Research Of User Behavior Analysis Method For Mass Power Consumption Data

With the popularity of smart meters,the residential electricity data can be collected efficiently.Data mining technology is used to identify residential electricity load model and analyze user power consumption behavior and it has guiding significance for load forecasting,load control and time sharing price making.At the same time,the scale of the smart meter is growing,which poses new challenges to the storage and computing performance of the traditional power data.In recent years,the clustering algorithm and deep learning algorithm based on machine learning theory and its improved algorithm have been widely used in residential electrical behavior analysis and load pattern recognition.In view of the problem that the existing power load model focuses on industry,agriculture,and other large commercial users,but not on the residents,this thesis put forward a method for the research of the residential electrical load model based on clustering and deep belief network.Firstly,For a fixed feature weight is not flexible enough and the problem that K-means clustering needs to determine the number of clusters in advance,and makes corresponding improvements.So dynamically different weights to attributes are assigned and the number of clusters is adjusted dynamically.Than considering the challenge of mass power consumption data,this thesis needs to consider how to reduce the computational complexity of data mining as much as possible.If the power load patterns of each household in a residential district are classified one by one,it will increase the computational complexity of data processing.Therefore,only one kind of typical power load pattern is classified,which can represent the load patterns of all the residents,thus reducing the complexity of data processing.In addition most of the current clustering algorithms are used to classify the load patterns,whose result is in a variety of clutter,the purpose is not clear enough to increase the operating costs of the power company.This thesis adopts the deep belief network classifier to classify residential electricity load pattern according to the classification standard.Lastly,the experiment was conducted according to the electricity data,and it proved the validity and superiority of the model based on improved k-means clustering algorithm and the deep belief network classifier.In addition,there are many iterations in the machine learning algorithm.Spark,a distributed memory computing framework in cloud computing,can efficiently process iterated data and improve the execution performance of the algorithm.Aiming at the problem of high computational complexity of deep belief network,By studying the architecture of cloud computing system and adopting the Spark distributed memory computing framework in cloud computing technology,the deep belief network is improved in parallel to improve the execution performance of the algorithm.Than Cloudera's CDH5 version is used to build the cloud computing platform on the laboratory server,and the performance of the proposed algorithm is tested.