A stochastic machine learning based approach for observability enhancement of automated smart grids

This paper develops a machine learning aggregated integer linear programming approach for the full observability of the automated smart grids by positioning of micro-synchrophasor units, taking into account the reconfigurable structure of the distribution systems. The proposed stochastic approach presents a strategy occurring in several stages to micro-synchrophasor unit positioning based on the load level and demand in the system and based on the pre-determined sectionalizing and tie switches. Such a technique can also deploy the zero-injection limitations of the model and reduce the search space of the problem. Moreover, a novel method based on whale optimization method (WOM) is introduced to simultaneously enhance the reliability indices in order to specify the optimum topology for each phase and reduce the costs of power losses and customer interruptions. Although the problem of micro-synchrophasor placement is formulated in an integer linear programming framework, the restructuring technique is resolved on the basis of the WOM heuristic approach. Considering the uncertainty due to the metering devices or forecast errors, a stochastic framework based on point estimation is deployed to handle the uncertainty effects. The simulation and numerical results on a real system verify that the proposed method assures visibility of the distribution network pre and post reconfiguration in the time horizon of the planning. Furthermore, the results show that the system observability can be guaranteed at different load levels even though the system experiences different reconfiguration and topologies.