A hybrid traffic controller system based on flower pollination algorithm and type-2 fuzzy logic optimized with crow search algorithm for signalized intersections

In Turkey and in the world, both the population and the number of vehicles tend to increase. This increase also causes an increase in the demand for transportation, a density in urban road networks and therefore an increase in vehicle delays (Cakici and Murat 2019). Vehicle delay is one of the important problems affecting daily life. For this reason, reducing vehicle delays and at the same time managing traffic safely has become an important study topic in traffic engineering. Different signal control approaches such as fixed time and traffic actuated have been developed to solve this problem. However, it is observed that these control systems are insufficient to provide the balance between both traffic safety and maximum intersection performance. This is the main reason why many researchers have been interested in investigating the applicability of different techniques based on particularly artificial intelligence in intersection control. A wide range of techniques such as artificial neural networks (ANN), fuzzy logic (FL), meta-heuristic algorithms have been implemented and effective results have been obtained compared to traditional approaches.

The researchers are focused on the development of control systems based on time or phase optimization for minimizing intersection performance indicators such as delay, number of stops, emission, queue length. Trabia et al. (1999) developed a control system that decides on the extension of the current signal timing based on the FL approach according to real-time traffic demands. They compared this system with the traffic-actuated controller and found that they could reduce the average delay by 9.5%. Ceylan and Bell (2004) developed a system that optimizes signal timing based on genetic algorithm (GA) and determined the system performance index (PI) by means of TRANSYT software. Results of comparison with the mutually consistent (MC) solution approach showed that the GA approach is more effective. Chen and Xu (2006) applied the particle swarm optimization (PSO) algorithm to optimize both the membership functions and the rules of the FL. They showed that the optimized FL signal controller could significantly reduce the delay in time-varying traffic demands. Balaji and Srinivasan (2009) developed an urban traffic signal control system with Type-2 FL approach. They compared the developed system with the adaptive control and Type-1 Fl control systems according to 25 different intersection data in Singapore. They could reduce the delay value by approximately 66%. He and Hou (2012) determined the signal durations using the ant colony algorithm (ACO). Delay, number of stops, and intersection capacity values were used as the PIs. When compared with the Webster model and the GA approach, it was revealed that the proposed approach was superior to the Webster model and performed better than the GA approach. Thus, they were able to provide intersection control with lower delays and stops. Sabetghadam et al. (2012) developed a Type-2 FL traffic signal controller and made a performance comparison based on queue length. In their comparison with fixed-time control, they showed that they could reduce queue length by approximately 40%. Gacovski et al. (2012) developed a fuzzy traffic controller and used stop times and queue length as input parameters. By optimizing signal times and phase order, they improved intersection performance and reduced stop time, delays, and queue length compared to the fixed-time control. Araghi et al. (2014) determined green times by using the neural network (NN) and the FL approaches in traffic signal control. They used the queue length in each link as the input parameter. They compared the NN and FL approaches with the fixed-time control system and significantly reduced the total delay. Dell’Orco et al. (2014) determined the signal timing with the controller they developed using the artificial bee colony (ABC) algorithm. They compared the ABC approach with GA and hill climbing (HC). It was found that ABC had better performance and this method could reduce delays by 2.4–2.7%. Bi et al. (2014) developed a new controller via type-2 FL whose membership functions and rule base optimized by differential evolution (DE). They compared this approach with type-1 FL and fixed-time signal control. They improved intersection performance by decreasing delay, queue length, and stop rate. In addition, they were able to enhance the vehicular throughput rate. Odeh et al. (2015) developed an adaptive traffic signal system using the FLC approach hybridized with GA. They reported that the hybrid system performed 31% better than the conventional FLC approach and achieved a 34% improvement over the traditional traffic signal controller. Araghi et al. (2015) optimized the NN and adaptive neuro-fuzzy inference system (ANFIS) traffic signal controllers using the cuckoo search (CS) algorithm. They made a performance comparison with the Q-learning and fixed-time controllers and revealed the performance of the CS algorithm. They found that the NN, ANFIS, and Q-learning controllers showed 44%, 39%, and 35% improvement compared to fixed-time control, respectively. Long et al. (2015) developed a traffic controller based on queue lengths, delay, and stop times via Q-learning algorithm. when compared the developed model with fixed-time control according to two different scenarios, it was seen that the queue lengths, delay, and stop times were significantly reduced. Yu et al. (2016) determined the cycle length and effective green times by optimizing traffic capacity, delay, stops, and exhaust emission parameters with GA. They performed their analysis with the Vissim simulation program and revealed that the method could effectively reduce delays and stops and improve capacity. Doğan and Akgüngör (2016) developed a FL traffic controller and optimized the FL controller with the DE algorithm. In two levels of optimization, they determined the minimum and maximum points of the membership functions (MFs) in the first level and rearranged the MFs in the second level to achieve the lowest delay at specified intervals. They were able to improve intersection performance by 52%, 48%, and 14%, respectively, according to low, medium, and high traffic conditions. Araghi et al. (2017) improved the control system performance by optimizing the interval type2-FL traffic signal controller (IT2ANFIS) with simulated annealing (SA), GA, and CS algorithms. They achieved 31% improvement with the CS approach, 17% improvement with the GA approach, and 3% improvement with the SA approach compared to fixed-time control. Khooban et al. (2017) optimized the type-2 FL system with the modified backtracking search algorithm (MBSA) and determined the signal timing and phase. They demonstrated that the new system reduced queue length and average delays and thus is superior to the fixed-time and type-1 FL control approaches. Zargari et al. (2018) developed internal–external traffic metering strategy (IETMS) model for high-priority congested networks using the SA algorithm and imperialist competitive algorithm (ICA). After the application of the novel model where traffic lights are optimized according to queue length, average speed and delay could be improved by 14% and 19%, respectively. Cakici and Murat (2019) developed a traffic control system using the DE algorithm. This approach in which the signal duration and phase were optimized was compared with fixed-time and the adaptive control systems. The average vehicle delay was able to reduce by 28–42% compared to the fixed-time control and 3–38% compared to the adaptive control. Korkmaz and Akgüngör (2021a) developed a traffic control system using the FPA algorithm. This approach in which the signal timing and phase were optimized was compared with fixed-time and the fuzzy logic control (FLC) systems. The average vehicle delay was able to reduce by 12–30% compared to the fixed-time control and 0.77–3% compared to the FLC system.

