Traffic signal scheduling system and method based on road section congestion information

Abstract

The invention discloses a traffic signal scheduling system and method based on road section congestion information, and relates to the technical field of traffic management and control. Comprising the following steps: S1, data acquisition and preprocessing: acquiring original traffic data of a road section in real time through a multi-source data acquisition unit, the original traffic data comprising traffic flow, vehicle speed, queuing length, occupancy and travel time; meanwhile, external environment data are obtained, wherein the external environment data comprise weather information, holiday and festival information and traffic event information; traffic data and external environment data are acquired in real time through the multi-source data acquisition unit, data quality is guaranteed through cleaning, denoising and standardized preprocessing, and the problem that data fusion of an existing system is insufficient is solved; by means of an adaptive weight learning model, historical traffic data, real-time data and environmental data are combined to dynamically adjust a congestion evaluation index weight, and the defect that a fixed or empirical weight is difficult to adapt to traffic dynamic changes is overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of traffic management and control, in particular to a traffic signal scheduling system and method based on road congestion information. BACKGROUND

[0002] With the acceleration of urbanization and the continuous growth of motor vehicle ownership, urban traffic congestion problems are becoming increasingly serious, which has a significant impact on economic and social development and residents' lives. As a key means to alleviate urban traffic congestion, the intelligentization and refinement level of traffic signal control scheduling strategy directly affect road traffic efficiency. Existing traffic signal scheduling systems mostly rely on fixed or experience-set congestion evaluation indicators (such as traffic flow, occupancy rate, average speed, queue length, etc.) to judge the congestion status of road sections, and adjust the signal timing accordingly.

[0003] However, actual traffic flow has high dynamicity and complexity, and the reflection ability of these single or combined indicators on the "true" congestion level of road sections and the influence weight on traffic efficiency are dynamically changing under different time periods, different road section types, or even under the influence of different weather or unexpected events. For example, during peak hours, even if the queue length is moderate, but the lane saturation is extremely high, it may mean potential overflow risk; while during off-peak hours, similar queue length may represent completely different congestion levels. Fixed weight or experience-set indicator system is difficult to capture these subtle changes, resulting in insufficient adaptability of signal scheduling strategy, often failing to achieve the best traffic flow optimization effect, and even may exacerbate local or regional congestion due to misjudgment. In addition, the existing systems still have room for improvement in data fusion and intelligent decision-making, and it is difficult to fully utilize the increasingly diverse traffic perception data.

[0004] Therefore, there is an urgent need for a traffic signal adaptive scheduling method and system that can overcome the above-mentioned defects by dynamically adjusting the congestion evaluation indicator weight through intelligent learning mechanism, so as to achieve more accurate and intelligent traffic signal adaptive scheduling. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a traffic signal scheduling system and method based on road congestion information, which solves the problem of poor signal scheduling adaptability and ineffective traffic flow optimization caused by the difficulty of fixed or experience-set congestion indicator weight to adapt to traffic dynamic changes, insufficient data fusion and intelligent decision-making compared with the prior art.

[0006] To achieve the above purpose, the present application is implemented by the following technical scheme: a traffic signal scheduling method based on road congestion information, comprising: S1, data acquisition and preprocessing: real-time acquisition of original traffic data of the road section through a multi-source data acquisition unit, the original traffic data including traffic volume, speed, queue length, occupancy rate and travel time; at the same time, external environment data is acquired, including weather information, holiday information and traffic event information; data cleaning, data denoising and data standardization are performed on the original traffic data to obtain preprocessed real-time traffic data; S2, congestion index weight learning: based on historical traffic data, preprocessed real-time traffic data and external environment data, an adaptive weight learning model is used to dynamically adjust the weights of each congestion evaluation index in the original traffic data; S3, dynamic weight congestion index calculation: according to the preprocessed real-time traffic data and the dynamic weight combination output by the adaptive weight learning model, the dynamic weight congestion index of the current road section is calculated; S4, signal timing optimization: taking the dynamic weight congestion index as an input parameter, combining the topological structure information and traffic flow model of the current road network, the current optimal traffic signal timing scheme is generated through the signal timing optimization module; S5, signal control and feedback: the optimal traffic signal timing scheme is sent to the traffic signal controller for execution, and the traffic signal controller accurately controls the red-green light switching of the traffic signal light; and the actual execution state of the traffic signal light and the traffic running effect perceived by the intersection are used as feedback data for iterative optimization of the adaptive weight learning model, to continuously improve the scheduling accuracy.

[0007] Further, the data acquisition and preprocessing comprises: Real-time acquisition of original traffic data through a video detector, a geomagnetic detector, a loop detector and a floating car data receiving module; Accessing external environment data through an external environment information interface; Performing data cleaning operation on the original traffic data to remove abnormal values and noise data; Performing data deduplication and merging operation on the cleaned data to eliminate duplication and fuse multi-source same type data; Performing data standardization processing on the deduplicated and merged data to convert data of different dimensions to a unified scale.

[0008] Further, the congestion index weight learning comprises: The historical traffic data, preprocessed real-time traffic data and external environment data are input into the adaptive weight learning model as a training data set; The adaptive weight learning model learns and identifies the contribution of congestion evaluation indexes to the real congestion state of the road section under different traffic situations through a training process based on the training data set; The adaptive weight learning model outputs a dynamic weight combination that accurately matches the current traffic situation in real time based on learning and recognition results.

