The Effects of Autonomous Vehicles on Traffic Efficiency and Energy Consumption

Abstract

This paper studies how autonomous vehicles (AVs) influence future traffic and energy consumption when adopting various route guidance strategies, which can be divided into global information strategies and local information strategies according to the scope of information collection. A mixed traffic flow model is established. Then, an autonomous vehicle model is improved to investigate the impact of six route guidance strategies. The selected strategies are specifically evaluated in a double-route scenario, considering both single-exit and dual-exit configurations. Three indicators are chosen to evaluate traffic efficiency, including traffic flow, average speed and quantity of vehicles. Consumption per unit flux is the indicator of energy consumption level. Simulation results show that autonomous vehicles can improve traffic efficiency and reduce energy consumption.

1. Introduction

As urban areas continue to grow, the proliferation of motor vehicles has experienced a persistent escalation. Traffic congestion and energy consumption have gradually become two major problems faced by traffic managers [1,2]. To enhance traffic efficiency and mitigate energy consumption, the field of transportation has embraced the development of Intelligent Transportation Systems (ITS) [3,4]. These systems effectively integrate cutting-edge science and technology into various aspects, including the management of transportation, the manufacturing of vehicles, and the control of services [5,6]. By strengthening the links between vehicles, roads, and users, ITS has become an integrated transportation system that ensures safety, network efficiency and energy saving [7]. ITS has six components, the Advanced Traffic Information System (ATIS) is the most critical among the six components. ATIS leverages traffic information and employ sophisticated strategies of route guidance to ascertain the best choice for drivers. Therefore, the that provides guidance information to the driver through the real-time road conditions is a principal component of ATIS. An efficacious strategy of route guidance has the potential to enhance both the efficiency of the traffic network and mitigate energy consumption [8].

Autonomous vehicles (AVs) have developed rapidly in recent years and have gradually become an important part of the future ITS. Leveraging the synergistic capabilities of artificial intelligence, advanced visual computing algorithms, cutting-edge radar technology, sophisticated surveillance devices and precise global positioning systems, computers possess the remarkable potential to engage in autonomous and secure operation of motor vehicles, obviating the requirement for human intervention. Rakhmanov and Wiseman [9] developed an efficient method for compressing GNSS data for autonomous vehicles, improving safety and reliability by facilitating real-time, error-tolerant communication with other vehicles and infrastructure. Ozdemir [10] introduces a machine learning-based algorithm using GNSS observables to improve positioning estimations by detecting and excluding non-line-of-sight and multipath reflections in urban UAV applications. Compared to human-driven vehicles (HVs), autonomous vehicles can avoid safety hazards that may cause traffic accidents, such as fatigue, long reaction time and blind spots. With further improvements in automation, traffic accidents will rarely occur. In addition, autonomous vehicles are equipped with the monitoring device and automatic cruise system that makes the vehicle run smoothly and improves energy efficiency. In 2020, Chen [11] studied a mixed AV and HV system to help improve the management and control of traffic.

In this paper, autonomous vehicles’ effects on the efficiency of traffic and energy consumption were studied when adopting different route guidance strategies. The vehicle is set to be powered by electric energy. Across various speeds and modes of operations, namely deceleration, acceleration, and cruising, the electric vehicle exhibits distinct energy consumption patterns. The evaluation of traffic network efficiency entails assessing the traffic flux, average speed, and vehicle count, while energy consumption is measured by the consumption per unit flux metric. Within the realm of microscopic traffic simulation, the cellular automaton paradigm has gained significant prominence for its widespread application [12,13], and the rules of these models are simple and easy to apply and can be used to describe the complex traffic phenomenon. More importantly, in the following paragraph, we will see that many researches on route guidance strategies are carried out based on cellular automata, which means the using of cellular automata can better extend the comparability of research. Therefore, the most popular Nagel–Schreckenberg model (NS model) is selected as the mechanism of vehicle running and has been modified to meet the characteristics of autonomous vehicles.

