Research On The Key Technology Of Optical Fiber Sensor For Power Transmission Lines Monitoring And Intelligent Prediction Of Icing

In 2019,State Grid proposed the strategic goal of constructing a Ubiquitous Electric Internet of Things(UEIOT),with the fundamental objective and eternal theme of ensuring the safe operation of the power system.The safe operation of the power system involves multiple aspects,including data monitoring and collection,security analysis,and automated control.Especially as the energy structure of the power grid becomes increasingly complex and electricity consumption becomes more scale-driven,the safe operation of the power system faces greater challenges.Constructing a stable and reliable intelligent power grid based on the UEIOT is a highly complex project.Transmission lines,as the lifeline of power transmission,play a crucial role in power distribution and exchange.However,the current monitoring of transmission line status still faces many problems,which cannot guarantee the safe operation of the lines.At the monitoring terminal level,the vast majority of the sensors deployed on transmission lines are electronic sensors,and their operational status is unstable in strong magnetic fields.In some regions with harsh and changeable climatic conditions,it is difficult to maintain the realtime working status of sensors powered by solar panels.Therefore,there is an urgent need to develop a highly reliable monitoring solution for transmission lines.Meanwhile,although the power grid company collects monitoring data of multiple parameters,there is a lack of effective analytical models to provide reliable support for accurately monitoring the safety of transmission lines.Thus,there is a pressing need for a system model that can process and analyze multiple parameters of monitoring data for transmission lines.To address the aforementioned problems,this dissertation researches the key technology of optical fiber sensors for power transmission lines monitoring and intelligent icing prediction.This dissertation focuses on the construction of an integrated transmission lines monitoring system for "lines monitoring-information collection-intelligent evaluation".It explores the monitoring and prediction of transmission line icing from the four aspects of bottom-level sensor design,freezing rain prediction model,hoarfrost prediction model,and comprehensive prediction model.The main research and innovation work are as follows:1、In the dissertation,we address the issue of poor reliability and low transmission efficiency of traditional transmission line status monitoring sensors by proposing and designing a fiber optic grating-based meteorological sensor and a fiber optic collimator-based wind speed and direction sensor for transmission lines.We create multiple types of FBGbased meteorological sensing units by utilizing the sensitivity of gratings to temperature and strain.The measurement range and accuracy of various sensors are in line with the corresponding provisions for "meteorological monitoring devices" in GB/T 35697-2017《General Technical Specifications for On-line Monitoring Devices for Overhead Power Transmission Lines》.Furthermore,we develop wind speed and direction sensing units that can accurately and effectively monitor in real-time the differences in optical power coupling under different wind speeds and directions using fiber optical collimators.Since the collimator uses a 1310nm light source while the FBG sensors use a C-band light source,wavelength division multiplexing technology can be used for co-fiber monitoring,which increases the monitoring range of a single fiber core and saves cable resources.Fiber optical sensors have been installed on several power transmission lines to verify their accuracy and stability.2、In response to the current problems of unclear icing conditions and incomplete theoretical models in transmission line icing,this dissertation analyzes the mechanism of line icing formation and proposes a solution to accurately monitor it using machine learning algorithms to simulate the icing model.To address the issue of the lack of effective intelligent analysis methods for multi-parameter monitoring data in power systems,this dissertation proposes and validates a transmission line glaze ice prediction model based on the Gaussian Process Regression(GPR)model.The impact of GPR parameters on model prediction is analyzed,and an optimization plan for the model is determined.After learning and training with data collected from the Jiangxi 220kV Qinwu line FBG meteorological sensor,the model can accurately predict the thickness of icing on different towers in the same region using micro-meteorological data.The Mean Square Error(MSE)and Mean Absolute Percentage Error(MAPE)are both less than 1%.3、To address the issue of a single prediction model for power transmission lines icing,this dissertation proposes and verifies two power transmission line prediction models for rime and glaze ice on GPR.To improve the accuracy of the icing prediction model in different micro-terrain environments,the conditions for selecting historical icing data as a new input parameter are studied and validated.1-minute historical data is selected as the new input parameter for predicting rime and glaze ice.Experiments were conducted on nine complete icing cycles(historical data collected from four lines in three provinces equipped with optical fiber sensors),and the results showed that the proposed models can effectively predict rime and glaze ice,with a correlation coefficient R>0.99 and MSE and MAPE less than 2%.4、Aiming at the poor timeliness and flexibility of single icing model prediction,and the lack of comprehensive analysis and prediction schemes for various icing models.The training set selection and production methods that can improve the robustness of the model are studied.Some data sets are reassembled and combined to form a new training set.The training set selection and production methods that can improve the robustness of the model are studied.The advantages and applicable conditions of various machine learning algorithms are re-examined.In-depth study of mainstream machine learning algorithms.The applicable conditions and parameter adjustment methods for parameter optimization training of GPR,support vector machine(SVM),random forest(RF),and neural network models belonging to regression algorithm,classification algorithm,and ensemble learning algorithm respectively have been studied.A complete compound prediction model optimization training program was determined,and finally,a machine learning model capable of comprehensive prediction of rime and glaze ice was obtained.The model can effectively predict various types of icing,and it is verified in nine icing cycle data.The results showed R>0.99,MSE,MAPE<4%.This solution only requires data from FBG meteorological sensors and can be combined with weather forecasts for broad icing prediction without the need for hardware sensing devices.