Aiming at the problem that the traditional single kernel weighted least square support vector regression（WLSSVR） localization algorithm has poor fitting degree of RSSI

a Wi Fi indoor localization algorithm based on improved multi-kernel WLSSVR（IMK-WLSSVR） is proposed. This algorithm uses the MKSVR model to build the nonlinear relationship between the RSSI values of finger-print and their locations. The polynomial kernel function and the Gaussian radial basis kernel function are used to build the multiple kernel function

and the normal distribution weight function is used to determine weights. The improved simulated annealing algorithm is utilized to obtain the optimal positioning model parameters; and improve positioning property of model. Finally

the localization model is used for positioning. The experimental results show that the positioning error is less than 2 m in 92. 2 % probability under the condition of 3 m grid and noise standard deviation of 3 d Bm

and the positioning precision is prior to weighted K-nearest neighbor（WKNN） algorithm and single-kernel WLSSVR method.