Smart city traffic indication system and working method thereof

Abstract

The invention relates to the technical field of smart city traffic, in particular to a smart city traffic indication system and a working method thereof. According to the technical scheme, the system comprises a multi-source heterogeneous data acquisition module, a data preprocessing and fusion module, a feature extraction and analysis module, a traffic flow prediction and signal timing decision module and a signal display and communication module, the acquisition frequency is dynamically adjusted according to the traffic flow. Traffic and meteorological data are dynamically collected through multiple types of sensors, the quality of data preprocessing and fusion is guaranteed, features are accurately extracted, signal timing is optimized through reinforcement learning, the display and communication module effectively transmits information and cooperates with surrounding traffic, emergency situations can be dealt with, tail gas emission is reduced, pedestrian waiting time is shortened, and traffic safety is improved. And a basis can be provided for long-term traffic planning, and the traffic efficiency, the environmental protection property and the pedestrian friendliness are comprehensively improved.

Description

Technical Field

[0001] The present invention relates to the field of smart city traffic technology, and in particular to a smart city traffic indication system and a working method thereof. Background Art

[0002] In smart city traffic indication systems, traffic flow prediction and signal timing optimization are often based on simple historical data statistics and fixed algorithm models. This approach is difficult to adapt to complex and changing traffic conditions. For example, in the event of sudden traffic accidents, large-scale events, weather changes and other special circumstances, traffic flow will fluctuate irregularly. Traditional methods lack comprehensive consideration of real-time dynamic factors and are difficult to accurately predict changes in traffic flow, which in turn leads to unreasonable signal timing. This may cause increased traffic congestion, long waiting times for vehicles, and low utilization of road resources.

[0003] In summary, the present application proposes a smart city traffic indication system and a working method thereof. Summary of the invention

[0004] The purpose of the present invention is to address the problem of insufficient accuracy of traffic flow prediction and signal timing optimization in the background technology, and to propose a smart city traffic indication system and a working method thereof.

[0005] First, as Figure 1 As shown, the present application provides a smart city traffic indication system, comprising: Multi-source heterogeneous data acquisition module: uses a variety of sensors to collect traffic intersections and surrounding traffic and meteorological data. The acquisition frequency is dynamically adjusted according to traffic flow, and the data is transmitted to the pre-processing and fusion module via the network; Data preprocessing and fusion module: including data cleaning submodule, normalization submodule and fusion submodule; Feature extraction and analysis module: including LSTM units and CNN units, multi-layer convolution and pooling, extracting and fusing spatial features; Traffic flow prediction and signal timing decision module: Based on reinforcement learning, including reinforcement learning agent environment simulator and reward evaluation submodule; Signal display and communication module: includes traffic lights and VMS, which communicate with surrounding traffic facilities to achieve coordinated control.

[0006] Optionally, the multi-source heterogeneous data acquisition module: Equipped with various types of sensors, including high-definition cameras, induction coils, radar speed guns, and meteorological sensors, which are distributed at traffic intersections and surrounding areas to collect comprehensive traffic-related data; High-definition cameras capture images, vehicle models, and license plate information of vehicles in all directions at the intersection in real time. Induction coils accurately detect the passing time and number of vehicles. Radar speedometers measure vehicle speeds. Meteorological sensors collect temperature, humidity, visibility, and precipitation meteorological data. The data collection frequency is dynamically adjusted according to the traffic flow, with high frequency collection during peak hours and appropriate reduction of frequency during off-peak hours. The collected data is transmitted to the data preprocessing and fusion module via high-speed wired networks and reliable wireless communication protocols.

[0007] Optionally, the data preprocessing and fusion module includes: The data cleaning submodule uses an outlier detection algorithm to identify and remove abnormal data caused by sensor failure and electromagnetic interference factors; The data normalization submodule uses different normalization methods for different types of data. For traffic flow and vehicle speed numerical data, a distribution-based normalization method is used, and for vehicle type classification data, a unique hot encoding method is used; The data fusion submodule fuses the spatiotemporal data from different sensors. Based on the spatiotemporal calibration algorithm, the data of different sensors are aligned in a unified spatiotemporal coordinate system. At the same time, the multi-source data fusion algorithm is used to fuse the repeated or complementary data at the same time and location. The multi-source data fusion algorithm adopts the Kalman filter fusion algorithm, and its state prediction equation is: The measurement update equation is: in, yes The predicted state vector at each moment, yes Estimate the state vector at each moment, is the state transition matrix, is the control input matrix, is the control input vector, is the measurement vector, is the measurement matrix, is the Kalman gain.

