Research On Adjoint Sensitivity Analysis Based Transient Stability Preventive Control Algorithm And Its Parallelization

As the first line of defense to ensure the safety and stability of power system,preventive control is an important research topic in power system analysis and control.Transient stability preventive control is to seek an economically optimal state,in which the system not only satisfies the static security constraints but also meets the transient stability requirements after the expected faults occurred.This problem is a large-dimension,nonlinear and nonconvex dynamic optimization problem.The differential-algebraic equations that descripts the transient process lead to the "curse of dimensionality" when dealing with large-scale system and more faults,which will enhances the computational complexity greatly.However,with the access of renewable energy sources such as wind power and photovoltaic power,modern power system’s operation mode and state changes frequently,which raises the requirements for real-time control of power system.Thus,the purpose of this paper is to get a reliable transient stability preventive control plan as quickly as possible.The main contents and results are as follows:1)The transient stability preventive control is modeled as the problem of Transient Stability Constrained Optimal Power Flow(TSCOPF).Constraints conversion technology is adopted to transform the original problem into an ordinary Nonlinear Programming problem(NLP),which can effectively relieve the "curse of dimensionality".A two-layer-structure is proposed to implement the algorithm:In simulation layer Adjoint Sensitivity Analysis(ASA)is adopted to achieve fast calculation of dynamic sensitivity,and very dishonest Newton method and Jacobian matrix reusing approach are applied to speed up calculation.The optimization layer is based on Predictor-corrector Interior Point Method(IPM)to solve the transformed NLP.BFGS method is adopted to generate Hessian matrix of transient stability constraints approximately.An extended-generating method is proposed to improve the accuracy and sparsity of the Hessian matrix according to the sparse characteristic.The test results of multiple systems show that the proposed algorithm has good convergence and high computational efficiency,and shows great speed advantage compared with other algorithm.2)Aiming at the two parts of the TSCOPF serial algorithm that cost most of time,which are ASA and modified equations solving in IPM,parallel computing is used to speed up calculation.Aiming at the difference of time consuming in the calculation of ASA for different faults,a dynamic assignment strategy based on task queue is proposed to improve the load balancing efficiency.Aiming at the increase of nonzero element in coefficient matrix of modified equations,the structure of coefficient matrix is analyzed and utilized,and a block-solving method is proposed to transfer the main task into the back substitutions of linear equations with multiple right terms.Then parallel computing can be used to improve solving speed.A TSCOPF parallel algorithm that contains two parallel parts is proposed.The test results show that the parallel algorithm has good load balancing efficiency and speed-up ratio,and can effectively reduce consuming time when solving TSCOPF problem with multiple contingencies in large-scale system,which can meet the requirements of online preventive control.