Research On Perception System Of Distribution Network And Short-term Load Forecasting

With the development of science and technology,electricity is increasingly demanded by industrial production activities and people’s daily lives.As an important part of the power system,the distribution network bears important tasks in the distribution and use of electrical energy.With the continuous increase of new loads on the demand side of the distribution network,the uncertainty of the distribution network is becoming more and more serious.The new load demand situation puts forward higher requirements on the accuracy of load forecasting.Short-term load forecasting is based on historical load data,with the establishment of a mathematical model that can fully reveal the load law as the core,in order to obtain the best forecasting results.To improve the accuracy of load forecasting,not only the data collected by the system is required to be accurate and comprehensive,but also the forecasting algorithm must be continuously optimized.Therefore,the use of the latest technologies in sensing,communication and data mining,etc,to collect load data that can represent the most real situation of the distribution network,to carry out load characteristics analysis and prediction has certain research value.In this thsis,the Lo Ra(Long Range)technology and Cat Boost(Categorical Features Gradient Boosting)algorithm were deeply analyzed and applied to the real-time perception system of distribution network and short-term load forecasting.Firstly,in view of the difficulty of obtaining comprehensive and accurate distribution network data in the traditional real-time perception system of distribution network communication method,a scheme of applying Lo Ra technology to the realtime perception system of distribution network is proposed,and a real-time perception system of distribution network based on Lo Ra is built.The results show that Lo Ra technology can meet the data collection requirements of the distribution network awareness system in terms of transmission distance,network capacity,power consumption,etc.The real-time perception system of distribution network based on Lo Ra can adapt to the complex working environment and realize "full coverage and full collection" of the distribution network,which can provide accurate and comprehensive data support for load forecasting.Secondly,according to several factors such as climate and date type that affect the distribution network load changes,the distribution network load characteristics analysis and selection were carried out,and the original data is preprocessed to provide data guarantee for the subsequent distribution network load forecast.After that,the basic principle steps and requirements for short-term load forecasting through load data are introduced.Finally,in view of the problems of over-fitting and conditional migration in traditional prediction algorithms,the advantages and disadvantages of BP neural network,decision tree,and GBDT(Gradient Boosting Decision Tree)algorithm are analyzed,the advantages of Cat Boost algorithm in load prediction was elaborated,and introduce Cat Boost algorithm into the field of short-term load forecasting.Combined with distribution network load data and its characteristic analysis,a short-term load forecast model based on Cat Boost algorithm is established,and the forecast results are obtained.Experimental results show that the load forecasting model of distribution network based on Cat Boost algorithm can get ideal forecasting results and can improve the accuracy of load forecasting.