Research On Short-term Load Forecasting And Design Of Data Platform Based On Spark

The real-time of electric energy makes the power demand change dynamically.Power production companies usually adjust their power production plans according to power demand to ensure the safe operation of the power system and the economic benefits of power companies.Therefore,how to accurately perform load forecasting becomes a problem worthy of study.The past decade has been a decade of vigorous development of smart grids.With the application of high-speed communication network and advanced sensing equipment,the realtime power consumption information of users can be obtained by power enterprises.The growth of data scale provides the possibility for higher-precision forecasting.At the same time,the huge amount of calculation also brings challenges to load forecasting.In the context of electric power big data,the traditional methods of load forecasting have problems such as slow processing speed and lack of accuracy.This paper conducts research on short-term load forecasting and electric power big data:1.A short-term load forecasting model combining STL,adaptive particle swarm optimization and support vector regression machine is proposed.Firstly,the load is divided into periodic item with day as cycle and residual item through STL,and then optimizes the parameter adjustment work of the support vector regression machine through particle swarm algorithm,and predicts the remaining items through the obtained model.The result is aggregated with the period term to obtain the final load forecast value.2.Based on big data technology,a power data platform is designed.In order to meet the requirements of the calculation and storage of electric power big data,a distributed cluster of one master and two slaves was built.The migration of collected data was completed through Sqoop,the data was cleaned through Hive,and the parallelization of prediction algorithms was realized through Spark.Finally,the obtained prediction results and statistical historical data are visually displayed through Tableau.Through the above research,this paper solves the rapid prediction of load data under big data power consumption samples,and verifies the accuracy and efficiency of the prediction model after parallelization.At the same time,the designed platform can also complete a series of tasks such as the collection,processing,aggregation and display of electric power big data.The visual display interface also makes the application of data more convenient.