Research On Data-driven Unit Commitment Optimization Under Dynamic Uncertainties

Unit commitment is one of the most important control process in power systems,which ensures the economical efficiency and stability of power systems by determining the on/off state and output of thermal units.Recent years,with the growing penetration of renewable energy such as wind power,uncertainties in the supply sides of power systems is increasing.At the same time,the popularity of new devices such as distributed power source,energy storage system and electric vehicle makes the energy consumption in power systems more flexible and changeable.Therefore,uncertainties in both supply and demand sides bring significant challenges to power system dispatch.Besides,with the development of smart grid,a large number of data about power plants and consumers will be collected to the dispatch center.In the new situation,how to utilize the real time collected data in system operation and construct unit commitment model to make power system operate economically,reliably and environment-friendly has become an important topic.This paper focuses on data-driven unit commitment optimization research.Firstly,to handle wind power and future load uncertainties,a data-driven approach for multiobjective unit commitment model was proposed.The non-parameter kernel density method was used to estimate the empirical distribution of wind power and future load.To make uncertainty representation serve for unit commitment,a new bandwidth selection method was proposed by optimize the bandwidth and system operation simultaneously.Besides,value-at-risk was utilized as a measurement for system reliability.Secondly,to capture the uncertainties of wind power and future load,a wind power interval forecast model based on deep learning methods was proposed.A novel criterionnamed prediction interval estimation error was proposed to make forecast models serve for unit commitment optimization.Finally,a rolling horizon unit commitment model was proposed.To integrate forecast models and unit commitment,a rolling forecast and rolling optimization framework was proposed.