Design And Implementation Of Grid Corporation Host Prediction System Based On Time Series Data Mining

With the continuous advancement of intelligence and informatization in Grid Corporation,building information systems has become a very important part of the Corporation’s business system.During the operation of the information system,a large amount of host indicator data is generated.The way of Corporation’s existing operation and maintenance is hard to quickly and efficiently analyze useful information from the massive host indicator data,and cannot guarantee the stable operation of the information system.Based on this,this thesis designs and implements a Grid Corporation host indicator prediction system based on time series data mining from the perspective of host indicator data prediction.The predicting system can make shortterm forecasts on the host indicator data in the information system and visualize the results,so that users can intuitively feel the future change trend of the host indicator.On this basis,anomaly detection is performed on the predicted data.If abnormal data is found,the system will generate an early warning message for users to view,reducing the possibility of abnormalities in the information system.The research content of this thesis is as follows:1.Research on the prediction model of host indicator based on difference calculation.In the Grid Corporation information system,there are some non-stationary and changing host indicator data.When analyzing non-stationary host indicator data,there will be a pseudo-regression problem.This thesis first uses ADF to test whether the host indicator data is stable,and for the data that fails the test,the difference operation is used to smooth it.Then use Xgboost to train its prediction model,and finally restore the prediction result.Compared with direct prediction,it can eliminate some pseudo-regression problems in the analysis process.2.Research on the host indicator prediction model based on EMD decomposition.In the Grid Corporation information system,there are some non-stationary and fluctuating host indicator data.It is difficult to directly predict this part of the data to achieve the desired result.EMD can decompose it into multiple relatively stable subsequences,so this thesis first uses EMD to decompose the host indicator data,and then trains the prediction model for each sequence obtained by the decomposition,and the model mainly uses LSTM.Finally,the output results of each model are added together to obtain the prediction results of the host indicator data.3.Design and implement a Grid Corporation host prediction system based on time series data mining.The system takes the proposed prediction algorithm model as the core and integrates host indicator data collection,host indicator data preprocessing,host indicator data prediction,intelligent early warning and other functions.It can automatically analyze the latest host indicator data without user intervention,and make predictions about it.While improving the efficiency of operation and maintenance,the stable operation of the information system is ensured to the greatest extent.