[0008] Optionally, the feature extraction and analysis module: Contains long short-term memory network units and convolutional neural network units. LSTM units are used to process time series data, and CNN units are used to extract features of image and space-related data. The LSTM unit has a multi-layer structure, with 3-5 hidden layers. The number of hidden units in each layer is dynamically adjusted between 100-500 according to the complexity of the traffic intersection. The activation functions of the forget gate, input gate, and output gate use adaptively adjusted nonlinear functions. The CNN unit has multiple convolutional layers and pooling layers. The size and number of convolution kernels are adaptively adjusted according to the input image resolution and feature complexity. The spatial features of vehicle models and queue lengths in vehicle images are extracted through convolution and pooling operations, and then fused with the time series features output by the LSTM unit. In the LSTM unit, The update formula of the hidden layer is as follows: Forget Gate: in, It is The output of the forget gate at time t, is the adaptively selected sigmoid function, It is The weight matrix of the forget gate of the layer, It is The previous hidden state of the layer, It is The current input of the layer, It is The bias term of the forget gate of the layer; Input Gate: in, It is The layer input gate is at time The output, It is The state of the candidate cell in the layer is at time The value of , It is The weight matrix for layer input gate and candidate cell state calculation, , is the corresponding bias term; Cell status update: Output Gate: in, It is The layer output gate is at time The output, It is The hidden state of the layer at time, It is The weight matrix of the layer output gate, It is Bias term for the output gate of the layer.

[0009] Optional, traffic flow prediction and signal timing decision module: Based on reinforcement learning architecture, including reinforcement learning agent, environment simulator and reward evaluation submodule; The reinforcement learning agent uses a deep Q network and its variants. The network structure has multiple hidden layers and the number of neurons is between 500 and 2000. It generates a traffic signal timing strategy based on the feature vector fused from LSTM and CNN. The strategy includes the duration of green light, yellow light, red light and phase sequence. The environment simulator simulates the dynamic changes of traffic intersections, taking into account the arrival, departure, queuing, and lane-changing behaviors of vehicles, and generates simulated traffic scenarios based on traffic rules and actual road conditions to provide a training environment for the agent; The reward evaluation submodule calculates the reward signal based on multiple evaluation indicators, including but not limited to the average waiting time of vehicles, road congestion, exhaust emissions, and pedestrian waiting time. The reward signal is calculated by weighted combination, and the weights are dynamically adjusted according to the city traffic management goals; In the reinforcement learning agent, the calculation of the reward signal is based on the following formula: in is the average waiting time of vehicles, is the degree of road congestion, which is comprehensively evaluated by the length and density of vehicle queues at intersections. is the exhaust gas emission, which is estimated based on vehicle type, residence time and emission model. It is the waiting time for pedestrians. , , , It is a weight coefficient that is dynamically adjusted according to traffic management objectives and environmental factors. Environmental factors include air quality and pedestrian flow. ; In the convolutional neural network unit, The output feature map of the convolutional layer The calculation formula is: in , are the feature map coordinates, , is the convolution kernel size, is the number of input channels, It is The convolution kernel weights of the convolutional layers, is the input feature map of the previous layer, It is The bias term of the convolutional layer, is the activation function.

[0010] Optional, signal display and communication module: It includes high-brightness, high-resolution traffic lights and variable message signs. Traffic lights can accurately display light signals of different colors and durations according to the instructions of the signal timing decision module. VMS is used to display real-time traffic information to road users. It has the ability to communicate with surrounding transportation facilities, including traffic sign systems at adjacent intersections, bus systems, and parking lot management systems, and realizes data interaction and collaborative control through short-range communication and 5G communication technology.

