Research Of Load Foresting Based On Big Power Data

As the advancement of China's electric power system reform,the electricity market environment will make a big difference.There will be more electricity companies entering the market in the future.In order to improve their core competitiveness,the power sales company needs to scientifically analyze the characteristics of the future load curve and accurately forecast the load in the future power market.And with the rapid development of the construction of Energy Internet and power data,the expansion of grid scale and the expansion of the amount of power information collection have objectively brought higher requirements on the accuracy of load forecasting and the computational efficiency of prediction algorithm.Load forecasting plays an important role in the normal operation and scheduling of the energy Internet.The load forecasting based on the power large data technology makes it more realistic for the comprehensive utilization of mass real-time data,large amount of historical data and weather meteorological data.In this paper,the detection interval of power load is smaller and smaller,which makes the load data change towards high dimension,as a result the difficulty of load curve clustering is increasing.In order to solve this problem,this paper proposes a method to map the data into high-dimensional space by using kernel method.The kernel principal component analysis and the reduction matrix are used to optimize the calculation of the method.Then the basis of the algorithm improvement and the deduction of the related theory are expounded.Finally,the experiment show that the effect of clustering number and output dimension on clustering result,and the improved clustering algorithm can effectively improve the accuracy of load curve clustering.Secondly,in order to solve the problem of repeated training of load data for massive data,this paper introduces the incremental method into the load forecasting.The data Updates are continually occurring over time,and as time goes by,the value of the data at one time becomes smaller and smaller.Once the forecasting model is established for power load,the model can not add the latest value with time,so that the forecast model can not reflect the latest time series information,in result reducing the accuracy of prediction.To solve this problem,this paper proposes a BP neural network to increase the level of hidden nodes to achieve BP neural network incremental training,through training the BP neural network with the latest data,the experiments results enhance the capacity of handling massive load data and load forecast accuracy.Finally,in order to make a better solution to the problem of power load data quantification and high dimension,this paper using the latest research hotspot Gradient Boosting Decision Tree algorithm,based on large data Spark platform,builds a load forecasting model,comparing with random forest algorithm.The experiment uses random forest and Gradient BoostingDecision Tree to train the load data separately.The results show that Gradient Boosting Decision Tree is better than random forest in the accuracy of load forecasting and the efficiency of calculation.