Improvements to ramp metering system in Jill Hayden Managing Consultant Intelligent Transport Systems Roger Higginson Senior Systems Engineer Intelligent Transport Systems Abstract The Highways Agency has installed ramp metering at over 8 sites in England, with evaluation results showing an average 13% reduction in journey times 1. Despite the success of the existing system, some potential improvements have been identified. Due to the complexity of the algorithms, the improvements were modelled using the micro-simulation package VISSIM 2. This paper describes the method and results of the modelling. The modelling exercise has shown that all the algorithms have worked as expected. It has also shown that all the algorithms designed could provide additional performance benefits over the existing system. Matthew Hall Managing Consultant Intelligent Transport Systems Background Sukhvinder Ubhi Project Sponsor Highways Agency Ramp metering was first trialled in England in 1986, with a small number of sites around Birmingham and then Southampton. From the outcome of these trials the Highways Agency (HA) commissioned Atkins to design a system for wider roll-out, and IPL & PEEK to implement it. This system was first installed in 25 and there are now over 8 sites in England. The existing system was not modelled prior to implementation and no sites were modelled before being installed, for two main reasons. Firstly the basic concepts of the algorithms (ALINEA on the main carriageway, queue management and queue override on the slip road) were well understood and had been successfully used elsewhere, giving confidence that the system would work. Secondly, the timescales did not allow it. Since the system was first installed, several improvements have been identified with the following potential benefits: A reduction in the amount of calibration required Improvements to the operational benefits Widening of the applicability of technique These improvements are described in detail in the paper Improvements to Ramp Metering System in England: Detailed Description of Algorithm Development 3. As these are new ideas and some of the solutions are quite complex, it was decided that modelling should be used to check their feasibility and that they would work as expected, before producing the final specifications. Although there are a range of micro-simulation modelling packages available such as PARAMICS, AIMSUN and DRACULA, the VISSIM software from Visual Solutions was chosen for this assessment due to its strength in junction modelling, both signalled and un-signalled. VISSIM is a microscopic traffic flow simulation model based on car following and lane change logic. Furthermore, it includes the Vehicle Actuated Programming (VAP) integrated package that enables complex algorithms to be modelled. 51
Improvements to ramp metering system in Methodology The project made use of an existing VISSIM model of the current ramp metering system at M1 Junction 33 southbound, which comprised: A geometrically correct road layout of the junction MIDAS loops and traffic signals accurately located compared to the real system on the road The ramp metering system algorithms, coded using VAP programming to represent the system The model was improved and recalibrated for the purpose of this project. This provided the base ramp metering model, against which the improvements could be compared. The improvements were each implemented in the VAP code and then calibrated. To compare the performance of the ramp metering system, with and without each improvement, various measures were used such as: Network performance Travel/delay times RM controller outputs Speed profile This enabled conclusions to be drawn about the success of each improvement. Calibration and validation of existing ramp metering The existing model was re-calibrated, for both pre and existing ramp metering conditions. The validation showed excellent comparisons between actual on-road and modelling results. For example, Figure 1 shows the comparison of upstream occupancies for existing ramp metering conditions, for real MIDAS data over two days, and the model. These results gave confidence that the model was suitable for the purposes of the project. The evaluation of the ramp metering system implemented at M1 J33 southbound, showed a 9.1% reduction in journey time on the main carriageway. The model results showed a 7.4% decrease which is of a similar