A novel model-based dynamic distributed state estimator is proposed using sensor networks. The estimator consists of a
filtering step – which uses a weighted combination of sensors information – and a model-based predictor of the system’s
state. The filtering weights and the model-based prediction parameters jointly minimize both the bias and the variance of the
prediction error in a Pareto framework at each time-step. The simultaneous distributed design of the filtering weights and of
the model-based prediction parameters is considered, differently from what is normally done in the literature. It is assumed
that the weights of the filtering step are in general unequal for the different state components, unlike existing consensus-
based approaches. The state, the measurements, and the noise components are allowed to be individually correlated, but no
probability distribution knowledge is assumed for the noise variables. Each sensor can measure only a subset of the state
variables. The convergence properties of the mean and of the variance of the prediction error are demonstrated, and they hold
both for the global and the local estimation errors at any network node. Simulation results illustrate the performance of the
proposed method, obtaining better results than the state of the art distributed estimation approaches.