A method and system for intelligent real-time processing and analysis of big data

Abstract

The present application relates to the field of intelligent traffic management, and discloses a method and system for intelligent real-time processing and analysis of big data, which includes the following steps: obtaining structured, semi-structured and unstructured multimodal traffic data; preprocessing the data and extracting feature vectors; fusing feature vectors of different modes through a self-attention mechanism to generate a unified fusion feature; using a long short-term memory network to perform time series modeling on the fusion feature and predict traffic flow; based on the traffic flow prediction results, using a deep reinforcement learning algorithm to dynamically optimize the duration and phase of traffic lights, and combining a path recommendation algorithm to provide the optimal path for vehicles; applying the optimized signal control and path recommendation to a real-time traffic management system. The present invention can realize the intelligent fusion of multimodal data, accurate traffic flow prediction, and intelligent traffic signal control and path recommendation, thereby improving the real-time and efficiency of traffic management.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent traffic management, and in particular to a method and system for intelligent real-time processing and analysis of big data. Background Art

[0002] With the acceleration of urbanization, urban traffic problems are becoming increasingly prominent. The increase in large-scale vehicles has made the urban traffic network more complex, leading to increasingly serious problems such as traffic congestion, frequent accidents, and low travel efficiency. In order to effectively manage traffic flow, reduce congestion, and improve road traffic efficiency, intelligent traffic management systems have gradually become a research hotspot. Such systems rely on the collection, processing, and analysis of traffic data, and improve the level of intelligent traffic management by optimizing traffic signal control and route recommendation.

[0003] Current traffic management systems mainly rely on single or limited data sources, such as traffic flow sensors, cameras, etc. to obtain real-time traffic data. These systems usually use traditional rules or static methods based on historical data for signal light control and route recommendation. However, this approach has many limitations. First, due to the singleness and incompleteness of the data source, it is difficult for the existing system to accurately and comprehensively grasp the traffic conditions, especially when facing emergencies (such as accidents or bad weather), the response capability is insufficient, resulting in the inability to make timely and effective decisions. Secondly, most traditional signal light control and route recommendation schemes are based on static analysis of historical data, lacking the ability to accurately predict and dynamically adjust real-time changes in traffic flow, and therefore are inefficient when dealing with complex traffic environments.

[0004] In addition, when processing large-scale data (such as the Chinese patent application with publication number CN118397831A), existing traffic management systems often have difficulty in achieving real-time response due to limitations in computing resources and processing capabilities. Especially when the urban traffic network is large and the amount of data is growing rapidly, the system is difficult to ensure real-time analysis and optimization, resulting in delayed signal control and inaccurate route recommendations, further exacerbating traffic congestion. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a method and system for intelligent real-time processing and analysis of big data, which realizes accurate prediction of traffic flow and real-time signal control through multimodal data fusion and intelligent decision optimization, so as to improve the efficiency of urban traffic management and the ability to cope with complex traffic conditions.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for intelligent real-time processing and analysis of big data, comprising the following steps:

[0007] S1, based on traffic sensors, vehicle GPS, Internet devices and traffic cameras, real-time acquisition of multimodal traffic data, wherein the multimodal traffic data includes structured data, semi-structured data and unstructured data;

[0008] S2, preprocessing the multimodal traffic data to extract feature vectors;

[0009] S3, the extracted feature vectors are fused through the self-attention mechanism to generate a unified fused feature vector;

[0010] S4, using the time series model to model the unified fusion feature vector and generate real-time traffic flow prediction results;

[0011] S5, based on the traffic flow prediction result, using an intelligent decision-making algorithm to optimize traffic signal control and recommend routes, and obtain traffic signal control and route recommendation results;

[0012] S6, applying the traffic signal control and route recommendation results to real-time traffic management.

[0013] Preferably, the structured data includes traffic flow data, vehicle speed data and lane occupancy data collected by traffic sensors; the semi-structured data includes vehicle GPS data and user feedback data collected by vehicle GPS and Internet devices; the unstructured data includes real-time video data collected by traffic cameras and social media text data collected in real time by Internet devices.

[0014] Preferably, the step of preprocessing the multimodal traffic data comprises:

[0015] Use Gaussian mixture models for anomaly detection and data cleaning on structured data;

[0016] According to the appearance, size, shape and speed of objects captured by traffic cameras, real-time video data is processed in real time through convolutional neural networks to extract vehicle information and pedestrian information in traffic scenes;

[0017] Perform sentiment analysis and keyword extraction on social media text data based on natural language processing models.