[0009] Further, the adaptive weight learning model includes a reinforcement learning model, a deep learning model, or an adaptive optimization algorithm model. The reinforcement learning model regards traffic signal scheduling as a Markov decision process, dynamically adjusts the weights of each congestion indicator by interacting with the traffic environment to maximize traffic efficiency or minimize congestion, and takes the maximum traffic efficiency or minimum congestion as the reward function. The deep learning model constructs a deep neural network, takes multi-source traffic data and environmental data as input, extracts features through the neural network, and predicts the optimal weight combination. The adaptive optimization algorithm model adjusts the weights according to the performance feedback of the traffic system in each iteration to optimize the overall traffic system performance.

[0010] Further, the dynamic weight congestion index calculation includes: Receiving pre-processed real-time traffic data, the real-time traffic data has been standardized to obtain a set of standardized real-time congestion evaluation indicators; Receiving the dynamic weight combination output by the adaptive weight learning model, the dynamic weight combination contains the dynamic weights corresponding to each standardized real-time congestion evaluation indicator; Performing weighted summation operation on the set of standardized real-time congestion evaluation indicators and the dynamic weight combination to obtain a dynamic weight congestion index, which reflects the comprehensive congestion degree of the current road section.

[0011] Further, the signal timing optimization includes: Taking the dynamic weight congestion index as the core input, combining the road network topology information obtained from the map service, the traffic flow pattern information obtained from the historical database, and the short-term traffic flow prediction data obtained through the traffic prediction model; Using VISSIM simulation optimization, SCATS, SCOOT, or other signal control strategies based on reinforcement learning as optimization algorithms to perform traffic flow simulation and evaluation; According to the simulation evaluation results, the best signal timing scheme is generated to maximize the relief of congestion and improve the road capacity, which includes the green light duration of multiple phases, the cycle length and the coordination relationship between adjacent intersections.

[0012] Further, the signal control and feedback includes: After the traffic signal controller receives the optimal traffic signal timing scheme, it immediately performs red-green light switching control of the intersection traffic signal lights to ensure the real-time performance of the timing scheme. The traffic signal controller simultaneously collects actual traffic operation data of the intersection in the execution process, including actual vehicle flow, queue length and average vehicle speed; The collected actual traffic operation data is fed back to the adaptive weight learning model as retraining data of the model to form a closed-loop control, so as to continuously adjust and optimize the dynamic weight generation strategy according to the actual operation effect.

[0013] The application also provides a traffic signal scheduling system based on road congestion information, which is applied to the traffic signal scheduling method based on road congestion information and comprises: A multi-source data collection unit is configured to collect real-time original traffic data and external environment data of a road section; A data preprocessing and storage module is connected with the multi-source data collection unit and configured to receive and clean, denoise, standardize and structure store the original traffic data, and store historical traffic data; An adaptive weight learning model is connected with the data preprocessing and storage module and configured to dynamically adjust the weight of the congestion evaluation index based on the historical traffic data, real-time traffic data and external environment data; A congestion index calculation module is connected with the data preprocessing and storage module and the adaptive weight learning model and configured to calculate a dynamic weight congestion index according to the real-time traffic data and the dynamic weight; A signal timing optimization module is connected with the congestion index calculation module and configured to generate an optimal traffic signal timing scheme according to the dynamic weight congestion index, in combination with a road network topology and a traffic flow model; A traffic signal controller is connected with the signal timing optimization module and configured to receive and execute the optimal traffic signal timing scheme, accurately control the traffic signal lamp, and feed back actual traffic operation effect to the multi-source data collection unit and the adaptive weight learning model.

[0014] Further, the multi-source data collection unit comprises: A video detector is configured to acquire vehicle flow, vehicle speed, lane occupancy rate and queue length parameters in real time through video image recognition technology; A geomagnetic or loop detector is configured to collect instantaneous vehicle flow and lane occupancy rate; A floating car data receiving module is configured to receive GPS data from a floating vehicle to calculate road travel time and average speed; An external environment information interface is configured to access meteorological data, holiday information and traffic event report data.

[0015] Further, the multi-source data collection unit is deployed at key positions in a road network, the data preprocessing and storage module is deployed on a regional traffic management center or a cloud server, and the adaptive weight learning model is deployed on a computing server.

[0016] Compared with the prior art, the present application has the following advantages: The present application solves the problem of insufficient data fusion of existing systems by real-time acquisition of traffic data and external environment data through a multi-source data acquisition unit, ensuring data quality through cleaning, denoising, and standardization preprocessing; dynamically adjusts the weight of congestion evaluation indicators by combining historical traffic data, real-time data, and environmental data with the aid of a self-adaptive weight learning model, overcoming the defects of fixed or experienced weights that are difficult to adapt to traffic dynamics; calculates the congestion index based on dynamic weights to accurately reflect the real congestion degree, and combines the road network topology structure and traffic flow model to generate an optimal signal timing scheme through an optimization algorithm; at the same time, the optimization model is fed back by the actual traffic operation effect to form a closed-loop control, continuously improving the scheduling accuracy, effectively relieving traffic congestion, and improving the efficiency of road section traffic. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The method flowchart of the present application is shown in the figure; Figure 2 The system structure diagram of the present application is shown in the figure. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0019] Please refer to Figure 1 The present application provides a traffic signal scheduling method based on road congestion information, comprising: S1, data acquisition and preprocessing: real-time acquisition of original traffic data of the road section through a multi-source data acquisition unit, the original traffic data including traffic volume, speed, queue length, occupancy rate, and travel time; at the same time, external environment data is acquired, including weather information, holiday information, and traffic event information; data cleaning, data denoising, and data standardization processing are performed on the original traffic data to obtain preprocessed real-time traffic data; S2, congestion index weight learning: based on historical traffic data, preprocessed real-time traffic data, and external environment data, a self-adaptive weight learning model is used to dynamically adjust the weight of each congestion evaluation indicator in the original traffic data; S3, dynamic weight congestion index calculation: according to the preprocessed real-time traffic data and the dynamic weight combination output by the self-adaptive weight learning model, the dynamic weight congestion index of the current road section is calculated; S4. Signal Timing Optimization: Using the dynamic weighted congestion index as an input parameter, combined with the current road network topology information and traffic flow model, the signal timing optimization module generates the current optimal traffic signal timing scheme. S5. Signal Control and Feedback: The optimal traffic signal timing scheme is sent to the traffic signal controller for execution, which precisely controls the switching of traffic lights. The actual execution status of the traffic lights and the traffic operation effect perceived at the intersection are used as feedback data for iterative optimization of the adaptive weight learning model to continuously improve scheduling accuracy.