order; this gave confidence in the model. The model also showed that an overall journey time reduction of 4.2% was achieved for both main carriageway and slip road. Occ [%] 35 3 25 2 15 5 Upstream Occupancy Comparison by Day 15:45 16: 16:15 16:3 16:45 17: 17:15 17:3 17:45 18: 18:15 18:3 18:45 19: Time Figure 1 - Comparison of upstream occupancies for model and real data Modelling of improvements The following sections describe the results of modelling the four key algorithms. Although the algorithms are only briefly explained here, detailed descriptions are available in the paper Improvements to Ramp Metering System in England: Detailed Description of Algorithm Development. Table 1 - Comparison of total vehicle journey times for auto-signal timings 25/1/8 31/1/8 Model Auto-calibration of signal timings (Auto-Sig) The main purpose of this algorithm was to reduce the significant time and effort required to calibrate the traffic signal timings for each release level. If the actual flow achieved by a release level is lower than the required flow, then the red time is reduced, and vice versa. The modelling showed that overall performance in terms of journey times through the junction is significantly improved, as shown in Table 1. Pre-RM 8.6-84.5-93.1-934.2-7.4% 112.6 33.3% 46.8-4.2% Auto-Sig 9.8 -% 5.6 24.9% 6.3-7.9% 8 7 6 5 3 2 Speed Profile - Average of Random Seeds 9 12 15 18 2 2 27 3 33 36 39 42 45 48 5 5 57 6 63 66 69 72 75 78 8 8 87 9 93 96 99 2 5 8 11 1 117 12 Figure 2 - Comparison of speeds with existing RM and with auto-sig HAQ24 52
Improvements to ramp metering system in It can be seen that both the main carriageway and slip road journey times are reduced, compared to the existing ramp metering scenario. This is because having accurate release flows allows more efficient operation of both the ALINEA and queue management algorithms. The modelled overall journey times are improved by 7.9% from the preramp metering scenario, compared with a modelled improvement of 4.2% with existing ramp metering. This is an excellent result because as well as fulfilling the main objective of successfully reducing calibration time, the algorithm should also significantly reduce journey times. Figure 2 shows the comparison of the speed throughout the peak, for the existing ramp metering (in red), and the auto-signal timings scenario (in green). It can be seen that speeds are higher with autosignal timings for the whole of the peak, which explains the significant improvement in journey times. The modelling showed that even when the red times for all release levels are initially set to the same value (15s), the algorithm calibrates the signal timings fairly successfully by the end of one day of operation (iteration 1 in Figure 3). The system improves further to converge to sensible signal timings within a week of being installed (iteration 7). Auto-calibration of Queue Management Parameters (Auto-QM) This algorithm dynamically updates the values of two key parameters in the QM algorithm; the desired combined queue occupancy o descq and the queue management gain factor K poqm. This is designed to remove the significant effort required to calibrate the proportional occupancy queue management algorithm. Figure 4 shows the comparison of the speed throughout the peak, for the existing ramp metering (in red), and the auto-qm scenario (in green). Time in Seconds 2. 19. 18. 17. 16. 15. 14. 13. 12. 11.. 9. 8. 7. 6. 5. 4. 3. 2. Red Time by Release Level for Iterations 1 2 3 4 5 6 7 8 9 Iteration Figure 3 - Variation in red times per release level for first days operation 8 7 6 5 3 2 Speed Profile - Average of Random Seeds 9 12 15 18 2 2 27 3 33 36 39 42 45 48 5 5 57 6 63 66 69 72 75 78 8 8 87 9 93 96 99 2 5 8 11 1 117 12 Figure 4 - Comparison of speeds with existing RM and with Auto-QM Table 2 - Comparison of total vehicle journey times for Auto-QM The figure shows that the Auto-QM scenario is slower to break down than base ramp metering, makes a quicker minor recovery but is then slower to recover in the latter periods. This algorithm has an opposite effect to the Auto-sig algorithm; the main carriageway journey time is slightly worse than with base ramp metering but the slip road is significantly better (because the queue is managed more efficiently). This results in a slight benefit overall, as shown in Table 2. HAQ2 Pre-RM 8.6-84.5-93.1-934.2-7.4% 112.6 33.3% 46.8-4.2% Auto-Sig 938.4-7.% 3.6 22.6% 41.9-4.7% RL1 RL2 RL3 RL4 RL5 RL6 RL7 RL8 RL9 RL 53