[0018] Preferably, the step of fusing the extracted feature vectors through a self-attention mechanism to generate a unified fusion feature comprises:

[0019] Extract feature vectors of structured data, real-time video data, and social media text data respectively, and add tags, wherein the structured data is marked as feature vectors , the real-time video data is a feature vector , the social media text data is marked as a feature vector ;

[0020] Based on the feature vector , eigenvector and the eigenvector Generate query vector , key vector Sum value vector ,in ;

[0021] The feature vector is calculated by the following attention weight calculation formula , eigenvector and the eigenvector The attention weights between:

[0022]

[0023] in, is the key vector The dimension of , is the feature vector , eigenvector and the eigenvector The attention weight between

[0024] Based on the feature fusion calculation formula, the feature vector is fused by weighted summation. , eigenvector and the eigenvector Perform feature fusion to generate a unified fusion feature vector , the feature fusion calculation formula is:

[0025]

[0026] in, is a unified fusion feature vector, is the value vector of the eigenvector, is the feature vector , eigenvector and the eigenvector The attention weight between .

[0027] Preferably, the time series model is a long short-term memory network, which is used to capture the time dependency of traffic flow data. The calculation formula of the time series model is:

[0028]

[0029] in, is the time step The hidden state of is the time step The hidden state of is the time step The unified fusion feature vector.

[0030] Preferably, in step S5, the step of using an intelligent decision-making algorithm to optimize traffic signal control and recommend routes based on the traffic flow prediction result includes:

[0031] S51, define basic elements of deep reinforcement learning, wherein the basic elements include state space , Action Space And the reward function ;

[0032] The state space : Traffic condition parameters including traffic flow, lane occupancy rate, and vehicle queue length at the current traffic intersection;

[0033] The action space : Including the phase and duration of traffic lights, as well as the switching duration of green, red and yellow lights;

[0034] The reward function for:

[0035]

[0036] in, is the weight factor, For the The waiting time of a vehicle, is the vehicle's speed, is the number of vehicles observed simultaneously in the current traffic signal control area, that is, the total number of all vehicles queuing or passing in the signal light control area under the current traffic conditions;

[0037] S52, based on the basic elements, obtaining the duration and phase of the traffic light, and optimizing the duration and phase of the traffic light using a deep Q network algorithm:

[0038] S521, at each time step , according to the time step The state space , select the time step Action Space Adjustment strategies as signal lights;

[0039] S522, after executing the action, obtain the time step The reward function , and observe the new time step The state space , update the value of depth Q according to the Bellman equation, which is:

[0040]

[0041] in, is the learning rate, is the discount factor, is the time step The state space and action space The value of depth Q;

[0042] S523, during the continuous training process, by continuously optimizing the value of depth Q, the optimal traffic light duration and phase strategy is gradually learned to reduce the average waiting time of vehicles and improve traffic efficiency.

[0043] Preferably, the route recommendation adopts a federated learning method to collaboratively train a route recommendation model through vehicle data from multiple regions and generate personalized route recommendations.

[0044] Preferably, the method adopts an architecture that combines edge computing with cloud computing, where the edge nodes are responsible for preprocessing local data and the cloud nodes are responsible for complex calculations and model training of global traffic data.

[0045] The present invention also provides a system for intelligent real-time processing and analysis of big data, comprising:

[0046] Data collection module, used to obtain structured data, semi-structured data and unstructured data;

[0047] A data processing module is used to clean and extract features from the acquired multimodal data;

[0048] Data fusion module, used to fuse multimodal data through self-attention mechanism;

[0049] Time series prediction module, used for time series modeling and prediction of traffic flow based on fusion features;

[0050] Decision-making module, used for intelligent traffic signal control and route recommendation based on traffic flow prediction.

[0051] Preferably, the system further comprises:

[0052] The resource scheduling module is used to dynamically allocate resources between edge nodes and cloud nodes to ensure optimal efficiency of real-time processing.

[0053] The present invention provides a method and system for intelligent real-time processing and analysis of big data, which has the following beneficial effects:

[0054] 1. The present invention dynamically weights and fuses the features of structured, semi-structured, and unstructured data through the self-attention mechanism, and can flexibly adjust the importance of different data sources. This mechanism ensures that in different traffic scenarios, the system can prioritize the most representative data according to actual needs, make full use of multimodal data sources, and improve the accuracy of analysis and the rationality of decision-making.

[0055] 2. The present invention uses the long short-term memory network (LSTM) to perform time series modeling on the fused traffic data, which can capture the long-term and short-term dependencies in traffic flow. Compared with traditional static prediction methods, LSTM performs better in dealing with complex and dynamic traffic environments, and can generate accurate traffic flow prediction results, ensuring that traffic light control and route recommendations are based on reliable prediction data.

