The Prediction System Of Power Emergency Supplies Based On The Relevance Vector Machine

Typhoons often occur in coastal areas of China,which leads to frequent destruction of power grid.To repair the power grid in time,provide normal power supply to users,emergency materials for power grid repairment is crucial.While catastrophe is abrupt and the degree of damage is unpredictable,it causes some variety and quantity for emergency materials reserve insufficient while some redundant,all these leading to the safety coefficient would decline and cause waste of funds.At present,the prediction for emergency materials of power grid is mostly based on experience and some statistical methods,most of these methods are subjective assessment of emergency materials based on the typhoon situation which lack of science.Found through research,the research data provided by Hainan Power Grid is small sample type.The Relevance Vector Machine(RVM)is adapted to solve this problem and has been widely used in the areas of speech,image processing,medical diagnosis,fault diagnosis and forecasting.With the analysis of the damage of power facilities in various parts of Hainan by typhoon events,in this paper,we come up with a power emergnecy materail prediction model based on RVM to predict demand through extract feature of typhoon events,the main contents are as follows:(1)Based on the analysis of the demand characteristics of power emergency supplies after the typhoon event,the main influencing factors of the demand are determined:the distance from typhoon center,rainfall intensity,wind speed,typhoon intensity,typhoon duration,local grid scale value.Then characteristic vector of emergency power supply has been put forward based on the typhoon disaster.(2)In this paper,we first extract the feature vector of power emergency demand in typhoon disaster,then,use genetic algorithm(GA)to optimize the kernel parameters and come up with a prediction model of electric power emergency based on simple kernel relevance vector machine(RVM).(3)Because the simple kernel prediction model is sensitive to data,it has poor generalization ability and poor robustness.In order to improve the generalization ability and prediction accuracy of the model,this paper proposes a prediction method based on the combined kernel function learning to estimate the weights of the various kernel functions by using the adaptive ability of different kernel functions.(4)Typhoons events continue to occur and the data could be constantly obtained along with time series,prediction models need to be continually learned.While above RVM prediction model relics on batch learning and does not have on-line learning ability.Therefore,we propose a prediction method based on online RVM,which is based on the fast marginal likelihood algorithm,it can guarantee the accuracy of the model and online adaptive ability.The experimental data set is obtained from the information of historical typhoon emergency power supply demanded by Hainan Power Grid Company,the South China Sea typhoon network and weather network,we analyse the prediction model by experiments.The experimental results show that the method proposed in this paper can predict the demand of emergency materials well.Finally,C#and SQL Server 2005 is used to implement the system.