Improvements to ramp metering system in Merge Control The Merge Control algorithm attempts to prevent saturation of the merge area which can be caused by the high flows requested by queue management and queue override algorithms at certain sites. A new set of loops is placed downstream of the stop-line within the nose area (approximately 2m downstream of the stop line for this model) to determine when the merge area is congested. When this occurs, the merge control algorithm limits the release levels used by the traffic signals, i.e. it limits the flow from the signals. The journey times on both the main carriageway and on the slip road are lower than for existing ramp metering, with an overall very significant benefit (see Table 3). Figure 5 shows the comparison of speeds throughout the peak, for the existing ramp metering (in red), and the Merge Control scenario (in green). The figure shows that the Merge Control model does not break down as far and is quicker to recover. This algorithm showed the largest overall benefits in the modelling. These benefits will only be gained at a relatively small number of sites where the merge area is regularly saturated by the slip road flow. At junctions with no regular merge problem, incidents may occasionally cause a saturated merge area and on those occasions Merge Control would also be beneficial. AINEA Cascaded with Demand-Capacity (ACDC) The ACDC algorithm combines the best characteristics of the ALINEA and Demand Capacity (DC) algorithms, by cascading them (putting the output of the ALINEA equation into the DC equation). This was expected to provide performance benefits. Table 3 - Comparison of total vehicle journey times for Merge Control 8 7 6 5 3 2 Speed Profile - Average of Random Seeds Figure 6 shows the comparison of speeds throughout the peak, for the existing ramp metering (in red), and the ACDC scenario (in green). This shows that the ACDC model is slower to break down and slightly quicker to recover, although it does break down further than the base. The journey times on the main carriageway are slightly lower and on the slip road are marginally lower than for existing, with an overall slight benefit (see Table 4). Pre-RM 8.6-84.5-93.1-934.2-7.4% 112.6 33.3% 46.8-4.2% Auto-Sig 888.1-11.9% 2.5 21.3% 99.6-9.4% 8 7 6 5 3 2 Speed Profile - Average of Random Seeds 9 12 15 18 2 2 27 3 33 36 39 42 45 48 5 5 57 6 63 66 69 72 75 78 8 8 87 9 93 96 99 2 5 8 11 1 117 12 Figure 5 - Comparison of speeds with existing RM and with Merge Control 9 12 15 18 2 2 27 3 33 36 39 42 45 48 5 5 57 6 63 66 69 72 75 78 8 8 87 9 93 96 99 2 5 8 11 1 117 12 Figure 6 - Comparison of speeds with existing RM and with ACDC HAQ27 HAQ35 54
Improvements to ramp metering system in Table 4 - Comparison of total vehicle journey times for ACDC Pre-RM 8.6-84.5-93.1-934.2-7.4% 112.6 33.3% 46.8-4.2% Auto-Sig 922. -8.6% 112.2 32.7% 34.2-5.4% Table 5 - Predicted journey time savings with four improvements Algorithm Improvement JT Improvement in model - above 4.2% with existing RM Auto-Sig 3.7% 8 Auto-QM.5% 8 Merge Control 5.2% 12 ACDC 1.2% 8 Number of sites Conclusions The VISSIM modelling has shown that all algorithms developed work as expected in the model, which provides confidence that they could be implemented successfully. It has also been shown that all four algorithms provided additional performance benefits over and above the existing system. Table 5 shows the predicted journey time savings for each algorithm. While the results of the modelling do not guarantee the same results on the road, they do give increased confidence that greater journey time benefits can be achieved. Acknowledgements This paper draws on the work undertaken for Highways Agency Network Services Directorate and is published with the permission of the Highways Agency. The views contained in this paper are those of the authors and not necessarily those of the Highways Agency. 55
Improvements to ramp metering system in References 1. 2. 3. Ramp Metering Operational Assessment, Highways Agency Report, April 28, http://www.highways.gov.uk/knowledge/17375.aspx VISSIM Visual Solutions software, www.vissim.com Higginson R, Hayden Dr J, Charton T, Ubhi S, Improvements to Ramp Metering System in England: Detailed Description Of Algorithm Development, in Proceedings of ITS World Congress 29 (Stockholm, 21-25 September 29).