Abstract: Based on the nonlinear and fluctuating characteristics of pork price, this paper proposes a pork price prediction approach using variational modal decomposition (VMD) and Bayesian optimization-based bidirectional long short-term memory (BiLSTM). VMD decomposes the data into subsequences with simple fluctuations, which are then used in BiLSTM. Bayesian optimization is adopted to optimize the number of neurons, learning rate, and batch size of the first and second hidden layers of the BiLSTM network model. Experimental results show that the proposed VMD-BO-BiLSTM method outperforms traditional single LSTM and BiLSTM models in terms of mean absolute error, root mean square error, mean absolute percentage error, and determination coefficient. It has higher accuracy and applicability for pork price prediction.

Key words: bidirectional long short-term memory (BiLSTM), Bayesian, pork price forecast, variational modal decomposition, hyper-paramete