Identification of interacting bad data in the framework of the weighted least square method

The identification of multiple interacting bad data, arising in the framework of static state estimation, is commonly handled by the largest normalized residual criterion. However, this technique may lead to faulty results when the bad data are of the conforming type. In the present work, the identification problem is formulated as a non-linear optimization with mixed variables. Its solution is found by means of combinatorial optimization methods such as branch-and-bound, genetic algorithms and tabu search techniques. All these approaches consist of three successive steps: generation of a tentative bad data identification, solution of the corresponding state estimation problem and memorization of already considered cases. To speed up the state estimation solution, the possible use of sensitivity techniques is also considered. It is shown that the efficient storage of solved cases and the breadth of the search play a critical role in determining the efficiency of the procedures. The proposed approaches were applied to the identification of multiple interacting bad data with reference to the IEEE test systems as well as to an actual network of Italian origin.