Intelligent control method and system for traffic lights

Abstract

The invention discloses a traffic signal lamp intelligent control method and system, and relates to the technical field of intelligent traffic systems, and the method comprises the steps: collecting traffic operation and environment data through deploying a sensor group in a target traffic area signal lamp; a self-adaptive analysis model including anomaly detection and congestion prediction is obtained through a traffic scheduling platform, traffic conditions are analyzed, and congestion parameters are obtained. And starting an emergency response mechanism according to the parameters, analyzing and determining signal lamp control strategy parameters, and finally realizing adaptive regulation and control of the traffic signal lamp. The technical problems that the real-time response capability of an existing traffic signal lamp control system is insufficient, and traffic abnormity and congestion cannot be effectively predicted are solved, and through real-time traffic data analysis and intelligent prediction, the real-time response capability of the traffic signal lamp system is enhanced, and the traffic safety is improved. And traffic abnormity and congestion can be effectively predicted and dealt with.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent transportation systems, and in particular to an intelligent control method and system for traffic lights. Background Art

[0002] With the acceleration of urbanization, traffic flow has increased dramatically, and the problems of traffic congestion and frequent accidents have become more prominent. Traditional traffic light control systems usually rely on fixed signal cycles or manually set rules, lacking the ability to respond to changes in traffic flow in real time and dynamically adjust, and are difficult to cope with emergencies and peak traffic pressure. These systems cannot effectively predict and identify traffic anomalies, congestion and other conditions, resulting in low traffic efficiency and delays. Therefore, how to improve the control efficiency of traffic lights and reduce traffic congestion through intelligent means has become a problem that needs to be solved urgently in urban traffic management.

[0003] At present, relevant technologies still have technical problems such as insufficient real-time response capability in traffic light control systems and inability to effectively predict traffic anomalies and congestion. Summary of the invention

[0004] The present application solves the technical problems of insufficient real-time response capability and inability to effectively predict traffic anomalies and congestion in existing traffic light control systems by providing a traffic light intelligent control method and system.

[0005] The present application provides a traffic light intelligent control method, including:

[0006] A sensor group is deployed on the traffic lights in the target traffic area, and the traffic operation data information and the traffic environment data information are collected and acquired through the sensor group; a traffic adaptive analysis model is acquired through a traffic dispatching platform, and the traffic adaptive analysis model is composed of a traffic anomaly detection sub-model and a traffic congestion prediction sub-model; a traffic condition analysis is performed on the traffic operation data information and the traffic environment data information based on the traffic adaptive analysis model to acquire traffic operation congestion parameters; a traffic emergency response mechanism is initiated based on the traffic operation congestion parameters; a traffic light control strategy analysis is performed on the traffic operation congestion parameters based on the traffic emergency response mechanism to determine the traffic light control strategy parameters; and the traffic lights in the target traffic area are adaptively controlled based on the traffic light control strategy parameters.

[0007] This application provides a traffic light intelligent control system, including:

[0008] A sensor group deployment module, the sensor group deployment module is used to deploy a sensor group on the traffic lights in the target traffic area, and collect traffic operation data information and traffic environment data information through the sensor group; a traffic adaptive analysis model acquisition module, the traffic adaptive analysis model acquisition module is used to obtain a traffic adaptive analysis model through a traffic dispatching platform, and the traffic adaptive analysis model is composed of a traffic anomaly detection sub-model and a traffic congestion prediction sub-model; a traffic condition analysis module, the traffic condition analysis module is used to perform traffic condition analysis on the traffic operation data information and traffic environment data information based on the traffic adaptive analysis model, and obtain traffic operation congestion parameters; a traffic emergency response mechanism startup module, the traffic emergency response mechanism startup module is used to start the traffic emergency response mechanism according to the traffic operation congestion parameters; a signal light control strategy analysis module, the signal light control strategy analysis module is used to perform signal light control strategy analysis on the traffic operation congestion parameters based on the traffic emergency response mechanism, and determine the traffic light control strategy parameters; an adaptive regulation module, the adaptive regulation module is used to perform adaptive regulation on the target traffic area signal lights based on the traffic light control strategy parameters.

[0009] The intelligent control method and system of traffic lights proposed in this application first deploys a sensor group at the lights in the target traffic area to collect traffic operation and environmental data. An adaptive analysis model including anomaly detection and congestion prediction is obtained through the traffic dispatch platform to analyze traffic conditions and obtain congestion parameters. The emergency response mechanism is initiated based on these parameters, and then the parameters of the traffic light control strategy are analyzed and determined, and finally the adaptive control of traffic lights is realized. Through real-time traffic data analysis and intelligent prediction, the real-time response capability of the traffic light system is enhanced, and the technical effect of being able to effectively predict and respond to traffic anomalies and congestion is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the accompanying drawings of the embodiment of the present invention will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the system according to the embodiment of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.

[0011] Figure 1 A flow chart of a traffic light intelligent control method provided in an embodiment of the present application;

[0012] Figure 2 A schematic diagram of the structure of the intelligent control system for traffic lights provided in an embodiment of the present application.

[0013] Explanation of the reference numerals: sensor group deployment module 10 , traffic adaptive analysis model acquisition module 20 , traffic condition analysis module 30 , traffic emergency response mechanism activation module 40 , traffic light control strategy analysis module 50 , adaptive control module 60 . DETAILED DESCRIPTION

[0014] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0015] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0016] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms "first\second" involved are merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "including" and "having" and any variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0017] The present application embodiment provides a method for intelligently controlling a traffic light. Figure 1 As shown, the method includes:

[0018] Step S100, deploy a sensor group on the traffic lights in the target traffic area, and collect and obtain traffic operation data information and traffic environment data information through the sensor group. Specifically, firstly, the sensor group deployment planning is carried out, and the geomagnetic, video, meteorological, acoustic and other sensor types are selected according to the characteristics of the target traffic area, and the sensor is reasonably arranged above or around each lane of the traffic light. Professionals install and debug to ensure that the position, angle, line are correct and the parameters are accurately calibrated. Then the traffic operation data is collected. The geomagnetic sensor senses the change of the vehicle magnetic field and records the entry and exit time, calculates the traffic volume of each lane, and the video camera tracks the speed of the vehicle feature point displacement according to the image processing algorithm combined with the frame rate and resolution. At the same time, the head and tail of the queued vehicles are monitored to calculate the queue length, and the data is transmitted to the traffic control center at a certain interval. In addition, traffic environment data is also collected. The meteorological sensor measures temperature, humidity, light, wind speed and direction, etc. The acoustic sensor collects traffic noise to indirectly reflect the flow and driving conditions. All environmental data are also transmitted to the control center in real time to comprehensively analyze the traffic conditions and environmental conditions, and provide information support for management decisions.

