Abstract: by strong electromagnetic interference and dynamic loads，as well as the insufficient monitoring accuracy due to single feature extraction，a real time monitoring method based on cloud edge collaboration is proposed. A three layer collaborative architecture consisting of user side，edge side，and cloud center is constructed. Under cloud edge collaboration，multi dimensional features including volatility，trend，and variability are extracted from the energy consumption data. Taking these multi dimensional features as inputs， a BP neural network based prediction network and a Variational Autoencoder based reconstruction network are fused for preliminary anomaly detection，and a Peaks Over Threshold model is introduced to set dynamic thresholds，thereby achieving accurate determination of energy consumption states. Experimental results show that the monitoring relative error of this method is below 2.5%，and it can effectively distinguish different energy consumption states. The proposed method significantly improves the accuracy of real time energy consumption monitoring for electrical equipment in coal preparation plants and has good application value.