Research On Short Term Power Load Forecasting For Smart Campus

While the large power grid is gradually moving towards intelligence,digitization and informatization,the planning and construction of smart campus is also in-depth.As a key link,the rationality of power distribution plays an important role in the construction and improvement of smart campus.As the power load forecasting can provide an effective guidance scheme for power distribution,this paper takes smart campus as the engineering application background,in order to improve its short-term power load forecasting accuracy as the research purpose.Firstly,a campus typical scene analysis method based on pcag-km is given;secondly,a combination method which is used to meet the short-term power load forecasting based on snap-scheme is given;finally,a simulation example is given.A short-term load forecasting scheme for typical scenarios of smart campus is presented in this paper,and a visual platform for power load forecasting of smart campus is designed and implemented.(1)A typical scene clustering method based on PCAG-KM is given.With the continuous development of smart campus construction,the power load in campus is becoming more and more diversified.The power load of each building in campus is not only related to each other,but also has its own uniqueness.Therefore,it is of great significance to mine and summarize the typical scenes of buildings in the campus,and establish prediction models for each typical scene in the follow-up prediction.However,the load data in smart campus usually has the characteristics of multi-dimensional and multi noise.The traditional clustering algorithm has limitations in this application,and can not provide accurate and stable clustering results.In the process of clustering,the quality of clustering algorithm is very important for the accuracy of typical scene mining.Therefore,this paper proposes a typical scene analysis method based on pcagkm.Firstly,the correlation among the features in the original data multi-dimensional space is removed,and then the optimal number of typical scenes K is determined after processing it into the low dimensional space.Then the optimal initial centroid is determined by using the global optimization ability of GA.Finally,the algorithm is applied to a university in Northeast China to analyze the typical power consumption scenarios.The results show that the algorithm has better ability to select typical scenarios,and has the anti noise and stability required by practical engineering.(2)A STLF model based on snap-scheme is given.Aiming at the problem that traditional models are difficult to meet the accuracy and stability requirements of current power load forecasting tasks,the combination model of STLF based on snapscheme is given.Firstly,a VMD algorithm with snap-scheme(VMDsnap)is given to overcome the shortcomings of the VMD algorithm,such as the K value which has an impact on the performance of subsequent forecasting tasks and the IMF which can not be decomposed by adding the algorithm to restore to the original sequence.Secondly,a VMDsnap algorithm which can effectively take into account the history is constructed by combining VMDsnap algorithm with LSTM-network.Finally,the model is applied to four regional energy markets in Australia.The results show that the combined forecasting model not only has higher forecasting accuracy than other traditional models,but also has certain generalization ability.(3)A short-term load forecasting scheme for smart campus is proposed and a visualization platform for power load forecasting is designed and implemented.In view of the diversity of power load in various power consumption scenarios in modern smart campus,the traditional mathematical statistics based forecasting model and single machine learning forecasting model are difficult to meet the requirements of forecasting accuracy and the robustness of the model needs to be improved,a shortterm load forecasting scheme for typical scenarios of smart campus is given.The VMDsnap-LSTM prediction model is established for K typical power consumption scenarios,which can not only consider the rules of historical load data in smart campus,but also consider the differences of power consumption habits between typical scenarios and other scenarios.The results show that compared with other forecasting methods,the accuracy of this forecasting method is improved greatly.Then,based on the prediction scheme,the visualization platform of power load forecasting in smart campus is designed and implemented,which provides load data analysis service and short-term load forecasting service with visual interactive interface,and more effectively supports power dispatching,operation and maintenance,equipment repair and other power tasks in smart campus.