[0011] In a second aspect, the present application provides a working method of a smart city traffic indication system, which is applied to the smart city traffic indication system described in the first aspect, and includes the following steps: Multi-source data collection: Various sensors in the multi-source heterogeneous data collection module are used to collect traffic and meteorological multi-source data at traffic intersections and their surroundings. The collection frequency is automatically adjusted according to the real-time traffic flow during the data collection process; Data preprocessing and fusion: The collected data is transmitted to the data preprocessing and fusion module. The abnormal data is first removed through the data cleaning submodule, and then the data is normalized using the data normalization submodule. Finally, the spatiotemporal data of different sensors are fused through the data fusion submodule. Feature extraction and fusion: The fused data is input into the feature extraction and analysis module. The LSTM unit processes the time series data, and the CNN unit processes the image and spatial data. The features extracted by the two are then fused to obtain a comprehensive feature vector. Traffic flow prediction and signal timing strategy generation: The reinforcement learning agent in the traffic flow prediction and signal timing decision module generates the initial traffic signal timing strategy based on the comprehensive feature vector, while the environment simulator simulates the traffic scenario based on the actual traffic rules and road conditions; Strategy evaluation and adjustment: The reward evaluation submodule calculates the reward signal based on the average waiting time of vehicles, road congestion, exhaust emissions, and pedestrian waiting time evaluation indicators, and feeds it back to the reinforcement learning agent. The agent adjusts the traffic signal timing strategy based on the reward signal; Signal display and regional coordination: Traffic signals and real-time traffic information are displayed to road users through traffic lights and variable message signs in the signal display and communication module, while communicating with surrounding traffic facilities.

[0012] Optionally, when the LSTM unit processes data, an adaptive time window strategy is adopted according to the periodic changes in traffic flow, and the time window length is dynamically adjusted between 10-60 minutes to better capture the long-term and short-term changes in traffic data.

[0013] Optionally, during the training process, the reinforcement learning agent adopts a priority-based experience replay mechanism to prioritize experience samples according to their importance, and gives priority to important experience samples for training to improve training efficiency and model convergence speed. The importance of experience samples is determined based on the size of the reward signal and the distance in time from the current decision.

[0014] Optionally, the signal display and communication module exchanges traffic flow and congestion information every 5-10 minutes during the communication process with surrounding traffic facilities, and coordinates the traffic signal timing of the entire area through a distributed model predictive control algorithm.

[0015] Compared with the prior art, the present invention has at least one of the following beneficial technical effects: Traffic and meteorological data, including vehicle images, numbers, speeds and weather information, are collected through a variety of sensors. The collection frequency is adjusted dynamically with traffic flow, and traffic intersections and their surrounding data can be fully and efficiently acquired, providing rich information for traffic management. The data preprocessing and fusion module can effectively remove abnormal data, reasonably normalize different types of data, and fuse multi-source spatiotemporal data through advanced algorithms to ensure high quality and integrity of the data, making the data more suitable for subsequent analysis and decision-making.

[0016] The feature extraction and analysis module that combines LSTM and CNN can fully explore the temporal and spatial features in traffic data. LSTM captures the long-term dependence and temporal correlation of traffic data, and CNN extracts vehicle-related spatial features. The combination of the two provides a more accurate and in-depth basis for traffic flow prediction and signal timing. The traffic flow prediction and signal timing decision module based on reinforcement learning can generate more reasonable signal timing strategies based on complex traffic conditions. The reward evaluation that considers multiple factors prompts the intelligent agent to continuously optimize, balance vehicle waiting, road congestion, exhaust emissions, and pedestrian waiting time, etc., improve traffic efficiency, and reduce congestion and pollution.

[0017] The signal display and communication module can clearly display traffic signals and real-time information, and achieve regional coordinated control by communicating with surrounding traffic facilities, optimizing regional traffic flow and avoiding local congestion affecting the overall traffic conditions. In the face of special situations such as sudden accidents, construction or large-scale events, the system can quickly adjust signal timing to guide vehicles to bypass congestion and effectively alleviate traffic pressure, which is difficult to achieve with traditional systems. By reducing vehicle idling time to reduce exhaust emissions, while shortening pedestrian waiting time, improving the safety and convenience of pedestrians crossing the road, and achieving the dual benefits of environmental protection and pedestrian friendliness. The present invention dynamically collects traffic and meteorological data through multiple types of sensors, ensures the quality of data preprocessing and fusion, accurately extracts features, optimizes signal timing through reinforcement learning, and effectively conveys information and coordinates surrounding traffic through display and communication modules. It can respond to emergencies, reduce exhaust emissions and pedestrian waiting time, and provide a basis for long-term traffic planning, comprehensively improving traffic efficiency, environmental protection, and pedestrian friendliness. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a principle block diagram of a smart city traffic indication system; Figure 2 The present invention is a flow chart of a working method of a smart city traffic indication system. DETAILED DESCRIPTION

[0019] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0020] Example 1 like Figure 1 As shown, a smart city traffic indication system proposed in the present invention includes a multi-source heterogeneous data acquisition module, a data preprocessing and fusion module, a feature extraction and analysis module, a traffic flow prediction and signal timing decision module, and a signal display and communication module. Each module is described in detail below.

