Wind power has the benefits of low cost, low emission, abundant resources, and renewability. The inherent randomness, intermittency, and fluctuation of wind power bring about the volatility of wind power generation. Ameliorating the precision of wind speed prediction has great significance. This study aimed to put forward a hybrid model of the hunter–prey optimization (HPO) and the long short-term memory (LSTM) neural network based on improved complementary ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) to acquire the exact wind speed forecast. First, the ICEEMDAN–HPO–LSTM model used ICEEMDAN to preprocess the raw wind speed sequence and then used the HPO–LSTM model to forecast each decomposed subsequence. Ultimately, the last predicted outcomes of the original wind speed sequence were attained by synthesizing all prediction subseries. Five comparison models were established based on three sets of data with different sequence lengths in Inner Mongolia, China, to test the dependability and utility of the model, and the advantages of the model were proved. The findings displayed that (1) the constitution of ICEEMDAN decomposition and HPO–LSTM could ameliorate the behavior of wind velocity forecast; and (2) the average values of the mean absolute error, mean absolute percentage error, root mean square error, and determination coefficient (R²) of the three datasets were 0.22411, 4.60277%, 0.27590, and 0.99719, respectively. The proposed prediction model can be used for wind speed forecasts.