A universal freeway incident detection framework is a task that remains unfulfilled despite the promising approaches that have been recently explored. The need for an operationally successful incident detection and management system as a vital component of any advanced traffic management system, is well established and recognized. Only recently however, researchers and practitioners have begun to increasingly realize that for an incident detection framework to be universally operational and successful, it needs to fulfill all components of a set of recognized needs. It is the objective of this research to define those universality requirements and produce an incident detection framework that possesses the potential to fulfill them.

A new potentially universal freeway incident detection framework has been proposed, developed and evaluated. The research effort was started by defining a comprehensive set of requirements that any universal incident detection algorithm or framework should fulfill. Among these requirements, an incident detection algorithm needs to be operationally accurate, automatically transferable, and capable of automatically adapting to changes in the freeway environment. This set of universality requirements was used as a template against which all algorithms within the scope of this study have been evaluated. Three major incident and loop detector databases were heavily utilized, two of which are unprecedented real databases collected from two major freeway sites in California and Minnesota, namely the Alameda County's I-880 freeway database and the Minneapolis' I-35W database. The universality of the most well known existing incident detection algorithms was tested using the above databases. Serious lack of the universality, particularly transferability, was detected in all existing algorithms. Prior to the development of the new universal framework, limits on acceptable performance were elicited from TMC surveys conducted as part of this effort. Preliminary investigation of two promising advanced neural networks, namely the LOGICON and the PNN, was conducted. The PNN was more appealing due to its universality potential. The PNN was modified using a principal components transformation layer that resulted in performance enhancements. This together with its potential universality, led to the choice of the modified PNN for in-depth development. The in-depth development stage was divided into three phases. The first was the extraction of a new and improved input feature set that produced more distinct classes in the input feature space. The new features enhanced the transferability of the PNN and made the framework more compliant with the universality requirements. The second phase was the on-site real time retraining of the PNN after transferability, a phase that produced near optimal classification results and detection performance. The third phase was the development of a post processor output interpreter that linked the isolated 30 second outputs of the PNN and produced a sequentially updated probabilistic measure of existence of an incident in the field. The overall PNN-based framework was found to be fully compliant with the entire set of universality requirements. Finally, a new approach for training a multi-smoothing-parameter version of the PNN was investigated. The approach utilized genetic algorithms for optimizing the selection of the smoothing parameters. Obtained results indicated an improvement in performance over the single smoothing parameter PNN but at the expense of longer training time.

The superiority and universality of a particular advanced neural network model, namely the PNN, was concluded in this research, as compared to the Logicon and the MLF neural networks, as well as existing conventional freeway incident detection algorithms. Adding the principal components transformation layer to the PNN was found to enhance its performance. Although the genetically optimized version of the PNN showed better transferability, both versions showed equally good performance after retraining. The PNN was concluded to be more practical for TMC implementation due to its instantaneous training capabilities.

In this research, the authors introduce and define a universal incident detection framework that is capable of fulfilling all components of a set of recognized needs. An algorithm is presented that has the potential to fulfill the defined universality requirements. It is a modified form of a probabilistic neural network (PNN) that utilizes the concept of statistical distance. The first part of the report presents a definition of the attributes and capabilities that a potentially universal freeway incident detection framework should possess. The second part discusses the training and testing of the PNN. The third section evaluates the PNN relative to the proposed universality template. In addition to a large set of simulated incidents, the authors utilize a large real incident database from the I-880 freeway in California to comparatively evaluate the performance and transferability of different algorithms including the PNN.