The route guidance strategies can be divided into global information strategies and local information strategies according to the scope of information collection.

This study considers two distinct two-route scenarios: one with symmetrical routes and a single exit and another with symmetrical routes and two exits. These scenarios serve as the testing grounds for applying diverse route guidance strategies. The simulation outcomes unveil that, in terms of traffic efficiency within a mixed model, the strategies that leverage global information exhibit superior performance. Regarding energy consumption, autonomous vehicles outperform their human-driven counterparts, albeit with negligible disparities observed across the various route guidance strategies.

This paper unfolds as such: We delve into an extensive review in Section 2, which is conducted on the existing literature pertaining to route guidance strategies and energy consumption. Section 3 presents the pertinent definitions required for the subsequent analysis. Subsequently, in Section 4, we delineate and scrutinize the outcomes from the simulations. Lastly, Section 5 brings the discussion to a close by summarizing the key findings and drawing overall conclusions.

2. Literature Review

2.1. Route Guidance Strategies

Route guidance strategies play a crucial role in ATIS. The traffic control center disseminates guidance information through the advisory display at the entrance, guiding vehicles that are newly arrived to select appropriate routes. Over the years, extensive research has been conducted in this domain. In 2000, Wahle [14] introduced the travel time route guidance strategy (TTS). The TTS utilizes vehicle travel time as feedback information. Building upon this, Lee K [15] proposed the mean velocity strategy (MVS) in 2001. These advancements have significantly contributed to the evolution of route guidance methodologies. In 2005, Wang [16] proposed the CCS. This strategy first defined the concept of congestion coefficient and congestion cluster which had also been applied to other route guidance strategies. Chen [17] proposed the EFS later, which modified the calculation method of the congestion coefficient proposed by Wang [16] through an exponential function. Chen [18] also proposed two traffic flux route guidance strategies, including the time flux route guidance strategy (TFS) and space flux route guidance strategy (SFS). After that, the VLS was proposed in 2012 [19], which calculates the previous vehicleś distance to the entrance on each route at each time step. Dong [20] proposed vehicle number route guidance strategy (VNS) based on the number of vehicles. In a seminal publication by Wu in 2017 [21], a strategy of route guidance based on ant pheromone dynamics was introduced. Inspired by the behavior of ants, vehicles were conceptualized as unique entities analogous to these insects, while their traffic information was akin to the influential pheromone trails left by ants. By leveraging this novel perspective, the proposed strategy aimed to optimize traffic routing by mimicking the adaptive decision-making processes observed in ant colonies.

In the era of autonomous vehicles, the quest for enhancing traffic efficiency has led researchers to explore innovative approaches. Yuan [22] embarked on a comprehensive investigation of mixed traffic flow characteristics. Notably, Liu [23] made a seminal contribution in 2017 by introducing an advanced cellular automaton model that incorporated refined lane shifting strategies tailored to moderate and aggressive lane changing behaviors. Extensive simulation studies were conducted, revealing the transformative potential of integrating autonomous vehicles into the existing road traffic ecosystem. The results showcased substantial improvements in traffic flow dynamics, including noteworthy enhancements in road capacity and free-flow speeds. This breakthrough research underscores the significant role of autonomous vehicles in shaping the future of transportation systems.