[0020] Specifically, in the practical application of this traffic signal dispatching method based on road segment congestion information, the first step is data collection and preprocessing. Multi-source data acquisition units operate synchronously, acquiring raw traffic data in real time, including traffic flow, vehicle speed, queue length, occupancy rate, and travel time. Simultaneously, they also collect external environmental data such as weather information, holiday information, and traffic event information. The raw traffic data is then processed: first, outliers and noise are removed for data cleaning; then, denoising is performed to further improve data quality; finally, data standardization is performed to unify data of different dimensions to the same scale, resulting in preprocessed real-time traffic data, laying a reliable data foundation for subsequent analysis. Next, congestion index weight learning is performed, using historical traffic data, preprocessed real-time traffic data, and external environment data as basic data, which are then input into an adaptive weight learning model. This model can dynamically adjust the weights of each congestion assessment index in the original traffic data based on different time periods, road segment types, and changes in the external environment. For example, during peak hours, the weight of lane occupancy in congestion assessment is increased, while during off-peak hours, the weights of indicators such as queue length are adjusted accordingly. This more accurately reflects the true congestion status of road segments under different scenarios, solving the problem that the existing fixed-weight index system cannot adapt to dynamic changes in traffic flow. The dynamic weighted congestion index is then calculated. Based on preprocessed real-time traffic data and the dynamic weight combination output by the adaptive weight learning model, the dynamic weighted congestion index of the current road segment is obtained through weighted summation. This index comprehensively reflects the congestion level of the current road segment, providing an accurate basis for signal timing. The formula for calculating the dynamic weighted congestion index is: ; in, This represents the dynamic weighted congestion index, with a value range of [0,1]. The larger the value, the higher the degree of congestion on the road segment. Indicates the first The dynamic weights of each congestion assessment indicator, all the sum of which is 1, output by the adaptive weight learning model; denotes the standardized congestion evaluation index, such as the standardized traffic flow, speed, etc., with a value range of [0, 1]; denotes the standardized congestion evaluation index, such as the standardized traffic flow, speed, etc., with a value range of [0, 1]; denotes the total number of congestion evaluation indexes, which is 5 in this embodiment; corresponding to traffic flow, speed, queue length, occupancy rate, and travel time; In the signal timing optimization link, the dynamic weight congestion index is used as a key input parameter, and the topological structure information and traffic flow model of the current road network are combined. The signal timing optimization module comprehensively analyzes these information to generate the current optimal traffic signal timing scheme, ensuring that the scheme can adapt to the current traffic conditions to the greatest extent and improve road traffic efficiency. Finally, signal control and feedback are performed. The optimal traffic signal timing scheme is sent to the traffic signal controller for accurate control of the red-green light switching of the traffic signal light. The actual execution state of the traffic signal light and the traffic running effect perceived at the intersection are collected as feedback data and fed back to the adaptive weight learning model for iterative optimization of the model. Through such a closed-loop feedback mechanism, the scheduling accuracy is continuously improved, urban traffic congestion is effectively alleviated, and traffic operation effect is improved.

[0021] In this embodiment, data collection and preprocessing include: Raw traffic data is obtained in real time through video detectors, geomagnetic detectors, loop detectors, and floating car data receiving modules; External environmental data is accessed through an external environmental information interface; Data cleaning operations are performed on the raw traffic data to remove outliers and noise data; Data deduplication and merging operations are performed on the cleaned data to eliminate duplicates and fuse multiple sources of the same data; Data standardization processing is performed on the deduplicated and merged data to convert data of different dimensions to a unified scale.

[0022] Specifically, in the data collection link, various detection devices and modules are used to obtain raw traffic data. Video detectors capture traffic flow, speed, queue length, and occupancy rate information in real time by shooting road videos using image recognition technology. Geomagnetic detectors and loop detectors are buried under the road to collect instantaneous traffic flow and lane occupancy rate data by sensing vehicles. The floating car data receiving module receives GPS data sent by traveling vehicles to calculate travel time and average speed of road segments. At the same time, through the external environmental information interface, weather information provided by the meteorological department, holiday information released by the government system, and traffic event information reported by the traffic management department are accessed. In the data preprocessing process, first, the original traffic data obtained is subjected to data cleaning operation, preset abnormal value judgment rules are adopted, for example, traffic flow and speed data exceeding the normal range are identified as abnormal values, chaotic data generated due to equipment failure are regarded as noise data, and these abnormal values and noise data are removed. Then, the data after cleaning is subjected to data deduplication and merging operation, the same type of data collected by different detection devices is compared, the repeated data entries are eliminated, then the multi-source same type of data is fused to form a more comprehensive and coherent data set. Finally, the data after deduplication and merging is subjected to data standardization processing by Min-Max standardization, and the traffic flow, speed, queue length and other data of different dimensions are converted to a unified scale (such as [0, 1] interval), so as to facilitate subsequent unified analysis and calculation, and improve the usability and accuracy of the data.

