应用近似动态规划算法求解安全约束随机动态经济调度问题

摘要

In view of the problem that the computing time for solving a security-constrained stochastic dynamic economic dispatch(DED)model of large-scale power system is too long,a method using approximate dynamic programming algorithm is proposed to quickly solve security-constrained stochastic DED problem of power system without pumped storage plants.A virtual storage model(VSM)for the stochastic DED problem is established.In this model,system”s positive spinning reserve capacity of current time interval is taken as storage variable,then the system”s state transition equation of relations between adjacent time intervals is established,including the security constraints of transmission lines and sections.Based on wind farms”power output forecast curves,error scenarios are generated by the Latin hypercube sampling method.In order to train the value functions of each time interval by scanning the error scenarios one by one,the quadratic programming model of each period should be solved successively.After getting the convergence value functions,the final generation dispatch scheme will be obtained by solving the optimal dispatch model with the forecast scenario.This method has realized the decoupling of each scenario and each time interval of the stochastic DED model by decomposing a large-scale of optimization problems into a series of small-scale optimization problems,effectively increasing the computing speed of solving the stochastic DED model of large-scale power system.Test results on a provincial power system demonstrate the feasibility and effectiveness of the proposed method by comparison with the scenario method and robust optimal dispatch method.%针对大电网安全约束随机动态经济调度(DED)问题的求解时间太长，提出了应用近似动态规划算法快速求解不含抽水蓄能电站电网的安全约束随机DED问题的方法。建立了随机DED问题的虚拟存储器模型，以系统的正旋转备用容量作为存储变量，构建系统相邻时段的状态转移方程，并考虑了各输电线路和断面的安全约束。以风电场日前功率预测曲线为基础，通过拉丁超立方抽样产生风电场出力的误差场景，并逐一场景递推求解每个时段的二次规划模型以对各个时段的值函数进行训练，形成收敛的值函数，再代入预测场景求解以获得最终的优化调度方案。该方法实现了对随机DED模型各个场景和各个时段的解耦求解，将一个大规模优化问题分解为一系列的小规模优化问题，有效提高了对大电网随机DED模型的求解速度。以某一实际省级电网为算例，通过与场景法和鲁棒优化调度方法的比较验证了所提出模型和求解方法的正确有效性。