Using multiple models is a common approach for describing a large variety of physical systems. Alas, the optimal algorithm for estimating the state vector in a multiple-model system demands computation resources the grow exponentially in time. Therefore, there is an immense importance in creating techniques for estimation that are sub-optimal but are computationally efficient. On of the popular systems is the IMM (Interacting Multiple Model) algorithm, which balances performance and computation. In this algorithm we assume that the transition matrix between the different models is completely known. This assumption is rather presumptuous, and in practical cases this matrix needs to be estimated in real time. We tried to improve the estimation technique, under conditions where some of the system’s parameters are unknown, in comparison to the classic IMM algorithm.
The solution is based on estimating the transition matrix according to the probabilities vector of the IMM algorithm. The TPM is estimated in each time step and compared to an initial fixed value and the correct value, in manners of MSE. This was tested along different cases of noise and correct values of TPM. The solution has two variations both tested in the according manner. The solution showed up to 53% in reducing the gap of MSE between the correct TPM and a fixed initial value of TPM. Results varied greatly upon conditions of noise, and each variation is better in different cases.