基于随机矩阵理论和改进粒子群优化-深度置信网络的无功优化

配电网无功优化是使电网能安全稳定运行的重要问题，本文将深度学习应用于配电网无功优化上，提出基于改进的粒子群优化 (Particle swarm optimization, PSO)深度置信网络（Deep Belief Networks，DBN）模型的配电网无功优化方法。首先，利用随机矩阵理论将配电网运行过程中产生的环境以及电气数据构造五种随机矩阵，每种矩阵构建8种特征指标，共提取40种特征指标；其次，以随机矩阵理论所提取的各特征指标为输入，以变压器分接头档位和节点的无功补偿容量作为输出，利用改进的PSO优化DBN网络的初始权重，建立PSO-DBN模型，学习各特征指标输入与变压器分接头档位以及节点无功补偿这两种控制策略之间的关系，得出相应的无功优化策略。最后，以改进的IEEE33节点系统作为算例仿真对象，验证了所提方法可降低网损和减少电压波动。

Reactive power optimization based on random matrix theory and improved PSO-DBN

Reactive power optimization of the distribution network is an important issue of safe and stable operation in distribution network. Deep learning is applied to the reactive power optimization of the distribution network, and an improved particle swarm optimization (PSO) -deep belief network (DBN) method in reactive power optimization is proposed. First, the random matrix theory of five random matrices from the environment and electrical data from the operation of the distribution network. 8 characteristic indexes are constructed in each matrix, a total of 40 characteristic indexes are extracted; Second, the random matrix theory is used to input of extracted characteristic index. the transformer tap position and the reactive power compensation capacity of the node are used as output. Third, the improved PSO algorithm is used to optimize the initial weight of the DBN network, a PSO-DBN model is established. The relationship between characteristic indicators and reactive power optimization strategies are leaned. So, the corresponding reactive power optimization strategies is obtained. Finally, the improved IEEE33-node system is used as an example simulation case. This case is verified the method in this paper can reduce network loss and reduce voltage fluctuations.