Two-stage Robust Day-ahead Dispatching Optimization Of Distribution Network Based On Load And Wind Power Forecasting

As the power the deepening of power system reform,the promotion degree of power spot market and the penetration rate of intermittent distributed generation have been significantly improved.The traditional deterministic simulation of load,wind power,photovoltaic power and electricity price can not longer satisfy the basic requirements of present distribution network operation tasks.In the increasingly complex background,how to effectively improve the prediction accuracy of uncertain factors and dispatching decision-making ability of distribution network is a challenge that must be overcome by the fine control technology of distribution network.For this reason,this paper builds a short-term prediction model with certain prediction accuracy as the data base,and then uses the robust optimization technology to establish the uncertain decision model,and then reasonably formulating the day-ahead dispatching of the distribution network to achieve the lean power distribution and ensure the stable and economic operation of the power grid.This paper first builds an optimal power flow algorithm based on second-order cone relaxation technique,which is suitable for active and reactive power flow calculation.Then,the Mc Cormick convex envelope method,auxiliary variable method and Big-M method are applied to deal with the nonlinear factors in the model,so as to effectively transform the dispatching optimization of distribution network into mixed integer linear programming problem,and lay a foundation for the uncertain decision making model.For the prediction model,based on the point prediction model of neural network,this paper simply and effectively transforms the point prediction results into high-quality interval prediction through the proportional coefficient method.In the aspect of data mining,this paper uses de invalidation processing technology to eliminate redundant data information,uses normalization technology to effectively avoid over saturation of neurons,and uses similar-day method of choosing to screen out effective information with high correlation with forecast day from historical data.In the aspect of model improvement,genetic algorithm is used to determine the initial weights and thresholds of the prediction network to minimize the prediction error caused by human.Aiming at the problems of complex calculation,strong hypothesis and long prediction time in the existing interval prediction methods,this paper proposes a simple and effective interval construction method.Firstly,the comprehensive index for evaluating the quality of the prediction interval is given.With the help of the optimization algorithm,the optimal proportional coefficient is obtained.the optimal proportion coefficient is obtained,and then the same proportion of the point prediction result is obtained through the proportion coefficient Finally,a high quality prediction interval is constructed.taking a certain area and wind farm in Fujian as an example,the simulation results show that the similar-day method of choosing and improved neural network can effectively reduce the prediction error;the width of prediction interval provided by proportional coefficient method is small enough,and the prediction interval can basically include the observation value of the prediction points,which can effectively provide data support for the two-stage robust optimization algorithm.For uncertain decision-making model,the interval form is used to describe the distribution of uncertain factors according to the short-term prediction results.Based on the zero-sum game,Benders decomposition method and outer approximation method,a two-stage robust optimization model is established.Aiming at the min-max-min problem in the model,Benders Decomposition Method and strong duality theory are used to divide the problem into primary and sub problems for iterative solution.The dual cone and outer approximation methods are introduced to solve the second-order cone constraints and bilinear elements of the subproblem.In the first stage,the variable is the Boolean decision variable of the charging and discharging state of the energy storage device;in the second stage,the power distribution of the energy storage device,the controllable distributed power supply and the power purchasing power of the main network adapt to the worst uncertain environment.Finally,an improved IEEE 33-node power distribution system is taken as an example to simulate.The simulation results show that this model not only meets the requirements of the accuracy of the slack error,but also can formulate the most reasonable power distribution scheme of the micro gas turbine,energy storage device and reactive power compensation device,so as to provide a decision-making plan for the day-ahead dispatching decision makers of the distribution network to give consideration to economy and conservatism.