1. Introduction

1.1. Background

With the acceleration of urbanization, traffic congestion and accidents have become major issues faced by many large cities worldwide. Traditional traffic signal control systems often rely on fixed signal cycles, lacking the ability to adapt to real-time changes in traffic flow, which leads to inefficiencies and wasted resources. With the rapid development of Artificial Intelligence (AI) technologies, more and more intelligent traffic systems are beginning to employ AI algorithms for traffic signal control. Especially during peak hours and at complex intersections, AI technology has shown significant advantages. Intelligent traffic signal control systems can dynamically adjust signal cycles based on real-time data analysis, improving traffic flow, reducing congestion, and optimizing traffic resource allocation.

1.2. Problem Statement

Although some cities have started to incorporate AI technologies into traffic management, the application of AI in traffic signal control is still in its infancy, with many issues yet to be resolved. Existing AI-based signal control systems face challenges related to algorithm complexity, data quality requirements, and system integration difficulties. Furthermore, the non-linear nature of traffic flow and the complexity of unexpected events mean that the accuracy and reliability of AI models still need to be improved. Therefore, it is of significant research value to explore and analyze the potential and limitations of AI in intelligent traffic signal control.

1.3. Research Objective

This paper aims to explore the application of AI in intelligent traffic signal control, focusing on how AI technologies can improve traffic flow, reduce congestion, and enhance traffic efficiency through real-time data analysis and adaptive optimization. By analyzing and experimenting with existing AI control systems, the paper seeks to reveal the potential of AI in traffic signal control, identify the challenges it faces, and propose directions for improvement and development to promote the widespread application of AI technologies in urban traffic management.

2. Methodology

2.1. Research Design

The overall framework of this study consists of three main components: data collection, AI model design and application, and experimental evaluation. First, real-world traffic data (such as traffic flow, vehicle speed, and accident information) will be collected to serve as input for the intelligent traffic signal control system. Next, a deep reinforcement learning (DRL)-based intelligent traffic signal control system will be used, utilizing a Deep Q-Network (DQN) model for signal timing optimization. The DQN model learns the optimal signal timing strategy through continuous interaction with the traffic environment, maximizing traffic flow, reducing congestion, and improving efficiency. The core idea of the system is to dynamically adjust the signal light cycles based on real-time traffic conditions, rather than using fixed-duration signal cycles. Finally, the system will be experimentally validated through simulation platforms and real-world road networks to assess its effectiveness in improving traffic flow, reducing congestion, and enhancing safety.

2.2. Data Collection and Analysis

The data collection for this study primarily relies on a combination of real-time traffic monitoring devices, including sensors, traffic cameras, and roadside equipment, integrated within the city’s intelligent traffic system. The data sources are a mix of open-source traffic flow datasets, as well as proprietary data collected by our research team in collaboration with local traffic authorities. These data sources provide key inputs such as traffic flow, vehicle speed, and accident occurrences from multiple intersections across the city road network.

To ensure the robustness and accuracy of the data, we collected information over several days, capturing traffic conditions during both peak and off-peak hours. This allows us to account for fluctuations and patterns in traffic flow at different times, ensuring diverse and representative training samples for the AI model.

The data processing involves several steps. First, we clean the data by removing any missing values, outliers, or duplicate records to ensure dataset accuracy. Next, we perform feature extraction to derive essential features like average traffic flow, vehicle passage time, and speed variations. These processed features serve as the input variables for training the AI model, ensuring effective optimization of the intelligent traffic signal control system.

2.3. Artificial Intelligence Model Selection and Application

This study adopts Deep Q-Networks (DQN), a deep reinforcement learning (DRL) algorithm, as the core model to dynamically adjust traffic signal timing based on real-time traffic flow data. DQN operates within a reinforcement learning framework, where the system selects an appropriate signal control strategy at each time step and updates its strategy based on feedback signals, such as traffic flow and travel time. The model gradually optimizes signal control by continuously learning from interactions with the environment, which helps in reducing congestion and improving traffic efficiency. The choice of DQN is motivated by its ability to handle high-dimensional state spaces and complex traffic environments. Moreover, DQN’s capacity to process non-linear, time-varying traffic flow features through the expressiveness of neural networks makes it particularly suitable for dynamic traffic signal control applications.

