Accurate forecasting of natural gas prices and trade flows is an important research topic in the field of energy economics. The results are directly related to the business decisions of energy companies and national energy strategy planning. However, due to the dynamic and multidimensional complexity of the natural gas market, traditional statistical models and machine learning methods have obvious limitations when dealing with nonlinear time series. To this end, this paper proposes a prediction method based on a hybrid deep learning model (CNN-BiLSTM), which fully combines the local feature extraction capability of the convolutional neural network (CNN) with the long-term dependence modeling advantage of the bidirectional long short-term memory network (BiLSTM) to improve prediction accuracy and robustness. This paper constructs a high-quality input feature set covering temporal features, economic variables and external events based on global natural gas market data from 2010 to 2023, and ensures data quality through pre-processing steps such as data cleaning and normalization. Experimental results show that the CNN-BiLSTM model is significantly superior to traditional models such as ARIMA and SVM and the single deep learning model LSTM in terms of indicators such as root mean square error (RMSE) and mean absolute error (MAE). In addition, the model performs well in capturing the local dynamics and long-term trends of time series, and is particularly adaptable under market volatility conditions. This study not only provides a novel solution to the problem of complex time series prediction, but also provides technical support for scientific decision-making in the natural gas market. Future research can further optimise the model efficiency and expand its application to other energy markets.