Application of automatic differentiation in power system trajectory sensitivity analysis

Trajectory sensitivity is one of the most important analysis tools for power system dynamic performance assessment and control. Calculating Jacobian matrix has been identified as one of the most computational-intensive and error-prone algorithmic procedures. Automatic differentiation (AD) is able to efficiently generate derivatives of a specified function without any additional hand coding work. Application and improvement of AD in trajectory sensitivity is investigated in this paper. A novel partial-decoupled AD strategy is developed. The proposed approach enables a module-based design separating dynamic component models from time-domain simulator. Maintainability and flexibility in software architecture can be achieved. Numerical results demonstrate that the application of AD retains numerical accuracy and achieves high computational performance compared with numerical finite differentiation.