[0019] Step S200, obtain the traffic adaptive analysis model through the traffic dispatching platform, the traffic adaptive analysis model consists of a traffic anomaly detection sub-model and a traffic congestion prediction sub-model. Specifically, the traffic dispatching platform first integrates a large amount of historical traffic data from multiple sources such as the traffic management department database and the road sensor network, and pre-processes it by cleaning and denoising, laying the foundation for model acquisition. Then obtain the traffic anomaly detection sub-model, divide the normal and abnormal traffic condition sample sets from the pre-processed data by combining rules and machine learning, screen and mark them with indicators such as vehicle flow, and then train them with a feedforward neural network, optimize the parameters after multiple iterations, evaluate them with a test set, adjust the model structure if the performance threshold is not reached, etc., until the model is determined to meet the needs. Then obtain the traffic congestion prediction sub-model, divide the time window and mark it after stabilizing the time series data of indicators such as vehicle flow, train it with a recurrent neural network and its variants, adjust the parameters by minimizing the loss function, evaluate the error index with a validation set, and improve the model hyperparameters accordingly, and determine the model after meeting the standards. Finally, the two sub-models are combined to determine the input and output integration method to make them work together. The anomaly detection results help congestion prediction, and the congestion prediction results also optimize anomaly detection, thereby constructing a traffic adaptive analysis model to provide support for traffic light control and management decisions.

[0020] In a possible implementation, a traffic adaptive analysis model is obtained through a traffic dispatching platform, and the traffic adaptive analysis model is composed of a traffic anomaly detection sub-model and a traffic congestion prediction sub-model. Step S200 further includes step S210, and a historical traffic operation data set is obtained through the traffic dispatching platform. Specifically, the traffic dispatching platform has a powerful data collection and integration capability. It establishes a real-time communication connection with traffic sensors (such as geomagnetic sensors, cameras, microwave radars, etc.) distributed at various road nodes in the city, and continuously collects traffic operation data of different sections and intersections over a long period of time (covering various seasons, working days and holidays for several years), including detailed records of parameters such as vehicle flow, vehicle speed, vehicle occupancy rate, and headway. At the same time, the platform is also connected to the database of the traffic management department to obtain relevant data such as historical traffic accident information, road construction records, and the implementation of traffic control measures. These data from different sources and in different formats are cleaned, denoised, and formatted in a unified manner to remove invalid or erroneous data points and fill in missing values ​​to ensure that the data is complete and accurate. Ultimately, it is integrated into a comprehensive, systematic, and structured historical traffic operation data set, providing a rich and reliable data foundation for subsequent model training.

[0021] Step S220, a feedforward neural network is used to perform sample identification training on the historical traffic operation data set to generate a traffic anomaly detection sub-model. Specifically, from the historical traffic operation data set, rules are set based on professional knowledge and experience in the field of traffic, and combined with data analysis technology, feature vectors that can reflect the normal and abnormal state of traffic are extracted. For example, a sudden and substantial drop in traffic volume, a sharp drop in vehicle speed, and a rapid increase in vehicle occupancy are used as characteristic manifestations of abnormal traffic, and the corresponding data sets are marked as abnormal samples; and data sets with indicators such as traffic volume, vehicle speed, and vehicle occupancy rate that fluctuate stably within the normal range are marked as normal samples, thereby constructing a normal traffic condition sample set and an abnormal traffic condition sample set. A feedforward neural network architecture is constructed, which includes an input layer, one or more hidden layers, and an output layer. The input layer receives the feature vector of traffic data after preprocessing and feature extraction, and the hidden layer performs complex nonlinear transformation and feature learning on the input data through weighted connections between neurons, and the output layer outputs a classification result indicating whether the traffic condition is abnormal (usually 0 is used to indicate normal and 1 is used to indicate abnormal). During the training process, a large amount of labeled sample data is input into the feedforward neural network in batches. Based on the back-propagation algorithm, the weights and bias parameters of the network are continuously adjusted by minimizing the cross entropy loss function between the predicted results and the true labels, so that the network gradually learns a pattern that can accurately distinguish between normal and abnormal traffic conditions. At the same time, techniques such as Dropout regularization and early stopping are used to prevent overfitting and ensure the generalization ability of the model. After multiple iterations of training and optimization, when the model's accuracy, recall rate, F1 value and other evaluation indicators on an independent validation data set reach a satisfactory performance level, the traffic anomaly detection sub-model is finally determined and generated. The model can detect and warn abnormal situations in traffic operation in real time and accurately, providing strong support for traffic management departments to take timely response measures.

[0022] Step S230, using a recurrent neural network to perform sample identification training on the historical traffic operation data set to obtain a traffic congestion prediction sub-model. Specifically, for the historical traffic operation data set, firstly perform a stationarity test and preprocessing operation on the time series data of key traffic parameters such as vehicle flow and vehicle speed, such as using a difference method, seasonal decomposition and other methods to convert a non-stationary series into a stationary series, so as to better perform modeling and analysis. Then, according to the formation and development law of traffic congestion, the time series data is divided into different time windows (such as 5 minutes, 10 minutes, etc.), and the data in each time window is marked, for example, the time window in which the vehicle flow continues to rise, the vehicle speed continues to decline, and the vehicle occupancy rate gradually increases is marked as a congestion trend sample, and the time window in which the vehicle flow and vehicle speed are relatively stable is marked as a smooth trend sample, thereby constructing a traffic trend status sample set, wherein each sample not only contains the traffic data features in the current time window, but also contains the historical traffic data features of the previous time windows, so as to reflect the time dependence and trend of traffic congestion. The prediction model architecture is constructed using recurrent neural networks (RNNs) and their variants, such as long short-term memory networks (LSTMs) or gated recurrent units (GRUs). Due to its unique structural design, recurrent neural networks are able to process data with time series characteristics. By introducing recurrent connections between neurons, they can memorize and utilize historical information, thereby predicting future traffic congestion conditions. The constructed traffic trend condition samples are input into the recurrent neural network in sequence. The network learns the changing patterns and trends of traffic congestion based on the input historical traffic data characteristics, predicts the degree of traffic congestion (such as mild congestion, moderate congestion, severe congestion, etc.) or the probability of congestion in the future (such as the next 15 minutes, 30 minutes, etc.), and outputs the prediction results. During the training process, the weight and bias parameters of the network are adjusted by minimizing the mean square error loss function between the prediction results and the true labels to optimize the model performance. In order to prevent overfitting and improve the stability of the model, adaptive learning rate adjustment strategies (such as Adam optimization algorithm) and regularization techniques (such as L1 and L2 regularization) are adopted. The hyperparameters of the model (such as the number of hidden layers, the number of neurons, the time step, etc.) are adjusted by evaluating indicators such as mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) on an independent validation data set. After repeated training and verification, when the error indicators of the model on the validation set meet the accuracy requirements of practical applications, the traffic congestion prediction sub-model is finally determined and obtained. The model can predict future traffic congestion in advance according to the current traffic operation status and historical trends, and provide an important basis for traffic management departments to formulate reasonable traffic diversion and signal light control strategies, help alleviate traffic congestion, and improve traffic operation efficiency.

