Linear Time Invariant State Space System Identification Using Adam Optimization

In this paper, a new state space system identification method is proposed. Adam optimizer is used as to optimize the state and input matrices of linear time invariant systems. A method of solving state space model numerically is presented and is used to build the TensorFlow graph of the identification method to get the gradients from a loss function for both matrices and to run the optimizer. The simulation results show that proposed method can succeed even when data is disturbed with random noise.