On the quality of pseudo-measurements for distribution system state estimation

With the increasing penetration of distributed energy resources, Smart Grids control applications are always more dependent on monitoring data. Pseudo-measurements are used in Distribution System State Estimation to allow estimating the operating conditions of the system also when the number of field measurements is limited. Since the accuracy of the estimation depends on the quality of the pseudo-measurements, in this paper the factors that affect this quality are investigated and the performance of a machine learning-based approach for pseudo-measurements generation is evaluated. Starting from real data collected from the Forschungszentrum Jiilich campus, a dataset is engineered, and the considered data coding approach is presented. Finally, different neural models based on multi-layer perceptron are presented, and their performances are compared with those of trivial alternatives.