[0021] In this embodiment, the multi-source heterogeneous data acquisition module uses a variety of sensors to collect traffic and meteorological data at the traffic intersection and its surrounding areas. The acquisition frequency is dynamically adjusted according to the traffic flow, and the data is transmitted to the pre-processing and fusion module via the network. The multi-source heterogeneous data acquisition module is equipped with a variety of types of sensors, including high-definition cameras, induction coils, radar speed meters, and meteorological sensors, which are distributed at the traffic intersection and its surrounding areas to collect comprehensive traffic-related data; Among them, high-definition cameras capture real-time images, vehicle models, and license plate information of vehicles in all directions at the intersection, induction coils accurately detect the passing time and number of vehicles, radar speedometers measure vehicle speeds, and meteorological sensors collect temperature, humidity, visibility, and precipitation meteorological data; In addition, the frequency of data collection is dynamically adjusted according to traffic flow, with high frequency collection during peak hours and appropriate frequency reduction during off-peak hours. The collected data is transmitted to the data preprocessing and fusion module via high-speed wired networks and reliable wireless communication protocols; The multi-source heterogeneous data acquisition module is equipped with various types of sensors, covering various aspects of information such as traffic and weather. High-definition cameras, induction coils, and radar speed guns obtain traffic data from different angles and can accurately capture various status information of vehicles. The addition of meteorological sensors allows the system to consider the impact of weather on traffic, providing a rich data foundation for subsequent accurate decision-making. At the same time, the acquisition frequency is dynamically adjusted according to traffic flow, which can not only ensure that sufficient data is obtained during peak hours to reflect complex road conditions, but also avoid unnecessary data redundancy during off-peak hours, thereby improving the efficiency and pertinence of data collection.

[0022] In this embodiment, the data preprocessing and fusion module includes a data cleaning submodule, a normalization submodule and a fusion submodule, specifically including: The data cleaning submodule uses an outlier detection algorithm to identify and remove abnormal data caused by sensor failure and electromagnetic interference factors; The data normalization submodule uses different normalization methods for different types of data. For traffic flow and vehicle speed numerical data, a distribution-based normalization method is used, and for vehicle type classification data, a unique hot encoding method is used; The data fusion submodule fuses the spatiotemporal data from different sensors. Based on the spatiotemporal calibration algorithm, the data of different sensors are aligned in a unified spatiotemporal coordinate system. At the same time, the multi-source data fusion algorithm is used to fuse the repeated or complementary data at the same time and location. The multi-source data fusion algorithm adopts the Kalman filter fusion algorithm, and its state prediction equation is: The measurement update equation is: in, yes The predicted state vector at each moment, yes Estimate the state vector at each moment, is the state transition matrix, is the control input matrix, is the control input vector, is the measurement vector, is the measurement matrix, is the Kalman gain.

[0023] The data cleaning submodule in the data preprocessing and fusion module uses advanced outlier detection algorithms to effectively remove abnormal data caused by sensor failure or electromagnetic interference, thereby ensuring data quality. Different types of data use different normalization methods to make the data more comparable and stable in subsequent processing. The data fusion submodule uses spatiotemporal calibration and multi-source data fusion algorithms to fuse spatiotemporal data from different sensors, improving data integrity and consistency, avoiding data duplication or conflict, and providing more reliable data support for subsequent analysis and decision-making.

[0024] Feature extraction and analysis module: including LSTM units and CNN units, multi-layer convolution and pooling, extracting and fusing spatial features; the feature extraction and analysis module specifically includes: Contains long short-term memory network units and convolutional neural network units. LSTM units are used to process time series data, and CNN units are used to extract features of image and space-related data. The LSTM unit has a multi-layer structure, with 3-5 hidden layers. The number of hidden units in each layer is dynamically adjusted between 100-500 according to the complexity of the traffic intersection. The activation functions of the forget gate, input gate, and output gate use adaptively adjusted nonlinear functions. The CNN unit has multiple convolutional layers and pooling layers. The size and number of convolution kernels are adaptively adjusted according to the input image resolution and feature complexity. The spatial features of vehicle models and queue lengths in vehicle images are extracted through convolution and pooling operations, and then fused with the time series features output by the LSTM unit. In the LSTM unit, The update formula of the hidden layer is as follows: Forget Gate: in, It is The output of the forget gate at time t, is the adaptively selected sigmoid function, It is The weight matrix of the forget gate of the layer, It is The previous hidden state of the layer, It is The current input of the layer, It is The bias term of the forget gate of the layer; Input Gate: in, It is The layer input gate is at time The output, It is The state of the candidate cell in the layer is at time The value of , It is The weight matrix for layer input gate and candidate cell state calculation, , is the corresponding bias term; Cell status update: Output Gate: in, It is The layer output gate is at time The output, It is The hidden state of the layer at time, It is The weight matrix of the layer output gate, It is Bias term for the output gate of the layer.

