Dear all,
This week's DMAS & CMS talk will be on Wednesday, 2pm-4pm. The speaker is Ms Ho Yun Li and Ms Cheu Yea Loon from the Department of Civil Engineering and Department of Mathematical and Actuarial Science respectively, UTAR.
Date: 25 April 2012 (Wednesday)
Time: 2pm - 3pm
Venue: SD 102, Setapak Campus, UTAR, Kuala Lumpur.
Speaker: Ms Ho Yun Li
Title: Bi-objective Optimization of Exclusive Bus Lane Allocation and Scheduling in Urban Cities
Date: 25 April 2012 (Wednesday)
Time: 3pm - 4pm
Venue: SD 102, Setapak Campus, UTAR, Kuala Lumpur.
Speaker: Ms Cheu Yea Loon
Title: Sample Size Requirements in Validating the Asymptotic Results for Generalized Linear Models for Poisson-Distributed Data
Abstract (Ho Yun Li):
Exclusive bus lane is a strategy that reserves a lane of the roads for buses travel. It could improve the buses level of service by increasing their schedule adherence and travel speed. Nevertheless, the strategy draws two important decisions making problems which are the lane allocation and scheduling problems. The lane allocation determines which links/roads in the network that needs to have the exclusive bus lane while the scheduling problem determines the start time and the duration of the strategy. To find the optimal solutions for these problems, a bi-level bi-objective optimization problem could be formulated. The upper level is the bi-objective optimization model that aims to minimize the total travel time of general purpose traffic and buses simultaneously. The lower level is a microscopic traffic simulation model that could simulate drivers (general purpose traffic) response based on the lane allocation and scheduling solutions. The proposed model is solved by a hybrid non-sorting genetic algorithm (NGSA II) –Paramics. An illustrative case study of Klang Valley region is developed to test and evaluate the proposed methodology. Results show that both objectives (i.e. the total travel time of general purpose traffic and buses) are contradicted in which Pareto solutions are obtained.
Abstract (Cheu Yea Loon):
In recent decades, Generalised Linear Models (GLMs) have provided a means of analysing the data from various non-Normal distributions that Normal-based procedures such as Analysis of Variance and multiple regression are not suitable as an analysis tool. In GLMs, the link between the observations and the model is not as simple as it is in regression and ANOVA. The results we have on the variances of the parameter estimators in GLMs are 'asymptotic' or large sample results. This study attempts to find out the sample size needed so that the asymptotic results are reasonable valid, specifically for the Poisson log-linear models. In addition, the study aims to find out as well whether the sample size required relies on the values of the parameters, values of the predictors and number of observations assigned at each design point. Two types of Poisson log- linear model are involved in this study, namely the two-parameter Poisson log- inear model and the three-parameter Poisson log-linear model.
For two-parameter log-linear models, the results of the study indicate that 100 observations are basically sufficient for the asymptotic results to be reasonable valid but this depends on the values of the parameters and predictors that are involved in the model. Different values of the parameters and predictors may require a higher number of observations for the asymptotic results to be valid. This criterion also applies to the three-parameter log-linear models. The values of the parameter and predictors are crucial factors that affect the sample size required for the asymptotic results to be valid.
All are welcome.
Thank you very much.
Regards,
Event Committee.
Time: 2pm - 3pm
Venue: SD 102, Setapak Campus, UTAR, Kuala Lumpur.
Speaker: Ms Ho Yun Li
Title: Bi-objective Optimization of Exclusive Bus Lane Allocation and Scheduling in Urban Cities
Date: 25 April 2012 (Wednesday)
Time: 3pm - 4pm
Venue: SD 102, Setapak Campus, UTAR, Kuala Lumpur.
Speaker: Ms Cheu Yea Loon
Title: Sample Size Requirements in Validating the Asymptotic Results for Generalized Linear Models for Poisson-Distributed Data
Abstract (Ho Yun Li):
Exclusive bus lane is a strategy that reserves a lane of the roads for buses travel. It could improve the buses level of service by increasing their schedule adherence and travel speed. Nevertheless, the strategy draws two important decisions making problems which are the lane allocation and scheduling problems. The lane allocation determines which links/roads in the network that needs to have the exclusive bus lane while the scheduling problem determines the start time and the duration of the strategy. To find the optimal solutions for these problems, a bi-level bi-objective optimization problem could be formulated. The upper level is the bi-objective optimization model that aims to minimize the total travel time of general purpose traffic and buses simultaneously. The lower level is a microscopic traffic simulation model that could simulate drivers (general purpose traffic) response based on the lane allocation and scheduling solutions. The proposed model is solved by a hybrid non-sorting genetic algorithm (NGSA II) –Paramics. An illustrative case study of Klang Valley region is developed to test and evaluate the proposed methodology. Results show that both objectives (i.e. the total travel time of general purpose traffic and buses) are contradicted in which Pareto solutions are obtained.
Abstract (Cheu Yea Loon):
In recent decades, Generalised Linear Models (GLMs) have provided a means of analysing the data from various non-Normal distributions that Normal-based procedures such as Analysis of Variance and multiple regression are not suitable as an analysis tool. In GLMs, the link between the observations and the model is not as simple as it is in regression and ANOVA. The results we have on the variances of the parameter estimators in GLMs are 'asymptotic' or large sample results. This study attempts to find out the sample size needed so that the asymptotic results are reasonable valid, specifically for the Poisson log-linear models. In addition, the study aims to find out as well whether the sample size required relies on the values of the parameters, values of the predictors and number of observations assigned at each design point. Two types of Poisson log- linear model are involved in this study, namely the two-parameter Poisson log- inear model and the three-parameter Poisson log-linear model.
For two-parameter log-linear models, the results of the study indicate that 100 observations are basically sufficient for the asymptotic results to be reasonable valid but this depends on the values of the parameters and predictors that are involved in the model. Different values of the parameters and predictors may require a higher number of observations for the asymptotic results to be valid. This criterion also applies to the three-parameter log-linear models. The values of the parameter and predictors are crucial factors that affect the sample size required for the asymptotic results to be valid.
All are welcome.
Thank you very much.
Regards,
Event Committee.
No comments:
Post a Comment