Towards loss minimization in power distribution systems using AI, the WatDist algorithm

Abstract

As electric power distribution systems continue to grow in size and complexity, Distribution Automation schemes become more attractive. One of the features that is desirable in an automated system is feeder reconfiguration for loss reduction. Reducing losses can result in substantial savings for a utility. Other benefits from loss reduction include released system capacity, and possible deferral or elimination of capital expenditures for system improvements and expansion. There is also improved voltage regulation as a result of reduced feeder voltage drop.

System reconfiguration is accomplished using existing switches in the network. For a given system, there will be a switching pattern that minimizes system losses. However, if there are N switches in a network, there are 2N possible switching combinations, and the challenge of finding the optimum switching pattern to minimize losses becomes formidable as the number of switches increases.

In this thesis, a novel algorithm, WatDist, is introduced to solve the network reconfiguration for loss minimization problem. The proposed techniques is based on artificial intelligence techniques applied to constraint satisfaction optimization problems. A critical review of earlier methods is presented to highlight their shortcomings. Computer simulations using WatDist demonstrate its advantages, including a high success rate in finding the global optimum, the final solution being independent of the initial configuration, and assurance that any solution offered will have a radial configuration with all loads connected and no constraint violations. A cost/benefit analysis demonstrates the significant contribution of the algorithm to distribution system analysis and operation.