[0023] Step S240, the traffic anomaly detection sub-model and the traffic congestion prediction sub-model are combined to obtain the traffic adaptive analysis model. Specifically, when combining the two sub-models, the input-output relationship and the collaborative working mechanism between them are first clarified. The output result of the traffic anomaly detection sub-model (whether the traffic is abnormal) is used as an important input feature of the traffic congestion prediction sub-model. When a traffic anomaly (such as a traffic accident, temporary road control, etc.) is detected, the traffic congestion prediction sub-model can adjust its prediction strategy and parameters in time according to the type and location of the anomaly, and more accurately predict the development trend and impact range of traffic congestion under abnormal conditions. For example, if a traffic accident is detected on a certain road section, the traffic congestion prediction sub-model will focus on the traffic data changes of the road section and its surrounding areas, and combine the congestion evolution law under similar accident scenes in historical data to make a more accurate prediction of future congestion conditions. The output result of the traffic congestion prediction sub-model (traffic congestion degree and trend) is also input into the traffic anomaly detection sub-model as feedback information to help it further optimize the threshold and rules of anomaly detection and improve the accuracy and timeliness of anomaly detection. For example, if it is predicted that a certain area will be severely congested, the traffic anomaly detection sub-model will correspondingly increase its sensitivity to abnormal changes in traffic data in the area, and discover potential abnormal factors that may cause more serious congestion in advance, such as small abnormal fluctuations in vehicle speed, sudden concentration of traffic flow, etc., so as to issue abnormal warnings more timely. By establishing such a close information interaction and collaborative working mechanism, the traffic anomaly detection sub-model and the traffic congestion prediction sub-model are organically integrated to form a complete traffic adaptive analysis model, which can comprehensively consider various complex factors and dynamic changes in traffic operation, conduct real-time, comprehensive and accurate analysis and prediction of traffic conditions, provide efficient and reliable support for intelligent control of traffic lights and traffic management decisions, effectively improve the operating efficiency and intelligence level of urban transportation systems, better deal with various traffic problems such as traffic congestion and accidents, and ensure the smoothness and safety of urban transportation.

[0024] In a possible implementation, a feedforward neural network is used to perform sample identification training on the historical traffic operation data set to generate a traffic anomaly detection sub-model. Step S220 further includes step S221, which performs sample identification on the historical traffic operation data set to obtain a normal traffic condition sample set and an abnormal traffic condition sample set. Specifically, it is necessary to deeply analyze the historical traffic operation data set and formulate scientific sample identification rules based on traffic engineering expertise and long-term experience. For normal traffic condition samples, the traffic volume should be stably and reasonably distributed within the design capacity of the road, the vehicle speed should be in the normal driving speed range of the road section, the vehicle queue length is short or there is no queue, and the driving trajectory is stable and compliant. For example, during non-peak hours on urban main roads, the traffic volume is 3,000 to 4,000 vehicles per hour, the average speed is 50 to 60 kilometers per hour, and each lane is running smoothly without frequent starts and stops or backlogs. The data time period that meets these conditions will be marked as a normal sample and aggregated into a normal sample set. Abnormal traffic condition samples are identified by detecting significant changes and abnormal features in traffic data, such as when a lane is blocked for an instant due to a traffic accident or when traffic volume rises sharply after a large event, when vehicle speed drops to less than half of normal due to road construction, or when the length of vehicle queues continues to grow for multiple cycles due to traffic light failures. Combined with event records, the data in the corresponding time period is marked as abnormal samples to construct an abnormal sample set. The precise construction of these two sample sets provides key basic data for subsequent model training, ensuring that the model can grasp the essential difference between normal and abnormal traffic conditions.

[0025] Step S222, a feedforward neural network is used to perform detection training on the normal traffic condition sample set and the abnormal traffic condition sample set to obtain an initial traffic anomaly detection model. Specifically, a feedforward neural network architecture is constructed, including input, hidden and output layers. The number of input layer nodes depends on the selected traffic feature dimensions such as vehicle flow, vehicle speed, vehicle occupancy rate, vehicle queue length, etc. If these four features are selected, four nodes are set. The number of hidden layers (try 2 to 3 layers first) and the number of nodes per layer (such as 10 to 20) are determined through experimental optimization. The layers are connected by a weight matrix, and the neurons use the ReLU function to introduce nonlinear factors to enhance the expressiveness. The output layer has 1 node, which is activated by the Sigmoid function, and the output value of 0 to 1 represents the probability of traffic anomaly. The normal and abnormal traffic condition sample sets are divided into training sets and validation sets in a ratio of 80% and 20%, and the training set sample data is input into the network in batches after normalization and other preprocessing. During iterative training, the network forward propagates the prediction results based on the input sample features, compares them with the true labels of the samples (normal is 0, abnormal is 1), and calculates the loss value based on the difference using the binary cross entropy loss function. The network weights and biases are adjusted based on the gradient information through back propagation. After 100 to 500 iterations (depending on the data scale and convergence), the training is stopped when the performance indicators on the validation set meet the standards, and the initial traffic anomaly detection model is obtained.

[0026] Step S223, calculate the loss of the initial traffic anomaly detection model to obtain traffic loss data. Specifically, during the training process, the initial traffic anomaly detection model is continuously trained and optimized using the training subset. At the same time, after each training cycle (epoch), the model is verified using the validation subset, and the loss value of the model on the validation subset is calculated as part of the traffic loss data. The loss calculation uses the same loss function as in the training process (such as the binary cross entropy loss function). For each sample in the validation subset, the predicted output of the model and the true label are substituted into the loss function formula for calculation to obtain the loss value of the sample. Then, the loss values ​​of all samples in the validation subset are averaged to obtain the average loss value of the model on the validation subset in the current training cycle. The changes in these average loss values ​​over the training cycle are recorded to form a traffic loss data sequence. These loss data reflect the degree of deviation between the model's prediction results on the validation set and the true label. By analyzing the changing trend of the loss data, we can understand the training status and performance of the model. For example, if the loss data shows a continuous downward trend, it means that the model is constantly optimizing and improving; if the loss data tends to stabilize or starts to rise at a certain stage, it may mean that the model has problems such as overfitting or falling into a local optimal solution, and corresponding measures need to be taken to make adjustments.

[0027] Step S224, optimize the initial traffic anomaly detection model based on the traffic loss data to generate the traffic anomaly detection sub-model. Specifically, the initial traffic anomaly detection model is optimized based on the obtained traffic loss data. When the loss data presents an overfitting feature, that is, the loss value of the validation set decreases first and then increases, and the accuracy of the training set is high while the accuracy of the validation set is low, regularization techniques are used, such as L1 regularization (adding the L1 norm term of the weight vector to the loss function to make some weights approach zero to reduce the complexity of the model) or L2 regularization (adding the L2 norm term of the weight vector to make the weight distribution uniform and prevent it from being too large), and at the same time appropriately increase the diversity of training data, such as random transformation of the original data, expanding the number of abnormal samples, etc. to improve the generalization ability of the model. If the loss data shows that the model converges slowly or falls into a local optimum, such as the loss value decreases very slowly and multiple rounds of training fail to meet expectations, the optimization algorithm parameters can be adjusted, such as using the Adam optimization algorithm (dynamically adjusting the learning rate to promote convergence based on the first-order moment and second-order moment estimates of the model parameter gradient), or fine-tuning the network structure, such as increasing or decreasing the number of hidden layers, adjusting the number of hidden layer nodes, etc., and then retraining and verifying, observing the changes in loss data and continuously adjusting the optimization strategy until the model has an accuracy of more than 90%, a recall rate of more than 80%, and an F1 value of more than 0.85 on the verification set (depending on actual needs). At this time, the model is the optimized traffic anomaly detection sub-model, which can accurately and reliably detect traffic anomalies, provide timely and effective assistance for traffic management decisions, and ensure the safety and smoothness of the traffic system.

