Calibration of a FOSIM Model for a Section of the N2 Freeway in Cape Town: Weaving Section at an Off-Ramp
| dc.contributor.advisor | Zuidgeest, Marcus | |
| dc.contributor.author | Dlamini, Bongani | |
| dc.date.accessioned | 2024-07-04T14:10:37Z | |
| dc.date.available | 2024-07-04T14:10:37Z | |
| dc.date.issued | 2024 | |
| dc.date.updated | 2024-07-02T14:11:24Z | |
| dc.description.abstract | Driver behaviour differs from person to person depending on the driver's behavioural characteristics, the vehicle used and road geometry. Traffic models must account for this variation in behaviour to properly analyse this complex reality. Typically, the variation is modelled by distinguishing between different ‘vehicle-driver combinations', whose behaviour is established by the settings of model parameters. Adjusting the parameters to make the model deliver realistic results is called model calibration. This is done by comparing the model results with field observations and adjusting the parameters systematically until the model results and empirical data align as much as possible. Subsequently, model validation is carried out to test the applicability of the calibrated parameters on a dataset separate from that used in the calibration process. This research is aimed at calibrating and validating a microscopic simulation model for a section of the N2 freeway using FOSIM software. FOSIM, a software originally developed to simulate Dutch freeway traffic, is used by road authorities in The Netherlands and some parts of the UK to test freeway capacity. Its simplistic, intuitive, and user-friendly nature makes it a more efficient and cost-effective tool for assessing traffic management and operations and would be of great benefit to the South African market and transport authorities. Since the software is calibrated and validated for the Dutch freeway scenario, it needs to be tested and adapted to the local context in order to be applied on South African freeways. The selected network of analysis falls within Section 1 of the National Road N2 (Settlers Way) in Cape Town between the M4 off-ramp and Hospital Bend (M3) in the in-bound direction. Peak hour traffic volumes were obtained using CCTV footage supplied by the Cape Town Freeway Management System over a period of three days. Travel time and speed data for the same dates was obtained from floating car runs and supplemented by CCTV observations. Based on this collected data, the Measures of Effectiveness (MoEs) examined in the study were traffic volumes, speed and travel time. The modelling commenced with coding the road geometry and the applicable traffic demand. The calibration process was carried out using a series of steps, the first of which was a sensitivity analysis. A sensitivity analysis was conducted to determine the relative influence of each of the parameters on the simulated output. Through this analysis, it was discovered that 36% of the parameters did not have a significant influence on the simulated output and were thus omitted from the calibration. On the other hand, the parameters that showed significant effect were carried over to the next step which included establishing the search range in which the optimum parameter values lie. This was done using the Golden Selection Method (GSM), which is an iterative, singleparameter search technique that seeks to minimize the standard error between the simulated output and empirical data. The initial model, which was run with default parameters, showed good results when assessed on traffic volumes. This model had a GEH statistic of less than 5 for 97% of the simulated values and an R-squared value of 0.88, both signifying a very high correlation between the simulated and empirical traffic volumes. However, travel time and speed exhibited high discrepancies between the empirical average and model output, with all the simulated values falling outside the acceptable range and the average percentage error reaching 40.5% - clearly indicating problems with the model. When the local conditions were calibrated based on floating car run data, travel time and speed results were improved significantly with only 21% of simulated values falling out of the acceptable range. The average error decreased from 40.5% to 3.8%. On the other hand, the traffic volume MoE worsened, with the GEH being less than 5 for only 63% of the simulated values and the R-squared consequently decreased to 0.42. The model was then advanced further to be calibrated globally by adjusting model parameters. After several iterations and multiple improvements to the model, the final model with the optimum parameter values demonstrated a significant improvement in the simulated volumes. The average GEH statistic decreased from 4.56 to 2.2. The GEH was also less than 5 for 97% of the simulated values, which was an improvement from the 63% obtained in the previous model. Furthermore, the R-squared increased from 0.42 in the previous model to 0.82. Travel time and speed also exhibited improved results when compared to the previous model, with only 1.6% of the simulated values falling out of the acceptable range as opposed to 21% in the previous model. Following the calibration process, the model which best estimated the empirical