Research On Key Technologies Of Intelligent Operation And Maintenance Of Power Information System

With the advancement of the national smart grid information construction work,diversified computer technologies have been applied to the grid construction.Companies in various grid provinces have invested a large amount of information equipment and information systems to carry power services.Various equipment and systems integrate huge and heterogeneous In the IT environment of the electric power intranet,the traditional manpower-based operation and maintenance method has been stretched,and the work of electric power IT operation and maintenance personnel is facing huge challenges.At present,AIOps intelligent operation and maintenance technology has become the main solution for the operation and maintenance of huge heterogeneous IT environments due to its advantages of automation,precision and efficiency,low manual participation,and high operation and maintenance input-output ratio.There is a lack of effective experience and guidance on maintenance issues,there are problems such as low data labeling rate of electric power big data,and unclear intelligent operation and maintenance scenarios,which lead to difficulties in the implementation of AIOps in the field of electric power IT operation and maintenance.How to seamlessly connect the key technologies of AIOps with the traditional power IT environment,clarify the landing scenarios,and achieve the expected results has become a research hotspot of common concern in the power industry and AIOps academic fields.In this regard,the main work and innovations of this paper are as follows:(1)Aiming at the problem that it is difficult to accurately estimate the operating status of power marketing servers in a densely integrated environment,a server operating status prediction method MAT-Trans that can resist abnormal data is proposed.Considering the strong autocorrelation of equipment operating status and the noise disturbance of monitoring data,a correction mechanism for outliers in training data is proposed.Aiming at the high-dimensional characteristics of monitoring data,a data feature filtering/selection mechanism is proposed.Through Peel The low correlation coefficient performs feature screening for the simplification of model computational complexity,which speeds up the convergence speed of model training.Combined with the attention mechanism to train the prediction model to predict the running status of the server.Finally,a comparative experiment was set up,and the index values of 60 time steps were predicted on the 90% confidence interval,and the ND value reached 0.0126.At the same time,the advantages of the proposed algorithm were verified on other comparative indexes.(2)Aiming at the problem of receiving traffic prediction of the core router of the power information intranet with high amplitude jitter,a deep learning traffic prediction algorithm TD2 AG based on the topology feature screening mechanism is proposed.In the actual operation of the power information intranet,the network traffic has a strong correlation with the actual power grid business,and the amount of specific business volume will directly affect the surge and decline of traffic in the network,resulting in unpredictability of the power information intranet network traffic.The proposed method first conducts correlation analysis on the original data through causality testing,filters out attributes that are less relevant to the predicted target,and reduces the dimension of the training data,and then extracts features from the data through topological sequence analysis to make the dynamic changes of the data explicit.The characteristics of the model,strengthen the capture and learning of the high-level change characteristics of the data,and finally connect the data to the GRU network for training.Finally,by comparing the traditional methods and the analysis of experimental results,the effectiveness of the model proposed in this paper on the power intranet traffic prediction problem is verified.Accuracy,compared with traditional methods on MAPE and R2 indicators,shows the advancement of the proposed method.(3)Aiming at the application server of the financial management and control system of the power company,an anomaly detection method NCaps combining unsupervised learning and deep learning is proposed.Traditional static threshold detection technology has the problem that high threshold is prone to false positives and low threshold is ineffective.After a fault occurs,fault identification increases the time for fault repair,which leads to long-term interruption of power services and other livelihood issues.A power information system operation and maintenance monitoring data labeling method combining unsupervised learning and expert experience is proposed to label various abnormal patterns of potential power information servers during operation,and to label power data by combining unsupervised learning mining and human expert experience It is more objective and complete,and the training data is more accurate.The feature dimensionality reduction and explicitization of the training data are carried out through the convolutional network.Finally,the Capsule network is trained to detect and classify the abnormalities of the power information equipment.The proposed method is divided into 15 types of abnormalities.In terms of problems,the accuracy of abnormal detection has reached more than 95%,and the practicability of the proposed method has also been confirmed in terms of F1 value and recall rate in comparison with traditional methods.To sum up,this paper explores the implementation of AIOps intelligent operation and maintenance in the operation and maintenance of electric power information systems.By proposing the corresponding intelligent operation and maintenance algorithm for the positioning of intelligent operation and maintenance scenarios in the power IT environment,it innovatively and efficiently solves the problem of traditional operation and maintenance.By verifying the rationality and efficiency of the proposed method on the actual power production data set,it provides an effective solution for the implementation of AIOPs intelligent operation and maintenance technology in the field of power information system operation and maintenance,and has important theoretical significance and application value.