[0028] In a possible implementation, a recurrent neural network is used to perform sample identification training on the historical traffic operation data set to obtain a traffic congestion prediction sub-model, and step S230 further includes step S231, performing time series processing and sample identification on the historical traffic operation data set to obtain a traffic trend status sample set. Specifically, a stationarity test is performed on the historical traffic operation data set, and methods such as Augmented_Dickey-Fuller (ADF) test are used to determine whether the time series data of key traffic parameters such as vehicle flow, vehicle speed, and vehicle occupancy rate are stable. If the data is not stable, a suitable transformation method is used to make it stable, such as the difference method, to perform difference calculation on the data at adjacent time points, eliminate the trend and seasonal factors in the data, and make the mean and variance of the sequence no longer change with time. For data with obvious seasonal characteristics, seasonal difference or seasonal decomposition technology can also be used to decompose the data into trend terms, seasonal terms, and residual terms, and each term is analyzed and processed separately to better capture the inherent laws and characteristics of the data. According to the formation mechanism of traffic congestion and the rules in historical data, the processed time series data is divided into different time windows, such as 10 minutes as a window unit. Then, the sample identification is carried out by analyzing the changing trend and characteristics of the traffic data in each time window. If in a time window, the traffic volume shows a continuous upward trend, the speed gradually decreases, and the vehicle occupancy rate continues to increase, and combined with the traffic event information in the surrounding area during the time period (such as whether there are large-scale activities, road construction, etc.), it is judged that the traffic conditions in the window have a trend towards congestion, then the data in this window is marked as a congestion trend sample; on the contrary, if the traffic volume is relatively stable, the speed remains in the normal range, and the vehicle occupancy rate does not change much, it is marked as a smooth trend sample. In this way, the entire historical data set is traversed and marked, and finally a traffic trend condition sample set is formed, in which each sample contains not only the traffic data characteristics in the current time window, but also the historical traffic data characteristics of the previous time windows, so as to fully reflect the time dependence and trend of traffic congestion, and provide a representative and predictive data basis for subsequent model training.

[0029] Step S232, using a recurrent neural network to perform prediction training on the traffic trend condition sample set, and obtain an initial traffic congestion prediction model. Specifically, a recurrent neural network (RNN) architecture is constructed, including input, hidden and output layers, and the hidden layer has recurrent neurons with memory functions to retain the transmitted information in the time series. The number of input layer nodes depends on the selected traffic data feature dimensions and the number of historical time windows. For example, if the three features of vehicle flow, vehicle speed and vehicle occupancy rate are selected and the first three time windows are considered, the number of input layer nodes is 9. The number of hidden layers (initial 1 to 2 layers) and the number of nodes per layer (about 10 to 20) can be experimentally tuned, and the layers are connected by a weight matrix. The neuron activation function can select tanh or ReLU function to enhance the expression and learning ability. The number of output layer nodes depends on the prediction target. If the predicted congestion level is divided into 3 categories, 3 nodes are set to use softmax to activate the output of each probability value. If it is predicted whether it is congested, 1 node is set to use Sigmoid to activate the output of the congestion probability. Then, the traffic trend condition sample set is divided into a training set and a validation set at 70% and 30%, respectively. The training set sample data is preprocessed by normalization and then input into the RNN in sequence. At each time step, the network obtains the current hidden layer output according to the current input features and the output state of the hidden layer in the previous step through the recurrent neuron update mechanism. After completing the sequence input, the traffic congestion prediction result is obtained. Then, the prediction result is compared with the true label, and the loss value is calculated based on the difference using the cross entropy (multi-classification) or mean square error (continuous or binary classification) loss function. The weights and bias parameters of each layer of the network are adjusted according to the gradient information using the back propagation algorithm. After 200 to 100 iterations (depending on the data scale, network complexity and convergence), the training is stopped when the performance index of the validation set meets the standard, and the initial traffic congestion prediction model is obtained.

[0030] Step S233, based on the initial traffic congestion prediction model, loss calculation optimization is performed to obtain the traffic congestion prediction sub-model. Specifically, during the training of the initial traffic congestion prediction model, the training subset is continuously used for training, and the validation subset is used for validation after each training cycle. The loss value on the validation subset is calculated, and the same loss function as that used in training (such as cross entropy or mean square error function) is used. The predicted output and true label of each sample are substituted into the formula to calculate the loss value, and then the average is calculated to obtain the average loss value of the cycle. The changes of these values ​​over the cycle are recorded to form a loss data sequence. The trend can be analyzed to understand the training status and performance of the model. If it continues to decline, it is being optimized and improved. If it is stable or rising, there may be problems such as overfitting or falling into a local optimal solution that need to be adjusted. According to the loss calculation results, a targeted optimization strategy is adopted for the model. If there is overfitting, such as the validation set loss value first decreases and then increases, and the training set accuracy is high while the validation set is low, L1 or L2 regularization can be used to add the corresponding weight vector norm term to reduce complexity and balance weights, while increasing the diversity of training data, such as randomly transforming the original data, expanding the sample size of different scenarios, and enhancing generalization ability. If the convergence is slow or falls into a local optimum, such as the loss value decreases slowly and does not reach a satisfactory level, the optimization algorithm parameters can be adjusted, such as using the Adam algorithm to dynamically adjust the learning rate, or fine-tune the network structure, such as increasing or decreasing the number of hidden layers, adjusting the number of nodes, changing the type of cyclic units, etc., and then retraining and verifying, observing the change in loss data and continuously adjusting until the model has an accuracy of more than 85% on the validation set, a root mean square error of less than 0.2, and a mean absolute error of less than 0.15 (depending on actual needs), and the optimized traffic congestion prediction sub-model is obtained to provide decision support for traffic management departments, alleviate congestion, and improve the level of intelligence in traffic operation and management.

[0031] Step S300, based on the traffic adaptive analysis model, the traffic operation data information and the traffic environment data information are analyzed for traffic conditions, and traffic operation congestion parameters are obtained. Specifically, the collected traffic operation data (such as vehicle flow, vehicle speed, vehicle occupancy rate, etc.) and traffic environment data (such as weather, lighting, road construction conditions, etc.) are pre-processed, the error data are checked and corrected, the operation data is normalized, and key features are extracted from the operation data, and the environmental data is converted into a form that can be processed by the model. Then the processed data is input into the traffic adaptive analysis model, and its anomaly detection sub-model determines the traffic anomaly, determines the location, time and impact range of the abnormal event, and the congestion prediction sub-model combines the operation and environment data and historical trends to predict the congestion degree and development trend of each section and intersection within a certain period of time in the future. Finally, the traffic operation congestion parameters are obtained according to the model output, including clarifying the congested sections and intersections, using the average speed reduction ratio, vehicle queue length, vehicle flow saturation, etc. to quantify the congestion degree, and estimate the congestion duration, so as to provide key data support for subsequent traffic management measures to control congestion and regulate traffic flow.

