This paper presents an improved implementation of automatic differentiation（AD） technique in interior-point optimal power flow（OPF）. Distinguished from the existing AD-based OPF algorithms， the proposed algorithm applies operator overloading AD technique and explores the characteristics that most elements in Jacobian matrix and Hessian matrix of constraints in rectangular coordinates are constants， and adds a subroutine to identify all constant derivates to avoid repeated computations. Numerical studies on large-scale test cases indicate that the proposed AD implementation can compete with hand codes in execution speed without loss of maintainability and flexibility， which demonstrates that the proposed AD-based OPF algorithm has on-line application potential.