Research On Key Technology Of On-line Attitude Monitoring Of Transmission Tower

As a support for overhead transmission lines,transmission towers play an important role in the safe operation of power systems.Due to the wide distribution of transmission lines in China,transmission towers will inevitably be built in bad geological areas such as goafs,riverbeds and hillsides where landslides or subsidence disasters are likely to occur,and the transmission towers will cause structural stress changes due to extreme weather.This may cause the transmission tower to tilt,deform or even collapse,which may cause large-area grid failures,which will ultimately affect people's normal production and life.However,due to the difficulty in accurately measuring the attitude data of the transmission tower,the measurement data is difficult to transmit over long distances and the attitude of the tower is difficult to predict accurately in time.The development of the on-line monitoring system for the transmission tower attitude is relatively slow and cannot meet the practical requirements of reliable operation of the power system.Aiming at the key technologies of on-line monitoring of transmission tower attitude,based on sensor data fusion method,Internet of Things technology,support vector regression and cyclic neural network theory,this dissertation studies the attitude measurement,preprocessing method and data transmission method of transmission tower and discusses the application of machine learning technology and deep learning technology in attitude prediction of transmission towers.Firstly,aiming at the problem of high cost and poor real-time performance of traditional transmission tower attitude measurement method,a MEMS inertial sensor based attitude measurement method for transmission tower is proposed.Accelerometer and gyroscope are used as the attitude measurement of transmission tower.The frequency characteristics of the two sensors for measuring the inclination data are analyzed,the accelerometer measurement tilt data is accompanied by high frequency noise,and the gyroscope measurement tilt data is accompanied by low frequency noise.Subsequently,this dissertation proposes a method of data acquisition and transmission of transmission tower based on Lo Ra wireless communication technology to realize long-distance transmission of sensor data.In order to solve the power supply problem of Lo Ra nodes,solar energy is used as the energy source of Lo Ra nodes,and super capacitor is used as the standby power supply,which realizes the uninterrupted power supply of the Lo Ra node without the sun light.In order to verify the reliability of the Lo Ra node transmitting the sensor data over long distance,the transmission distance of the Lo Ra signal and the power consumption of the Lo Ra node are experimentally tested.In addition,in order to obtain accurate transmission tower attitude data,this dissertation uses the complementary filter in the sensor data fusion method to preprocess the attitude data of the transmission tower measured by accelerometer and gyroscope.Through experimental comparison,the proposed preprocessing method based on complementary filter for transmission tower attitude data has higher processing precision and computational efficiency than principal component analysis,singular value decomposition,wavelet analysis method and Kalman filter.Finally,this dissertation analyzes the long and short time characteristics of the attitude data of the transmission tower,and proposes an online prediction method for the attitude of the transmission tower based on dynamic sliding window.Based on the traditional prediction model,two improved prediction models are proposed,namely sparse weighted least squares support vector regression method and sparse residual LSTM neural network.The two prediction models have better generalization ability and higher training efficiency than the prototypes.The hyperparameters of the prediction model are optimized by using the Double Chain Quantum Genetic Algorithm(DCQGA).The experimental results show that the sparse weighted least squares support vector regression method is suitable for scenarios with higher prediction efficiency.The sparse residual LSTM neural network is suitable for scenarios with higher prediction accuracy.