Research On Trajectory Position Prediction Algorithm Based On Deep Learning

With the expansion of the Internet,big data and other industries,mobile devices are used in almost every aspect of life.The location service platform provides users with convenient services while also acquiring a large number of user historical trajectories.In addition to the basic geographic location,the historical trajectory also contains other rich hidden information,such as user behavior habits,urban hotspot locations,and so on.The mining of trajectory data can extract valuable user behavior and other information,which can help users better.Route planning can be used in urban traffic management and commercial advertisement layout.However,with the changes in user travel modes,the rapid development of big data,and the explosive growth of massive data,how to efficiently process and use these trajectory data,which contains rich information such as geographic location,travel time,and speed,has become a problem faced by many researchers.problem.But now,with the rise and development of artificial intelligence,machine learning and deep learning technologies have made it possible to analyze massive data,so that valuable potential information can be efficiently mined from trajectory data.This paper combines deep learning technology to process trajectory data,and builds a neural network for the processed trajectory data sequence and trains it.The purpose is to learn the hidden semantic relationships in the long trajectory sequence,and to deeply mine the data features until the model achieves a good prediction effect.In the preprocessing of trajectory data,feature engineering methods in the field of machine learning are used to remove noise and abnormal points in the data,complete the trajectory and extract important locations.Based on the preprocessed data,the LSTM is improved,the location prediction problem is treated as a location generation problem,and the trajectory is mapped from the original space to the new multi-dimensional space through the embedding layer.A two-way LSTM with a static attention mechanism is constructed and Perform a training to get a preliminary prediction result,which is also regarded as the sequence generated by the generator in the adversarial generation network;finally,combine the SeqGAN adversarial generation network to strengthen the prediction result obtained by the LSTM,and use the convolutional neural network as a discriminator to make the data from the generator more close to real data.A large number of experimental results show that the trajectory position prediction algorithm based on deep learning proposed in this paper avoids the problems of discreteness and missing semantics in traditional mathematical methods,fully analyzes the temporal and semantic characteristics of trajectory sequences,improves long-term prediction capabilities,and better The information reflects the user's intentions such as time.