[0025] The combination of LSTM units and CNN units in the feature extraction and analysis module can fully explore the features in traffic data. Through its multi-layer structure and dynamically adjusted parameters, the LSTM unit can well capture the long-term dependencies and time correlations of data such as traffic flow and vehicle speed, and is very effective in processing traffic flow data with periodicity and dynamic changes. The CNN unit extracts features from images and spatially related data, such as vehicle type and queue length, through convolution and pooling operations, and integrates them with the time series features output by the LSTM unit, achieving a more comprehensive and in-depth understanding of traffic conditions, providing a more valuable basis for traffic flow prediction and signal timing decisions.

[0026] Traffic flow prediction and signal timing decision module: Based on reinforcement learning, including reinforcement learning agent environment simulator and reward evaluation submodule, specifically including reinforcement learning agent, environment simulator and reward evaluation submodule based on reinforcement learning architecture; Among them, the reinforcement learning agent uses a deep Q network and its variants. The network structure has multiple hidden layers and the number of neurons is between 500 and 2000. The traffic signal timing strategy is generated based on the feature vector fused from LSTM and CNN. The strategy includes the duration of green light, yellow light, red light and phase sequence. In addition, the environmental simulator simulates the dynamic changes of traffic intersections, taking into account the arrival, departure, queuing, and lane-changing behaviors of vehicles, and generates simulated traffic scenarios based on traffic rules and actual road conditions to provide a training environment for the agent; Furthermore, the reward evaluation submodule calculates the reward signal based on multiple evaluation indicators, including but not limited to the average waiting time of vehicles, the degree of road congestion, exhaust emissions, and pedestrian waiting time. The reward signal is calculated by weighted combination, and the weights are dynamically adjusted according to the urban traffic management objectives; In the reinforcement learning agent, the calculation of the reward signal is based on the following formula: in is the average waiting time of vehicles, is the degree of road congestion, which is comprehensively evaluated by the length and density of vehicle queues at intersections. is the exhaust gas emission, which is estimated based on vehicle type, residence time and emission model. It is the waiting time for pedestrians. , , , It is a weight coefficient that is dynamically adjusted according to traffic management objectives and environmental factors. Environmental factors include air quality and pedestrian flow. ; In the convolutional neural network unit, The output feature map of the convolutional layer The calculation formula is: in , are the feature map coordinates, , is the convolution kernel size, is the number of input channels, It is The convolution kernel weights of the convolutional layers, is the input feature map of the previous layer, It is The bias term of the convolutional layer, is the activation function The traffic flow prediction and signal timing decision module based on reinforcement learning can generate more reasonable signal timing strategies according to complex traffic conditions. The reinforcement learning agent uses deep Q network and its variants to generate traffic signal timing strategies based on the feature vectors fused from LSTM and CNN, which can adapt to various complex traffic scenarios. The environmental simulator can simulate the dynamic changes of traffic intersections and provide a close-to-real training environment for the agent. The reward evaluation submodule comprehensively considers multiple evaluation indicators to calculate the reward signal, prompting the agent to continuously optimize the signal timing strategy to balance multiple factors such as vehicle waiting time, road congestion, exhaust emissions and pedestrian waiting time, thereby improving traffic efficiency and reducing traffic congestion and environmental pollution.

[0027] Signal display and communication module: including traffic lights and VMS, which communicate with surrounding traffic facilities to achieve coordinated control, including high-brightness, high-resolution traffic lights and variable information signs. Traffic lights can accurately display light signals of different colors and durations according to the instructions of the signal timing decision module. VMS is used to display real-time traffic information to road users; It has the ability to communicate with surrounding transportation facilities, including traffic sign systems at adjacent intersections, bus systems, and parking lot management systems, and realizes data interaction and collaborative control through short-range communication and 5G communication technology.