2.2. Energy Consumption

The complex connection between a vehicle’s various functioning states at different velocities and its usage of energy has been the focus of considerable interest. Notably, in 2002, Ahn [24] pioneered the development of a range of energy models that captured the nuanced interplay between the speed, acceleration and energy consumption. In 2005, Chang and Morlok [25] presented a direct derivation of the theory that fuel consumption in land transport vehicles is minimized by constant-speed operation, confirmed by tests using a performance simulator. In Silva’s research, three models (EcoGest, CMEM, and ADVISOR) were evaluated for estimating fuel consumption [26]. In 2009, Song’s research [27] aimed to create a model to assess traffic management’s impact on fuel efficiency using data from 26 light-duty vehicles, which was then applied in various case studies. Odhams [28] investigated factors affecting heavy goods vehicles’ energy consumption and explored the effect of different vehicle configurations, concluding larger, fully-loaded vehicles are more energy-efficient, and regenerative braking can reduce fuel consumption. By the availability of signal information that may be provided through vehicle-to-infrastructure communication, Rakha and Kamalanathsharma [29] devised a schema to enhance the fuel efficiency of automobiles as they near a junction with traffic signals. These models were mainly used in gasoline and diesel-powered vehicles [30]. Subsequently, Shankar and Marco [31] proposed an innovative framework that could predict an electric vehicle’s (EV) energy consumption or a plug-in hybrid electric vehicle’s (PHEV) range of zero emissions across a route. In a seminal analysis conducted by Yao in 2013 [32], a comprehensive examination was undertaken to explore the nuanced distinctions in energy consumption factors for EVs across different road types. Building upon this investigation, Yao further extended the research in 2014 [33] by devising energy consumption rate models that accounted for diverse operational modes, including acceleration, deceleration, cruising and idling. These models were derived from meticulously collected data obtained through rigorous chassis dynamometer testing, adhering to the rigorous standards set forth by the New European Driving Cycle. Zhang and Yao [34] formulated a comprehensive method for approximating power usage in electric vehicles, grounded in both physical and statistical understanding and tailored for real-world driving patterns.

According to the review above, this paper chose three global information strategies and three local information strategies, namely CCS, MVS, SFS, VLS, APS and EFS. The consumption was based on the data collected by Yao [33] in 2014.

3. Related Definitions

3.1. NS Cellular Automaton Model

The cellular automaton, a formidable mathematical construct employed to replicate actual environments considering the tiniest details, has garnered significant attention within the scientific community [35,36,37,38]. Distinguished by its temporal and spatial discretization, this model assigns finite discrete states to each cell in a grid [39]. Notably, the NS model, recognized as the preeminent cellular automaton model, has enjoyed widespread adoption in diverse domains [40]. Characterized by its four fundamental rules [41] encompassing (a) acceleration, (b) deceleration, (c) randomization and (d) motion, the NS model has proven indispensable in comprehending intricate traffic dynamics, enabling comprehensive analyses and simulations that illuminate the multifaceted nature of vehicular systems.

3.1.1. Primary Mixed Vehicles Model

In the primary mixed vehicles model, AVs and HVs follow the rules of the NS model. Random brake is the deceleration behavior caused by the driver’s reaction time. For HVs, they will randomly slow down at this step with a certain probability and generally p takes 0.25. For AVs, vehicles are controlled by computer so their random brake probability equals 0.

3.1.2. Improved Mixed Vehicles Model

In the improved mixed vehicles model, the HVs follows the rules of NS model, and the rules of AVs can be decomposed into two steps.

(a)

Speed update

At time t, if the i-th vehicle on the road is an AV, the vehicle can collect the speed information of the front vehicle, and the rule is Equation (1).

$$v_i(t+1)=min\{v_i\left(t\right)+1,v_{max},d_i\left(t\right)+v_{x_{i+1}}(t+1)\}$$

(1)

where $v_{max}$ is the maximum speed allowed, d presents the longitudinal separation distance between two vehicles, $d_i\left(t\right)$ is the distance of the vehicle i with its front vehicle in the time step t, $v_x$ is a virtual speed and $v_{x_{i+1}}(t+1)$ represents the update speed of the front vehicle of vehicle i, that is, the update speed of vehicle $i+1$ in the next time step. When the previous vehicle of the i-th vehicle is an autonomous vehicle, its virtual speed is defined as Equation (2). Equation (2) not only considers the vehicle directly in front but also incorporates information from the vehicle even further ahead. This allows autonomous vehicles, when forming a platoon, to update their speed based on the state information of multiple preceding autonomous vehicles. This reflects the role of vehicle-to-vehicle communication among autonomous vehicles.