[0032] Step S400, based on the traffic congestion parameters, start the traffic emergency response mechanism. Specifically, after obtaining the traffic congestion parameters, first evaluate and grade them, and divide the congestion conditions into different levels according to the preset standards based on the average speed reduction ratio, vehicle queue length, traffic flow saturation and other indicators. Then start the emergency response measures and operate in coordination, release congestion information through traffic broadcasts, smart APPs and electronic display screens, and provide detour suggestions for drivers; deploy police forces to key congestion points according to the congestion situation, and traffic police manually command at intersections to optimize the traffic order; implement the bus priority strategy, set up special lanes or priority signals for it, and ensure the green channel for emergency vehicles; the traffic management department also cooperates with road maintenance, meteorology and other departments, and coordinates to adjust the plan if the construction causes congestion, and prepares in advance in bad weather, so as to alleviate congestion, ensure smooth traffic and emergency response capabilities.

[0033] Step S500, based on the traffic emergency response mechanism, the traffic congestion parameters are analyzed for the traffic light control strategy, and the traffic light control strategy parameters are determined. Specifically, the traffic emergency response mechanism integrates the traffic congestion parameters and the road network information, presents the congestion distribution by means of GIS technology, and evaluates the causes of congestion in combination with the traffic environment data, so as to provide a scenario basis for the formulation of the traffic light strategy. Then, the solution space of the traffic light control strategy is constructed, and the value range and combination mode of each traffic light parameter are clarified, such as the reasonable range of the green light duration, phase sequence, cycle duration, etc. Then, a fitness function is designed with the average vehicle delay time, the number of stops, the utilization rate of road traffic capacity, the change rate of queue length, etc. as considerations, and a comprehensive index is formed by assigning weights to each factor to measure the pros and cons of the strategy. Finally, an optimization algorithm such as a genetic algorithm is used to search for the best in the solution space, and after iterative operations such as selection, crossover, and mutation, the optimal parameter combination is found, that is, the traffic light control strategy parameters are determined, including the duration and phase sequence of each time period of the traffic lights at each intersection, and the parameters will be adjusted according to the real-time changes of traffic to ensure smooth traffic.

[0034] In a possible implementation, a signal light control strategy analysis is performed on the traffic operation congestion parameters based on the traffic emergency response mechanism to determine the traffic light control strategy parameters. Step S500 further includes step S510, in which a signal light control strategy analysis is performed on the traffic operation congestion parameters based on the traffic emergency response mechanism to construct a signal light control strategy solution space. Specifically, after the traffic emergency response mechanism is activated, the traffic operation congestion parameters are first analyzed in depth. These parameters cover key information such as vehicle flow, vehicle speed, vehicle queue length, location and range of congested road sections, etc. Combined with the topological structure of the traffic network, including the number of intersections, lane distribution, road connection relationships, etc., a comprehensive analysis of the control strategy of the signal light is carried out. For each traffic light intersection that needs to be regulated, determine the range of adjustable control parameters. For example, the range of green light duration can be initially set based on historical traffic data and road capacity. Generally speaking, there may be a basic green light duration range during non-congested periods, and in congested conditions, this range can be appropriately extended or shortened. The red light duration is also adjusted accordingly to ensure that traffic flow in each direction can be reasonably allocated time. The phase setting of the signal light is also considered. For example, according to the proportion of traffic flow in each direction in different time periods, determine whether to use a simple two-phase control method (such as releasing east-west straight and left turns at the same time, and the same for north-south directions) or a more complex three-phase control method (adding a separate left turn phase) or even more phases. Through comprehensive consideration and range determination of these parameters, a solution space containing multiple possible combinations of signal light control strategies is constructed, providing a basis for subsequent optimization decisions.

[0035] Step S520, evaluate and analyze the signal light control target, and construct a signal light control fitness function. Specifically, the signal light control target is clarified, mainly focusing on relieving traffic congestion, improving road traffic efficiency, reducing vehicle waiting time and number of stops, etc. In order to quantitatively evaluate the degree of achievement of these targets, a signal light control fitness function is constructed. For the target of relieving traffic congestion, the reduction rate of the length of the vehicle queue in the congested section can be included in the fitness function as an important indicator, and the corresponding score weight is given by comparing the changes in the queue length before and after adopting different signal light control strategies; for improving road traffic efficiency, the total number of vehicles passing through the intersection per unit time can be used as a measurement indicator, that is, the total number of vehicles passing through the intersection in each direction within a certain time interval is counted, and the score is given according to the degree of closeness to the road design traffic capacity; in terms of reducing vehicle waiting time and number of stops, the average waiting time and average number of stops of vehicles at the intersection are calculated for evaluation. The shorter the waiting time and the fewer the number of stops, the higher the fitness function score. At the same time, taking into account the differences in importance between different objectives, the weight distribution of each indicator in the fitness function can be determined through expert experience or data analysis. For example, if the congestion problem is more prominent in the current traffic conditions, the weight of the vehicle queue length reduction rate can be appropriately increased, thereby constructing a fitness function that can comprehensively and accurately reflect the advantages and disadvantages of the traffic light control strategy, so as to facilitate the subsequent effective evaluation and comparison of various strategies in the solution space of the traffic light control strategy.

[0036] Step S530, using the traffic light control fitness function to perform control optimization analysis on the traffic light control strategy solution space, and determine the traffic light control strategy parameters. Specifically, an optimization algorithm (such as a genetic algorithm, a simulated annealing algorithm, a particle swarm optimization algorithm, etc.) is used to search and iteratively optimize in the constructed traffic light control strategy solution space. Taking the genetic algorithm as an example, firstly, a group of initial traffic light control strategy individuals are randomly generated, each individual represents a specific combination of traffic light control parameters (such as green light duration, red light duration, phase setting, etc.), and these individuals are substituted into the traffic light control fitness function to calculate their respective fitness values. Then, according to the size of the fitness value, a selection operation is adopted to select individuals with higher fitness as the parent generation, and a crossover operation is performed to simulate the gene recombination process in biological inheritance to generate new offspring individuals, and at the same time, a mutation operation is performed on some individuals with a certain probability to introduce new gene features to avoid the algorithm from falling into a local optimal solution. After multiple rounds of selection, crossover and mutation operations, the individuals in the population are continuously updated so that they gradually evolve in the direction of higher fitness. During the iteration process, the changes in the fitness function value and the convergence trend of the population are continuously monitored. When the preset stop condition is met (such as reaching the maximum number of iterations or the change in the fitness function value is less than a certain threshold), the iteration is stopped. At this time, the combination of signal light control strategy parameters corresponding to the individual with the highest fitness in the population is the determined traffic light control strategy parameters. These parameters will be transmitted to the actual traffic light control system to achieve precise control of the traffic lights to cope with the current traffic congestion and improve traffic operation efficiency and smoothness. With the dynamic changes in traffic conditions, the above process can be repeated again to adjust and optimize the signal light control strategy parameters in real time to ensure the effectiveness and adaptability of traffic management.

[0037] In a possible implementation, the traffic light control fitness function is used to perform control optimization analysis on the traffic light control strategy solution space to determine the traffic light control strategy parameters, and step S530 further includes step S531, selecting multiple traffic light control strategy parameters in the traffic light control strategy solution space. Specifically, the traffic light control strategy solution space is determined by the previous analysis of traffic congestion parameters and combined with traffic network information, and covers a set of multiple possible control parameter value ranges such as the green light duration range, red light duration range, traffic light phase combination mode, cycle duration range, etc. at each traffic light intersection. In this solution space, multiple combinations of traffic light control strategy parameters are selected using random sampling or a specific sampling strategy as the initial candidate solution set. For example, for a small traffic network with four intersections, each intersection has three possible phase settings, the green light duration is in the range of 20 to 60 seconds, the red light duration is adjusted accordingly, and the cycle duration is in the range of 80 to 120 seconds, 10 to 20 groups of different parameter combinations are selected by random generation. Each group of parameters clearly defines the specific values ​​of the green light start time, green light duration, red light duration and phase switching sequence of each intersection in a signal cycle, forming the initial multiple signal light control strategy parameters. These parameter combinations will enter the subsequent evaluation and optimization process to provide diverse starting points for finding the optimal signal light control strategy.

