Transmission line ice coating prediction model based on EEMD feature extraction

Transmission line icing is a common natural phenomenon, but it is the most dangerous factor that severely threatens the safety and stability of the power grid operation. Transmission line icing involves many factors, including temperature, humidity, wind speed, light intensity, wire tension, pressure, and wind deflection angle. Because of the high dimensionality, nonlinearity, multi-modality, and heterogeneity of the data generated by these factors, it is difficult to establish an accurate prediction model based on these data adopting traditional data mining methods. How to establish an accurate and effective new model of transmission line icing prediction has become a key problem to be addressed urgently. To address these problems, the paper collects the data monitored by the China Southern Power Grid Online Monitoring System from 2011 to 2016 to study the prediction model of the icing level of the transmission lines. Since the values affecting the icing level are dynamically changing with time, this paper first uses the time series analysis method to process the icing data and proposes an ensemble empirical mode decomposition (EEMD) method to adaptively decompose the meteorological and mechanical data, which reduces the impact of noise and outliers in high-dimensional data, and maximizes the use of the inherent law of time-frequency to effectively analyze icing data. The feasibility of this method is verified with real data. The experimental results show that the prediction model based on EEMD time-frequency is more accurate than the prediction model based on the original data. Compared with the five prediction models as random forest, support vector machine, BP neural network, Elman neural network, and Bayesian network, the accuracy has increased by 0.47%, 2.93%, 1.85%, 0.92%, and 1.86%, respectively. In addition, this new method is more sensitive to the serious situation of icing on the transmission lines. Compared with the prediction model based on the original data, this method improves the accuracy of prediction for icing level 4 and 5 by 17.5%, 16.67%, 50%, 3.13%, and 10.26%, respectively.