$$v_{x_{i+1}}(t+1)=min\{v_{i+1}\left(t\right)+1,v_{max},d_{i+1}\left(t\right)+v_{x_{i+2}}(t+1)\}$$

(2)

When the previous vehicle of the i-th vehicle is a human driving vehicle, then the human driving vehicle runs according to the NS model and then its virtual speed is the update speed of the next time step, as Equation (3)

$$v_{x_{i+1}}(t+1)=min\{v_{i+1}\left(t\right)+1,v_{max},d_{i+1}\left(t\right)\}$$

(3)

(b)

Position update

$$x_i(t+1)=x_i\left(t\right)+v_i(t+1)$$

(4)

Here, x refers to the cell number of a one-dimensional cell sequence, which can be understood as similar to the coordinate value. The speed here is actually the number of cells that the vehicle passes in a time step, so when the position is updated, the speed value is actually the variation amount of the cell number.

3.2. The Strategy of Route Guidance

The route guidance strategies can be divided into global information strategies and local information strategies. To secure the logic behind the contrast, our study has made some simple modifications to the route guidance strategies. Three global information strategies and three local ones were selected. The global information strategies group comprises CCS [16], MVS [15] and SFS [18]. The SFS proposed by Chen [18] only collected the spatial information of local cells. The local information strategies group compises VLS [19], APS [21] and EFS [17].

3.3. Two-Route Scenario

The pioneering work by Wahle [14] delved into the intricacies of the two-route scenario, an intriguing domain that has captivated researchers in the field. In this scenario, the entrance serves as the birthplace of a new vehicle, marking the commencement of each time step. The new vehicle will select the road according to the strategy of route guidance or randomly. When the vehicle cannot enter the road or has reached the end, it will be removed. In this scenario, the vehicles present two distinct types. Dynamic vehicles, characterized by a ratio denoted as $sdyn$, pertain to entities that adhere to prescribed route guidance strategies. In contrast, static vehicles, with a ratio of $1-sdyn$, represent a cohort that opts for route selection through random means.

In this paper, the scenario of two symmetric routes with one exit and two exits is studied and is shown in Figure 1 and Figure 2, and black vehicles represent static vehicles (which do not follow the route guidance), while green ones are dynamic vehicles (which follow the route guidance). Under the one-exit scenario, at each time step, the exit only allows one vehicle to pass. Therefore, we took the export rules proposed by Chen [19]: when the vehicles on the road reach the exit and meet the speed conditions off the road at the same time, the two vehicles are bound to interfere. Therefore, we need to make export rules as follows:

3.4. Energy Consumption

The consumption data were based on the data collected by Yao [33] in 2014. At different speed levels, the energy consumption of various operation modes (acceleration, deceleration, cruise) of the vehicle is shown in

Table 1. Consumption of various operation modes (Unit: W*S.)

Speed Level	Decelerate	Cruise	Accelerate
0	-	1528.9	1948.223
1	1804.714	1984.677	6651.58
2	1706.3	3531.064	14,594.53
3	1183.213	6379.672	-

. The speed levels are divided according to the New European Driving Cycle (NEDC).

4. Simulation Results

In the scenario of two symmetric routes, each path is designed to span 2000 segments, where each segment measures 7.5 m. Therefore, the total length is 15 km. In the NS model, to ensure accuracy, we perform 110,000 iterations for each route guidance strategy simulation and only retain the last 10,000 steps of the data. Then we repeat the process 20 times to take the average of all. To make the diagram of the simulation more readable, we specify the following unit abbreviations: N represents the number of vehicles passing, N/S represents the vehicle number per step, C/S represents the speed unit (cells per step) and W*S/F is the unit of average traffic flow power consumption and represents the watt * step per traffic flow unit.

4.1. Primary Mixed Vehicles Model

In this part, each autonomous vehicle is set to obey the strategy of route guidance. That is, the ratio of autonomous vehicles equals $sdyn$, and the ratio of human-driven vehicles is $1-sdyn$. The random brake probability of human-driven mode is 0.25, while the one of autonomous vehicles is 0.