[0038] Step S532, using the signal light control fitness function to evaluate the fitness of the multiple signal light control strategy parameters, and obtain the fitness of multiple control strategy parameters. Specifically, the multiple signal light control strategy parameters selected in the previous step are respectively substituted into the pre-constructed signal light control fitness function for evaluation. The fitness function comprehensively considers multiple key indicators such as the degree of traffic congestion relief, the improvement of road traffic efficiency, and the reduction of average vehicle waiting time. For example, for the degree of traffic congestion relief, a quantitative evaluation is performed by comparing the changes in the length of vehicle queues on congested sections before and after the adoption of a certain set of signal light control strategy parameters. If the queue length is significantly reduced, a higher score is given on this indicator; for road traffic efficiency, the ratio of the total number of vehicles passing through each intersection within a certain period of time to the theoretical traffic capacity of the road is statistically calculated, and the higher the ratio, the higher the score; the average vehicle waiting time is calculated through the data recorded by the vehicle detection equipment at each intersection, and the shorter the waiting time, the higher the score. According to the weight distribution of these indicators in the fitness function (the weight can be determined by regression analysis of historical data or expert experience), the fitness value corresponding to each combination of traffic light control strategy parameters is calculated, thereby obtaining multiple control strategy parameter fitness. These fitness values ​​intuitively reflect the pros and cons of each group of traffic light control strategy parameters under the current traffic conditions, providing a clear comparison basis for subsequent iterative optimization.

[0039] Step S533, based on the fitness of the multiple control strategy parameters, iterative optimization is performed in the solution space of the traffic light control strategy to determine the traffic light control strategy parameters. Specifically, according to the obtained fitness of the multiple control strategy parameters, an optimization algorithm (such as selection, crossover, and mutation operations in the genetic algorithm, or the cooling search strategy in the simulated annealing algorithm, and the particle speed and position update mechanism in the particle swarm optimization algorithm, etc.) is used to iteratively optimize in the solution space of the traffic light control strategy. Taking the genetic algorithm as an example, firstly, a selection operation is performed according to the fitness value, and a combination of traffic light control strategy parameters with higher fitness is selected as the parent individual. The parent individual has a higher "good gene" ratio and is more likely to produce excellent offspring individuals. Then, a crossover operation is performed on the parent individuals to simulate the gene exchange process in biological inheritance, and some control parameters of different parent individuals are combined to generate new offspring individuals, thereby increasing the diversity and search range of the solution space. At the same time, a mutation operation is performed on some individuals with a certain probability to randomly change the values ​​of certain control parameters to prevent the algorithm from falling into a local optimal solution and explore other potential high-quality areas in the solution space. After each iteration, the traffic light control fitness function is reused to evaluate the fitness of the newly generated individuals (i.e., the new traffic light control strategy parameter combination), and the selection, crossover and mutation operations are continued according to the evaluation results, and this cycle repeats. As the number of iterations increases, the individuals in the solution space continue to evolve in the direction of higher fitness. When the preset stop conditions are met (such as reaching the maximum number of iterations, the change in the fitness function value is less than a certain threshold, or the optimal solution is no longer improved after multiple consecutive iterations), the traffic light control strategy parameter combination corresponding to the individual with the highest fitness in the solution space is the final traffic light control strategy parameter. These parameters will be applied to the actual traffic light control system to achieve precise control of traffic lights, effectively alleviate traffic congestion, and improve traffic operation efficiency. As the traffic conditions change in real time, the optimization process can be started again to dynamically adjust and optimize the traffic light control strategy parameters to maintain the good operation of the traffic system.

[0040] Step S600, adaptively control the traffic lights in the target traffic area based on the traffic light control strategy parameters. Specifically, firstly, the determined traffic light control strategy parameters are parsed, and connected with the target traffic area signal light control system to ensure that it can receive and adjust according to the parameters, and upgrade the software and hardware when necessary. Then, the traffic conditions in the target area are continuously monitored in real time, and the operation data such as vehicle flow, vehicle speed, queue length, and environmental data such as weather, construction, and special events are collected by geomagnetic sensors, cameras, etc., and transmitted to the management center control system, which determines the traffic condition category. Then, the adaptive adjustment mechanism is started according to the monitoring feedback data. When the vehicle flow increases, the queue is long, and the congestion increases, the intersection signal light parameters are automatically adjusted, such as increasing or decreasing the green light duration in each direction, and flexibly changing the phase sequence. In special circumstances (such as bad weather, sudden accidents), the cycle duration, phase time, and yellow light duration are also adjusted accordingly to guide vehicles to detour, realize dynamic matching and optimization of signal lights and traffic conditions, and improve traffic operation efficiency and emergency response capabilities.

[0041] In a possible implementation, the target traffic area signal lights are adaptively regulated based on the traffic light control strategy parameters, and step S600 further includes step S610, controlling and monitoring the target traffic area signal lights based on the traffic light control strategy parameters to obtain signal light control feedback parameters. Specifically, when the traffic light control system regulates the signal lights in the target traffic area according to the established traffic light control strategy parameters, a comprehensive monitoring mechanism is synchronously started. High-precision vehicle detection equipment, such as geomagnetic sensors, video cameras, etc., is set at each signal light intersection to collect key traffic data such as vehicle flow, vehicle speed, and vehicle queue length passing through the intersection in real time, and record the operating status information such as the phase switching time, green light duration, red light duration, and yellow light duration of the signal lights. These data are quickly transmitted back to the monitoring platform of the traffic management center using the communication network, and the collected massive data are integrated and refined through data processing and analysis technology to extract key indicators that can reflect the control effect of the signal lights as signal light control feedback parameters. For example, calculate the average delay time of vehicles, that is, count the total time all vehicles wait for traffic lights at intersections divided by the total number of vehicles; count the saturation of each intersection, measured by the ratio of actual traffic volume to the designed traffic capacity of the road; analyze the changing trend of vehicle queue length, and observe the increase or decrease of queue length before and after the implementation of the traffic light control strategy, etc. These traffic light control feedback parameters will provide objective and accurate data support for subsequent strategy optimization, helping traffic management personnel to deeply understand the implementation effect and existing problems of the current traffic light control strategy, so as to make timely adjustments and improvements.