[0023] In the present embodiment, the congestion index weight learning includes: The historical traffic data, the preprocessed real-time traffic data and the external environment data are input into the adaptive weight learning model as a training data set; The adaptive weight learning model learns and identifies the contribution degree of the congestion evaluation index to the real congestion state of the road section under different traffic situations based on the training data set through a training process; The adaptive weight learning model outputs a dynamic weight combination accurately matched with the current traffic situation in real time according to the learning and identification results.

[0024] Specifically, when learning the congestion index weight, first, the historical traffic data, the preprocessed real-time traffic data and the external environment data are integrated to construct a training data set, which is input into the adaptive weight learning model. The historical traffic data contains traffic operation data of each road section at different dates and different time periods, providing rich historical experience reference for the model; the preprocessed real-time traffic data can reflect the instant traffic condition of the current road section; the external environment data supplements the external factor information affecting the traffic; After receiving the training data set, the adaptive weight learning model starts the training process. In the training process, the model analyzes the data in depth, learns and identifies the contribution degree of each congestion evaluation index to the real congestion state of the road section under different traffic situations. For example, in the bad weather situation such as rainy day, the influence of the speed index on the congestion state will increase, and the model can identify this feature; in the situation where a traffic event occurs, the contribution degree of the queue length index will be significantly improved, and the model can also accurately capture it; Based on the learning and identified results, the adaptive weight learning model outputs a dynamic weight combination that accurately matches the current traffic situation in real time. When the traffic situation changes, such as transitioning from off-peak to peak hours or from sunny to rainy weather, the model quickly adjusts the weights of each congestion evaluation indicator to ensure that the weight combination always adapts to the current traffic situation, providing accurate weight basis for subsequent dynamic weight congestion index calculation, thereby improving the accuracy of road congestion state judgment.

[0025] In this embodiment, the adaptive weight learning model includes a reinforcement learning model, a deep learning model, or an adaptive optimization algorithm model. The reinforcement learning model considers traffic signal scheduling as a Markov decision process, learns through interaction with the traffic environment, and dynamically adjusts the weights of each congestion indicator with maximum traffic efficiency or minimum congestion as the reward function. The deep learning model constructs a deep neural network, takes multi-source traffic data and environmental data as input, extracts features through the neural network, and predicts the optimal weight combination. The adaptive optimization algorithm model adjusts the weights according to the performance feedback of the traffic system in each iteration to optimize the overall traffic system performance.

[0026] Specifically, when the adaptive weight learning model uses a reinforcement learning model, it considers the traffic signal scheduling process as a Markov decision process. In this process, the model continuously interacts with the traffic environment, constantly tries different weight combinations of congestion indicators, and adjusts the strategy according to the feedback from the traffic environment. At the same time, maximum traffic efficiency or minimum congestion is set as the reward function. When a certain weight combination is adopted, if the traffic efficiency is improved or the congestion level is reduced, the model will receive a positive reward, otherwise a negative reward. Through continuous interaction and reward feedback, the model will gradually learn the optimal weight of congestion indicators that can achieve the best reward under different traffic conditions, achieving dynamic adjustment. The formula of the reinforcement learning model is: ; wherein, represents the immediate reward value of the reinforcement learning model; , represents the reward coefficient, which is positive, used to balance the influence weight of traffic efficiency and congestion level, and can be preset according to the actual traffic scene; represents the traffic efficiency indicator, such as the normalized average speed of the road segment, with a value range of [0, 1], and the larger the value, the higher the traffic efficiency; represents the congestion level indicator, which can use the dynamic weight congestion index , with a value range of [0, 1], and the larger the value, the higher the congestion level; If a deep learning model is used, a deep neural network is constructed, which typically includes an input layer, hidden layers, and an output layer. The input layer receives multi-source traffic and environmental data and transforms it into a format that the neural network can process. The hidden layer uses multiple neurons to perform feature extraction and complex calculations on the data, uncovering potential correlations and patterns. The output layer then predicts the optimal combination of congestion index weights based on the results of the hidden layer's calculations. During training, a large amount of historical data is used to train the neural network, and the network parameters are continuously adjusted using the backpropagation algorithm to improve the accuracy of the model's predictions, enabling it to accurately output appropriate weight combinations based on real-time input data. When using an adaptive optimization algorithm model, in each iteration, the model first generates a traffic signal timing scheme based on the current congestion index weight combination and applies it to the actual traffic scenario. Then, it collects traffic system performance data, such as road segment speed and vehicle queuing time, as performance feedback. Next, based on this performance feedback, it determines whether the current weight combination can achieve optimal traffic system performance. If not, it adjusts the weights of each congestion index according to preset optimization rules. Through continuous iterative adjustments, the weight combination is gradually optimized, continuously improving the overall traffic system performance to adapt to constantly changing traffic conditions. The weight iteration formula is: ; in, Indicates the first After the nth iteration Weights of each congestion assessment indicator; Indicates the first During the nth iteration Weights of each congestion assessment indicator; This represents the learning rate, which ranges from [0,1]. It is used to control the step size of weight adjustment to avoid excessive adjustment that could lead to model instability. Indicates the first The weight adjustment of each congestion assessment indicator is calculated based on traffic system performance feedback; for example, when performance improves... When positive, performance degrades. It is negative.

