Research On Optimal Dispatch And Resource-load Uncertainties Of Power Systems

With the large-scale integration of high-proportioned renewable energy,e.g.,wind and solar generation,and the initiatively connection of demand-side resources,e.g.,controllable loads,distributed energy storages and electric vehicles,the long-term sustainability and flex-ibility of modern power systems have enhanced hugely.However,with the cost of instant volatility and uncertainties.The increasingly complex power systems make the tradition-al power dispatch diffcult to be conducted efficiently.The dispatch information of power system operation becomes more diffcult to be obtained through fast computation.Also,the uncertainties from both resources and loads have an unknown negative effect on the power system operation and dispatch.In response to these problems,more computationally effi-cient power system dispatch control or optimization algorithms will be required.And it is crucial to model and analysis the uncertainties from both resources and loads from a prob-abilistic perspective,in order to transform the deterministic solution of power dispatch,to meet the new requirments of power dispatch under uncertainties.Based on these,this s-tudy focuses on the difficulties faced by power system dispatch under the situtions of high integration of uncertain renewable energy and demand-side resources.The main research contents and results are shown as below:According to the transient operation of power systems,a predictive control model for real-time economic dispatch with the consideration of transient economic benefit and op-eration security has been proposed.Solution of the predictive control model is a dynamic receding horizon optimization process.Conventional analytical solution of predictive con-trol is based on the partial derivative,which requires inverse matrix calculation and thus is comutationally heavy.The large matrix also often involves ill-conditioned data,which re-sults in failure of matrix inversion.So the conventional analytical solution can not satisfy the computational requirments of real-time economic dispatch.We propose a backwards square completion based model predictive control solution in this article.This method avoids the large matrix inversion by performing the square completion on predictive input recursively,and reduce the computational complexity of conventional analytical solution from O(n~3H_u~3)to O(n~3Hu)with respect to the control horizon Hu.The numerical simulations on IEEE standard power system cases verify the outstanding feasibility and computational efficiency of the proposed backwards square completion solution,which can improve the computation speed of real-time economic dispatch problem when the control horizon Hu is large.The fast solution of power system optimal power flow problem has been proposed.The optimal power flow problem,which is crucial for the power system dispatch,can be modeled as a nonconvex optimization problem with equality and inequality constraints in mathematics.Based on the Powerball optimization method,we improve the traditional itera-tive optimization solution methods like interior point method and Newton-Raphson method,and propose the fast solution algorithms for optimal power flow problem and power flow calculation,respectively.This provides methodological support for the fast computation and analysis in power systems.We then conduct numerical simulations on multiple widely accepted standard power system cases with different scales,to verify the proposed algo-rithms can help reduce the iterations required by the solution process of optimal power flow problem,thus accelerate the convergence speed.The simulation results on power flow cal-culation testify the proposed algorithm can accelerate the power flow solution,and at the same time,better deal with the ill initial conditions of power systems.The accelerated computing of power system probabilistic optimal power flow prob-lem with consideration of the wind power uncertainties and correlations from multiple wind farms has been studied.Real wind power data from multiple wind farms probabilisticly is modeled by the multivariate Gaussian mixture model.Then the model is improved ac-cording to the truncated characteristics of wind power distribution.Based on the traditional Markov chain Monte Carlo sampling method,the pseudo random numbers are replaced by low-discrepancy quasi random numbers generated by quasi Monte Carlo method,to enhance the computational efficiency of sampling procedure.A complete probabilistic optimal pow-er flow accelerated computing method is designed based on Monte Carlo simulation.Nu-merical simulations of modeling,sampling and probabilistic optimal power flow calculating are conducted on multiple publicly available real wind power datasets and widely accepted power system cases.Results indicate that the proposed probabilistic model has high mod-eling accuracy,and the proposed sampling algorithm can perform the probabilistic optimal power flow calculation efficiently.Probabilistic representation of power uncertainties based on nonparametric Bayesian modeling and inference is performed,and the data driven probabilistic optimal power flow online computing framework is proposed.According to the nonparametric Bayesian theory,a Dirichlet process mixture model is proposed for power data,and is improved according to the distribution characteristics of wind power.By adopting variational Bayesian inference,probabilistic representation of power uncertainties with high accuracy is performed without prior knowledge of the mixture component numbers.With the help of designed sampling algorithm,a data driven probabilistic optimal power flow online computing framework is proposed,which can handle the constantly updated power data,automatically complete the probabilistic modeling and probabilistic optimal power flow calculation.Thus the proposed framework can provide methodological support for the decision and risk assessments during the power system dispatch procedure under the power uncertainties.Above research results focus on the transient-state control,steady-state optimization and probabilistic analysis,respectively,and enhance the ability of dealing with difficulties faced by power dispatch while renewable energy and demand-side resources initiatively integrate into power systems.Finally,this thesis summarizes all the contents,and discusses the future research.