Applications of artificial intelligence approaches on traffic signal control have been carried out for a long time and successful results have been obtained. Especially with the recent developments, studies on determining the signal duration and phase plan are continuing rapidly. Some recent studies are given in Table 1.

The main motivation of this study is the development of a dynamic control system that can respond quickly to changes in traffic and allows intersection control with lower delay value. With many studies in the literature, it is seen that different artificial intelligence-based approaches have achieved successful results in intersection control. However, in most of the existing control systems, the signal timing optimization approach has been used, but it is not sufficient by itself to increase the intersection performance. At the same time, the phase plan also has an impact on the intersection performance and it is very important to create a dynamic phase plan. Therefore, it is aimed to develop a dynamic traffic control system that can adapt to changing traffic conditions and optimize the signal duration and phase plan. For the proposed dynamic traffic control system to perform better than the existing systems, the phase plan and signal times have been optimized as well as the phase sequence has been provided to be changeable. With this system developed in a modular structure, the effects of phase and time optimization can be clearly demonstrated. Traffic contains uncertainties due to its stochastic nature. A fuzzy logic approach is preferred to develop a control system that will adapt to these uncertainties. It is also known that the type-2 FL approach exhibits an effective performance in traffic control where there are many uncertainties. At the same time, effective optimization methods are needed to optimize the signal duration and adapt the fuzzy logic approach to different traffic situations. Although many different optimization methods have been used in this problem, fast and effective performance algorithms have been preferred. Flower pollination and crow search algorithms have shown successful results in many engineering problems. These algorithms have performed better than the widely used algorithms such as GA, PSO, DE, and they are able to reach solutions quickly. FPA is used to determine the phase plan that reveals the minimum delay value in the phase optimization module. As the objective function, minimizing the delay per vehicle has been determined and the HCM 2000 delay formula has been preferred in the calculation of the delay per vehicle. At the same time, the determination of the optimum cycle length of the determined phase plan is carried out in this module. In addition to the phase optimization, phase selection is also provided and which phase will be prioritized is determined in the determined phase plan. The phase order has been determined according to the number of vehicles belonging to the phases and has been created in such a way that priority has been given to the phase with the highest number of vehicles. In the time optimization module, it is decided to increase or decrease the signal times of an isolated intersection with a certain intersection geometry, phase plan, and optimum cycle length by a Type-2 fuzzy logic approach. Optimizing the fuzzy logic method according to different traffic situations allows intersection control with lower delay values and allows the system to adapt to every traffic situation. For this reason, membership functions of the fuzzy logic approach are optimized using the crow search algorithm (CSA). In summary, the performance of a new hybrid traffic signal controller, which determines the phase plan and optimum cycle length with flower pollination algorithm (FPA) and the change in signal timing with Type-2 FL optimized with CSA, has been investigated in this study. The most important contribution of this study and its distinction from other studies is that the signal timing, phase plan, and sequence are optimized together. In addition, the hybrid signal control system is the first control approach that can change the signal timing, phase plan, and sequence at the same time by constantly controlling traffic changes. On the other hand, FPA and CSA algorithms are not used alone or together in traffic control systems. Another innovative aspect of this study is to obtain better performance results by using both algorithms together and to examine the performances of the algorithms.

