Coarse-grained Parallel Computing Of The Scheduling Decision In Large Scale Power System

The scale of power system is expanding gradually with the fast development of science and technology. Large-scale computing problems, which can be analyzed by mature methods in the single processor, emerged in dispatching decision field of power system. However, the disharmony between the requirement of engineering application and too much time consuming in serial computing mode is becoming more and more prominent. Therefore, it's very urgent for us to do some research on improving the calculation efficiency to meet the requirements of large-scale engineering application. In recent years, processor's power density spiking reveals that the performance of single processor upgrading has been basically reach the physical limits. So I introduced the multi-processor computing technology in this paper to improve the computational efficiency of the large-scale power system. The main research of this paper is a coarse-grained parallel method of the Monte-Carlo simulation of probabilistic load flow(PLF), the unit commitment(UC) problem and the dynamic optimal power flow(OPF).Firstly, this paper introduces the basic principle of parallel computing. And then, this paper focused on the analysis of OpenMP, MPI and the "MPI+OpenMP" parallel model. The typical problems that may exist in the parallel processing, such as load imbalance and data conflict, have been studied and the corresponding solutions have been raised. And then, this paper gives the calculation meth of the evaluation index of parallel computing. Based on the Amplifier XE, this paper also gives the optimization method of parallel program. The optimal number of processors for coarse-grained symmetrical parallel model has been researched based on the parallel server.Based on the OpenMP parallel programming model, a coarse-grained parallel computing method of the PLF calculation based on the Monte-Carlo simulation method has been proposed. Firstly, this paper introduces the basic principles of the Newton-Raphson method of power flow calculation and the method of the Monte-Carlo simulation. The probability model of wind power and load has also been proposed at the same time. And then, this paper gives the solution to the problem of random number and the scope of variables which can be faced in the process of parallel processing. This paper also gives the process of the coarse-grained parallel computing method of the PLF calculation based on the Monte-Carlo simulation method. The simulation results show that the efficiency of the large-scale power system PLF calculation has been improved.Based on the MPI parallel programming model, a parallel computing method for solving network security constrained UC problem has been proposed. Firstly, the detailed mathematical model of the UC problem has been given. Owing to the consideration of the valve point effect of the thermal power unit, this paper adopts a double-layer genetic algorithm to solve the UC problem. A coarse-grained parallel solving method has been proposed to shorten the time-consuming of the double-layer genetic algorithm. That is parallel computing of the UC problem and retains many feasible solutions based on the MPI model without considering the coupling constraints between each period. Finally, this paper uses the genetic algorithm to search the best result in the feasible solution which can meet the coupling constraints of the UC problem. The degree of compression of the feasible solution space of the inner constraint is much larger than the outer constraint of the double layer genetic algorithm after each period of time has been decoupled. In order to further improve the quality of solutions of the inner layer genetic algorithm, a chromosome initialization method based on mathematical expectation has been proposed. At the same time, in order to improve the diversity of the double layer genetic algorithm to improve the quality of the overall solution, the crossover probability and mutation probability is set in a range of variables. The IEEE118 system simulation results show that the proposed coarse-grained parallel solving method of the UC problem can obtain a considerable speedup and parallel efficiency.Based on the determined UC and the "MPI+OpenMP" double-layer parallel model, a double-layer parallel computing method of dynamic optimal power flow has been proposed in this paper. Firstly, this paper gives a comparison of the model of unit commitment problem and dynamic optimal power flow. And then, this paper uses the genetic algorithm to solve the dynamic optimal power flow problem. In order to improve the computational efficiency of dynamic optimal power flow, the optimal power flow problems of each period have been parallelized. At the same time, in order to maximize the resource utilization rate of the computer and the efficiency of parallel computing, the chromosome decomposition based coarse-grained parallel solving method of the genetic algorithm has been proposed in this paper. That is the "MPI+OpenMP" double-layer parallel model, the parallel computing of each period of optimal power flow in the outer layer is realized through the MPI model, and the parallel genetic algorithm is realized by the OpenMP model. Finally, the simulation results of IEEE 118 system show that the double-layer parallel computing model can realize "the second speed" of the algorithm and further improve the parallel efficiency within the range of the performance of the parallel server.