Abstract: The complexity of offshore environment makes it difficult for a single forecasting model to adapt to the intermittency and fluctuation of offshore wind speed. The offshore wind speed sequence is decomposed into low frequency sub-sequence and high frequency sub-sequence by orthogonal wavelet transform（OWT）. Then the decomposed low frequency sub-sequence is predicted by long and short term memory network（LSTM） model,and the high frequency sub-sequence is predicted by autoregressive comprehensive moving average（ARIMA）model. The two prediction sequences are combined to form a complete wind speed prediction result. Finally,the wind speed of typical days in different seasons is predicted by single model and combined model respectively,and the prediction curves of single model,combined model and real value are compared and analyzed. The results show that compared with the single model,the combined model can improve the accuracy and stability of the prediction.