This section of the study provides a summary of the aims of the study and the review of the literature. In the second section, the FPA, CSA and Type-2 FL are explained and details of the developed traffic Simulation Program and hybrid traffic signal controller are given. In the third section, the performance indicators of the proposed approach are presented using statistical analyses. The results of the study are given in section four.

Minimizing the delay and maximizing the capacity in intersection control is the main objective. Therefore, due to the fact that the traffic is stochastic, an effective intersection control can be achieved by changing the signal timing or phase plan. Generally, optimization of the signal timing approaches is applied and the effect of the phase plan is ignored. Optimizing the signal timing, phase plan and sequence according to changing traffic conditions will create a more effective system than existing systems. A hybrid control system has been developed in order to reveal both the effects of signal timing optimization, phase optimization, and the performance of a dynamic control approach that can manage stochastic traffic. This system has been developed in a modular structure. Thus, the effects of phase and signal timing optimizations could be clearly demonstrated. FPA has been used to determine the phase plan that reveals the minimum delay value in the phase optimization module. At the same time, the determination of the optimum cycle length of the determined phase plan is carried out in this module. In the signal timing optimization module, it is decided to increase or decrease the signal timing with the Type-2 fuzzy logic approach at the isolated intersection whose geometry, phase plan, and optimum cycle length is determined. CSA is used to optimize the FL approach according to different traffic situations. The details of the FPA, CSA, and Type-2 FL approach and the literature summary of the algorithms are briefly explained in the 1st, 2nd, and 3rd subsections of this chapter. Information about the simulation program used is given in the 4th subsection. In the last part, detailed information about the developed signal control system method is given.

Performance comparison was made according to different traffic situations and intersection geometries. Nine different scenarios including low, medium and high traffic volumes were constructed. In addition, delay values were obtained by performing simulations according to four different intersection geometries including three legs and four legs. Intersection geometries are given in Figs. 11, 12. Assumptions for simulations are as follows.

In the simulation of traffic situations, the effect of different vehicle types on the delay was also taken into account by considering five different vehicle types. Cars, minibuses, buses, lorries and trucks with different lengths and acceleration–deceleration speeds were used. When creating traffic situations, 80% cars, 10% minibuses, 5% buses, 3% lorries and 2% trucks were identified as vehicle type distribution. The traffic conditions used for 3 and 4-leg intersections are given in Tables 9 and 10.

The control parameters of the algorithms are important for performance. A constant value (2000) was used in the number of iterations for algorithms, and all algorithms were run accordingly. In the FPA approach, two different control parameters are used: population size (N) and probability switch (P) parameters. Yang (2012) showed that the probability switch between 0.05 and 0.95 and population size between 10 and 50 was available, and examined the performance of different values. He stated that 25 for N and 0.8 for P were the most suitable operating conditions. There are three control parameters in the CSA approach: population size (N) and awareness probability (AP) and flight length (fl). (Askarzadeh 2016) stated that the AP parameter should be between 0.05 and 0.3 and the fl parameter should be between 1.5 and 2.5. By comparing the performances of different values, it showed that 0.05 for AP and 2 for fl were the best results. The control parameter values of the algorithms are used according to the suggested ones. Simulations of the HTSC, Fix-Time, Type-1FL-TSC (Doğan and Akgüngör 2016) and FPA-TSC (Korkmaz and Akgüngör 2021a) systems in which used only the phase optimization module were performed according to the traffic conditions created in four different intersection geometries. Since vehicle arrival in the simulation environment varies by seed number, obtaining delay values by a single seed number will result in misleading results. Therefore, each traffic situation was simulated according to 20 different seed numbers randomly selected, and the average of 20 different delay values obtained was determined as the delay value for each traffic situation. The distribution of delay values for traffic situations according to the simulation results is given in Figs. 13, 14, 15 and 16.