The DQN model comprises three key components: state space, action space, and reward function. The state space includes multi-dimensional data representing the current traffic conditions, such as traffic flow, traffic light status, vehicle passage times, and even accident occurrences. These factors contribute to the model's understanding of the intersection's current status and help inform decisions about signal adjustments. The action space includes various strategies for controlling traffic signals, such as adjusting the duration of red, green, and yellow lights, or even changing the entire signal cycle to optimize traffic flow. These actions are selected based on the current state and are intended to maximize overall traffic efficiency. The reward function is designed to provide feedback based on the outcomes of the selected actions. It includes traffic flow, accident frequency, and congestion index. Efficient signal timings are rewarded, while unreasonable control decisions that lead to congestion or accidents are penalized. The reward function thus plays a crucial role in training the model to learn optimal signal control strategies.

In recent years, several studies have explored the application of DRL algorithms, particularly DQN, for intelligent traffic signal control. Razack et al. introduced a DQN-based approach to optimize traffic signal control in urban environments [1]. Their study highlighted how DQN could effectively learn traffic patterns and adjust signal timings to improve traffic flow, especially in densely populated areas. The authors found that the model was able to handle complex traffic scenarios and provide adaptive signal control based on real-time data. Similarly, Rasheed et al. conducted a comprehensive review of DRL applications for traffic signal control, noting the growing interest in using DQN for adaptive signal optimization [2]. Their work emphasized that while the technology shows great promise, challenges remain in terms of real-time processing and system integration.

Pan et al. demonstrated the practical implementation of deep Q-learning for adaptive traffic signal control [3]. Their case study involved optimizing traffic signal strategies for a busy urban intersection and showed how the DQN model could reduce waiting times and improve traffic throughput under varying traffic conditions. They noted that implementing such systems in real-world scenarios would require addressing data quality issues and the computational complexity of the algorithm. Furthermore, Zeng et al. explored the use of deep recurrent Q-learning for traffic signal control, showing how incorporating recurrent neural networks could better capture the temporal dynamics of traffic patterns and further improve signal optimization [4].

These studies demonstrate the potential of DQN and related DRL algorithms in transforming urban traffic management systems. By enabling traffic signal systems to learn from traffic patterns and adjust signal timings accordingly, these approaches can significantly enhance the efficiency and safety of urban transportation networks. However, challenges remain in optimizing the algorithms for real-time operation and integrating them with existing traffic infrastructure, as discussed in the literature.

Overall, this study builds upon these foundational works by applying DQN to optimize traffic signal control, demonstrating its ability to handle complex, time-varying traffic environments, and showing how it can lead to more efficient and safer urban traffic management systems.

The following table 1 presents the AI model’s application framework and specific implementations:

Table 1: AI Model Framework and Components.

By the collaborative work of these model components, DQN can learn and continually optimize its signal timing strategies, maximizing traffic flow and minimizing congestion.

2.4. Experimental Setup and Evaluation Metrics

This study will conduct preliminary tests using traffic simulation platforms (such as VISSIM) to simulate traffic flow and signal control performance under different traffic densities. Additionally, the AI signal control system will be deployed in real-world road networks to evaluate its effectiveness through real data collection. The experimental scenarios will include traffic conditions during peak and off-peak hours, as well as traffic flow at complex intersections.

For the real-world experiments, the testing will take place in several key areas of Beijing, including the Xidan District, the CBD area in Chaoyang, and several major intersections, such as those on the Third Ring Road and feeder roads, as well as key junctions like T-junctions and crossroad intersections. These areas were chosen due to their high traffic volume, complex intersection layouts, and significant congestion problems, making them ideal for testing the effectiveness of the AI signal control system.

Data collection will be carried out using various traffic monitoring devices, including ground sensors, roadside cameras, and vehicle GPS data. These devices will collect real-time data on traffic flow, vehicle speed, queue length, and accident occurrences. In particular, high-definition cameras and radar sensors will be used to monitor traffic flow at intersections, capturing key data such as vehicle passage times and signal light cycle changes, with the data transmitted to a central control system for processing.

