Short-term (less than 1 h) forecast of the power generated by wind turbines in a wind farm is extremely challenging due to the lack of reliable data from meteorological towers and numerical weather model outputs at these timescales. A robust deep learning model is developed for short-term forecasts of wind turbine generated power in a wind farm using the state-of-the-art temporal convolutional networks (TCN) to simultaneously capture the temporal dynamics of the wind turbine power and relationship among the local meteorological variables. An orthogonal array tuning method based on the Taguchi design of experiments is utilized to optimize the hyperparameters of the proposed TCN model. The proposed TCN model is validated using twelve months of data from a 130 MW utility-scale wind farm with 86 wind turbines in comparison with some of the existing methods in the literature. The power curves obtained from the proposed TCN model show consistent improvements over existing methods at all wind speeds. •Deep learning model based on state-of-the-art TCN to predict the total wind power.•Multi-step prediction of wind power for 0, 10, 20, 30, 40 and 50 min ahead.•OATM to optimize the hyperparameters of the deep learning models.•Comparison of multi-step ahead TCN model against LSTM, CNN + LSTM and MLR models.