AUTHOR=Ding Can , Zhou Yiyuan , Ding Qingchang , Wang Zhenyi 

TITLE=Loss Prediction of Ultrahigh Voltage Transmission Lines Based on EEMD–LSTM–SVR Algorithm

JOURNAL=Frontiers in Energy Research

VOLUME=Volume 10 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/energy-research/articles/10.3389/fenrg.2022.811745

DOI=10.3389/fenrg.2022.811745

ISSN=2296-598X

ABSTRACT=Line loss prediction of Ultra-high voltage (UHV) transmission lines is the key to ensuring the safe, reliable and economical operation of the power system. However, the strong volatility of line loss brings challenges to the prediction of transmission line loss. For more accurate prediction, this paper uses ensemble empirical mode decomposition (EEMD) to decompose the line loss, and proposes the EEMD-LSTM-SVR prediction model. First of all, this paper performs feature engineering on power flow, electric energy, and meteorological data, and extracts the exponentially weighted moving average (EWMA) feature from the line loss. After the integration of the time dimension, this paper mines the curve characteristics from the time series and constructs a multi-dimensional input dataset. Then, through EEMD, the line loss is decomposed into high-frequency, low-frequency and random IMFs. These IMFs and the standardized multi-dimensional data set together constitute the final input dataset. In this paper, each IMF fusion dataset is sent to LSTM and SVR models for training. In the training process, the incremental cross-validation method is used for model evaluation, and the grid search method is used for hyperparameter optimization. After evaluation, the LSTM algorithm predicts high-frequency IMF1, 2 and random IMF4,5; the SVR algorithm predicts low-frequency IMF6, 7 and random IMF3. Finally, the output value of each model is superimposed to obtain the final line loss prediction value. And the comparative predictions were performed using EEMD-LSTM, EEMD-SVR, LSTM, and SVR. Compared with the independent prediction models EEMD-LSTM, EEMD-SVR, the combined EEMD-LSTM-SVR algorithm has a decrease in MAPE% by 2.2, 25.37, respectively, which fully demonstrates that the combined model has better prediction effect than the individual models. Compared with SVR, the MAPE% of EEMD-SVR decreases by 11.16. Compared with LSTM, the MAPE% of EEMD-LSTM is reduced by 32.72. The results show that EEMD decomposition of line loss series can effectively improve the prediction accuracy and reduce the strong volatility of line loss. Compared with the other four algorithms, EEMD-LSTM-SVR has the highest R-Square of 0.9878. Therefore, the algorithm proposed in this paper has the best effectiveness, accuracy and robustness.