Research On Prediction Method Of Release Time Interval Of Big Location Data Based On Deep Learning

Real-t ime release o f big locat ion data stat ist ics can help intelligent transportatio n systems realize management and dispatch,help users plan reasonable travel t ime and routes,alleviate traffic congest ion and reduce unnecessary waste.Locat ion big dat a has the characterist ics o f rando mness,nonlinearit y,and temporal and spat ial correlation.It is difficult to effect ively obtain the essent ial characterist ics o f big locat ion data with the help o f tradit ional short-term traffic forecast ing methods.Deep learning is data-driven as its essential feature,which is part icularly suitable for the analysis and mining o f nonlinear data models in a big data environment.This thesis applies deep learning techno logy to the modeling o f the release process o f big locat ion data to realize the effect ive predictio n o f the release t ime interval of big locat ion data.The release process o f big locat ion data conforms to t ypical t ime series characterist ics,and the hidden t ime correlat ion can be extracted fro m the massive historical big locat ion data with the help of deep learning models.This thesis proposes a big locat ion data release interval predict ion method based on the LSTM model.First,the historical locat ion big data is converted into a reasonable release time interval sequence by adapt ively adjust ing the sampling;Then,an LSTM deep learning model is constructed to mine the deep features of the historical release interval sequence,and predict the subsequent release interval;Through experiments and analysis of the actual big location data set,it proves the feasibilit y and effect iveness of the release interval predict ion method based on deep learning in the big locat ion data stat ist ics release process.In order to improve the accuracy o f big locat ion data release time interval sequence predict ion,through further analysis of the time do main characterist ics o f big locat ion data,a t ime interval sequence release predict ion method based on pattern deco mposit ion is proposed.Introduce the extremely overlapping discrete wavelet transform to deco mpose the time interval series,and obtain the periodic characterist ics and fluctuat ion characterist ics o f the interval series;according to different time do main characterist ics,different deep learning models are selected for predict ion;The above predict ion results are fused and the final predict ion result is obtained through the reverse reconstruct ion o f the maximum overlap discrete wavelet transform.Experiments on large data sets of actual locations show that the t ime interval sequence release predict io n method based on pattern deco mposit io n proposed in this thesis has higher predict ion accuracy than the single-model deep learning predict ion method.