[0042] Step S620, based on the signal light control feedback parameters, the traffic light control strategy parameters are optimized and adjusted, and the traffic light optimization control strategy parameters are obtained to adaptively optimize and control the target traffic area signal lights. Specifically, according to the signal light control feedback parameters obtained in the previous step, the intelligent optimization algorithm is used to optimize and adjust the existing traffic light control strategy parameters. First, an optimization objective function based on the signal light control feedback parameters is established. For example, the objectives of shortest average vehicle delay time, balanced intersection saturation and close to the optimal value, stable or reduced vehicle queue length, etc. are included in the function, and different weights are assigned to each objective according to actual traffic management needs. Then, optimization tools such as genetic algorithms, simulated annealing algorithms, and particle swarm optimization algorithms are used to search and iteratively optimize within the preset range of traffic light control strategy parameters. Taking the genetic algorithm as an example, the current traffic light control strategy parameters are encoded as individuals of the initial population, and the fitness value of each individual is calculated according to the optimization objective function. Through genetic operations such as selection, crossover, and mutation, new individuals are continuously generated to gradually approach the optimal solution. During the iteration process, the weight coefficients in the optimization objective function are continuously adjusted to balance the relationship between different optimization objectives. For example, during periods of severe traffic congestion, the weight of the vehicle queue length reduction objective is appropriately increased to prioritize the alleviation of congestion; while during periods of relatively stable traffic flow, more attention is paid to reducing the average vehicle delay time and balancing the intersection saturation. After multiple iterations of optimization, when the preset stop conditions are met (such as reaching the maximum number of iterations, the change in the optimization objective function value is less than a certain threshold, etc.), the traffic light control strategy parameters corresponding to the optimal individual are the traffic light optimization control strategy parameters. These optimized parameters are applied to the signal light control system of the target traffic area to achieve adaptive optimization and regulation of the signal light, so that the control of the signal light can be dynamically adjusted according to the changes in real-time traffic conditions, continuously improve traffic operation efficiency, reduce vehicle waiting time and congestion, and improve the service level and operation stability of the entire traffic network. As the traffic conditions continue to evolve, the above optimization and adjustment process is repeated to ensure that the signal light control strategy is always in the best state.

[0043] The embodiment of the present application adopts the method of deploying a sensor group at the traffic lights in the target traffic area to collect traffic operation and environmental data. The traffic dispatching platform obtains an adaptive analysis model including anomaly detection and congestion prediction, analyzes traffic conditions and obtains congestion parameters. The emergency response mechanism is activated according to these parameters, and then the signal light control strategy parameters are analyzed and determined, and finally the adaptive control of traffic lights is realized, achieving the technical effect of enhancing the real-time response capability of the traffic light system through real-time traffic data analysis and intelligent prediction, and being able to effectively predict and respond to traffic anomalies and congestion.

[0044] In the above, refer to Figure 1 The intelligent control method of traffic lights according to the embodiment of the present invention is described in detail. Figure 2 The intelligent control system of traffic signal lights according to an embodiment of the present invention is described.

[0045] The intelligent control system for traffic lights according to the embodiment of the present invention is used to solve the technical problems of insufficient real-time response capability and inability to effectively predict traffic anomalies and congestion in the existing traffic light control system. Through real-time traffic data analysis and intelligent prediction, the real-time response capability of the traffic light system is enhanced, and the technical effect of being able to effectively predict and respond to traffic anomalies and congestion is achieved. The intelligent control system for traffic lights includes: a sensor group deployment module 10, a traffic adaptive analysis model acquisition module 20, a traffic condition analysis module 30, a traffic emergency response mechanism startup module 40, a signal light control strategy analysis module 50, and an adaptive control module 60.

[0046] The sensor group deployment module 10 is used to deploy sensor groups on traffic lights in a target traffic area, and to collect and acquire traffic operation data information and traffic environment data information through the sensor groups.

[0047] The traffic adaptive analysis model acquisition module 20 is used to acquire the traffic adaptive analysis model through the traffic dispatch platform. The traffic adaptive analysis model consists of a traffic anomaly detection sub-model and a traffic congestion prediction sub-model.

[0048] The traffic condition analysis module 30 is used to perform traffic condition analysis on the traffic operation data information and the traffic environment data information based on the traffic adaptive analysis model to obtain traffic operation congestion parameters.

[0049] The traffic emergency response mechanism activation module 40 is used to activate the traffic emergency response mechanism according to the traffic congestion parameters.

[0050] The traffic light control strategy analysis module 50 is used to perform a traffic light control strategy analysis on the traffic congestion parameters based on the traffic emergency response mechanism to determine traffic light control strategy parameters.

[0051] The adaptive control module 60 is used to adaptively control the traffic light in the target traffic area based on the traffic light control strategy parameters.

[0052] The specific configuration of the traffic adaptive analysis model acquisition module 20 will be described in detail below. As described above, the traffic adaptive analysis model is acquired through the traffic dispatching platform, and the traffic adaptive analysis model is composed of a traffic anomaly detection sub-model and a traffic congestion prediction sub-model. The traffic adaptive analysis model acquisition module 20 further includes: a historical traffic operation data set acquisition unit, which is used to acquire a historical traffic operation data set through the traffic dispatching platform; a sample identification training unit, which is used to perform sample identification training on the historical traffic operation data set using a feedforward neural network to generate a traffic anomaly detection sub-model; a traffic congestion prediction sub-model acquisition unit, which is used to perform sample identification training on the historical traffic operation data set using a recurrent neural network to obtain a traffic congestion prediction sub-model; and a traffic adaptive analysis model acquisition unit, which is used to combine the traffic anomaly detection sub-model and the traffic congestion prediction sub-model to acquire the traffic adaptive analysis model.

[0053] Among them, a feedforward neural network is used to perform sample identification training on the historical traffic operation data set to generate a traffic anomaly detection sub-model, and the sample identification training unit further includes: a traffic condition sample set acquisition sub-unit, the traffic condition sample set acquisition sub-unit is used to perform sample identification on the historical traffic operation data set to obtain a normal traffic condition sample set and an abnormal traffic condition sample set; a detection training sub-unit, the detection training sub-unit is used to perform detection training on the normal traffic condition sample set and the abnormal traffic condition sample set using a feedforward neural network to obtain an initial traffic anomaly detection model; a loss calculation sub-unit, the loss calculation sub-unit is used to perform loss calculation on the initial traffic anomaly detection model to obtain traffic loss data; a model optimization sub-unit, the model optimization sub-unit is used to optimize the initial traffic anomaly detection model based on the traffic loss data to generate the traffic anomaly detection sub-model.

[0054] Among them, a recurrent neural network is used to perform sample identification training on the historical traffic operation status data set to obtain a traffic congestion prediction sub-model, and the traffic congestion prediction sub-model acquisition unit further includes: a traffic trend status sample set acquisition sub-unit, the traffic trend status sample set acquisition sub-unit is used to perform time series processing and sample identification on the historical traffic operation status data set to obtain a traffic trend status sample set; an initial traffic congestion prediction model acquisition sub-unit, the initial traffic congestion prediction model acquisition sub-unit is used to perform prediction training on the traffic trend status sample set using a recurrent neural network to obtain an initial traffic congestion prediction model; a loss calculation optimization sub-unit, the loss calculation optimization sub-unit is used to perform loss calculation optimization based on the initial traffic congestion prediction model to obtain the traffic congestion prediction sub-model.

