Several Typical Models Of Deep Learning And Applications In Temperature Estimation

Deep learning is a broader class of machine learning methods based on data representation.Its emergence not only promotes the development of machine learning,but also accelerates the innovation of artificial intelligence.In recent years,many derivable models of deep learning have been proposed.The paper mainly studies several typical models of deep learning,and applies them to temperature estimation.The main works of this paper are as follows.We study the unsupervised learning and supervised learning models of deep learning respectively.Firstly,it discusses restricted Boltzmann machine,deep belief network,auto-encoder,sparse auto-encoder and denoising auto-encoder.And it explores their structure,principle,advantages and disadvantages of these unsupervised learning models in detail.Secondly,it investigates several supervised learning models including convolutional neural network,recurrent neural network and deep stacked network.And it evaluates and analyzes the model structure,working principle,advantages and disadvantages.Several typical models of deep learning are analyzed and compared.Neural network,deep belief network and convolutional neural network model are applied to the handwriting digits recognition task.This thesis first studies the influence of dropout technology and weight decay strategy on neural network models.Then it disusses the effects of learning rate and epoch on the deep belief network model and convolutional neural network model.Finally,experimental results show that these two deep learning models have better recognition performance than traditional neural network.Neural network,deep belief network,stacked denoising auto-encoders and convolutional neural network models are applied to temperature estimation respectively.Firstly,it establishes a neural network model with temperature as the research objectives.We evaluate the performance of the neural network model,which based on the number of hidden layer neurons,the number of hidden layers,the dropout technology and the weight decay strategy respectively.Under the optimal number of hidden layer and hidden layer neurons,the neural network model obtains better performance in estimating temperature with dropout technology.Secondly,the deep belief network and the stacked denoising auto-encoder model are established respectively,and the influence of the number of hidden layer and hidden layer neurons on the experimental results has been discussed.Then,convolutional neural network model is built for the temperature datasets of multiple stations.The experimental results show that the model has better estimated performance.Finally,the inference results of the above four models are compared,and they all achieve better estimation performance.