Design And Implementation Of Power Load Forecasting Platform Based On Spark

The development of social economy is inseparable from the reliability guarantee of power grid.Accurate load forecasting plays an important role in ensuring the stable dispatching of power system and its economic,safe and reliable operation,which is the most critical part of distribution network planning.With the development of power grid intelligence,the data in power grid is becoming more intelligent and information-based,which provides massive data samples for power system load forecasting and lays a foundation for improving the accuracy of load forecasting.Based on the big data distributed computing cluster,this paper proposes a study of power load forecasting based on combined deep learning algorithm,which requires the stability and reliability of the system,fast operation speed and high accuracy.The main research contents are as follows.(1)Research on distributed computing based on big data technologyAccording to the characteristics of massive and heterogeneous power grid data,a distributed big data cluster based on Hadoop and spark is constructed to process power grid big data.By deploying Hadoop big data ecosystem,the whole process of power grid data collection,transmission,processing,calculation and storage is realized.The data mining and analysis of power load forecasting is realized by spark engine based on memory computing and its core components.(2)Research on power load forecasting based on combined deep learning algorithmThe factors affecting the power load are divided into timing factors and non-sequential factors.Therefore,this paper proposes a new method of forecasting electric load based on pre training GRU-Light GBM,which is used to model both temporal and non-sequential features.The bi-directional circular GRU network is used as the time series feature extractor,considering the influence of load factors in historical period and future period on the predicted load,fully extracting the effective potential relationship between the load feature data.The extracted time series features are combined with the non-time series features which affect the power load,and put into the Light GBM model which has good effect on the regression forecasting problem and fast operation speed for model training and forecasting.Especially,in the design of algorithm,this paper fully considers that load forecasting is a multi-cycle task.Different time periods provide important information support in all aspects of power grid operation,so an algorithm framework is proposed to adapt to different periods of load forecasting.In the case that the algorithm framework does not change,the characteristics of different dimensions of the change can achieve the corresponding prediction,which is very efficient and fast.(3)Research on Visualization Technology of load historical data and load forecasting resultsIn order to make the platform users more convenient and intuitive to understand the data,the power load forecasting platform is designed to solve the problem of power related data visualization.The platform uses Java technology to build a data interaction and visualization platform based on B / S architecture,which provides a convenient way of human-computer interaction and reduces the complexity of operation.