Numerical observability analysis based on network graph theory

This paper presents a numerical method for topological observability analysis of a measured power system. By floating-point operations on the echelon form of a rectangular test matrix, which is based on network graph properties, observability and maximal observable islands are determined. A minimal set of pseudo measurements, which make an unobservable network barely observable, is selected in a noniterative manner. The existing numerical methods are based on the number of zero pivots that may appear during the factorization of the measurement Jacobian or the gain matrix. Due to round-off errors, the zero pivots may be misclassified. The problem becomes more severe when the number of injection measurements is large, resulting in a great disparity of values in Jacobian or gain matrix. In the proposed method, the test matrix consists of +/-1 values, it is numerically better conditioned and zero pivots are identified more accurately. By topological processing of the flow measured branches and by removing the redundant injection measured nodes that are incident only to flow measured branches or branches which form loops with flow measured branches, a reduced test matrix is created with fewer nonzero elements than the Jacobian or the gain matrix, resulting in less computational effort. The method details are illustrated by various test systems.