Extreme Learning Approach for Blood Glucose Estimation

摘要

This paper proposes an extreme learning machine approach for performing the blood glucose system estimation. The extreme learning machine consists of a unitary matrix in the first layer, a set of nonlinear activation functions and a weight vector in the second layer. Here, the radical basis functions are used as the nonlinear activation functions. The joint design problem of these three sets of parameters are a nonconvex optimization problem. An iterative algorithm based on the joint genetic algorithm and the linear programming is applied to find a near global optimal solution of the nonconvex optimization problem. The simulation results show that the proposed method is effective for performing the blood glucose estimation.