Research on Low-Frequency Oscillation Situational Awareness and Prediction of Power System Based on L2R-DLR-LGDMM Method

Abstract

With the development of power grid technology, low-frequency oscillation has become a key issue affecting the stable operation of the power system. In view of the difficulty of accurate prediction of low-frequency oscillation situational awareness of power grid and the difficulty of considering less environmental factors, the L2R-DLR-LGDMM method is proposed to predict the situation awareness of the comprehensive indicators of power grid, which combines with Topsis normalization, L2 regularization (L2R), dynamic learning rate (DLR) and Logistic gradient descent with momentum factor (LGDMM). Firstly, the original data collected by the phasor measurement unit (PMU) is analyzed and processed. A variety of indicators are fused into a comprehensive indicator, the attribution degree is calculated by substituting the safety prediction indicator, and the fuzzy evaluation matrix is established to construct a comprehensive indicator dataset. Secondly, the comprehensive indicator dataset is substituted into the L2R-DLR-LGDMM prediction method, the model parameter gradient is calculated according to the comprehensive indicator dataset, and the learning rate and momentum factor size are adaptively adjusted during the training process. After the difference between the two gradient changes is less than the set minimum value, the proposed method stops the iteration process and best parameters are obtained. This proposed method can use comprehensive indicators to achieve the situational awareness prediction of low-frequency oscillation. Finally, the proposed method has a better prediction effect on the low-frequency oscillation of the New England 10-machine and 39-node system and the Western Electricity Coordinating Council (WECC) 146-machine and 243-node system. The prediction accuracy of the comprehensive indicator is higher than that of other methods, and the prediction accuracy can reach 86.74% and 95%, respectively. Therefore, the proposed method provides a reference for the stable operation of power systems.

Keywords

• Situational awareness
• logistic momentum gradient descent method
• comprehensive indicator forecasting
• L2 regularization
• dynamic learning rate