<p>Electric vehicle charging stations often operate in complex environments with strong electromagnetic interference, which severely compromises the accuracy and robustness of pulse width modulation (PWM) signal feature extraction. To address this issue, this paper proposes a robust method for extracting instantaneous frequency-domain features of PWM signals under heavy interference. First, a linear fitting model for the power characteristics of charging stations and a spatiotemporal dynamic load monitoring model are established, forming a collaborative monitoring framework from the macro-system level to the micro-signal level. Based on this framework, a multiscale time–frequency feature extraction model is developed by integrating wavelet packet energy spectrum, autocorrelation-windowed short-time Fourier transform (STFT), and complementary ensemble empirical mode decomposition (CEEMD). This approach overcomes the limitations of traditional methods, such as insufficient resolution and mode mixing in noisy conditions. Experiments conducted on a public charging pile dataset show that, compared with existing typical methods, the proposed method extracts instantaneous frequency-domain features of PWM signals that are closer to the true values, achieving an improvement in feature extraction accuracy of more than 3%. The curve of current variation rate exhibits higher stability, along with enhanced feature steadiness and class separability under noisy conditions. This method provides an effective technical solution for status monitoring and early fault diagnosis of charging piles in heavily disturbed environments.</p>