Research On Job Failure Prediction Algorithm Of Agricultural Information Cloud Platform

With the development of agricultural informatization technology,agricultural data has been gradually presented characteristics such as large scale,various types,low value density,high complexity,and regional characteristics.Therefore,the storage capacity and computing capacity of the processing of agricultural big data are increasingly demanded.Applying cloud computing technology to agricultural information services to provide better storage capacity,computing capacity,and information sharing capabilities for agricultural information services is a great boost to agricultural information technology.Among them,cloud platform resource management is an important part of improving resource utilization,and is a key technology for the realization of agricultural information cloud.However,because of hardware and software failures,insufficient job scheduling resources,and node failures,jobs cannot be successfully run,resulting in extended runtime and wasted resources.How to predict the termination status of a job and take corresponding scheduling measures is an urgent problem to be solved in cloud resource management optimization.Based on the statistical analysis of cloud platform job failure factors,this paper proposes a network model based on depth automatic coding for the complex load of cloud platform.And using the structural relationship between tasks,a prediction method based on multi-task learning is proposed.The main work and contributions are as follows:(1)The resource management framework of the agricultural information cloud service platform was proposed.Through the task manager in the cloud resource management,the task of the agricultural information cloud platform is predicted.For the task that is predicted to fail,a corresponding scheduling strategy is adopted to avoid resource waste..(2)Aiming at the problem that it is difficult to accurately define the characteristics of complex load data,a failure job prediction method based on the depth automatic encoding extreme learning machine model is proposed to automatically extract features of the original dynamic sequence of resource consumption for long tasks.Experimental results show that the accuracy of prediction reaches 98%,and the accuracy and time performance are better than other prediction method.(3)In the process of job execution,the correlation between tasks under the same job,a failure job prediction method based on the multi-task learning model was proposed to improve the generalization performance of the prediction,and achieved 97.3% accuracy.The paper proposes a method for predicting failure jobs,effectively improving the efficiency of resource utilization in agricultural information cloud platforms,and shortening the operation time of operations.It has important engineering significance.