[0028] The signal display and communication module can accurately and clearly display traffic signals and real-time traffic information to road users through high-brightness, high-resolution traffic lights and variable information signs, improving the visibility and readability of traffic information. The communication capability with surrounding traffic facilities, through short-range communication, 5G and other technologies to achieve data interaction and collaborative control, can optimize the traffic flow of the entire area, avoid overall congestion caused by local optimization, and achieve traffic coordination and optimization on a larger scale.

[0029] The present invention can quickly adapt and adjust the signal timing strategy in the face of special situations such as sudden traffic accidents, temporary road construction or large-scale activities because the system comprehensively considers a variety of traffic and environmental factors. When an accident occurs and causes traffic congestion on a certain section of road, the system can timely adjust the signals of the surrounding intersections through the information obtained by the sensor, guide vehicles to bypass the congested area, and relieve traffic pressure. This adaptive ability to sudden situations is difficult for traditional traffic indication systems to have. In the process of optimizing traffic flow to reduce congestion, since the reward evaluation submodule considers factors such as exhaust emissions and pedestrian waiting time, the system will automatically tend to reduce the vehicle's stay time and idling time at the intersection, thereby reducing exhaust emissions and improving air quality. At the same time, reasonable signal timing can also reduce pedestrian waiting time and improve the safety and convenience of pedestrians crossing the road, which is often overlooked in the design of traditional traffic indication systems, or it is difficult to take into account both at the same time. During long-term operation, the large amount of data accumulated by the system can provide strong support for urban traffic planning. By analyzing data such as traffic flow changes and congestion conditions in different time periods and areas, we can discover traffic bottlenecks and potential optimization space, and provide decision-making basis for long-term planning of transportation infrastructure such as road construction, bus route adjustments, and parking lot planning. This is an important value of the system beyond its daily traffic indication function.

[0030] Example 2 See also Figure 2 As shown, the working method of the smart city traffic indication system includes the following steps: Multi-source data collection: Various sensors in the multi-source heterogeneous data collection module are used to collect traffic and meteorological multi-source data at traffic intersections and their surroundings. The collection frequency is automatically adjusted according to the real-time traffic flow during the data collection process; Data preprocessing and fusion: The collected data is transmitted to the data preprocessing and fusion module. The abnormal data is first removed through the data cleaning submodule, and then the data is normalized using the data normalization submodule. Finally, the spatiotemporal data of different sensors are fused through the data fusion submodule. Feature extraction and fusion: The fused data is input into the feature extraction and analysis module. The LSTM unit processes the time series data, and the CNN unit processes the image and spatial data. The features extracted from the two are then fused to obtain a comprehensive feature vector. When the LSTM unit processes the data, it adopts an adaptive time window strategy based on the periodic changes in traffic flow. The time window length is dynamically adjusted between 10 and 60 minutes to better capture the long-term and short-term changes in traffic data.

[0031] Traffic flow prediction and signal timing strategy generation: The reinforcement learning agent in the traffic flow prediction and signal timing decision module generates the initial traffic signal timing strategy based on the comprehensive feature vector. At the same time, the environmental simulator simulates the traffic scene according to the actual traffic rules and road conditions. During the training process, the reinforcement learning agent adopts a priority-based experience playback mechanism to prioritize the experience samples according to their importance, and gives priority to important experience samples for training to improve the training efficiency and model convergence speed. The importance of the experience samples is determined according to the size of the reward signal and the distance in time from the current decision.

[0032] Strategy evaluation and adjustment: The reward evaluation submodule calculates the reward signal based on the average waiting time of vehicles, road congestion, exhaust emissions, and pedestrian waiting time evaluation indicators, and feeds it back to the reinforcement learning agent. The agent adjusts the traffic signal timing strategy based on the reward signal; Signal display and regional coordination: Traffic lights and variable information signs in the signal display and communication module are used to display traffic signals and real-time traffic information to road users, while communicating with surrounding traffic facilities. During the communication process with surrounding traffic facilities, the signal display and communication module exchanges traffic flow and congestion information every 5-10 minutes, and coordinates the traffic signal timing of the entire area through the distributed model predictive control algorithm.