To ensure the accuracy and comprehensiveness of the data, the experiment will span a period of two weeks, during which data will be collected across peak hours, off-peak hours, and special event scenarios (such as accidents). The focus during peak hours will be on analyzing traffic flow during morning and evening rush hours, while off-peak hours will help assess the system’s performance in optimizing signal timing during periods of lighter traffic. Special event scenarios, like traffic accidents, will also be incorporated to test the system’s adaptability in real-time.

The evaluation metrics will include:

• Traffic Smoothness: Vehicle throughput and traffic capacity.

• Travel Time: The time required for vehicles to pass through intersections.

• Congestion Index: Traffic density and queue length.

• Accident Frequency: Accident occurrence and severity.

These metrics will help assess the effectiveness of the AI signal control system in improving overall traffic flow, reducing congestion, and enhancing safety in various traffic conditions.

3. Results

3.1. Experimental Results

Through experiments conducted on traffic simulation platforms and real-world road networks, the main results were obtained. The AI-based signal control system was compared with traditional signal control systems to assess its performance under different traffic conditions.

Table 2: Performance Metrics Comparison: Traditional vs. AI Signal Control System

Table 2 demonstrates the performance comparison between the traditional and AI signal control systems. The AI system improves traffic flow by 7%, from 85% in the traditional system to 92%, indicating better utilization of road capacity. It also reduces average travel time by 25%, from 120 seconds to 95 seconds, showing that the AI system can efficiently optimize signal timing, leading to reduced delays. In terms of congestion, the AI system lowers the congestion index by 26.67%, from 0.75 to 0.55, highlighting its ability to alleviate traffic bottlenecks through dynamic signal adjustments. Furthermore, the accident rate is reduced by 40%, from 0.05 to 0.03, suggesting that optimized signal timing contributes to safer traffic conditions. These results indicate that the AI signal control system significantly outperforms the traditional system, offering substantial improvements in traffic flow, travel time, congestion, and safety, particularly during peak hours when dynamic adjustments are most effective.

3.2. Performance Analysis

The AI signal control system demonstrates clear advantages in improving traffic flow, reducing congestion, and enhancing efficiency. Unlike traditional fixed-timing systems, the AI system adapts in real-time to fluctuations in traffic conditions. This dynamic adjustment is especially effective during peak periods, where the system helps alleviate congestion by optimizing signal timings. Additionally, the AI system can quickly respond to unforeseen events, such as accidents, by adjusting the signal cycle to minimize disruption.

Table 3: AI System Performance under Various Traffic Conditions

The AI system’s flexibility allows it to effectively manage peak-time congestion and recover quickly from sudden disruptions, such as accidents or rapid changes in traffic flow.

3.3. Statistical Data and Model Validation

To validate the effectiveness of the AI signal control system, detailed experimental data and statistical analyses were provided. The results indicate significant improvements in traffic flow, congestion reduction, and accident reduction. Below is a statistical breakdown of experimental data:

Table 4: Comparison of travel time and accident reduction

Table 4 provides a statistical comparison of travel time and accident reduction between the traditional and AI signal control systems in various experimental scenarios. In peak hours, the AI system reduces travel time by 40 seconds (from 140 seconds to 100 seconds), leading to a 20% increase in flow and a 35% reduction in accidents. During off-peak hours, the AI system still shows improvement, reducing travel time by 10% (from 100 seconds to 90 seconds), increasing flow by 10%, and lowering accidents by 25%. In incident scenarios, such as accidents or disruptions, the AI system cuts travel time by 40 seconds (from 150 seconds to 110 seconds), improving flow by 15% and reducing accidents by 40%. Statistical analysis confirms that the AI system performs significantly better than the traditional system, particularly in terms of travel time, flow improvement, and accident reduction. The AI system's ability to dynamically adjust to traffic fluctuations and unforeseen events not only increases efficiency but also enhances safety, making better use of available resources even during low traffic periods. Overall, the AI signal control system outperforms traditional systems across various traffic conditions, particularly by reducing congestion and improving safety.

4. Discussion

4.1. Advantages of AI in Intelligent Traffic Signal Control

The application of Artificial Intelligence (AI) in intelligent traffic signal control offers significant advantages over traditional traffic management systems. AI can process real-time traffic data and dynamically adjust signal timings to improve traffic efficiency, reduce congestion, and shorten travel times. Unlike traditional fixed-timing signal systems, which are rigid and unable to adapt to changing traffic conditions, AI systems continuously learn from traffic patterns and adjust in real-time. This adaptability is particularly beneficial in managing both regular traffic and unexpected events such as accidents or traffic surges.

