Research On Multi-source Data-driven Methods For Power Equipments Condition Prediction And Power Grid Resilience Enhancement

Affected by the geographical environment,typhoon disasters occur frequently in summer in the eastern and southern coastal areas of China,resulting in difficulties in power equipment operation status assessment and maintenance and even structural failures,which seriously restrict the reliability of power supply.Building a power system with high penetration of renewable energy is an inevitable requirement for the low-carbon and high-quality development and clean energy transformation in China.The development of large-scale offshore wind power has become an important strategic support for the transformation of the energy structure in the eastern and southern coastal areas of China.The operation characteristics of large-scale offshore wind power are vulnerable to typhoons,and the spatio-temporal variation characteristics of typhoons lead to a multi-stage spatio-temporal correlation of power grid resilience in typhoon scenarios.It is urgent to combine intelligent sensing,artificial intelligence and other technologies to mine power,meteorological,geographic and other multisource data information,and carry out research on power equipment operation state prediction based on data-driven methods and grid resilience enhancement methods in typhoon scenarios.The main contributions are summarized as follows,1.An imputation and prediction method for the missing data of power multivariate time series is proposed based on the Generative Adversarial Network architecture and bi-directional Long Short Term Memory network.Considering the difference in time series characteristics of different power data,this paper analyzes the data missing characteristics of multivariate time series and the temporal correlation between multivariate variables.The data imputation and prediction with various missing conditions are completed by utilizing differential training and testing methods.Represented by the monitoring data of dissolved gases in transformer oil,the imputation and prediction for the missing data of power multivariate stationary time series is realized;Furthermore,a forecasting strategy for power multivariate nonstationary time series is proposed,which combines similar day extraction and local mean decomposition methods,and realizes the imputation of missing data and dayahead forecasting of such power time series represented by power load data;The case studies verify the feasibility and effectiveness of the method applied to the imputation and prediction for variate power data,which provides a theoretical basis for multisource time series data cleaning and reliable prediction in typhoon scenarios.2.An explainable bi-level fault diagnosis method for power transformers monitoring datasets with small sample data problems is proposed.A bi-level diagnosis model is constructed with the aim of class-imbalanced sample anomaly detection and fault type identification considering the small sample size problem of power transformer gas-in-oil condition monitoring data;An automatic hyperparameters tuning method is derived based on Bayesian optimization,and the custom objective functions are designed aiming at the different diagnostic targets at each level;An explainable method for the fault diagnosis method is constructed,which could quantify the impacts of selected fault features on diagnosis results and the interaction between the selected fault features;The case studies verify the accuracy of fault diagnosis method in fault detection and fault type identification,obtain explainable feature selection results,which provide support for power equipment operation status identification and the scheduling strategies arrangement of operation and maintenance with high timeliness requirements in typhoon scenarios.3.A method for resilience evaluation and prediction of key power components in typhoon scenarios is proposed based on spatial-temporal analysis.Combined with the spatial-temporal distribution and variation characteristics of wind speed in the typhoon wind field,a prediction model of typhoon wind speed and typhoon center moving path based on Temporal Fusion Transformers architecture is established,which utilizes multi-source spatial-temporal information such as typhoon historical record data,meteorology and geography;With the analysis of the spatial-temporal correlation characteristics between typhoon prediction results and offshore wind turbines,the characteristics of wind speed variation with height of large-scale offshore wind turbine in the typhoon wind field are revealed,and the power output prediction model of offshore wind turbine considering the wake effect is further proposed;A prediction model of power transmission lines failure probability in typhoon scenarios is proposed based on the wind speed and moving path prediction results of the typhoon;The case studies verify the effectiveness of the proposed method in offshore forecasting offshore wind power output and failure probability of power transmission lines in typhoon scenarios,which is conducive to realizing multi-dimensional state prediction of power system operation with the help of meteorological information in typhoon scenarios.4.A model and method using offshore wind power to actively enhance power system resilience in typhoon scenarios are proposed.Based on the typhoon prediction and power components operation status prediction results,the coupling and difference between the operation state of offshore wind power and coastal onshore power system in time and space dimensions are revealed,and a multi-stage resilience enhancement strategy of power systems integrated with offshore wind power considering three stages of before,during and after typhoons is proposed;The typhoon attack scenarios are divided combined with the interval prediction results of typhoon wind speed and moving path.Considering the uncertainty of transmission line fault caused by the typhoon,a three-level unit combination robust optimization model to enhance the resilience of power systems with large-scale offshore wind power is proposed;Given the flexible reserve supporting ability of large-scale offshore wind power,a multi-stage power system reserve coordination strategy for offshore wind power and dispatchable generators along with the resilience enhancement model is constructed;The case studies verify that the proposed multi-stage strategy with large-scale offshore wind power can effectively enhance power system resilience in typhoon scenarios to achieve the safe and economic operation.Focusing on the critical needs of low-carbon development and the resilience enhancement of the power system with large-scale offshore wind power in typhoon scenarios,this paper integrates the model-driven and data-driven methods based on artificial intelligence technology to mine the status monitoring data,which realizes the operation status prediction and fault diagnosis of power equipment,improves the ability of the power system with large-scale offshore wind power in coastal areas to withstand typhoon disasters and provides theoretical guidance and technical support for natural disaster prevention and mitigation in the process of energy structure transformation in China.