Proper prediction of traffic flow parameters is an essential component of any proactive traffic control system and one of the pillars of advanced management of dynamic traffic networks. In this paper, we present a new short term traffic flow prediction system based on an advanced Time Delay Neural Network (TDNN) model, the structure of which is optimized using a Genetic Algorithm (GA). After presentation of the model's development, its performance is validated using both simulated and real traffic flow data obtained from the California Testbed in Orange County, California. The model predicts flow and occupancy values at a given freeway site based on contributions from their recent temporal profile as well the spatial contribution from neighboring sites. Both temporal and spatial effects were found essential for proper prediction. An in-depth investigation of the variables pertinent to traffic flow prediction was conducted examining the extent of the "look-back" interval, the extent of prediction in the future, the extent of spatial contribution, the resolution of the input data, and their effects on prediction accuracy. Results obtained indicate that the prediction errors vary inversely with the extent of the spatial contribution, and that the inclusion of three loop stations in both directions of the subject station is sufficient for practical purposes. Also, the longer the extent of prediction, the more the predicted values tend toward the mean of the actual, for which case the optimal look-back interval also shortens. Interestingly, it was found that coarser data resolution is better for longer extents of prediction. The implication is that the level of data aggregation/resolution should be comparable to the prediction horizon for best accuracy. The model performed acceptably using both simulated and real data. The model also showed potential to be superior to such other well-known neural network models as the Multi layer Feed-forward (MLF) when applied to the same problem. Keywords: Traffic Flow Prediction, Neural Networks, Genetic Algorithms, Traffic Management.

A significant body of research on advanced techniques for automated freeway incident detection has been conducted at the University of California, Irvine (UCI). Such advanced pattern recognition techniques as artificial neural networks (ANNs) have been thoroughly investigated and their potential superiority to other techniques has been demonstrated. Of the investigated ANN architectures, two have shown the best potential for real-time implementation: namely, the Probabilistic Neural Network (PNN), (Abdulhai and Ritchie 1997), and the Multi-Layer-Feed-Forward Neural Network (MLF), (Cheu and Ritchie 1995). This project extended existing freeway incident detection research conducted under both PATH and under the ATMS Testbed Research Program, to operationalizes its principal findings. The most prosmising neural network, the PNN, was integrated into the UCI testbed for on line operation on the testbed network in Southern California. The PNN incident detection system was re-coded in Java, to facilitate network communications and platform-independent operation. A Java-based graphical user interface has been developed. The GUI components include a display of the probabilistic neural network (PNN), the current input to the PNN, a sliding window display of the output (the computed incident probability every time step) and a sliding button to allow the user to specify the desired misclassification cost ratio. The GUI code is in the form of a Java Applet object and has a modular structure that makes it easier to incorporate possible future modifications and extensions. The PNN algorithm itself was then translated from C to Java as a stand alone application object and was interfaced to the GUI applet running on the same host. The GUI display is updated each time a new output is computed by the PNN. The PNN algorithm and the GUI display update run as separate threads of control in Java; this concurrency leads to better utilization of CPU resources. A new module for computing the principal component transformation of the volume and occupancy inputs was developed to replace using statistical packages for this transformation. This was needed for maximum portability and independence of the overall system. Another module for computing volume and occupancy historical Averages for different Times and Locations (ATLs.) was alsodeveloped to prepare the ATLs from real freeway data. The PNN and GUI were tested and correct operation was confirmed with sample inputs from data files. The whole package was interfaced to a remote C++ CORBA client program that acquires online CalTrans traffic data from a CORBA server in the Testbed. Communication modules were added to the CORBA client program as well as the PNN to enable online volume and occupancy data from different freeway sections to be sent from the CORBA client to the PNN at a specific rate (once every 30s). The data are sent to the PNN using a reliable TCP/IP streams sockets connection. An on-line retraining module was developed as well. This module enables the TMC operator to initiate retraining on recently captured incident data, on-line without disturbing the operation of the system. The PNN was then started on-line on a 5 mile section of the 405 freeway, for on line monitoring and testing. The overall on-line operation of the PNN was demonstrated to Caltrans engineers from D12. Currently, efforts are in progress to expand the network coverage to enhance the odds of capturing incidents. On line evaluation will be performed next.