High Performance Design of Traffic-Flow Prediction Using Artificial Neural Network

Traffic flow prediction is an important topic to be tackled in Intelligent Transportation Systems (ITS) especially after the obvious increase in the traffic flow of vehicles on the roads and highways in the 21st century. Hence, it is vitally important to present accurate and fast approaches to predict traffic flow. The rapid prediction will improve the overall efficiency of traffic and reduce road issues like heavy congestion and road accidents. In this paper, an optimized design of traffic flow prediction using Artificial Neural Network (ANN) is proposed to achieve high performance, efficient timing and resources usage. The design provide accurate prediction for traffic flow rapidly, without using complex machine learning models. The proposed design is implemented using the Multi Layer Perceptron (MLP) Neural Network (NN) model on ZYNQ System On Chip (SOC). In addition, the MLP NN model is implemented with different memory architectures & using floating point and fixed point formatted data to determine which model is more efficient and what is the case to use each model. It is proven that the MLP NN implemented on the field programmable gate array (FPGA) produces better timing results than the software version of the MLP NN implemented on the Graphics processing unit (GPU). The Block Random Access Memory (BRAM) architecture with fixed point data implemented on ZYNQ board improved the timing of the software by 79% for the 50 neurons model and 28% for the 100 neurons model. Moreover, the area usage for BRAM architecture with 50 neurons is reduced by 18% as compared to the other Double data rate (DDR) memory architecture. Finally, the power consumption is reduced by 3.9% for the 50 neurons model and by 5.8% for the 100 neurons model.