[0027] In this embodiment, the calculation of the dynamic weighted congestion index includes: Receive preprocessed real-time traffic data. The real-time traffic data has been standardized to obtain a set of standardized real-time congestion assessment indicators. It receives a dynamic weight combination output by an adaptive weight learning model, which includes dynamic weights corresponding to each standardized real-time congestion assessment index. The dynamic weight congestion index is obtained by weighted summation operation of the set of standardized real-time congestion evaluation indicators combined with dynamic weights, and the dynamic weight congestion index reflects the comprehensive congestion degree of the current road section.

[0028] Specifically, in the calculation of the dynamic weight congestion index, first, the preprocessed and standardized real-time traffic data are received, which constitute the set of standardized real-time congestion evaluation indicators. Each indicator in the set has a unified scale, eliminating the influence of dimensional differences and facilitating subsequent weighted calculation. At the same time, the dynamic weight combination output by the adaptive weight learning model is received, which contains dynamic weights corresponding to each indicator in the set of standardized real-time congestion evaluation indicators. These weights are dynamically adjusted according to the current traffic situation and can accurately reflect the influence of each indicator on the congestion degree of the road section. Then, each indicator in the set of standardized real-time congestion evaluation indicators is multiplied by the corresponding dynamic weight to obtain the weighted value of each indicator. Finally, the weighted values of all indicators are added to obtain the dynamic weight congestion index by weighted summation operation. This index takes into account the importance of each congestion evaluation indicator in the current situation and can accurately reflect the comprehensive congestion degree of the current road section, providing a reliable basis for subsequent signal timing optimization.

[0029] In this embodiment, signal timing optimization includes: The dynamic weight congestion index is used as the core input, combined with the road network topology information obtained from the map service, the traffic flow pattern information obtained from the historical database, and the short-term traffic flow prediction data obtained through the traffic prediction model. VISSIM simulation optimization, SCATS, SCOOT, or other signal control strategies based on reinforcement learning are used as optimization algorithms for traffic flow simulation and evaluation. According to the simulation evaluation results, the best signal timing scheme is generated to maximize the relief of congestion and improve the capacity of the road section. The scheme includes the green light duration of multiple phases, the cycle length, and the coordination relationship between adjacent intersections.

[0030] Specifically, in the signal timing optimization process, the dynamic weight congestion index is used as the core input parameter, and the topology structure information of the current road network is obtained from the map service, including the connection relationship of the road, the distribution of the intersection, the number of lanes, etc. The traffic flow pattern information is retrieved from the historical database to understand the variation of the road traffic flow at different times and on different days. The short-term traffic flow prediction data is obtained by predicting the future traffic flow within a short period of time through the traffic prediction model. Then, according to the actual demand and traffic scene characteristics, a suitable optimization algorithm is selected, such as VISSIM simulation optimization, SCATS, SCOOT or other signal control strategies based on reinforcement learning. VISSIM simulation optimization, SCATS and SCOOT are all mature existing technologies, and will not be described here. Taking VISSIM simulation optimization as an example, the obtained data is input into the VISSIM simulation software to construct a simulation model similar to the actual traffic scene. Different signal timing schemes are simulated to simulate the driving process of vehicles on the road, record the travel time, queue length, delay of vehicles and other data, and evaluate and analyze these data; According to the simulation evaluation results, the signal timing scheme that can most effectively alleviate congestion and improve road capacity is selected. This scheme includes the green light time of multiple phases, and reasonably allocates green light time according to the flow and congestion of vehicles in different directions; determine the cycle length to ensure that the traffic flow in each direction can be effectively dredged in one cycle; clarify the coordination relationship between adjacent intersections, so that the signal lights of adjacent intersections are switched in coordination, reduce the delay of vehicles between intersections, and improve the traffic efficiency of the entire road network.

[0031] In this embodiment, signal control and feedback include: After the traffic signal controller receives the optimal traffic signal timing scheme, it immediately performs red-green light switching control of the intersection traffic signal lights to ensure the real-time performance of the timing scheme; The traffic signal controller also collects actual traffic operation data at the intersection during execution, including actual traffic flow, queue length and average speed; The collected actual traffic operation data is fed back to the adaptive weight learning model as retraining data for the model to form a closed-loop control and continuously adjust and optimize the dynamic weight generation strategy according to the actual operation effect.

[0032] Specifically, after receiving the optimal traffic signal timing scheme, the traffic signal controller will immediately start the execution program, accurately control the red-green light switching of the intersection traffic signal lights according to the green light time, red light time, yellow light time and phase switching sequence specified in the scheme, ensure that the timing scheme can be applied to actual traffic control in a timely and accurate manner, ensure the real-time performance of signal control, and avoid traffic chaos caused by delayed scheme execution; While executing the timing scheme, the traffic signal controller collects real-time traffic operation data of the intersection in the execution process through connected detection devices such as video detectors, geomagnetic detectors, etc. These data include actual traffic volume, i.e. the number of vehicles passing through each direction of the intersection per unit time; queue length, i.e. the queue length of vehicles waiting for traffic in each direction of the intersection; average speed, i.e. the average speed of vehicles within the intersection range, etc., to record the traffic conditions in the execution process of the traffic signal completely; Then, the collected actual traffic operation data is fed back to the adaptive weight learning model in time as retraining data of the model. If the model is a reinforcement learning model, the feedback data is used to update and in the formula , so as to adjust the reward value to optimize the weight; if the model is an adaptive optimization algorithm model, the feedback data is used to calculate in the formula , to realize weight iteration. The adaptive weight learning model will optimize and improve its weight adjustment strategy in combination with these new feedback data, relearn the importance of each congestion index under the current timing scheme execution, and then adjust the dynamic weight generation method. Through such a closed-loop control, the model continuously adjusts and optimizes the dynamic weight generation strategy according to the actual operation effect, continuously improves the accuracy and adaptability of the model, and thus further improves the precision of traffic signal scheduling.