4.1.1. Results of Different Route Guidance Strategies

We have divided the six route guidance strategies into two groups, namely, the global information strategy and the local information strategy. The former comprises CCS, MVS and SFS; the latter comprises VLS, APS and EFS. In Figure 3a, as $0<sdyn<1$, when adopting local information strategies, the traffic flow has always shown an upward trend with the increase in $sdyn$, and the average speed of the vehicles is gradually decreased and the vehicle numbers increased badly. When adopting a global information strategy, the traffic flow increases first and then decreases, the average speed rises slowly and the vehicle numbers perform a steady trend. In Figure 3b, vehicle speeds of local information strategies are much slower than that of global ones. Conversely, Figure 3c shows the opposite result that adopting local information strategies caused higher vehicle numbers. According to Figure 3a–c, the result of local information strategy performs a congestion state with high flow and low speed. Obviously, to improve traffic efficiency, global information strategies are better, of which CCS is the best choice.

Considering that total consumption of vehicles is related to traffic flow, Figure 3d shows the consumption per unit flux. Global information strategies always performed better than local ones. In each group, the lines of different route guidance strategies almost coincided. Therefore, when $0<sdyn<1$, using CCS, MVS or SFS can save energy.

In the scenario with two symmetric routes and one exit, according to Figure 3, when $sdyn=1$, the results of each route guidance strategy are similar. When $0<sdyn<1$, CCS can meet the requirements of improving traffic efficiency and saving energy.

As $0<sdyn<1$, in Figure 4a, MVS is an exception. When adopting MVS, traffic flux increased first and then decreased badly, while the results of other route guidance strategies increased all the time. In Figure 4b, the average speed of vehicles when adopting local information strategies is far lower than those of global information strategies. Figure 4c shows a large difference between the two groups; that is, vehicle numbers of local information strategies are much higher. Therefore, the results of local information strategies performed a congestion state with high flow and low speed. Conversely, when adopting CCS and SFS, the result performed a high-flow and high-speed state, in which SFS is a little better.

Figure 4d shows the results of consumption per unit flux, which were also divided into two groups. Global information strategies performed better than local information strategies. As $sdyn$ increases, the gap between the two groups becomes smaller. When $sdyn=1$, the simulation results produced by each route guidance strategy are almost the same. Therefore, when $0<sdyn<1$, using CCS, MVS or SFS can save energy.

In the scenario with two symmetric routes and two exits, according to Figure 4, when $sdyn=1$, the results of each route guidance strategy are similar. When $0<sdyn<1$, SFS can meet the requirements of improving traffic efficiency and saving energy.

4.1.2. Comparison between the Primary Mixed Vehicles Model and the Human-Driven

Traffic efficiency and energy consumption can be affected when AVs and HVs are mixed. In this part, when adopting various route guidance strategies, the results of mixed traffic and pure HVs are compared. The purpose is to study whether the introduction of autonomous vehicles will improve traffic conditions. In the mixed model, each autonomous vehicle is set to obey the strategy of route guidance so the ratio of AVs is equal to $sdyn$.

Figure 5 shows the simulation results when using global information strategies, namely CCS, MVS and SFS, in the scenario with two symmetric routes and one exit. Figure 6 shows the results of local information strategies, namely VLS, APS, EFS. In Figure 5, when adopting global information strategies, the mixed model always performs a higher traffic flow, higher speed, more vehicle numbers and lower consumption per unit flux than the HV model does as $sdyn$ increases. On the other hand, when using the global route guidance strategy, i.e., a single-exit road compared with a double-exit road, numerically, the mixed traffic flow model in the double-exit road number is greater than the single-exit road, and for the two cars, the average speed is not high, showing that the environment for road capacity did not reach the best application in the dual-exit environment, and the mixed traffic flow model can make the road accommodate more vehicles and can reduce congestion.

