Research And Application Of User Housing Vacancy Prediction Based On Power Big Data

In the face of the power industry is generating a large amount of electricity consumption data at any time,mining reliable and effective information from massive data is currently an important task.From the perspective of ordinary households,electricity consumption data and housing vacancy are inseparable,and the model predicts housing vacancy to further expand the depth and breadth of services in the power industry.Based on the power data generated by the electricity supply companies,we design a fusion algorithm based on the support vector machine(FASVM),a housing vacancy prediction model as well as a platform based on LSTM neural network focusing on different scenarios.Aiming at the problem that the traditional machine learning algorithms have low accuracy in prediction,this thesis proposes a two-tier fusion algorithm FASVM.The first layer of FASVM is trained by the base classifiers,which consist of logistic regression and decision tree model.Next FASVM is retrained by the support vector machine in the second layer based on the output of the previous layer.It is worth noting that FASVM can further explore the correlation among data features,and can effectively improve the accuracy of user housing vacancy prediction.Considering the time series correlation among users’ consumption data,which is not insufficiently considered by traditional machine learning models,this thesis proposes a housing vacancy prediction model based on the LSTM neural network,where we use the LSTM to exact the correlation among data features of input data over a long period,so as to continuously improve the prediction effect.Through sufficient experimental,we find the prediction results of that the FASVM algorithm and the LSTM neural network model are 3% and 10.1% higher than the traditional machine learning.FASVM algorithm and LSTM neural networks have significant advantages in the field of user housing vacancy prediction.However,compared with the FASVM,the LSTM neural network model has a large number of network layers,a large amount of computation,and high requirements for computer parallel processing.Thus the LSTM neural network model is more suitable to process large-scale user power consumption data.For small-scale data analysis,FASVM algorithm has more advantages,and it also maintains good performance in terms of time efficiency,and is highly practical.In addition,this thesis designs a housing vacancy prediction system based on LSTM neural network model,which mainly includes housing vacancy prediction of single-user and batch users.The system interface is friendly and intuitive.Such system can assist the government and other relevant departments in fully understanding the situation of local housing vacancy and making relevant decisions support for the real estate industry.