Research On Virtual Power Plant Optimization Scheduling And Bidding Strategy Model Based On Distributionally Robust Optimization

In response to the dual challenges of depleting fossil fuel reserves and the increasingly severe environmental issues,China is actively promoting the development of renewable energy,committed to achieving a green transformation of the energy structure.However,the output of renewable energy is susceptible to weather and environmental factors,and the uncertainty of its power generation presents significant challenges to the stable operation of the power system.In this context,virtual power plant(VPP)has emerged as a solution.By comprehensively coordinating various distributed energy resources,it not only facilitates the absorption of renewable energy but also significantly enhances the overall efficiency of energy utilization.Nevertheless,VPP face a highly challenging problem: how to coordinate distributed energy resources with different technical characteristics and operating modes while ensuring the safety of system operation and maximizing overall benefits.In response to this problem,this paper takes the optimization scheduling and bidding strategy model of VPP as the research object.It thoroughly investigates unit commitment(UC)modeling methods,VPP optimization scheduling models based on distributionally robust optimization(DRO),VPP load resource assessment algorithms,and bidding strategy models based on DRO.The main research findings are as follows:(1)A systematic and universal UC problem modeling method is proposed,and a tight three-period UC problem model is constructed using this modeling method.First,by introducing additional state variables,the operational states of the units are described comprehensively,and a high-dimensional UC model with strong tightness is constructed.Then,leveraging the linear correlation between state variables,the model is projected from the high-dimensional space to a lowdimensional space to ensure simplicity.Based on this modeling method,tight three-period unit output upper bound constraints and three-period up/down ramping constraints are proposed,leading to the construction of a new threeperiod UC model.Finally,the proposed three-period UC model is tested on 56 test instances.The experimental results show that compared to the four commonly used UC models,the proposed model has the strongest tightness and the highest computational efficiency.(2)A multi-stage semi-anticipative optimization scheduling model for VPP based on DRO is proposed.Firstly,by comprehensively analyzing historical and forecast data of renewable energy,a moment-based semi-anticipative ambiguity set is constructed,which is then used to formulate a multi-stage DRO model for the VPP scheduling problem.Moreover,principal component analysis is employed to project the multi-stage DRO model into a low-dimensional space,effectively reducing the computational complexity of the model.Finally,the proposed multi-stage semi-anticipative DRO model and its corresponding projection model are tested on six IEEE test cases.Experimental results indicate that the multi-stage semi-anticipative model can dynamically adjust the robustness and economy of the decision-making scheme according to the credibility of the renewable energy forecast data.Additionally,the projected model exhibits excellent computational performance.(3)A non-intrusive load resource assessment algorithm based on DRO feature extraction technology is proposed.Firstly,by integrating DRO theory and linear regression algorithm,a noise-resistant linear regression algorithm based on DRO is developed.Then,based on this algorithm,feature extraction algorithms are constructed for the upper and lower bounds of appliance power,power fluctuation features,and minimum active time features,respectively.These algorithms are capable of extracting key feature information from noisy load data,providing effective information support for load disaggregation and resource assessment.Subsequently,inspired by the modeling approach of the UC model,power features constraints,new state variable constraints,new linear state transition constraints,new minimum active time constraints,and a new objective function are proposed,leading to the formulation of a novel nonintrusive load disaggregation model based on mixed-integer programming.By integrating the load feature extraction algorithm with the non-intrusive load disaggregation model,precise assessment of load resources for power users has been achieved.Finally,the performance of the proposed non-intrusive load resource assessment algorithm is tested on three different datasets.Experimental results show that the algorithm demonstrates satisfactory computational efficiency and accuracy.(4)A new two-level optimization model for VPP bidding strategies and a multi-stage DRO semi-anticipative bidding strategy model for VPP in the dayahead energy market are constructed.Firstly,based on the research findings of the UC model,tight three-period unit output upper bound constraints and threeperiod up/down ramping constraints are introduced into the traditional two-level model,thereby obtaining a new two-level optimization model for the bidding strategy problem of price-making virtual power.Subsequently,by utilizing dual theory,the big M method,and the proposed framework of the multi-stage DRO semi-anticipative model,a multi-stage DRO semi-anticipative bidding strategy model for price-making VPP is formulated.Finally,the proposed semianticipative bidding strategy model is tested on three test cases.The experiments demonstrate that the model can provide effective decision-making solutions for VPP.