We propose a new process identification method that combines the two methods of the relay feedback to activate the process and the backward integrals to estimate the model parameters. Novel deviation variables are introduced to incorporate the case that the initial part of the process is unsteady-state without sacrificing the dynamic information included in the initial part, while the previous approaches assign zero-weighting to the initial parts, resulting in loss of the dynamic information included in the initial part. The final cyclic-steady-state part of the process input and output data is chosen as the reference of the deviation variables. The proposed method can estimate the model parameters analytically by using the backward integrals and the least squares method.