Multivariate Time Series House Price Forecast Method Based On Neural Network

In recent years,China's real estate industry has become an important part of the national economy.Due to the rigid demand for people's living and the financial attribute of investment,the house has received important attention from society.How to accurately forecast house price has also become an important research topic.Early time series forecast of house price used one-dimensional price for autoregressive forecast,which didn't take into account the influence of multivariate features related to the house price,but these multivariate features were interrelated with house price on the timeline.Therefore,to use all of the multivariate features of house price and the information of time dimension,improving the accuracy of house price forecast,this thesis forecasts the house price from two aspects: time change and multivariate features that affect house price.The main research contents of this thesis are as follows:1.This thesis uses multivariate time series house price forecast method based on neural network in Beijing,paying attention to the short-term local changes and the long-term dependency trend changes of the house price.First,one-dimensional convolution neural network is used to automatically extract the multivariate time series features related to house price.Then,the multivariate time series features are convoluted with multiple convolution kernels to obtain the feature vectors.Finally,the feature vectors are sent into the Long Short-Term Memory model,and the method is used to forecast house prices.In this thesis,the multivariate time series house price forecast method based on neural network is used to carry out experiments on Beijing house price dataset.,and the experimental results showed that the method can extract the information of multivariate time series features effectively,and fit the trend of short-term local changes and long-term dependence of house price.2.On the basis of the multivariate time series house price forecast method based on neural network,to further improve the learning ability of the model in the short-term inflection point changes of house price and reduce the forecast error,this thesis introduced the attention mechanism into the method and assigned different weights to each time step.Firstly,the attention distribution of each time step was calculated by using the full connection layer,and then the corresponding weight was given to each time step to enhance the fitting ability of the house price on the timeline at the inflection point.Finally,the model with the attention mechanism was trained to forecast the house price.This thesis used the multivariate time series house price forecast method based on neural network which introduced the attention mechanism to further experiment on Beijing house price data set,and the experimental results show that this method can not only learn the long-term dependence trend of house price change but also learn the inflection point information of short-term local changes of the house price further,improving the accuracy of house price forecast and the generalization ability of the method.