data measurement was validated. The validation results showed that even though the established parameters produced an excellent calibration model, the same parameters failed to produce sufficiently good validation results. The average GEH value had increased to 3 and was less than 5 for only 77% of the simulated values compared to 97% in the calibration model. The travel time and speed were, however, well within the acceptable range. In order to overcome the unsatisfactory GEH results obtained, the calibration parameters had to the fine-tuned further over several models and iterations. The final calibration and validation models both passed the acceptance test on all three MoEs. Statistical analysis also showed that the models were acceptable. The FOSIM model was thus successfully calibrated and validated, and the objectives of the research were duly met. Some of the take-homes from the findings of this study include the observation that using a single MoE in model calibration can be problematic and misleading. This is exhibited by the initial model that produced excellent volume predictions while travel time and speed were extremely inaccurate. The research also showed that fine-tuning the parameters to explain the calibration too well leads to overfitting the model which causes other problems. This explains why a good calibration model failed when it was initially validated. Finally, it is recommended that future studies explore a sensitivity analysis which will consider interactive effect between model parameters. It is also recommended that a more generalized models with an acceptable margin of error be developed to avoid overfitting and to enable the models to be applicable on a wider network range | |
| dc.identifier.apacitation | Dlamini, B. (2024). <i>Calibration of a FOSIM Model for a Section of the N2 Freeway in Cape Town: Weaving Section at an Off-Ramp</i>. (). ,Faculty of Engineering and the Built Environment ,Department of Civil Engineering. Retrieved from http://hdl.handle.net/11427/40350 | en_ZA |
| dc.identifier.chicagocitation | Dlamini, Bongani. <i>"Calibration of a FOSIM Model for a Section of the N2 Freeway in Cape Town: Weaving Section at an Off-Ramp."</i> ., ,Faculty of Engineering and the Built Environment ,Department of Civil Engineering, 2024. http://hdl.handle.net/11427/40350 | en_ZA |
| dc.identifier.citation | Dlamini, B. 2024. Calibration of a FOSIM Model for a Section of the N2 Freeway in Cape Town: Weaving Section at an Off-Ramp. . ,Faculty of Engineering and the Built Environment ,Department of Civil Engineering. http://hdl.handle.net/11427/40350 | en_ZA |
| dc.identifier.ris | TY - Thesis / Dissertation AU - Dlamini, Bongani AB - Driver behaviour differs from person to person depending on the driver's behavioural characteristics, the vehicle used and road geometry. Traffic models must account for this variation in behaviour to properly analyse this complex reality. Typically, the variation is modelled by distinguishing between different ‘vehicle-driver combinations', whose behaviour is established by the settings of model parameters. Adjusting the parameters to make the model deliver realistic results is called model calibration. This is done by comparing the model results with field observations and adjusting the parameters systematically until the model results and empirical data align as much as possible. Subsequently, model validation is carried out to test the applicability of the calibrated parameters on a dataset separate from that used in the calibration process. This research is aimed at calibrating and validating a microscopic simulation model for a section of the N2 freeway using FOSIM software. FOSIM, a software originally developed to simulate Dutch freeway traffic, is used by road authorities in The Netherlands and some parts of the UK to test freeway capacity. Its simplistic, intuitive, and user-friendly nature makes it a more efficient and cost-effective tool for assessing traffic management and operations and would be of great benefit to the South African market and transport authorities. Since the software is calibrated and validated for the Dutch freeway scenario, it needs to be tested and adapted to the local context in order to be applied on South African freeways. The selected network of analysis falls within Section 1 of the National Road N2 (Settlers Way) in Cape Town between the M4 off-ramp and Hospital Bend (M3) in the in-bound direction. Peak hour traffic volumes were obtained using CCTV footage supplied by the Cape Town Freeway Management System over a period of three days. Travel time and speed data for the same dates was obtained from floating car runs and supplemented by CCTV observations. Based on this collected data, the Measures of Effectiveness (MoEs) examined in the study were traffic volumes, speed and travel time. The modelling commenced with coding the road geometry and the applicable traffic demand. The calibration process was carried out using a series of steps, the first of which was a sensitivity analysis. A sensitivity analysis was conducted to determine the relative influence of each of the parameters on the simulated output. Through this analysis, it was discovered that 36% of the parameters did not have