With the acceleration of urbanisation, intelligent transportation systems (ITS) have gradually become an important means of alleviating traffic congestion and improving travel efficiency. Traffic flow prediction is one of the key tasks in intelligent transportation systems, but complex time series characteristics and nonlinear fluctuations pose challenges to prediction models. Traditional ARIMA and SVR models have limitations in capturing long-term and short-term dependencies and processing multi-modal data, while deep learning techniques provide a new solution for dynamic traffic flow prediction. This study proposes a dynamic prediction model that combines time series data with deep learning. Through multi-modal data fusion and the optimisation of deep learning models (LSTM and Transformer), high-precision traffic flow predictions are achieved. According to experimental data, the developed model can handle complicated traffic conditions and unforeseen occurrences better than existing approaches, and it also greatly outperforms them in terms of forecast accuracy and resilience. The research results provide technical support for real-time traffic regulation and resource optimisation in intelligent transportation systems, and demonstrate the broad application prospects of deep learning in the construction of future smart cities.