In terms of state estimation， there are great differences between the innovation graph technique and the conventional methods on obtaining the topology configuration and identifying bad data. This paper is concerned with the observability of topology configuration， the correlation of measurement data， the detectability and identifiability of the bad data in the innovation graph technique. In the technique， the observability is determined by the requirement of obtaining the reckoned innovation vector. The detectability and identifiability can be analyzed by the feature of the innovation difference vector. An optimized sequence for identifying the related bad data is proposed. An analysis on this technique and the test on IEEE 30-bus system show that the bad data identifiability can be analyzed qualitatively. The innovation graph technique is robust in identifying anomalies owing to its low requirement on measurement redundancy.