Research On Parallel Algorithm For Economic Dispatch Of Power Systems With Large-scale Wind Power Integration

Wind energy is the most promising renewable energy source for large-scale applications,and wind power has been vigorously developed and applied in recent years.However,when random wind power is connected to the existing power systems,the modeling and solving of dynamic economic dispatch problem are facing enormous challenges.At present,transform the stochastic dynamic economic dispatch problem into the scenario-based model or sample average approximation model by simulating the randomness characteristics of wind power based on the Monte Carlo sampling method,is the most basic modeling method and the benchmark method for evaluating the accuracy of other stochastic dispatching models.But the scenario-based model and the sample average approximation model are large-scale linear or quadratic programming models,which will face Curse of Dimensionality problems if being solved directly.This paper focuses on the parallel solution method for scenario-based model and sample average approximation model of stochastic dynamic economic dispatching of power systems with large-scale wind power integration.Firstly,we use the Monte Carlo sampling method to establish the scenario-based model for stochastic dynamic economic dispatching problems of power systems with large-scale wind power integration.The model aims to minimize the cost of power generation.By intro-ducing scenes transfer constraints,the stochastic optimization problem can be transformed into a deterministic optimization problem.After using Dantzig-Wolfe decomposition algo-rithm to decouple the model according to the scenarios,the large-scale problem with tens of millions of variables and constraints can be decomposed into a series of sub-problems with only tens of thousands of variables and constraints and a smaller main problem.In the itera-tion,the improved sub-gradient method is used to improve the convergence of the algorithm,and the Grid Computing Facility on GAMS platform is used to construct a parallel computing framework for quickly solving sub-problems.The proposed method not only reduces the re-quirement of computer memory to solve the problem that can not be calculated before,but also improves the efficiency of solving the high-dimensional problem and thus has engineer-ing application value.Taking the IEEE 39-node system and a provincial power system as examples,the effectiveness of the parallel strategy of the Dantzig-Wolfe decomposition algo-rithm is verified.Then,the Monte Carlo sampling method is used to establish the sample average ap-proximation model for stochastic dynamic economic dispatching problems of power systems with large-scale wind power integration.The model aims to minimize the cost of purchasing electricity,and there are nonanticipativity constraints in the first period.The parallel simplex method is designed for the coefficient matrix of constraints which is highly sparse and has dual block-angular structure.By linearly transforming the large-scale basic matrix,the block calculation at key calculation steps is realized,and the computational performance of the sim-plex method is effectively improved.The algorithm is scalable.Parallel computing in large memory overhead and high time-consuming steps is realized by using parallel computing toolbox and parallel computing service in Matlab,which effectively reduces the memory re-quirements and reduces the calculation time.Finally,the effectiveness of the parallel simplex method is verified by the calculation and analysis of the IEEE 39-node system and a provin-cial power system.