[0033] The above specific embodiments are only several optional embodiments of the present invention. Based on the technical solutions of the present invention and the relevant inspirations of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A smart city traffic indication system, characterized in that: include: Multi-source heterogeneous data acquisition module: uses a variety of sensors to collect traffic intersections and surrounding traffic and meteorological data. The acquisition frequency is dynamically adjusted according to traffic flow, and the data is transmitted to the pre-processing and fusion module via the network; Data preprocessing and fusion module: including data cleaning submodule, normalization submodule and fusion submodule; Feature extraction and analysis module: including LSTM units and CNN units, multi-layer convolution and pooling, extracting and fusing spatial features; Traffic flow prediction and signal timing decision module: Based on reinforcement learning, including reinforcement learning agent environment simulator and reward evaluation submodule; Signal display and communication module: includes traffic lights and VMS, which communicate with surrounding traffic facilities to achieve coordinated control.

2. A smart city traffic indication system according to claim 1, characterized in that: The multi-source heterogeneous data acquisition module: Equipped with various types of sensors, including high-definition cameras, induction coils, radar speed guns, and meteorological sensors, which are distributed at traffic intersections and surrounding areas to collect comprehensive traffic-related data; High-definition cameras capture images, vehicle models, and license plate information of vehicles in all directions at the intersection in real time. Induction coils accurately detect the passing time and number of vehicles. Radar speedometers measure vehicle speeds. Meteorological sensors collect temperature, humidity, visibility, and precipitation meteorological data. The data collection frequency is dynamically adjusted according to the traffic flow, with high frequency collection during peak hours and appropriate reduction of frequency during off-peak hours. The collected data is transmitted to the data preprocessing and fusion module via high-speed wired networks and reliable wireless communication protocols.

3. A smart city traffic indication system according to claim 1, characterized in that: The data preprocessing and fusion module includes: The data cleaning submodule uses an outlier detection algorithm to identify and remove abnormal data caused by sensor failure and electromagnetic interference factors; The data normalization submodule uses different normalization methods for different types of data. For traffic flow and vehicle speed numerical data, a distribution-based normalization method is used, and for vehicle type classification data, a unique hot encoding method is used; The data fusion submodule fuses the spatiotemporal data from different sensors. Based on the spatiotemporal calibration algorithm, the data of different sensors are aligned in a unified spatiotemporal coordinate system. At the same time, the multi-source data fusion algorithm is used to fuse the repeated or complementary data at the same time and location. The multi-source data fusion algorithm adopts the Kalman filter fusion algorithm, and its state prediction equation is: The measurement update equation is: in, yes The predicted state vector at each moment, yes Estimate the state vector at each moment, is the state transition matrix, is the control input matrix, is the control input vector, is the measurement vector, is the measurement matrix, is the Kalman gain.

4. A smart city traffic indication system according to claim 1, characterized in that: The feature extraction and analysis module: Contains long short-term memory network units and convolutional neural network units. LSTM units are used to process time series data, and CNN units are used to extract features of image and space-related data. The LSTM unit has a multi-layer structure, with 3-5 hidden layers. The number of hidden units in each layer is dynamically adjusted between 100-500 according to the complexity of the traffic intersection. The activation functions of the forget gate, input gate, and output gate use adaptively adjusted nonlinear functions. The CNN unit has multiple convolutional layers and pooling layers. The size and number of convolution kernels are adaptively adjusted according to the input image resolution and feature complexity. The spatial features of vehicle models and queue lengths in vehicle images are extracted through convolution and pooling operations, and then fused with the time series features output by the LSTM unit. In the LSTM unit, The update formula of the hidden layer is as follows: Forget Gate: in, It is The output of the forget gate at time t, is the adaptively selected sigmoid function, It is The weight matrix of the forget gate of the layer, It is The previous hidden state of the layer, It is The current input of the layer, It is The bias term of the forget gate of the layer; Input Gate: in, It is The layer input gate is at time The output, It is The state of the candidate cell in the layer is at time The value of , It is The weight matrix for layer input gate and candidate cell state calculation, , is the corresponding bias term; Cell status update: Output Gate: in, It is The layer output gate is at time The output, It is The hidden state of the layer at time, It is The weight matrix of the layer output gate, It is Bias term for the output gate of the layer.

