Research On The Ultra Short-term Load Forecasting Method Based On The Improved Data Streams Technology And Wavelet Packet Decomposition

With the development of modern power system construction, national gridinterconnection is carrying out nowadays. The more complex grid structure andoperation mode adds more challenge and threat to the power system safety and electricenergy quality. Ultra short-term load forecasting uses the newest load information totrack the real-time variation of power load. It is the basic foundation and significantguideline for dynamic grid security detection, automatic generation control andemergency state strategy. Fast and accurate forecasting result can help electric powersector to regulate power frequency balance and protect power system safety andeconomical proficiency. Hence, it is key and meaningful to study the ultra short-termload forecasting method with real-time and accurate advantages.Based on physical mechanism of load structure, this thesis studies the relationshipbetween load characteristic and related influence factors. According to each influencefactor, influence regularity and mechanism of them on load variation are investigated.Research result shows that: time factor has large influence and presents obviousperiodicity; weather factor has some relationship with load variation; uncertain factor’sinfluence has strong fluctuation and its regularity is difficult to obtain. Hence, this thesisanalyze the corresponding load forecasting mode and method according to differentload variation regularities.It builds up ultra short-term load forecasting mode based on the improved on-linesegmentation of data stream. By using the data stream real-time processing technique toforecast ultra short-term load, its fast segmentation forecasting ability avoids repetitivemodeling, enhance real-time property. Then, it extracts short-term load forecastingresult with the impact of weather factor and load periodical characteristic and amendsthe segment point. This can not only increase history information utilization, but alsoimprove segment point forecasting accuracy and guarantee real-time efficiency. Finally,By actual example and simulation, the forecasting accuracy and speed of the model issuperior to other traditional ultra short-term load forecasting method, which also hasstrong adaptation during the abrupt varying weather.Besides, this thesis studies the ultra short-term load forecasting method with theconsideration of load random component. By the further disassembling of load randomcomponent using wavelet packet analysis, it can analyze random component characteristic. Based on the disassembled load subsequence component characteristic,each the forecasting mode is established. The forecasting data of each group isreconstructed to gain the final load forecasting result. By example analysis andsimulation, this method has comparatively higher forecasting accuracy and efficiency.