Electrical Load Identification Using Fast Continuous Wavelet Transform and Spatiotemporal SNN

Accurate and real-time electrical load identification technology has emerged as a critical requirement for ensuring electrical safety and enabling optimized energy management. However, traditional load identification methods have obvious limitations in feature extraction efficiency and computational energy consumption. This paper focuses on intelligent power load perception, proposing a Fast Continuous Wavelet Transform (FCWT) method based on Morlet wavelet. By adopting a frequency-domain zero-removal strategy to eliminate redundant computations, the proposed method reduces the computational complexity of frequency-domain convolution by 43.64% while maintaining high resolution. Furthermore, an FCWT-Spiking Neural Network (FCWT-SNN) framework is proposed by integrating FCWT and neural network techniques. The time-frequency features are fed into a LIF spiking neural network, which leverages spike sparsity to achieve low-power inference. The proposed architecture achieves 99.98% and 98.46% recognition accuracy on the open-source DataSet_SEIN_1 and PLAID datasets, respectively. Compared to conventional CNN architectures, the FCWT-SNN maintains comparable performance while reducing computational load by 97.3% and energy consumption by 98.5%. The methodology not only provides a high precision, low power technical solution for dynamic monitoring of power systems but also introduces novel research perspectives for time-domain signal feature extraction and identification.