CNN data-driven active distribution network: Integration research of topology reconstruction and optimal scheduling in multi-source uncertain environment

The access of high proportion of uncertain factors makes the operating conditions of active distribution network more complicated. In addition to the scheduling optimization of active management device, changing the network topology is also an effective way to improve the reliability of power supply and reduce cost. However, the integration of topology reconfiguration and scheduling makes the problem more complicated, and the solving speed is challenged. Therefore, an integrated decision-making method based on CNN data-driven is proposed. Firstly, by combining the conditions of network operation connectivity with the necessary conditions of topological radiation structure, the full constraint conditions of distribution network connectivity and radiation are obtained. Secondly, Monte Carlo stochastic method is used to simulate a variety of uncertain factors. By combining with the active management device scheduling model and taking the minimum total cost as the objective function, GUROBI solver is used to find optimal output of the active management device and optimal branch switch combination in various random scenarios, and a large amount of effective historical optimization data is obtained. The topology information is also marked as a topology label. Finally, the CNN is trained with node loads and uncertainty factors as input data and topology label and optimal scheduling as output data. The improved IEEE-33 node system is taken as a simulation example, and the results show that the topology reconfiguration can further reduce the power cost and improve the reliability of power supply on the basis of optimal scheduling; The integrated decision-making method based on CNN data-drive can quickly obtain the optimal output of active management device and the optimal topology without power flow calculation under various uncertain scenarios. The classification and regression prediction speeds of CNN are 886 times and 736 times faster than the GUROBI solver respectively.