Research And Implementation Of Prediction Method For Scheduling Optimization Timeliness Under The Cloud Computing Environment

Cloud computing task scheduling and resource allocation has always been a hot topic in current cloud computing research.With the development of large data and Internet of things,how to find the best results in the shortest time is the core problem of the cloud computing scheduling.The current various improved algorithms still seek to balance between algorithm complexity and performance,however it can't fundamentally resolve this contradiction.To solve the above problem,it is feasible to use the deep learning model to train the data of the swarm optimal scheduling and achieve a direct prediction of the scheduling results.But deep learning is due to its complex hierarchical structure and a large number of neuron nodes,the training time is longer than the traditional machine learning,how to improve training speed and optimize deep learning model are practical problem that must be solved.The thesis mainly research two aspects: One is to optimize the scheduling algorithm to obtain more accurate training samples;Another is to improve the training speed.The specific work is as follows:(1)To solve cloud computing task scheduling problems,a scheduling model is established,which is based on the shortest execution time of tasks.(2)To solve the problem of scheduling optimization,two optimization algorithms are used,One is on the basis of traditional bird swarm algorithm combining differential evolution algorithm and according to the behavior of individual birds flying into and fly-out swarm,we improved bird swarm algorithm(ISBSA).Another is on the basis of the new bat swarm algorithm adding dynamic factors,second-order oscillations and differential evolution algorithms.We propose Improved new bat swarm algorithm(ONBA).Then use these two improved algorithms to solve the multi-objective problem proposed in this paper.At the same time we compare the results of the two improved algorithms.(3)To solve the problem of the optimization efficiency for scheduling,this paper puts forward an improved deep learning model(IDBN),this paper Introduces adaptive learning rate to training RBM stage and reverse tuning stage in traditional DBN network,then using the above learning rate improvement algorithm to control the training times of the model to achieve the purpose of accelerating the training speed.finally we simulate the above improved DBN(IDBN)training method to verify that the method can speed up the training of the model.(4)Finally,using ISBSA and ONBA solve multi-objective task scheduling model to obtain scheduling data sets,then using data sets to train IDBN model.Using the trained IDBN model to predict the scheduling results of the task and compare with the traditional scheduling method in time to verify this paper proposed method has practical application value.