Short-term Parking Space Prediction Based On Wavelet ELM Neural Networks Method And DBN With Multitask Learning

With the development of country's economy,more and more private cars have entered thousands of ordinary families.The problems of city traffic is becoming more and more serious.Providing reasonable parking information can reduce the time of finding parking,relieve the traffic pressure,reduce air pollution and improve the utilization of parking space.So predicting real-time free parking space accurate is becoming the major research direction of solving above mentioned problems.There are always the situations that drivers find a parking space when they start but when they get,the parking space is full.This kind of prediction can avoid the situation which mentioned above and congestion sections.At present,we use BP Neural net and derivation to predict the parking information.First,we use phase-space reconstruction to reconstruct primary data in order to get training data of relationships within data.This step we called Data Preparation.Then the processed data are used to be the training data of neural network.The widespread adoption is Surface Web,such as Back Propagation,Wavelet neural network and Elman network.Most of the methods are based on improved method of BP.This text summarized the key issues of those methods and improved the methods in order to get fast and accurate prediction model.First,the text used wavelet ELM neural network to predict free parking spaces.In order to get the best prediction model,we combined veracity of wavelet prediction with rapidity of ELM.We use phase-space reconstruction to arrange the primary data to get the correlation of training data.And then using the wavelet analysis to get the sub-sequence patterns which are influenced by variety of factors.We used the fast accurate fitting ability of ELM to predict each subsequence.At last,we made an integration of the prediction of each subsequence and determined results of prediction.So far,many prediction methods mostly used the Surface Web,but rarely used deeper learning.So this text combined Multi-Task learning with DBN neural network to predict the problems we have researched.Multi-Task learning is a fitting data method.At the same time,Multi-Task learning can learn more tasks and extract the common features.First,we got the features of data by using the analytical ability of DBN neural network,then we improve the fitting ability of neural network by using the common characteristics between related tasks of Multi-Task learning.Finally we can improve the effect of prediction model.Experiments have shown that this method proved the results of prediction.This paper analyzed the results of the experiments.The prediction methods of this text have better effect of free parking problems and reach the expectation of free parking space.The algorithm in this text has big promotion in predictive ability.