In Figure 6, when $0<sdyn<0.9$, adopting local information strategies makes the mixed model show higher traffic flow, lower speed and more vehicle numbers than the HV model does. It can be inferred that the road is now in the congestion state because of the introduction of autonomous vehicles. When $sdyn>0.9$, the simulation results of the mixed model are obviously better than those of the HV model. Therefore, in a single-exit environment, when using a local route guidance strategy, the mixed traffic flow model aims to improve the traffic conditions on the road, and the proportion of autonomous vehicles is also conditional. Only when the proportion of autonomous vehicles exceeds 90% can the hybrid traffic flow model improve traffic efficiency and energy consumption better than traditional vehicle models. Compared to 70% in a dual-exit road environment, the mixed traffic flow model has higher application conditions in a single-exit environment.

Figure 7 shows the simulation results when using global information strategies, namely CCS, MVS and SFS, in the scenario with two symmetric routes and two exits. Figure 8 shows the results of local information strategies, namely VLS, AP and EFS. In Figure 7, the number of vehicles in mixed model is a little lower than those in the HV model. However, two route guidance strategies share higher traffic flow, higher speed and less consumption when adopting the mixed model. Considering both traffic efficiency and energy saving, the mixed model still performs better than the HV model under global information strategies.

In Figure 8, when $0<sdyn<0.7$, the mixed model causes a larger vehicle number and lower average speed than the HV model does, which means a congestion state has appeared on the road. At this time, the consumption per unit flux of each model is almost the same. When $sdyn>0.7$, the result of the mixed model became better than that of the HV model.

4.1.3. Comparison between the Primary AV Model and the HV Model

In this part, vehicle numbers and average speed of two kinds of models are studied. One is the primary AV model and another is the HV model. Each vehicle is set to obey the route guidance strategies, that is, $sdyn=1$.

The simulation results of the scenario with two symmetrical routes and a single exit, incorporating six distinct guidance strategies of routes, are presented in Figure 9a. The histogram graph represents the vehicle numbers of two models, denoted on the left y-axis. Conversely, the scatter plot displays the average speed of the two vehicle models, aligned with the right y-axis. When adopting the global information strategy, the average pace is nearly equivalent to the top velocity, and the number of vehicles is slightly greater than that in the HV model. Obviously, the autonomous vehicle improves the traffic efficiency of the road. When adopting the local information strategy, the average speed of the HV model is about 1.45, and the number of vehicles is almost two times greater than that of an autonomous vehicle model.

The simulation results of the scenario with two symmetrical routes and two exits, under the implementation of six distinct guidance strategies of routes, are depicted in Figure 9b. The average speed of the AVs is almost the maximum speed on the road, and the number of vehicles is also maintained at a stable level under the six strategies. Due to the characteristics of the autonomous vehicle which is not interfered with by human factors, the influence of route guidance strategy is very small. Therefore, when all the vehicles on the road are autonomous vehicles, the six strategies show very similar results. When adopting the global information strategy, the vehicle number of the AV model is slightly greater than that of the HV model, and the average speed is also higher. On the contrary, when adopting the local information strategy, the HV model has more vehicles, but the average speed on the road is maintained at about 2.1, about 70% of the maximum speed. This vehicle speed is generally acceptable. Therefore, when adopting the local information strategy, the vehicles in the AV model are flow freely and have higher speeds. In the single-exit scenario, a local information strategy is adopted, and the HV model performs worse than in the scenario with two symmetrical routes and two exits.

4.2. Improved Mixed Vehicles Model

4.2.1. Results of Different Guidance Strategies of Routes

Figure 10 shows the simulation results of the scenario with two symmetrical routes and one exit under six guidance strategies of the routes. The result of the local information strategy is slightly higher than that of the global information strategy. When adopting global information strategies, the system has a higher speed, more reasonable vehicle numbers and lower energy consumption. Figure 10b shows that the lowest average speed appears when $sdyn=0.5$ and the VLS strategy is adopted, which is about 1.9. According to the results in Section 4.1.1, the minimum speed of the primary mixed model is about 1.1. Figure 10c shows that the maximum vehicle number appears when $sdyn=0.6$ and the VLS strategy is adopted, which is about 930 compared with the simulation results in Section 4.1.1, in which the maximum vehicle number is about 1800. Therefore, the improved mixed traffic flow model improves the application efficiency of the local strategy.