In addition to improving traffic flow, AI can also optimize energy consumption. By minimizing unnecessary stops and reducing waiting times at traffic signals, AI helps decrease fuel consumption and vehicle emissions, contributing to environmentally friendly urban mobility. AI’s ability to manage these adjustments efficiently leads to a more sustainable and responsive traffic system.

Recent studies have highlighted the significant potential of AI-based systems for adaptive signal control. Agrahari et al. review how AI enhance the responsiveness and efficiency of urban traffic systems by adjusting signal timings based on real-time traffic data, reducing congestion, and improving environmental outcomes [5]. Similarly, Zheng proposed an intelligent signal optimization algorithm that leverages AI to optimize traffic flow, showing how AI can address both regular traffic conditions and unexpected disruptions, thus improving overall traffic management efficiency [6]. These studies underscore the effectiveness of AI in transforming urban transportation systems into smarter, more sustainable networks.

4.2. Challenges and Limitations

Despite the clear advantages, AI-driven traffic signal control faces challenges. One significant issue is the complexity of AI algorithms, which require substantial computational power and high-quality data to function effectively. For instance, the real-time data processing demands can strain system resources, especially in large cities with dense traffic networks. As pointed out by YANG Huiqin et al., the integration of automated control systems with communication networks requires sophisticated coordination, which may present scalability challenges when deployed in complex urban environments [7].

Additionally, data quality remains a major concern. AI-based traffic systems heavily rely on accurate, up-to-date data from sensors and cameras. Inconsistent or noisy data can negatively impact the system's performance. FEI Liqiong highlights that current urban traffic signal systems often face issues such as incomplete data, sensor malfunctions, and communication delays [8]. Moreover, the integration of AI with existing infrastructure is not straightforward, requiring extensive adjustments to traditional traffic control systems, which may incur high costs and disrupt existing traffic management frameworks. These challenges necessitate further research to enhance the robustness, scalability, and reliability of AI-based traffic control solutions.

4.3. Future Research Directions

Future research can focus on the integration of AI with the Internet of Things (IoT), utilizing smarter devices and sensors to enhance data collection capabilities and optimize signal control. Additionally, with the development of autonomous driving technology, AI signal control systems will be able to communicate directly with autonomous vehicles, leading to more efficient traffic flow management. The combination of AI with big data analytics will also help predict and optimize traffic patterns, further enhancing the intelligence of urban transportation systems.

5. Conclusions

5.1. Research Summary

This study demonstrated the significant potential of artificial intelligence (AI) in optimizing intelligent traffic signal control systems. The primary findings show that AI-based systems, particularly those utilizing deep reinforcement learning (DQN), outperform traditional fixed-timing systems in key areas such as traffic flow, congestion reduction, and accident prevention. The AI system’s ability to dynamically adjust signal timings based on real-time traffic conditions leads to smoother traffic flow, shorter travel times, and a reduction in congestion and accidents, particularly during peak hours and in response to unexpected events.

To fully capitalize on the advantages of AI in traffic management, cities should consider integrating AI-based signal control systems into their existing traffic infrastructure. Governments should prioritize investments in intelligent traffic sensors and real-time data collection systems, which are crucial for providing the data needed to train AI models. For urban areas with high congestion, AI systems could be introduced gradually, starting with critical intersections and expanding as the system proves effective. Furthermore, AI should be integrated with other smart city technologies such as connected vehicles and traffic monitoring platforms to create a more comprehensive and responsive traffic management ecosystem.

Despite the promising results, this study has limitations. The experiments were mainly conducted in simulation environments and on limited real-world data, which may not fully capture all traffic complexities. Additionally, the AI models require substantial computational resources, which may limit their scalability in some regions.

Looking ahead, the future of AI in intelligent traffic systems is bright. With the advancement of AI algorithms, increased computational power, and better data integration, AI can be expanded to optimize not only signal control but also traffic planning, route guidance, and public transportation systems. As AI technologies continue to evolve, their applications in smart city development will play a crucial role in creating more efficient, safer, and sustainable urban transportation networks.