a significant influence on the simulated output and were thus omitted from the calibration. On the other hand, the parameters that showed significant effect were carried over to the next step which included establishing the search range in which the optimum parameter values lie. This was done using the Golden Selection Method (GSM), which is an iterative, singleparameter search technique that seeks to minimize the standard error between the simulated output and empirical data. The initial model, which was run with default parameters, showed good results when assessed on traffic volumes. This model had a GEH statistic of less than 5 for 97% of the simulated values and an R-squared value of 0.88, both signifying a very high correlation between the simulated and empirical traffic volumes. However, travel time and speed exhibited high discrepancies between the empirical average and model output, with all the simulated values falling outside the acceptable range and the average percentage error reaching 40.5% - clearly indicating problems with the model. When the local conditions were calibrated based on floating car run data, travel time and speed results were improved significantly with only 21% of simulated values falling out of the acceptable range. The average error decreased from 40.5% to 3.8%. On the other hand, the traffic volume MoE worsened, with the GEH being less than 5 for only 63% of the simulated values and the R-squared consequently decreased to 0.42. The model was then advanced further to be calibrated globally by adjusting model parameters. After several iterations and multiple improvements to the model, the final model with the optimum parameter values demonstrated a significant improvement in the simulated volumes. The average GEH statistic decreased from 4.56 to 2.2. The GEH was also less than 5 for 97% of the simulated values, which was an improvement from the 63% obtained in the previous model. Furthermore, the R-squared increased from 0.42 in the previous model to 0.82. Travel time and speed also exhibited improved results when compared to the previous model, with only 1.6% of the simulated values falling out of the acceptable range as opposed to 21% in the previous model. Following the calibration process, the model which best estimated the empirical data measurement was validated. The validation results showed that even though the established parameters produced an excellent calibration model, the same parameters failed to produce sufficiently good validation results. The average GEH value had increased to 3 and was less than 5 for only 77% of the simulated values compared to 97% in the calibration model. The travel time and speed were, however, well within the acceptable range. In order to overcome the unsatisfactory GEH results obtained, the calibration parameters had to the fine-tuned further over several models and iterations. The final calibration and validation models both passed the acceptance test on all three MoEs. Statistical analysis also showed that the models were acceptable. The FOSIM model was thus successfully calibrated and validated, and the objectives of the research were duly met. Some of the take-homes from the findings of this study include the observation that using a single MoE in model calibration can be problematic and misleading. This is exhibited by the initial model that produced excellent volume predictions while travel time and speed were extremely inaccurate. The research also showed that fine-tuning the parameters to explain the calibration too well leads to overfitting the model which causes other problems. This explains why a good calibration model failed when it was initially validated. Finally, it is recommended that future studies explore a sensitivity analysis which will consider interactive effect between model parameters. It is also recommended that a more generalized models with an acceptable margin of error be developed to avoid overfitting and to enable the models to be applicable on a wider network range DA - 2024 DB - OpenUCT DP - University of Cape Town KW - Civil Engineering LK - https://open.uct.ac.za PY - 2024 T1 - Calibration of a FOSIM Model for a Section of the N2 Freeway in Cape Town: Weaving Section at an Off-Ramp TI - Calibration of a FOSIM Model for a Section of the N2 Freeway in Cape Town: Weaving Section at an Off-Ramp UR - http://hdl.handle.net/11427/40350 ER - | en_ZA |
| dc.identifier.uri | http://hdl.handle.net/11427/40350 | |
| dc.identifier.vancouvercitation | Dlamini B. Calibration of a FOSIM Model for a Section of the N2 Freeway in Cape Town: Weaving Section at an Off-Ramp. []. ,Faculty of Engineering and the Built Environment ,Department of Civil Engineering, 2024 [cited yyyy month dd]. Available from: http://hdl.handle.net/11427/40350 | en_ZA |
| dc.language.rfc3066 | Eng | |
| dc.publisher.department | Department of Civil Engineering | |
| dc.publisher.faculty | Faculty of Engineering and the Built Environment | |
| dc.subject | Civil Engineering | |
| dc.title | Calibration of a FOSIM Model for a Section of the N2 Freeway in Cape Town: Weaving Section at an Off-Ramp | |
| dc.type | Thesis / Dissertation | |
| dc.type.qualificationlevel | Masters | |
| dc.type.qualificationlevel | MSc (Eng) |