Study On Short-term Load Forecasting Of Intelligent Distribution Network

Intelligent distribution network，which is based on a flexible, reliable,efficientgrid structure and a high reliable security communication network,supports flexibleadaptive fault and self-healing,and meets the high permeability of distributedpower,energy components of access requirements and user requirements for powerquality. Load forecasting of distribution network is the basis for distribution networkplanning,the load forecasting accuracy will directly affect the planning result andpossible degree. In recent years, many scholars at home and abroad have doneextensive research on load forecasting, but with the power grid enterprises enteringthe electricity market, market economies has introduced new requirements on theaccuracy of load forecasting precision and load characteristics analysis technology,and thus the research on load forecasting is constant reinventing.Through the analysis of support vector regression principle, the support vectormachine has better performance and accuracy in small samples, generalization ofnonlinear problems. In order to reduce algorithm complexity,this article uses leastsquare support vector regression algorithm instead of the standard support vectorregression algorithm. In practical applications, least squares support vector machinealgorithm parameters are selected by experience,the parameter selection has a greatinfluence on prediction accuracy of the forcast results. Particle swarm optimizationalgorithm has global search capabilities to overcome the inefficient and blindness ofleast squares support vector machine parameter calculation.Short-term load forecast using the load sequence has both volatility and a specialperiodic, which can be as a superposition of multiple frequency component,and eachcomponent is showing a similar cycle changes,similar frequency characteristics andconsistent variation. Taking into account the analysis of wavelet has a uniqueadvantage in time-frequency domain,this article submits the combination forcastingmodel of wavelet transform and particle swarm optimizing the square support vectoralgorithm parameter.In selecting the same training sample and prediction samples,theprediction results are compared with BP neural network and a single support vectormachine forecasting method,simulation results show that the combination forcastingmethod has high precision and is an effective method for short-term load forecasting.