Trend prediction of Sichuan cuisine culture based on long short-term memory network

Sichuan cuisine, as an important representative of China's food culture, its development trend is influenced by regional culture, consumption demand and social changes. Traditional experience-driven forecasting methods have the problem of insufficient timeliness and accuracy in dynamic market environment. In this study, we propose a prediction model that integrates Bi-directional Long-Short Term Memory Network (BiLSTM) and multi-attention mechanism. By integrating multi-source time series data such as social media search index, restaurant transaction records and regional economic indicators, we construct a dynamic prediction framework for Sichuan cuisine popularity. The model uses bidirectional LSTM to capture historical trends and potential future associations, and combines attention mechanism to enhance key feature extraction capabilities. The experimental results show that the MSE of the model is reduced by 60.8%, 49.7% and 35.7% respectively compared with ARIMA, Prophet and one-way LSTM, especially in unexpected scenarios such as holiday peak consumption (56.7% lower than one-way LSTM). The study provides data-driven decision support for innovation direction selection and supply chain optimization of Sichuan cuisine industry, and verifies the effectiveness and application potential of deep learning technology in food culture trend prediction.