A Time Series Data Mining System Based On Deep Learning

The outer wall of the West-East Gas Pipeline in my country is protected by oxidation-reduction reaction.In order to check whether the protection of the outer wall of the pipeline is effective,it is only necessary to extract the HVDC（Higher Voltage DC）time series data in each section of the pipeline to complete the pipeline failure.Positioning.The occurrence of HVDC phenomenon represents the peeling off of the outer protective layer of the pipeline.This paper mainly uses deep learning methods to solve the classification problem of non-stationary time series data such as HVDC.Time series data classification methods based on deep learning mainly include CNN,LSTM,RNN,CNN + LSTM,CNN + RNN,etc.Compared with machine learning methods such as SVM and MLP,its performance is not only improved by 5%to 10%,but its algorithm complexity is also greatly reduced.Deep learning methods are not perfect in terms of versatility,stability and accuracy.In order to solve the above problems,a DCL neural network is proposed.DCL is a neural network with a 14-layer structure composed of Dense Net+CNN+LSTM.DCL is superior to other networks in stability and versatility.The accuracy of DCL classification in HVDC time series data reaches 91.25%.Compared with traditional analysis strategies,these deep learning techniques have more advantages in stability,versatility,classification and prediction accuracy.This paper makes the following six contributions in the research of time series data mining: 1.Based on JSD（Jensen Shannon Division: a time series segmentation method）,a new outlier detection algorithm is proposed: JSDFP（Jensen Shannon Find Point）.2.Based on QTA（Qualitative trend analysis: a qualitative trend analysis method that can be used for fault analysis and simulation verification）,a new feature extraction method for HVDC time series data is proposed: QTA-TS（Qualitative trend analysis Time Series）.3.Propose a time series data simulation method based on probability constraints:RT（Rand Time Series）,which can be customized to simulate five different types of time series data in the West-East Gas Pipeline Project.4.Propose a large-scale time series simulation system based on RT technology: TSS（Time-Series Simulation System）distributed system,which can simulate 5 different types of time series data on a large scale on hundreds of different nodes,The system can simulate the entire operation process of the entire West-East Gas Pipeline Project.5.Based on SGD（a method of policy verification based on VV＆A）,a new method of credibility evaluation is proposed: TM-SVF（Two or More Simulation Verification Frameworks）method,which mainly evaluates the credibility of the TSS system.6.Based on RNN,CNN,LSTM and other deep learning methods,a new time series data classification method: DCL deep learning method is proposed.This method mainly solves the classification of HVDC non-stationary time series.