As can be seen in Figs. 13, 14, 15, 16, the fixed-time control approach performed intersection control with the highest delay values. The Type-1 FL-TSC approach outperformed fixed-time controls, but fell behind although it showed closer delay values to the FPA-TSC and HTSC approaches. Although the FPA-TSC and HTSC approaches showed very close performance to each other, the HTSC performed intersection control with the lowest delay values. The performance comparison of the control systems was made according to nine different traffic conditions and evaluated in terms of mean percentage errors (MPE) which is given in Eq. 11. Comparisons were also made with the non-parametric test. Friedman test, one of the post-hoc tests, was used to determine the location of differences between control systems. Descriptive statistics of control systems are given in Table 11. Friedman test statistics are given in Table 12. The hypotheses of the equality of the medians are rejected by p values that are smaller than 0.05, and there is a significant difference between the ranking of the observations. The MPE values for four different intersection geometries are given in Table 13.

Here, n is the number of samples, \({{\text{Delay}}}_{r}\) is the delay value of the control system to be referenced, and \({{\text{Delay}}}_{c}\) is the delay value of the control system which is calculated the percentage error.

According to the descriptive statistics and Friedman test statistics in Tables 11 and 12, the HTSC has the lowest delay (objective function) value for all type intersections. The FPA-TSC has better delay values than fixed-time and Type-1FL-TSC, and it remains in second performance place. Fixed time controller has the lowest performance and it has been seen that artificial intelligence approaches and control systems show better results. Thus, it has been observed that the HTSC performs better than other control systems.

As can be seen from Table 13, compared to fixed-time, the Type-1 FL-TSC, FPA-TSC and HTSC approaches could achieve 8.64%, 10.73% and 11.59% improvement at three-leg intersection, respectively. The FPA-TSC approach achieved 1.67% improvement over Type-1 FL-TSC, and the best performing HTSC improved intersection performance by 2.70% and 0.80% respectively compared to Type-1 FL-TSC and FPA-TSC. The fact that there is a left turn pocket at the three-leg intersection affects the performance of the control systems. At the intersection geometry with approximately three-fold improvements, the delay value was reduced by up to 30%. The best performing HTSC approach decreased the delay value by 5.70% and 5.04%, respectively, compared to Type-1 FL-TSC and FPA-TSC approaches. Although Type-1 FL-TSC and FPA-TSC approaches show performance close to each other, 0.71% improvement shows the effect of phase optimization on intersection performance. Type-1 FL-TSC, FPA-TSC and HTSC approaches control the intersection with less delay in low and medium traffic conditions in four-leg intersection geometries, while their performance and delay values reach the fixed-time values in high traffic conditions. The main reason for this situation is the saturation of the intersection due to the high levels of traffic and the decrease in the effect of the control systems. The HTSC approach showed the best performance in four-leg intersections and achieved 4.83% and 1.47% better results than the FPA-TSC approach, which shows the closest performance to it. Statistical results obtained by applying different artificial intelligence methods and control approaches showed that better intersection performance can be achieved than fixed-time control approach. Optimizing and dynamizing the signal duration together with phase and sequence optimization has been found to have a significant impact on intersection performance, with a performance improvement contribution of up to 5.04%. In addition, according to the Type-1 FL-TSC approach, which shows an effective performance, the hybrid system developed using type-2 FL and algorithms provides an improvement between 2.5 and 5.7%.

When the effectiveness of the developed control approaches is compared according to the variance and standard deviation statistics, the HTSC system showed the lowest values in four intersection types. The combination of signal timing and phase optimization approaches provide intersection control at the lowest delay values. Also, using different algorithms together and optimizing the FL approach according to traffic conditions have been effective in obtaining the best results. The control system, which only performs phase optimization with the FPA approach, also produced better results than fixed-time and Type-1 FL-TSC systems.

In intersection control, it is important to operate the intersection quickly and effectively. Therefore, the main mission of the intersection control problem is to develop a system that can adapt to any traffic condition and to improve intersection performance indicators such as minimum delay and maximum capacity. It is known that traditional methods fail to accomplish this mission and new approaches based on artificial intelligence play an effective role in achieving this goal. In particular, when intersection control systems based on pioneering approaches of artificial intelligence methods such as GA, ACO and the FL are inadequate, new approaches are applied to intersection control and the deficiencies of the existing systems can be eliminated, thus improving performance. The main objective of this study is to develop more effective intersection control and to obtain effective results in intersection control using the FPA, CSA and Type-2 FL approaches among the current and high-performance algorithms. The applicability of the hybrid control system developed by combining the unique advantages of the FPA and CSA algorithms and the Type-2 FL approach was demonstrated, and the hybrid control system was compared with Type-1FL-TSC, which has a successful performance. The following conclusions were drawn from this study.