| The movement of the train makes the mathematical model of the traction network established during each sampling analysis change,thus forming the process of dynamic modeling and solution,namely,dynamic power flow calculation.Under the background of "one picture a day" mode promotion,it is particularly important to realize fast and accurate calculation of dynamic power flow of AC traction power supply system in each stage of system design,and it can also reproduce abnormal conditions in operation at the minimum cost.However,in the process of dynamic power flow solution,the solution of a series of large complex nonlinear equations takes a large proportion of time,and when the matrix order becomes larger,the proportion of time is more obvious,which affects the calculation efficiency.In order to solve the above problems,from the perspective of power flow algorithm,this paper adopts parallel calculation and block dynamic power flow to solve the speed.In this paper,a mathematical model of AC traction power supply system is established based on "multi-conductor chain network + power source".According to the principle of whether the derivative is needed or not,the algorithm for solving the model is divided into first-order convergent power flow algorithm and second-order convergent power flow algorithm,and the comparative analysis is made.There are two ways to solve the equations of the first order power flow algorithm,namely,the inverse and the block chase,the inverse formula of the block tridiagonal matrix formed by the first order power flow algorithm is derived,and the theoretical analysis shows that there is a strong "data dependent" relationship between the elements of the inverse matrix,and the single power flow calculation is difficult to be parallel.When the second-order convergence algorithms such as N-R method and improved PQ decomposition method are used to solve the power flow equations,it is necessary to simplify the multi-conductor model equivalent.Although it can speed up the calculation,some information is ignored during the simplification process,resulting in errors in the calculation results compared with the first-order convergence algorithm.The characteristics of the first order and second order algorithms for AC traction power supply are presented.A hybrid parallel computing model of time layer and traction layer was established based on the "operation diagram cross-section method",and the CPU-GPU heterogeneous computing framework was used for calculation.Firstly,in order to reduce the memory consumption and accelerate the generation speed of node admittance matrix,a non-repetitive generation method for multi-conductor chain model under dynamic power flow was proposed to generate node admittance matrix with full simulation time and full traction station.Secondly,GPU is used to solve the inverse matrix of the node admittance matrix in batch,so as to reduce the time consumption of matrix inverse,and the CPU thread pool technology is used to "apportion" each independent flow iteration task to multiple CPU cores for parallel computation.Finally,this paper gives the specific operation steps of the dynamic power flow calculation method of AC traction power supply system based on CPU-GPU heterogeneous computing framework,and the proposed parallel algorithm can accelerate the calculation.Based on the software development platform of "Visual Studio+CUDA",the parallel simulation software(ACTPS 2.0)of AC traction power supply system was designed and developed,and the dynamic power flow calculation method of AC traction power supply system accelerated by CPU-GPU heterogeneous computing framework was realized.In the example simulation part,the proposed algorithm is verified by two examples: the conventional distance AT power supply mode and the long-distance power supply mode of "main-station +cable +AT".Compared with the traditional serial method,the proposed algorithm has a speedup ratio of 5.28～121.3 times on the premise of ensuring accuracy,which has a certain engineering application value. |
PDF Full Text Request