5. A smart city traffic indication system according to claim 1, characterized in that: Traffic flow prediction and signal timing decision module: Based on reinforcement learning architecture, including reinforcement learning agent, environment simulator and reward evaluation submodule; The reinforcement learning agent uses a deep Q network and its variants. The network structure has multiple hidden layers and the number of neurons is between 500 and 2000. It generates a traffic signal timing strategy based on the feature vector fused from LSTM and CNN. The strategy includes the duration of green light, yellow light, red light and phase sequence. The environment simulator simulates the dynamic changes of traffic intersections, taking into account the arrival, departure, queuing, and lane-changing behaviors of vehicles, and generates simulated traffic scenarios based on traffic rules and actual road conditions to provide a training environment for the agent; The reward evaluation submodule calculates the reward signal based on multiple evaluation indicators, including but not limited to the average waiting time of vehicles, road congestion, exhaust emissions, and pedestrian waiting time. The reward signal is calculated by weighted combination, and the weights are dynamically adjusted according to the city traffic management goals; In the reinforcement learning agent, the calculation of the reward signal is based on the following formula: in is the average waiting time of vehicles, is the degree of road congestion, which is comprehensively evaluated by the length and density of vehicle queues at intersections. is the exhaust gas emission, which is estimated based on vehicle type, residence time and emission model. It is the waiting time for pedestrians. , , , It is a weight coefficient that is dynamically adjusted according to traffic management objectives and environmental factors. Environmental factors include air quality and pedestrian flow. ; In the convolutional neural network unit, The output feature map of the convolutional layer The calculation formula is: in , are the feature map coordinates, , is the convolution kernel size, is the number of input channels, It is The convolution kernel weights of the convolutional layers, is the input feature map of the previous layer, It is The bias term of the convolutional layer, is the activation function.

6. A smart city traffic indication system according to claim 1, characterized in that: Signal display and communication module: It includes high-brightness, high-resolution traffic lights and variable message signs. Traffic lights can accurately display light signals of different colors and durations according to the instructions of the signal timing decision module. VMS is used to display real-time traffic information to road users. It has the ability to communicate with surrounding transportation facilities, including traffic sign systems at adjacent intersections, bus systems, and parking lot management systems, and realizes data interaction and collaborative control through short-range communication and 5G communication technology.

7. A working method of a smart city traffic indication system, applied to the smart city traffic indication system as claimed in any one of claims 1 to 6, characterized in that: The following steps are involved: Multi-source data collection: Various sensors in the multi-source heterogeneous data collection module are used to collect traffic and meteorological multi-source data at traffic intersections and their surroundings. The collection frequency is automatically adjusted according to the real-time traffic flow during the data collection process; Data preprocessing and fusion: The collected data is transmitted to the data preprocessing and fusion module. The abnormal data is first removed through the data cleaning submodule, and then the data is normalized using the data normalization submodule. Finally, the spatiotemporal data of different sensors are fused through the data fusion submodule. Feature extraction and fusion: The fused data is input into the feature extraction and analysis module. The LSTM unit processes the time series data, and the CNN unit processes the image and spatial data. The features extracted by the two are then fused to obtain a comprehensive feature vector. Traffic flow prediction and signal timing strategy generation: The reinforcement learning agent in the traffic flow prediction and signal timing decision module generates the initial traffic signal timing strategy based on the comprehensive feature vector, while the environment simulator simulates the traffic scenario based on the actual traffic rules and road conditions; Strategy evaluation and adjustment: The reward evaluation submodule calculates the reward signal based on the average waiting time of vehicles, road congestion, exhaust emissions, and pedestrian waiting time evaluation indicators, and feeds it back to the reinforcement learning agent. The agent adjusts the traffic signal timing strategy based on the reward signal; Signal display and regional coordination: Traffic signals and real-time traffic information are displayed to road users through traffic lights and variable message signs in the signal display and communication module, while communicating with surrounding traffic facilities.

8. The working method of the smart city traffic indication system according to claim 7, characterized in that: When the LSTM unit processes data, it adopts an adaptive time window strategy according to the periodic changes in traffic flow, and the time window length is dynamically adjusted between 10 and 60 minutes to better capture the long-term and short-term changes in traffic data.

9. The working method of the smart city traffic indication system according to claim 7, characterized in that: During the training process, the reinforcement learning agent adopts a priority-based experience replay mechanism to prioritize the experience samples according to their importance, and gives priority to important experience samples for training, thereby improving the training efficiency and the model convergence speed. The importance of the experience samples is determined according to the size of the reward signal and the distance in time from the current decision.

10. The working method of the smart city traffic indication system according to claim 7, characterized in that: The signal display and communication module exchanges traffic flow and congestion information every 5-10 minutes during the communication process with surrounding traffic facilities, and coordinates the traffic signal timing of the entire area through a distributed model predictive control algorithm.

Images