Research On Short-term Load Forecasting Under The Big Data Environment For Intelligent Power

Load forecasting has always been an important work of power system.Accurate load forecasting can arrange the start and stop of power system generator sets economically and rationally,which is important to maintain the safe and stable operation of the power grid,maintain the normal production and life of the society and effectively reduce the generation cost.With the development of smart grid technology,advanced metering infrastructure(AMI)and a variety of monitoring systems are deployed on a large scale.Smart meter is an important part of AMI,and it can obtain the accurate user load during a certain time interval.The data generated by smart meters have fast speed and large size,and they are not analyzed deeply,which results in the waste of data.Therefore,it is of great significance to mine the values of user electricity data fully and study the short-term load forecasting under the intelligent power big data environment.Based on the characteristic that smart meters can obtain the detailed electricity data of users,this paper proceeds with a small amount of data and analyzes the similarity between users' behaviors of electricity consumption by load clustering.Then,this paper proposes a short-term load forecasting model based on OS-ELM,and simulation results have verified that the proposed model can improve the accuracy of load forecasting and further shows the relationship between clustering results and forecasting accuracy.In order to adapt to the big data environment of intelligent power,the short-term load forecasting model of parallel OS-ELM based on Spark is put forward,and it is proved that the model can still have a higher efficiency under the premise of guaranteeing the forecasting accuracy.The concrete work of this paper is as follows:1.Study the date type factor used in load clustering.The typical day loads of the user are calculated respectively according to different date types(ordinary workdays,day before holidays,holidays),and three typical day load curves are put together as the typical load curve.Then,the typical load curve is clustered to mine the similarity between user's behaviors.2.A short-term load forecasting model based on OS-ELM is proposed based on a small amount of user data.On the basis of load clustering,this load forecasting model is applied to different user clusters for load forecasting and system load forecasting is obtained by aggregating the partial load forecasting.The simulation experiment is carried out on the MATLAB platform,and it verifies that the proposed model is effective,and further shows the relationship between the forecasting accuracy and the number of clusters.3.For the massive user data,a short-term load forecasting model of parallel OS-ELM based on Spark is proposed.In order to adapt to the big data environment of intelligent power,the short-term load forecasting model of parallel OS-ELM based on Spark is put forward,and it is proved that the proposed parallel forecasting model can still have a higher efficiency under the premise of guaranteeing the forecasting accuracy.