拟将自适应鲁棒优化理论与方法引入主动配电网优化运行，将优化调度和实时控制集成一体，实现真正的主动控制和实时优化。1）基于集成学习和增量学习理论，建立可再生能源发电功率区间预测模型并开发预测模型在线自校正算法；2）建立三相不对称主动配电网高精度线性潮流模型，并研究基于逐次线性逼近技术的主动配电网鲁棒可行域估计方法；3）针对主动配电网无功优化、有功无功协调优化不同运行需求，分别建立基于线性、非线性自适应函数的自适应鲁棒优化模型，并开发相应的求解算法；4）根据配电网稀疏特性，建立主动配电网最优分区模型，并基于此开发主动配电网自适应鲁棒优化模型异步迭代分布式求解算法。

项目针对配电网中分布式可再生能源发电高比例渗透带来的不确定性问题，从自适应鲁棒优化的角度构建了配电网优化运行的技术框架：构建了双输出极限学习机模型，并基于增强学习的在线序列化，提出了具有在线自校正能力的可再生能源发电功率区间预测方法；基于复数空间及仿射数学理论，提出主动配电网运行鲁棒可行域估计方法，实现对分布式电源(distributed generation, DG)等调控设备安全运行范围的准确、高效预估；针对DG的MPPT/非MPPT等不同运行模式，探究逆变器的线性/非线性控制方法，从全局优化的角度构建配电网的自适应鲁棒无功/有功无功协调优化模型，提出不确定性环境下配电网高鲁棒性、低保守度运行策略；分析配电网的庞大规模、复杂运行工况以及高度稀疏特性，考虑配电网中不同节点间的耦合特性，提出避免人为因素介入的配电网分区方法，并开发了适用于求解自适应鲁棒无功优化/有功无功协调优化的分布式算法。 项目提出的配电网自适应鲁棒优化运行方法融合了主动控制与实时优化，实现了配电网运行安全性与经济性的统一。