In this article, we use the neural network (NN) to study bicycle's conflict avoidance behaviors when crossing the un-signalized intersections under mixed traffic flow conditions. We introduce the basic concepts and learning rules of NNs, the structure and algorithm principle of back propagation (BP) NN, and modeling ideas and steps of applying NN to construct the bicycle conflict-avoidance (BCA) behavior model. We will then construct three types of BPNN-based BCA behavior models at un-signalized intersections. Finally we will verify the NN-based bicycle's conflict avoidance behavior model at un-signalized intersections by field data.

Chapter Contents:

• 10.1 Introduction to artificial neural networks (ANNs)
• 10.1.1 Artificial neuron model
• 10.1.2 Transfer function
• 10.1.3 Learning method and learning rules
• 10.1.3.1 Learning methods
• 10.1.3.2 Learning rules
• 10.2 How BP neural network works
• 10.2.1 Structure of BP neural network
• 10.2.2 BPNN standard learning process
• 10.2.2.1 Mathematical variables
• 10.2.2.2 Learning process
• 10.2.3 Problems and improvement of BP algorithm
• 10.3 Modeling of bicycle conflict-avoidance behavior based on NN
• 10.3.1 Determine inputs and outputs
• 10.3.2 Data normalization
• 10.3.3 Learning sample division
• 10.3.4 Neural network structure
• 10.3.4.1 The number of hidden layers
• 10.3.4.2 The number of hidden layer nodes
• 10.3.5 Training algorithm and parameter selection
• 10.3.5.1 Training algorithm selection
• 10.3.5.2 Training parameters selection
• 10.4 The bicycle conflict avoidance model based on BPNN
• 10.4.1 BPNN-based bicycle conflict avoidance model in B–C conflict situations
• 10.4.2 BPNN-based bicycle conflict avoidance model in B–B conflict situations
• 10.4.3 BPNN-based bicycle conflict avoidance model in B–P conflict situations
• 10.4.4 NN-based bicycle conflict avoidance model considering gender and type of conflict object
• 10.5 Model simulation and verification
• 10.5.1 Model simulation
• 10.5.1.1 Model initialization state extraction
• 10.5.1.2 Simulation of bicycles' conflict avoidance behavior
• 10.5.2 Model verification
• 10.5.2.1 Model curve comparisons
• 10.5.2.2 Comparison of mathematical statistics
• 10.5.2.3 Discussion of model validation results
• 10.6 Summary
• References

Inspec keywords: road traffic; bicycles; neural nets; velocity; backpropagation; collision avoidance; traffic engineering computing; acceleration; road vehicles

Other keywords: NN-based bicycle conflict avoidance behavior model; field data; microscopic behavior simulation models; back propagation algorithm; mixed traffic flow conditions; BPNN-based BCA behavior models; NN-based bicycle collision avoidance behavioral model; modeling ideas; NN learning rules; un-signalized intersection crossing

Subjects: Neural nets; Traffic engineering computing