High Voltage Transmission Line Sag Prediction Model Based On Multi Time Series Deep Stacked SRU Recurrent Neural Network

With the rapid development of science technology and economy,various electrical equipment is constantly emerging,and the demand for power is increasing year by year.The power sector needs to increase power transmission capacity through dynamic capacity expansion.In the process of dynamic capacity expansion,if the sag of the high-voltage transmission line is too small,it will easily affect the mechanical safety of the line;if the sag is too large,it may easily cause discharge hazards.Therefore,accurate sag prediction for high voltage transmission lines has become a vital part of power transmission safety.The prediction of sags of high voltage transmission lines is based on historical data,and the changes of sags of high voltage transmission lines are explored based on the correlation of sag data related changes,attributes and power demand fluctuation.Because the traditional high voltage transmission line sag prediction model is difficult to solve the problem of nonlinearity and time correlation of high-voltage transmission line sag data at the same time,and the common deep learning model solves this problem with strong time series dependence and difficulty in parallel operation,and the model structure is relatively shallow.Therefore,this paper the proposed a sag prediction model for high voltage transmission lines based on multi time series deep stacked unidirectional and bidirectional Simple Recurrent Unit(SRU)network model.The main work of this thesis is as follows:(1)Aiming at the characteristics of high voltage transmission line sag data,this paper the proposed a sag prediction model for high voltage transmission lines based on multi time series deep stacked unidirectional and bidirectional Simple Recurrent Unit(SRU)network model.(2)In order to improve the quality of data,anomaly detection is performed using box plot analysis,and the mean value is used for replacement.At the same time,use the masking mechanism to deal with missing values.(3)For the traditional mathematical method and machine learning method,it isdifficult to solve the time correlation and nonlinear problem at the same time in the prediction.In this paper,the SRU recurrent neural network is used for prediction.The cell nucleus are connected to each other through residuals through self-loop weights,which can dynamically change the time scale to make it have long-term and short-term memory functions.The sigmoid activation function is used to control the nonlinear output of the time series.(4)Aiming at the problem that most studies only consider forward dependence to filter out part of the information easily,the structure is shallow and the accuracy is not high enough,a deep stacked bidirectional architecture is proposed to reduce the loss of information.At the same time,the Nadam optimization method is selected to reduce the memory consumption during training and testing,and can calculate different learning rates for different parameters.Experimental results based on real sag data show that the proposed a sag prediction model for high-voltage transmission lines based on multi time series deep stacked unidirectional and bidirectional Simple Recurrent Unit(SRU)network model is feasible and effective,and its performance is superior to Logistic Regression(LR)model,Artificial Neural Network(ANN)model,Back Propagation Neural Network(BPNN)model,deep stacked unidirectional and bidirectional Simple RNN recurrent neural network model,deep stacked unidirectional and bidirectional GRU recurrent neural network model,deep stacked unidirectional and bidirectional LSTM recurrent neural network model and deep stacked unidirectional SRU recurrent neural network model.