Neural-net based coordinated control of capacitors and ULTC transformer in daily operation of radial distribution systems

Abstract In this paper the problem of coordinated VAR/VOLT control of general radial distribution systems with lateral branches, in daily operations is considered, by using mechanically switched shunt capacitors and under-load tap-changing (ULTC) source transformers, feeding the radial network. Control variables are switchable, fixed or regulated capacitors and tap positions of the ULTC transformer, assuming the no-load tap-changing (NLTC) transformers within the network remain in fixed tap positions. In the first stage of the proposed approach, the training patterns required for the neural-net supervised learning process are generated by using a decoupled model for the coordinated VAR/VOLT control. The corresponding optimal daily schedules are used in the second stage, within the unsupervised/supervised concept, to synthesise complex mapping relating input load and source-voltage data to control variables. If discrepancy between forecasted and measured variables is greater than the prespecified tolerance threshold, the iterative decoupled algorithm for coordinated VAR/VOLT control is initiated to perform the correction of the schedule. The effectiveness of the proposed method is verified on the real 110 kV/35 kV/10 kV distribution system with 53 nodes. It has been demonstrated by this case study that the proposed approach preserves the optimization accuracy provided by the decoupled model for coordinated VAR/VOLT control and overcomes unnecessary calculations for similar load profiles.