Figure 11a shows an upward trend in the increase in $sdyn$ no matter which route guidance strategy is adopted. Global information strategies perform better than local ones. Figure 11b shows that the average speed of vehicles increases and finally approaches the top velocity; when the local information strategies are adopted, the average speed of vehicles first decreases and then increases. The lowest speed appears when $sdyn=0.3$ and the VLS strategy is adopted, which is about 2.1. According to the results in Section 4.1.1, the minimum speed of the primary mixed model is about 1.7.

Figure 11c shows that when adopting global information strategies, the vehicle number first decreases and then presents a stable and small increasing trend. When the local information strategies are adopted, the vehicle number first increases and then decreases. The maximum vehicle number also appears when $sdyn=0.3$ and the VLS strategy is adopted, which is about 840 compared with the simulation results in Section 4.1.1, in which the maximum vehicle number is about 1100. Therefore, the improved mixed model effectively dredges the congestion state of the road when adopting local information strategies. Figure 11d shows that when the global information strategies are adopted, the consumption per unit flux of the system decreases gradually with the increase in the $sdyn$, and then tends to be stable. When the local strategies are adopted, the consumption per unit flux first increases and then decreases.

In general, the improved mixed vehicles model performs better when adopting global information strategies with a higher speed, more reasonable vehicle numbers and lower energy consumption. When adopting local information strategies, compared with the previous mixed model, the improved mixed model shows a higher minimum speed and a lower maximum vehicle number. Therefore, the new mixed model improved traffic efficiency in this system. With only one exit at the end of the two roads, vehicles on different roads which arrive at the same time must queue up. Therefore, the simulation result in the one-exit scenario is slightly worse than that in the two-exit scenario, which is the negative impact caused by road environment on the whole road traffic condition.

4.2.2. Comparison of Three Models

Figure 12 shows the simulation results of the three models when adopting the global information strategies in the scenario with two routes and one exit. The improved model shows higher traffic flow, higher vehicle speed, a more stable and an appropriate number of vehicles and lower energy consumption. Figure 13 shows that the improved mixed model always has a higher speed and fewer vehicles than the HV model does. The minimum speed of the HV model is about 1.5, which is half of the maximum speed. At this time, the road traffic is not smooth. Therefore, the improved mixed model performs better in a the scenario with two routes and one exit.

When adopting the six route guidance strategies in a scenario with two routes and two exits, simulation results are similar. In Figure 14, the improved model shows higher traffic flow, higher vehicle speed, more stable and appropriate vehicle numbers and lower energy consumption. Figure 15 shows the simulation results of the three models when adopting the local information strategies in a scenario with two routes and two exits. In Section 4.1.2, there is a critical value of $sdyn$ which makes the previous mixed model perform better than the HV model. Figure 15 shows that the improved mixed model always has a higher speed and fewer vehicles than the HV model does. At this time, the minimum speed of the HV model is about 2.1, which is 70% of the maximum speed. The traffic condition on the road is not congested and can accommodate more vehicles. If the vehicle is required to drive at a higher speed, the improved mixed traffic flow is better. In general, no matter what the proportion of autonomous vehicles is, the improved mixed model has higher traffic flow, higher average speed, a more reasonable number of vehicles and lower energy consumption.