[0033] Please refer to Figure 2 , in this embodiment, a traffic signal scheduling system based on road congestion information is also provided, which is applied to the above-mentioned traffic signal scheduling method based on road congestion information, and includes: A multi-source data acquisition unit is configured to acquire real-time traffic data and external environment data of a road segment; A data preprocessing and storage module is connected with the multi-source data acquisition unit, configured to receive and clean, denoise, standardize and structure store the original traffic data, and store historical traffic data; An adaptive weight learning model is connected with the data preprocessing and storage module, configured to dynamically adjust the weight of the congestion evaluation index based on the historical traffic data, real-time traffic data and external environment data; A congestion index calculation module is connected with the data preprocessing and storage module and the adaptive weight learning model, configured to calculate a dynamic weight congestion index according to the real-time traffic data and the dynamic weight; A signal timing optimization module is connected with the congestion index calculation module, configured to generate an optimal traffic signal timing scheme according to the dynamic weight congestion index, in combination with a road network topology and a traffic flow model; Traffic signal controller: connected with the signal timing optimization module, used for receiving and executing the optimal traffic signal timing scheme, accurately controlling the traffic signal lights, and feeding back the actual traffic running effect to the multi-source data acquisition unit and the adaptive weight learning model.

[0034] Specifically, the multi-source data acquisition unit as the data acquisition source of the system will deploy corresponding detection equipment and modules at key road sections and intersections according to the layout of the road network and traffic monitoring needs, and real-time collect the original traffic data of the road section, such as traffic flow, speed, queue length, occupancy rate and travel time, etc., and also collect external environmental data such as weather information, holiday information and traffic event information, to provide comprehensive and real-time basic data for the system; The data preprocessing and storage module is connected with the multi-source data acquisition unit, and receives the original traffic data transmitted by the multi-source data acquisition unit. First, the original traffic data is cleaned to remove abnormal values and noise data; then, the noise is removed to further improve the data quality; then, the data is standardized to convert different dimensional data to a unified scale; finally, the processed data is stored in a structured manner for subsequent query and use. At the same time, the module also stores historical traffic data, which provides rich training samples and reference for the adaptive weight learning model; The adaptive weight learning model is connected with the data preprocessing and storage module, and can obtain historical traffic data, preprocessed real-time traffic data and external environmental data from the module. Based on these data, the model dynamically adjusts the weights of the congestion evaluation indexes through specific algorithms and training processes, so that the weights can adapt to different traffic situations and accurately reflect the influence of each index on the road congestion state; The congestion index calculation module is connected with the data preprocessing and storage module and the adaptive weight learning model, respectively, to obtain preprocessed real-time traffic data from the data preprocessing and storage module and dynamic weight combination from the adaptive weight learning model. Then, according to the dynamic weight congestion index calculation method, the real-time traffic data is combined with the dynamic weight combination to calculate the dynamic weight congestion index, which is used to measure the comprehensive congestion degree of the current road section; The signal timing optimization module is connected with the congestion index calculation module, and receives the dynamic weight congestion index transmitted by the congestion index calculation module. At the same time, the module also obtains the topological structure information and traffic flow model of the current road network, combines these information and the dynamic weight congestion index, and uses VISSIM simulation optimization, SCATS and other optimization algorithms to generate the optimal traffic signal timing scheme, to achieve the goal of relieving congestion and improving traffic efficiency; The traffic signal controller is connected with the signal timing optimization module, receives the optimal traffic signal timing scheme sent by the signal timing optimization module, and accurately controls the switching of the traffic signal according to the scheme. During the execution process, the traffic signal controller also collects the actual execution state of the traffic signal and the actual traffic operation effect of the intersection, and feeds back these data as feedback data to the multi-source data acquisition unit and the adaptive weight learning model respectively. The data fed back to the multi-source data acquisition unit can enrich the data reserve, and the data fed back to the adaptive weight learning model is used for iterative optimization of the model, so as to continuously improve the scheduling performance of the whole system.

[0035] In the embodiment, the multi-source data acquisition unit comprises: Video detector: real-time acquisition of traffic flow, speed, lane occupancy and queue length parameters through video image recognition technology; Magnetic or loop detector: used for collecting instantaneous traffic flow and lane occupancy; Floating car data receiving module: receiving GPS data from floating vehicles for calculating link travel time and average speed; External environment information interface: used for accessing meteorological data, holiday information and traffic event report data.

[0036] Specifically, the video detector in the multi-source data acquisition unit is installed on the crossbar or bracket above the road, and its lens is aimed at the monitoring road section or intersection. By continuously shooting video images of the traffic scene, the image recognition and processing technology is used to detect and track the vehicles in the video, so as to real-time acquire the traffic flow, i.e. the number of vehicles passing through the monitoring area per unit time; the speed, i.e. the speed of the vehicle on the monitoring road section; the lane occupancy, i.e. the time or space proportion of the vehicle on the lane; and the queue length, i.e. the length of the vehicle queue waiting at the intersection or road section, and other parameters; The magnetic or loop detector is buried under the lane of the road. When the magnetic detector detects the passing of a vehicle, the change of the geomagnetic field caused by the vehicle is captured by the sensor, and then the change of the magnetic field is converted into an electric signal, and the presence information of the vehicle is obtained through signal processing, so as to collect the instantaneous traffic flow, i.e. the number of vehicles passing through the detector at a certain time, and the lane occupancy data. The loop detector is realized by laying a loop coil under the lane. When a vehicle passes through the coil, it will change the inductance of the coil, causing the current in the circuit to change. By detecting this change of current, the presence of the vehicle is sensed, and the instantaneous traffic flow and lane occupancy are collected; The floating car data receiving module is deployed at traffic management centers or related data processing stations. This module receives GPS data, such as location, speed, and direction of travel, from floating vehicles equipped with GPS devices via wireless communication networks, such as 4G and 5G networks. Then, based on this GPS data and time information, it calculates the travel time for a road segment (the time required for a vehicle to traverse a certain road segment) and the average speed (the average speed of the vehicle during its journey through the road segment). The external environmental information interface establishes data connections with meteorological data systems of meteorological departments, government information systems, and traffic incident reporting systems of traffic management departments. Through this interface, real-time meteorological data, including temperature, rainfall, and visibility, can be accessed; holiday schedule information published by government information systems, such as the dates of statutory holidays and work schedule adjustments, can be accessed; and traffic incident reports submitted by traffic management departments, such as information on traffic accidents, road construction, and temporary traffic control, can be accessed. This external environmental data serves as input to the adaptive weight learning model, used to adjust the dynamic weight congestion index calculation formula. The formulas for reinforcement learning models and the parameters in the weight iteration formulas provide comprehensive external environment data support for the system.

