Symbolic partial mutual information is proposed in this paper, which is based on partial mutual information. This algorithm can be used to analyse the coupling between multivariate time series. We use this method to treat and analyse the sleeping multivariate bioelectricity signal (MBS) and wake one, it turns out that the coupling of wake MBS is obviously bigger than that of sleeping MBS. Finally hypothesis testing is done to prove that this method works and the average energy dissipation can be used as a parameter to detect nonequilibrium.