深度学习辅助约束辨识的电力市场快速出清方法

日前电力市场出清需要求解大规模安全约束经济调度问题，尽管实际采用线化处理方法，但需要考虑N－1场景下的大量安全约束，导致其规模庞大，难以快速求解。提出了一种深度学习辅助的日前市场快速出清方法，以满足快速出清计算场合的应用需求。首先，设计基于深度神经网络的安全约束经济调度模型计算框架，将深度学习技术应用于日前电力市场出清计算过程，兼顾安全约束经济调度模型的求解速度和求解精度；其次，提出面向起作用约束辨识的深度学习策略，从特征向量、深度神经网络结果处理2个方面，实现安全约束经济调度起作用约束集的辨识，并将其纳入日前市场出清模型，以简化模型的复杂度；最后，在接入新能源的IEEE 30标准测试系统中验证了所述方法的有效性。利用深度神经网络预辨识安全约束经济调度模型的起作用约束，有利于降低模型复杂度，提高日前市场出清的计算效率。

A Fast Power Market Clearing Method Based on Active Constraints Identification by Deep Learning

The day-ahead power market clearing needs to solve the security-constrained economic dispatch (SCED) problem. Although the SCED problem is a linear programming (LP) model, the model size is too large to be effectively solved because the massive security constraints in the N–1 scenarios need to be considered. Therefore, this paper proposes a fast clearing method for day-ahead power market based on the deep neural network. Firstly, a computation framework for SCED model based on deep neural network is designed, and embeds deep learning technology into the existing day-ahead power market clearing framework, which can effectively improve the solving speed of the SCED model without compromising precision. Secondly, a deep learning strategy is proposed for identification of active constraint sets, which can provide technical support for deep neural networks to effectively identify the active constraints of SCED from two aspects: feature vector design and efficient processing of the results of deep neural network. Finally, the effectiveness of the proposed method is verified in the IEEE 30 standard test system with renewable energy sources. The deep neural network is used to pre-identify the active constraints of the SCED model, which is beneficial to reduce the complexity of the model and improve the calculation efficiency of market clearing.