Power Analysis And Prediction Research Based On Big Data Technology

With the development of data storage technology and the popularity of intelligent monitoring devices,the scale and dimensions of grid power data have been expanded unprecedentedly.It is urgent to apply big data technology for in-depth mining and analysis to obtain higher power forecasting accuracy.Therefore,this paper takes grid electricity data as the research object,focusing on the cleaning of electricity data and the accurate prediction of electricity.The main research contents are as follows:1)Form a complete set of electricity data cleaning and standardization rules.Aiming at the problems of data outliers and data loss in the grid power database,an improved outlier detection and defect data recursive filling algorithm based on improved KNN algorithm is proposed.The maximum normalization is proposed for the problem of large difference in magnitude between power data.Standardized approach.2)Aiming at the problem of poor stability of the electricity curve and severe seasonal fluctuations,combined with the X-12-ARIMA program and the actual monthly fluctuation of China’s monthly electricity consumption,the industry’s electricity time series are seasonally adjusted to improve the stability of the electricity sequence.3)In order to reduce the complexity of modeling and calculation,an improved K-means clustering algorithm for power data is proposed.Aiming at the convergence problem of clustering algorithm,the optimal scheme of initial clustering center is given,including the screening of core clustering objects and the method of selecting the initial clustering center by using distance weights.In order to make the industries with similar fluctuation trends into the same category,the improved distance operator with weights is proposed,and the better clustering effect is obtained by the weighting of the Euclidean distance operator and the curve similarity operator.The contour coefficients are introduced to evaluate the clustering results,and the optimal distance operator and cluster number are obtained through calculation.4)Define the level of the electricity data,and propose a superior electricity prediction algorithm based on the seasonal decomposition and clustering results of the lower electricity curve.The season sequence is adjusted into the seasonal period item S and the trend cycle-random item TCI by seasonally adjusting the pre-processed industry electricity data.The seasonal period term is predicted by ARMA.The adjusted TCI curve needs to be clustered in the whole industry first,and then the prediction method which is most suitable for each class is selected according to the clustering result to predict separately.Finally,the predicted values of the seasonal components of each industry are combined to obtain the predicted value of the monthly electricity consumption in the region.