Research And Application Of Time Series Data Fusion Algorithms Based On Deep Learning

In recent years,time series has been widely used in the fields of transportation,Internet of things,smart home,finance and so on.The rapid growth of information collection equipment has diversified its sources.Data fusion technology can integrate different sources of data to obtain more consistent,more reliable and more accurate information than the original data,thus providing support for decision making.For time series data with complex structure,traditional fusion algorithms are difficult to express effective information.Deep learning automatically learns abstract feature representations of complex data without relying on prior knowledge,and its effect is significantly better than that of traditional fusion algorithms.Therefore,this paper mainly studies the deep learning-based time series data fusion algorithms.The main work of this paper is as follows:1.A hybrid neural network data fusion algorithm based on time series is proposed.For traditional data fusion algorithm,the fusion performance of time series data which have nonlinear,high noise and large-scale is poor.A hybrid neural network data fusion algorithm(namely SCLG algorithm)is proposed to solve this problem.Firstly,time series are decomposed and reconstructed by singular spectrum analysis algorithm to eliminate noise.Secondly,the spatial and short-term characteristics are extracted by deep convolutional neural network.Thirdly,the long-short-term memory neural network and gated recurrent unit neural network are introduced to learn the long-term memories of data in the time dimension.Finally,the fully connected layer is applied to integrate main information and output the final decision.The experimental results on the data sets of SP&500 and AQI show that the proposed algorithm is superior to the comparison algorithms in terms of fusion performance and stability.2.An eye-movement tracking data fusion algorithm driven by squeeze and excitation mechanism is proposed.Aiming at the problem that traditional data fusion algorithms have low fusion accuracy for eye movement tracking data in complex scenes.This paper proposes an eye movement tracking data fusion algorithm based on squeeze and excitation mechanism,namely TSCL algorithm.Firstly,the algorithm uses the method of mutual support between data to construct a new feature list.Then,the convolutional neural network driven by squeeze and excitation mechanism and the long short-term memory network are effectively combined to realize the fusion of eye movement tracking data.The experimental results of eight video sequences on OTB-100 Eye movement tracking data sets show that the fusion performance of TSCL is better than four classical deep learning models and two traditional fusion algorithms.3.An eye movement tracking data fusion algorithm based on data enhancement technology is proposed.Aiming at the existing eye movement tracking data with small data volume and complex structure,this paper proposes an eye movement tracking data fusion algorithm based on data enhancement technology,namely IGGA algorithm.Firstly,the algorithm uses cubic spline interpolation function to enrich the original data information and reduce the phenomenon of network overfitting.Then,the gated recurrent unit neural network and the attention mechanism are combined effectively to realize the fusion of eye movement tracking data.The experimental results show that the data enhancement technology can improve the accuracy of network fusion,and the performance of IGGA is better than that of the contrast algorithms in eye movement tracking fusion.