4.2.3. Comparison between the Primary AV Model and the Improved AV Model

When $sdyn=1$, the road is full of autonomous vehicles; for the two models before and after the improvement, the traffic flux and average speed are almost equal. In order to compare the performance of the improved mixed model and the primary mixed model more clearly, Figure 16a1–a4,b1–b4 show the simulation results of vehicle numbers and consumption per vehicle when $sdyn=1$. Simulation results show that the average energy consumption of each vehicle in the improved model is less than that in the original mixed model. In a two-route scenario with one exit, when CCS, MVS and SFS are adopted, the number of vehicles on the road increases by 10, 13 and 12, respectively; when the three local strategies of VLS, APS and EFS are adopted, the number of vehicles on the road increases by 15, 14 and 13, respectively. Using the improved mixed model in the two-route scenario with two exits, when CCS, MVS and SFS are adopted, the number of vehicles on the road increases by 23, 22 and 21, respectively; when the three local strategies of VLS, APS and EFS are adopted, the number of vehicles on the road increases by 8, 14 and 17, respectively. Therefore, the improved model can accommodate more vehicles and improve traffic efficiency.

5. Conclusions

In this paper, the effects of AVs on road efficiency and energy consumption under various route guidance strategies have been studied. Strategies were selected to apply to the scenario with two symmetrical routes and one exit and two exits. We have divided the six route guidance strategies into two groups, namely, the global information strategy and the local information strategy. Among them, CCS, MVS, and SFS belong to the global information strategy; VLS, APS, and EFS belong to the local one. The autonomous vehicles were set to obey route guidance strategies so that the ratio of AVs equals to $sdyn$. Firstly, considering the characteristics of AVs, a mixed traffic flow model of AVs and HVs is established. Then, the autonomous vehicle model is improved. The results of the improved mixed model, the primary mixed model and the human-driven vehicle model under the same simulation conditions are compared. Several important simulation results can be summarized: (1) The global information route guidance strategies can collect information about the whole road, so they can improve traffic efficiency of the mixed traffic flow model better than the local ones. (2) The primary autonomous vehicles model has a lower level of automation so the simulation results of a primary mixed model are not always better than that of human-driven vehicles model. In the scenario with two symmetrical route and one exit, the simulation results of the mixed model adopting local information strategies are better only when $sdyn>0.9$. In the scenario with two symmetrical routes and two exits, when adopting the local information strategies, simulation results of the mixed model are better only when $sdyn>0.7$. (3) When all the vehicles on the road are autonomous vehicles, compared with human driving vehicles, the traffic efficiency of the road is significantly improved. (4) In the two road scenarios, no matter which route guidance strategy is adopted, the improved mixed vehicles model performs better than that of the primary mixed model.

Two important conclusions can be summarized: (1) Using global information guidance strategies improves traffic efficiency and diminishes consumption of energy. (2) A higher level of automation and more quantity of autonomous vehicles both can improve traffic efficiency and diminish consumption of energy.

For the future work of this study, we believe that there are the following perspectives:

(1) Six path guidance strategies are applied in two symmetrical road environments. In the symmetrical road environment, the other settings of the two roads are the same, and static vehicles choose static information based on their own preferences at the entrance. Therefore, in the model in this paper, static vehicles randomly choose roads. Based on this, we can also consider the asymmetric road environment, in which the static car can have a preference about the road length and is thus more in line with the actual situation.

(2) The maximum velocity is a fixed value, but considering the situation corresponding to the reality, if there is school or community on the way, in order to ensure the safety of pedestrians, an additional lower velocity limit will be set in this section. Based on this, in the simulation model, the velocity limit bottleneck setting can be introduced. A certain section of the whole road is set as the bottleneck section, and the velocity limit of the bottleneck is lower than the global maximum velocity limit. The existence of the bottleneck will have a negative impact on the vehicle speed on the road, thus affecting the traffic efficiency of the whole road. Therefore, it can be discussed whether the autonomous vehicles can continue to enhance the traffic in bottleneck sections.

(3) The cellular automaton model in this paper is driven on the single lane section, but on the urban road, there are more multi-lane sections, so the number of lanes can be increased in the road model, and lane change rules can be introduced based on the micro traffic flow model to make it more practical so as to carry out more in-depth research.