Accurately predicting the turning movement flow at intersections is crucial for optimizing traffic fluidity. Existing traffic flow prediction models primarily focus on segment-level traffic, leaving a gap in the study of turning movement flow at intersections. The challenge in turning prediction lies in handling the nonlinear patterns of intersection turning movement while meeting real-time requirements. In this context, we propose a novel Parallel Hybrid Deep Learning Model (PHDLM) for accurately predicting turning movement flow at intersections. This model synergistically integrates Convolutional Neural Networks (CNN), attention mechanisms, Deep Neural Networks (DNN), and Bidirectional Gated Recurrent Units (BiGRU) to capture complex temporal patterns in traffic data. Through extensive experiments on the “Milton Turning Movement Flow Dataset,” we demonstrate that PHDLM outperforms established benchmarks in terms of accuracy, stability, and generalization capability.