<p>Electric vertical takeoff and landing (eVTOL) aircrafts powered by batteries promise to revolutionize urban mobility due to their advantages of being low-carbon, safe, and efficient. The state of battery directly affects whether eVTOL aircrafts can operate normally. Energy efficiency determines the flight performance of eVTOL aircrafts is an important indicator for evaluating battery performance. By combining multi-scale feature extraction with deep learning, a model integrating wavelet transform and an improved TimeMixer is built as three parallel structures to enhance the energy efficiency prediction capability of eVTOL aircraft batteries. This model is tested using a public eVTOL aircraft battery dataset containing 21,392 charge and discharge cycles for 22 cells. Considering eVTOL aircrafts operate in the mission cycles of takeoff, cruise, landing, and rest, a feature set consisting of 62 features is designed as the input for the models. A practical method which can simplify battery energy efficiency estimation is proposed to label samples. The test results show that the proposed model has a mean absolute percentage error (MAPE) of 0.153% for batteries working under dynamic high C-rate discharge conditions, which is significantly better than the compared model.</p>