In order to overcome the ill-posedness, two methods are studied One is to add prior information term to objective function, the other is to narrow solution space. In the first method, we compare Gauss and Laplace distribution to express uncertainties of prior information. It is indicated that Laplace type prior information is useful when the number of damaged points is small. The estimation performance, however, is bad when observation noise is considered or the number of the damaged point increases. In the second method, parameters which indicate location and level of damage are optimized. The performance of the damaged etectioni s much improved in the caset hat noise is considered and the number of damaged point is increased.
