Hybrid CNN-LSTM Network for Time-Varying Wide-area Composite Load Model Parameter Identification in Power Systems

Load modeling is a vital task for power system analysis and control, but most existing load models assume that the load parameters are constant and independent, which may not reflect the realistic and dynamic characteristics of the load. In this paper, a novel method is proposed to identify the time-varying and spatially distributed parameters of the time-varying wide area composite load model (TWACLM) using a residual convolutional neural network and long short-term memory (CNN-LSTM) network. The method consists of three steps: data denoising using wavelet denoising technique (WDT), sensitivity analysis using finite-difference methods, and parameter estimation using the residual CNN-LSTM network. The method is validated on the IEEE 39-bus system under different types of disturbance signals, and compared with other deep learning algorithms. The results show that the proposed method can achieve high accuracy, consistency, and efficiency in identifying the TWACLM parameters, and can capture the temporal and spatial features of the load data. The results underscore the potential of the proposed method to advance load modeling in power systems, offering robust and reliable parameter identification under various disturbance scenarios.