Research On The Interval Probability Estimation Method For Wind Power Ramp Events

With the increasingly severe energy and environmental problems,developing the renewable energy represented by wind energy has become a broad consensus of all mankind.With the gradual increase of wind power penetration,the inherent randomness and volatility of wind power generation have brought about negative impacts on the safe and stable operation of power system.The grid-connected wind power in China has the characteristics of large-scale and high concentration,in which context,the wind power ramp events(WPRE),characterized by the large fluctuations of wind power output in short periods,have attracted extensive attention of researchers in recent years.Realizing the accurate assessment and reliable early warning of WPRE is of great significance for improving the safety and stability of power grid operation,reducing operation costs,promoting wind power consumption as well as enhancing the economic and environmental benefits of wind energy.According to the different prediction objects,the researches on WPRE prediction can be divided into two categories.One is committed to predicting the values of ramp characteristics,such as the magnitude,duration,ramp rate and ramp timing,so as to warn and describe the severity of the approaching ramp events.The other regards WPRE as a multi-state random variable,predicts the occurrence probability of each ramp state,and identifies and warns the state with the highest occurrence probability.The traditional prediction methods can be roughly divided into the indirect and direct prediction methods.The former first predicts the power series,and then extracts the prediction results of ramp characteristics or ramp state.The latter mines the direct mappings from the weather prediction to the ramp characteristics or the occurrence probabilities of every ramp states.Currently,there are following issues in WPRE prediction:?The indirect prediction methods need to go through the weather-power conversion process,and it is difficult to get rid of the negative impact of the accumulated errors generated by this link.?In the direct prediction methods,due to the low frequency of ramp events,the collected historical observation samples are difficult to meet the requirements of adequacy,integrity and accuracy when mining the mapping relations.?The uniform threshold in ramp events definition cannot be applied to the different wind farms environments or the operating conditions of the same grid at different times.Firstly,this paper establishes a Naive Bayesian Network(NBN)to predict the interval probability distribution(IPD)of WPRE.This method constructs an NBN structure based on the meteorological elements that markedly affect the WPRE occurrence,and uses the extended Imprecise Dirichlet model to quantify the conditional dependences between meteorological elements and ramp events.Under the condition of obtaining the weather prediction,the IPD of the multi-state WPRE is estimated through the Bayesian network probability inference process.The method establishes a direct mapping from the weather prediction to the occurrence probabilities of every ramp state,as well as introduces the concept of imprecise probability and uses the interval probability prediction results to quantify the uncertainty involved in the statistical estimation based on finite samples.This method improves the credibility of the prediction results and enriches the WPRE warning information.In further research,with the Maximum Weighted Spanning Tree algorithm and the Greedy Search algorithm,the Bayesian network structure learning module is added to the prediction process.Based cm the dependence mining,the network structure that best fits the observation samples is automatically constructed,so as to flexibly present the differential relationships between the meteorological elements and the ramp events for different wind farms.The case studies based on the real measurements show that,by optimizing the network structure and the parameters automatically,the IPD results predicted by the proposed method can adaptively cater to the conservative,risky and neutral prediction attitudes predefined by the predictors.Besides,for the condition with scarce samples or even no samples,the proposed method can still present excellent prediction performance.