A novel intelligent recognition method based on the combination of PCA-SVM and feature extraction with sliding time-window was proposed to recognize each state in working cycle stages of hydraulic excavator. The pressure signals of two pumps were taken as recognizing object, and small waveform segments of each state in a working cycle of hydraulic excavator were intercepted by adopting sliding time-window method. The corresponding time domain and frequency domain parameters of all waveform segments were extracted and the eigenvalue normalization was carried out. The normalized set was used as input feature set after reduction of dimensionality with principal component analysis (PCA), then state classification was conducted by adopting the SVM algorithm, and the effects of sliding time-window width and overlap rate to the recognition accuracy were discussed, respectively. Besides, a real-time correction strategy was introduced to automatically check and correct the direct recognition results; the misjudgment results against the logic rule of excavator operation was verified, and the recognition accuracy was improved from 80.85% to 89.36%. Results prove that the proposed method can realize the state recognition of each stage in a working cycle of hydraulic excavator effectively and accurately.