[0037] In this embodiment, the multi-source data acquisition unit is deployed at key locations in the road network, the data preprocessing and storage module is deployed in the regional traffic management center or cloud server, and the adaptive weight learning model is deployed on the computing server.

[0038] Specifically, during the deployment of multi-source data acquisition units, they are deployed at key locations within the road network based on traffic flow distribution, critical node locations, and key traffic management needs. For example, video detectors, geomagnetic detectors, and loop detectors are installed at intersections of main urban roads, sections with high traffic volume, major roads leading into and out of the city, and hubs connecting different areas. Signal receiving points for floating car data receiving modules are also set up to ensure comprehensive and accurate collection of traffic data from these key locations, enabling timely monitoring of traffic conditions. The data preprocessing and storage module needs to handle large amounts of traffic data, including real-time collected data and historically stored data, which places high demands on computing power and storage capacity. Therefore, it is usually deployed in a regional traffic management center or on a cloud server. The regional traffic management center has dedicated hardware facilities and a network environment, which can provide stable operating conditions for the module and facilitate management and maintenance by staff. The cloud server has elastically scalable storage and computing capabilities, which can flexibly adjust resources according to changes in data volume to ensure efficient data preprocessing and secure data storage. It also allows other modules to access and retrieve the processed data via the network. The adaptive weight learning model needs to perform complex calculations and model training during operation, and has high requirements for computing resources, so it is deployed on a computing server. The computing server is equipped with high-performance CPU, GPU and sufficient memory, can quickly process a large amount of data, efficiently executes the training and weight adjustment algorithm of the model, ensures that the model can output accurate dynamic weight combination in time according to the input data, and meets the real-time and accuracy requirements of traffic signal scheduling. At the same time, the computing server can also provide stable calculation support for the iterative optimization of the model, and guarantee the continuous improvement of the performance of the model.

[0039] In summary, the present application acquires traffic data and external environment data in real time through a multi-source data acquisition unit, and guarantees data quality through cleaning, denoising and standardization preprocessing, solving the problem of insufficient data fusion of existing systems; with the help of an adaptive weight learning model, combining historical traffic data, real-time data and environment data, dynamically adjusting congestion evaluation index weight, overcoming the defect that fixed or experience weight is difficult to adapt to traffic dynamic changes; calculating congestion index based on dynamic weight to accurately reflect the real congestion degree, and combining road network topology structure and traffic flow model, generating an optimal signal timing scheme through an optimization algorithm; at the same time, feeding back the optimization model with the actual traffic operation effect, forming a closed-loop control, continuously improving the scheduling accuracy, effectively relieving traffic congestion, and improving the road section passing efficiency.

[0040] It should be noted that, in this text, relational terms such as first and second are used merely to distinguish one entity or action from another, without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variant thereof are intended to cover non-exclusive inclusions, so that a process, method, article, or apparatus that includes a list of elements does not only include those elements, but also includes other elements not explicitly listed, or inherent to such process, method, article, or apparatus.

[0041] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.​​​

Claims

1. A traffic signal dispatching method based on road segment congestion information, characterized in that, include: S1. Data Acquisition and Preprocessing: The raw traffic data of the road segment is acquired in real time through a multi-source data acquisition unit. The raw traffic data includes traffic flow, vehicle speed, queue length, occupancy rate, and travel time. At the same time, external environmental data is acquired, including weather information, holiday information, and traffic event information. The raw traffic data is cleaned, denoised, and standardized to obtain preprocessed real-time traffic data. S2. Congestion index weight learning: Based on historical traffic data, preprocessed real-time traffic data and external environment data, an adaptive weight learning model is used to dynamically adjust the weights of each congestion assessment index in the original traffic data. S3. Calculation of dynamic weighted congestion index: The dynamic weighted congestion index of the current road segment is calculated based on the preprocessed real-time traffic data and the dynamic weight combination output by the adaptive weight learning model. S4. Signal Timing Optimization: Using the dynamic weighted congestion index as an input parameter, combined with the current road network topology information and traffic flow model, the signal timing optimization module generates the current optimal traffic signal timing scheme. S5. Signal Control and Feedback: The optimal traffic signal timing scheme is sent to the traffic signal controller for execution, which precisely controls the switching of traffic lights. The actual execution status of the traffic lights and the traffic operation effect perceived at the intersection are used as feedback data for iterative optimization of the adaptive weight learning model to continuously improve scheduling accuracy.

