Research On CPU Utilization Early Warning Method Of Power Network Server Based On ARMA-BiLSTM

With the continuous development of information technology,the traditional manual operation and maintenance method has been difficult to meet the growing and increasingly complex business needs of enterprises.The automatic operation and maintenance technology can better solve this problem.The business system server of the automatic operation and maintenance platform will generate massive data services every day,which makes the CPU performance of the business system server face the problem of overload.Therefore,it is of great significance for the resource allocation of business system server to predict the CPU utilization of servers.The CPU utilization of server is affected by many objective factors,and its time series data is very complex,so it is difficult to get accurate prediction result using a single prediction model.A new method is proposed to predict the CPU utilization of server using the autoregressive moving average model combined with bi-directional long short-term memory(ARMA-BiLSTM)based on wavelet transform.First of all,the element average filling method and pauta criterion are used to preprocess the CPU utilization data and eliminate the abnormal data in the time series of server CPU utilization.Secondly,wavelet transform is used to decompose the CPU time series,and the stationary subsequence with detail fluctuation and the non-stationary subsequence with trend information are obtained.The ARMA model is used to model and predict the stationary subsequence which represents the detail fluctuation,and the Bi-LSTM neural network model is used to model and predict the non-stationary subsequence which contains the trend information.Finally,the wavelet transform is used to reconstruct the original signal,and the prediction results of each subsequence model are synthesized to obtain the final prediction result.Using the actual data of CPU utilization provided by the National Power Network Monitoring Center,the experiment is carried out and compared with BP neural network,single LSTM model and ARMA-LSTM composite model.The time series of CPU utilization is analyzed and predicted by using the proposed compound model.According to different servers,setting different early warning thresholds,combining with the prediction of CPU utilization,early warning of server overload,can effectively improve server efficiency.