[0055] The specific configuration of the signal light control strategy analysis module 50 will be described in detail below. As described above, based on the traffic emergency response mechanism, the signal light control strategy analysis is performed on the traffic operation congestion parameters to determine the traffic signal light control strategy parameters. The signal light control strategy analysis module 50 further includes: a signal light control strategy solution space construction unit, the signal light control strategy solution space construction unit is used to perform signal light control strategy analysis on the traffic operation congestion parameters based on the traffic emergency response mechanism to construct a signal light control strategy solution space; a signal light control fitness function construction unit, the signal light control fitness function construction unit is used to evaluate and analyze the signal light control target and construct a signal light control fitness function; a control optimization analysis unit, the control optimization analysis unit is used to perform control optimization analysis on the signal light control strategy solution space using the signal light control fitness function to determine the traffic light control strategy parameters.

[0056] Among them, the traffic light control fitness function is used to perform control optimization analysis on the traffic light control strategy solution space to determine the traffic light control strategy parameters, and the control optimization analysis unit further includes: a strategy parameter subunit, the strategy parameter subunit is used to select multiple traffic light control strategy parameters in the traffic light control strategy solution space; a fitness evaluation subunit, the fitness evaluation subunit is used to perform fitness evaluation on the multiple traffic light control strategy parameters using the traffic light control fitness function to obtain multiple control strategy parameter fitnesses; an iterative optimization subunit, the iterative optimization subunit is used to perform iterative optimization in the traffic light control strategy solution space based on the multiple control strategy parameter fitnesses to determine the traffic light control strategy parameters.

[0057] The specific configuration of the adaptive control module 60 will be described in detail below. As described above, the adaptive control module 60 further includes: a control monitoring unit, which is used to control and monitor the target traffic area lights based on the traffic light control strategy parameters to obtain the light control feedback parameters; and an adaptive optimization control unit, which is used to optimize and adjust the traffic light control strategy parameters based on the light control feedback parameters to obtain the traffic light optimization control strategy parameters to adaptively optimize the target traffic area lights.

[0058] The intelligent control system for traffic lights provided in the embodiments of the present invention can execute the intelligent control method for traffic lights provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0059] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0060] The above specific implementation manner does not constitute a limitation to the protection scope of the present application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application. In some cases, the actions or steps recorded in the present application can be performed in an order different from that in the embodiment and can still achieve the desired results. In addition, the process depicted in the accompanying drawings does not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A traffic light intelligent control method, characterized in that: The method comprises: Deploy a sensor group on the traffic lights in the target traffic area, and collect and obtain traffic operation data information and traffic environment data information through the sensor group; Acquire a traffic adaptive analysis model through a traffic dispatch platform, wherein the traffic adaptive analysis model is composed of a traffic anomaly detection sub-model and a traffic congestion prediction sub-model; Performing traffic condition analysis on the traffic operation data information and the traffic environment data information based on the traffic adaptive analysis model to obtain traffic operation congestion parameters; Initiate a traffic emergency response mechanism based on the traffic congestion parameters; Performing a traffic light control strategy analysis on the traffic congestion parameters based on the traffic emergency response mechanism to determine traffic light control strategy parameters; The traffic light in the target traffic area is adaptively controlled based on the traffic light control strategy parameters.

2. The intelligent control method for traffic lights according to claim 1, characterized in that: The method of obtaining the traffic adaptive analysis model through the traffic dispatching platform includes: Acquire a historical traffic operation status data set through the traffic dispatch platform; Using a feedforward neural network to perform sample identification training on the historical traffic operation data set to generate a traffic anomaly detection sub-model; Using a recurrent neural network to perform sample identification training on the historical traffic operation data set to obtain a traffic congestion prediction sub-model; The traffic anomaly detection sub-model and the traffic congestion prediction sub-model are combined to obtain the traffic adaptive analysis model.

3. The intelligent control method for traffic lights according to claim 2, characterized in that: The generating of the traffic anomaly detection sub-model comprises: Performing sample identification on the historical traffic operation condition data set to obtain a normal traffic condition sample set and an abnormal traffic condition sample set; Using a feedforward neural network to perform detection training on the normal traffic condition sample set and the abnormal traffic condition sample set to obtain an initial traffic anomaly detection model; Performing loss calculation on the initial traffic anomaly detection model to obtain traffic loss data; The initial traffic anomaly detection model is optimized based on the traffic loss data to generate the traffic anomaly detection sub-model.

4. The intelligent control method for traffic lights according to claim 2, characterized in that: The obtaining of the traffic congestion prediction sub-model comprises: Performing time series processing and sample identification on the historical traffic operation status data set to obtain a traffic trend status sample set; Using a recurrent neural network to perform prediction training on the traffic trend condition sample set to obtain an initial traffic congestion prediction model; The loss calculation is optimized based on the initial traffic congestion prediction model to obtain the traffic congestion prediction sub-model.

5. The intelligent control method for traffic lights according to claim 1, characterized in that: The determining of traffic light control strategy parameters includes: Based on the traffic emergency response mechanism, a signal light control strategy analysis is performed on the traffic congestion parameters to construct a signal light control strategy solution space; Evaluate and analyze the traffic light control objectives and construct the traffic light control fitness function; The traffic light control fitness function is used to perform control optimization analysis on the traffic light control strategy solution space to determine the traffic light control strategy parameters.

6. The intelligent control method for traffic lights according to claim 5, characterized in that: The determining of the traffic light control strategy parameters comprises: Selecting a plurality of signal light control strategy parameters in the signal light control strategy solution space; Using the signal light control fitness function to perform fitness evaluation on the plurality of signal light control strategy parameters to obtain fitness of the plurality of control strategy parameters; The traffic light control strategy parameters are determined by iteratively searching for the optimal solution in the traffic light control strategy solution space based on the fitness of the multiple control strategy parameters.

7. The intelligent control method for traffic lights according to claim 6, characterized in that: The adaptively regulating the target traffic area traffic light based on the traffic light control strategy parameters includes: Control and monitor the traffic lights in the target traffic area based on the traffic light control strategy parameters to obtain traffic light control feedback parameters; The traffic light control strategy parameters are optimized and adjusted based on the traffic light control feedback parameters, and the traffic light optimization control strategy parameters are obtained to perform adaptive optimization control on the traffic lights in the target traffic area.

8. Traffic light intelligent control system, characterized in that: The system is used to implement the intelligent control method for traffic lights according to any one of claims 1 to 7, and the system comprises: A sensor group deployment module, the sensor group deployment module is used to deploy a sensor group on a target traffic area signal light, and collect traffic operation data information and traffic environment data information through the sensor group; A traffic adaptive analysis model acquisition module, wherein the traffic adaptive analysis model acquisition module is used to acquire a traffic adaptive analysis model through a traffic dispatching platform, wherein the traffic adaptive analysis model is composed of a traffic anomaly detection sub-model and a traffic congestion prediction sub-model; A traffic condition analysis module, the traffic condition analysis module is used to perform traffic condition analysis on the traffic operation data information and the traffic environment data information based on the traffic adaptive analysis model to obtain traffic operation congestion parameters; A traffic emergency response mechanism activation module, the traffic emergency response mechanism activation module is used to activate the traffic emergency response mechanism according to the traffic operation congestion parameter; A traffic light control strategy analysis module, the traffic light control strategy analysis module is used to perform a traffic light control strategy analysis on the traffic congestion parameters based on the traffic emergency response mechanism to determine traffic light control strategy parameters; An adaptive control module is used to adaptively control the traffic light in the target traffic area based on the traffic light control strategy parameters.

Images