This study proposes an Artificial Neural Network (ANN) model and a Genetic Algorithm (GA) model to estimate the number of accidents (A), fatalities (F) and injuries (I) in Ankara, Turkey, utilizing the data obtained between 1986 and 2005. For model development, the number of vehicles (N), fatalities, injuries, accidents and population (P) were selected as model parameters. In the ANN model, the sigmoid and linear functions were used as activation functions with the feed forward‐back propagation algorithm. In the GA approach, two forms of genetic algorithm models including a linear and an exponential form of mathematical expressions were developed. The results of the GA model showed that the exponential model form was suitable to estimate the number of accidents and fatalities while the linear form was the most appropriate for predicting the number of injuries. The best fit model with the lowest mean absolute errors (MAE) between the observed and estimated values is selected for future estimations. The comparison of the model results indicated that the performance of the ANN model was better than that of the GA model. To investigate the performance of the ANN model for future estimations, a fifteen year period from 2006 to 2020 with two possible scenarios was employed. In the first scenario, the annual average growth rates of population and the number of vehicles are assumed to be 2.0 % and 7.5%, respectively. In the second scenario, the average number of vehicles per capita is assumed to reach 0.60, which represents approximately two and a half‐fold increase in fifteen years. The results obtained from both scenarios reveal the suitability of the current methods for road safety applications.

No abstract

The effectiveness of road traffic control systems can be increased with the help of a model that can accurately predict short-term traffic flow. Therefore, the performance of the preferred approach to develop a prediction model should be evaluated with data sets with different statistical characteristics. Thus a correlation can be established between the statistical properties of the data set and the model performance. The determination of this relationship will assist experts in choosing the appropriate approach to develop a high-performance short-term traffic flow forecasting model. The main purpose of this study is to reveal the relationship between the long short-term memory network (LSTM) approach's short-term traffic flow prediction performance and the statistical properties of the data set used to develop the LSTM model. In order to reveal these relationships, two different traffic prediction models with LSTM and nonlinear autoregressive (NAR) approaches were created using different data sets, and statistical analyses were performed. In addition, these analyses were repeated for nonstandardized traffic data indicating unusual fluctuations in traffic flow. As a result of the analyses, LSTM and NAR model performances were found to be highly correlated with the kurtosis and skewness changes of the data sets used to train and test these models. On the other hand, it was found that the difference of mean and skewness values of training and test sets had a significant effect on model performance in the prediction of nonstandard traffic flow samples.

This study aims at optimizing fuzzy logic controller (FLC) triangle membership functions (MFs) for different traffic volumes via differential evolution (DE). To achieve this goal, a new FLC with a red time limiter, which actually calculates green time and the extension time of traffic movement phase, is developed to control an intersection. Subsequently, this FLC is optimized with two levels, namely Level-1 and Level-2. Level-1 searches each fuzzy class’s minimum and maximum values (α and β) that generate the lowest average delay per vehicle with DE. Using DE Level-2 inherits Level-1 ranges and reshapes the MFs to explore lower delay values computed by Level-1. The proposed method is tested with nine different traffic scenarios. For each scenario, 15 different headways are applied for a four-leg isolated intersection. The results indicate that the intersection average performance is increased up to 52%, 48%, and 14% at 800, 1600, and 2400 veh/h total intersection volumes, respectively, after Level-1 optimizations. They also reveal that intersection control produces higher delay values in only four scenarios after Level-2 procedures. Consequently, it is shown that the DE has significant potential to optimize FLCs at the intersection signal control. In addition, tuning fuzzy class ranges is found to be more critical than the MF reshaping process in traffic control via FLCs.

No abstract