2. The traffic signal dispatching method based on road segment congestion information according to claim 1, characterized in that, The data acquisition and preprocessing include: Raw traffic data is acquired in real time using video detectors, geomagnetic detectors, loop detectors, and floating car data receiving modules. Access external environment data through the external environment information interface; Perform data cleaning operations on the raw traffic data to remove outliers and noisy data; Perform data deduplication and merging operations on the cleaned data to eliminate duplicates and integrate similar data from multiple sources; Data standardization is performed on the deduplicated and merged data to convert data of different dimensions to a unified scale.

3. The traffic signal dispatching method based on road segment congestion information according to claim 1, characterized in that, The learning of congestion indicator weights includes: Historical traffic data, preprocessed real-time traffic data, and external environment data are used as training datasets and input into the adaptive weight learning model. The adaptive weight learning model is based on a training dataset. Through the training process, it learns and identifies the contribution of congestion assessment indicators to the actual congestion status of road segments under different traffic scenarios. The adaptive weight learning model outputs dynamic weight combinations that precisely match the current traffic situation in real time based on the learning and recognition results.

4. The traffic signal dispatching method based on road segment congestion information according to claim 1, characterized in that, The adaptive weight learning model includes a reinforcement learning model, a deep learning model, or an adaptive optimization algorithm model. The reinforcement learning model treats traffic signal scheduling as a Markov decision process. Through interactive learning with the traffic environment, it uses the maximization of traffic efficiency or the minimization of congestion as the reward function to dynamically adjust the weights of various congestion indicators. Deep learning models construct deep neural networks, taking multi-source traffic data and environmental data as input, and extracting features and predicting the optimal weight combination through the neural network. The adaptive optimization algorithm model adjusts the weights based on the performance feedback of the traffic system in each iteration to optimize the overall traffic system performance.

5. The traffic signal dispatching method based on road segment congestion information according to claim 1, characterized in that, The calculation of the dynamic weighted congestion index includes: Receive preprocessed real-time traffic data. The real-time traffic data has been standardized to obtain a set of standardized real-time congestion assessment indicators. It receives a dynamic weight combination output by an adaptive weight learning model, which includes dynamic weights corresponding to each standardized real-time congestion assessment index. The standardized set of real-time congestion assessment indicators is combined with dynamic weights and weighted summation to obtain the dynamic weighted congestion index, which reflects the overall congestion level of the current road segment.

6. The traffic signal dispatching method based on road segment congestion information according to claim 1, characterized in that, The signal timing optimization includes: The dynamic weighted congestion index is used as the core input, combined with road network topology information obtained from map services, traffic flow pattern information obtained from historical databases, and short-term traffic flow prediction data obtained through traffic prediction models. VISSIM simulation optimization, SCATS, SCOOT, or other reinforcement learning-based signal control strategies are used as optimization algorithms for traffic flow simulation and evaluation. Based on the simulation evaluation results, the optimal signal timing scheme is generated to maximize the relief of congestion and improve the traffic capacity of the road segment. The scheme includes the green light duration, cycle length and coordination relationship of adjacent intersections for multiple phases.

7. The traffic signal dispatching method based on road segment congestion information according to claim 1, characterized in that, The signal control and feedback include: After receiving the optimal traffic signal timing scheme, the traffic signal controller immediately executes the red and green light switching control of the traffic lights at the intersection to ensure the real-time performance of the timing scheme. The traffic signal controller simultaneously collects actual traffic operation data at the intersection during the execution process, including actual traffic flow, queue length, and average vehicle speed; The collected actual traffic operation data is fed back to the adaptive weight learning model as retraining data to form a closed-loop control, prompting the model to continuously adjust and optimize the dynamic weight generation strategy based on the actual operation results.

8. A traffic signal dispatching system based on road segment congestion information, applied to the traffic signal dispatching method based on road segment congestion information described in claims 1-7, characterized in that, include: Multi-source data acquisition unit: used to collect raw traffic data and external environmental data of road sections in real time; Data preprocessing and storage module: connected to the multi-source data acquisition unit, used to receive and clean, denoise, standardize and structure the raw traffic data, and store historical traffic data; Adaptive weighted learning model: Connected to the data preprocessing and storage module, it is used to dynamically adjust the weights of congestion assessment indicators based on historical traffic data, real-time traffic data, and external environment data. Congestion Index Calculation Module: Connected to the data preprocessing and storage module and the adaptive weight learning model, it is used to calculate the dynamic weight congestion index based on real-time traffic data and dynamic weights. Signal timing optimization module: Connected to the congestion index calculation module, it is used to generate the optimal traffic signal timing scheme based on the dynamic weighted congestion index, combined with the road network topology and traffic flow model; Traffic signal controller: Connected to the signal timing optimization module, it receives and executes the optimal traffic signal timing scheme, precisely controls traffic lights, and feeds back the actual traffic operation results to the multi-source data acquisition unit and the adaptive weight learning model.

9. The traffic signal dispatching system based on road segment congestion information according to claim 8, characterized in that, The multi-source data acquisition unit includes: Video detector: It uses video image recognition technology to obtain parameters such as traffic flow, vehicle speed, lane occupancy and queue length in real time; Geomagnetic or loop detectors: used to collect instantaneous traffic flow and lane occupancy. Floating car data receiving module: Receives GPS data from floating vehicles to calculate segment travel time and average speed; External environment information interface: used to access meteorological data, holiday information and traffic incident report data.

10. The traffic signal dispatching system based on road segment congestion information according to claim 8, characterized in that, The multi-source data acquisition unit is deployed at key locations in the road network, the data preprocessing and storage module is deployed in the regional traffic management center or cloud server, and the adaptive weight learning model is deployed on the computing server.

Images