[0056] 3. The present invention uses a deep reinforcement learning (DRL) algorithm to dynamically optimize the duration and phase of traffic lights, with the goal of maximizing traffic efficiency and minimizing vehicle waiting time. At the same time, combined with a path recommendation algorithm, it can provide vehicles with personalized driving routes and guide them to avoid congested sections, thereby optimizing global traffic flow. This dual optimization significantly improves road traffic efficiency and reduces the risk of traffic congestion.

[0057] 4. The present invention can adaptively adjust the traffic light configuration and route recommendation strategy through continuous monitoring and feedback mechanism. The system continuously performs dynamic optimization based on real-time traffic data and vehicle feedback to ensure timely response and adjustment when traffic conditions change, thereby achieving continuous optimization of traffic management. This mechanism greatly improves the flexibility of the system and its ability to cope with complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0059] Figure 2 This is one of the system architecture diagrams of the present invention;

[0060] Figure 3 This is the second schematic diagram of the system architecture of the present invention.

[0061] Among them, 100, data acquisition module; 200, data processing module; 300, data fusion module; 400, time series prediction module; 500, decision module; 600, resource scheduling module. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] Please see attached Figure 1 -Attached Figure 2 The embodiment of the present invention provides a method for intelligent real-time processing and analysis of big data, which is mainly used in traffic management systems. By acquiring, processing and analyzing multimodal traffic data, traffic flow can be predicted in real time and traffic signal control and route recommendation can be optimized. The specific implementation of the method of the present invention is described in detail below.

[0064] S1. Obtaining multimodal traffic data

[0065] The effective operation of the traffic management system depends on the accurate collection and processing of multi-source and multi-modal data. In order to fully understand the traffic conditions, the method of the present invention integrates and analyzes structured, semi-structured and unstructured data sources. These data provide basic support for subsequent feature extraction, fusion and prediction. Specifically, multimodal data can capture the dynamic changes of the traffic system from different angles, among which structured data can reflect the basic characteristics of traffic flow, and semi-structured data and unstructured data provide supplementary information such as geography, vision and user feedback.

[0066] In this embodiment, the multimodal traffic data obtained includes the following types of data:

[0067] Structured data:

[0068] In this embodiment, structured data is collected by sensors installed on traffic sections, including traffic flow, vehicle speed, lane occupancy, etc. This type of data has a fixed format and a high sampling frequency. For example, traffic flow data can be expressed as the number of vehicles passing a certain detection point per unit time, vehicle speed data records the instantaneous speed of vehicles passing the detection point, and lane occupancy indicates the proportion of time a lane at the detection point is occupied.

[0069] Semi-structured data:

[0070] In this embodiment, semi-structured data mainly comes from the vehicle's GPS trajectory data and user feedback data from mobile applications. GPS data contains information such as the vehicle's geographic location and timestamp. This type of data is not completely standardized, but contains rich spatial location information. GPS data is collected as a set of spatiotemporal points. ,in and Represents latitude and longitude respectively, The user feedback data may contain information about traffic jams, accidents, or weather conditions, and its content is delivered in the form of unformatted text or tags.

[0071] Unstructured Data:

[0072] In this embodiment, the unstructured data mainly includes real-time video streams from traffic cameras and social media data. The video stream data comes from the cameras in the urban traffic monitoring system and provides visual information such as traffic volume, vehicle type, road conditions, etc. Social media data reflects real-time road conditions and emergencies through text or pictures uploaded by users.

[0073] Each frame of video data is represented as a matrix , where each element represents a pixel value, and the video stream can be viewed as a time series of frames Social media data is pre-processed using natural language processing technology to extract valuable keywords or semantic information.

[0074] In this embodiment, the method for obtaining the above multimodal data includes:

[0075] Data collection from sensor networks:

[0076] Structured data is automatically collected through sensors deployed on various traffic sections. The sensors monitor traffic flow, vehicle speed and lane occupancy in real time. The sensor equipment sends the collected traffic flow data to the system's data processing module at a fixed frequency to ensure the continuity and accuracy of traffic data.

[0077] GPS and user feedback data collection:

[0078] Semi-structured data is obtained through vehicle GPS devices and mobile applications. During driving, the vehicle will periodically upload the current location and timestamp through the GPS device, and these data will be stored and used for subsequent trajectory analysis. In the mobile application, users can upload traffic feedback, such as accidents, road congestion and other information. The system will receive these feedbacks in a semi-structured manner and combine them with other traffic data for analysis.

[0079] Collection of video data and social media data:

[0080] Unstructured data is collected in real time through video streams from urban traffic monitoring cameras. Video frames are sent to the data center in the form of continuous images, and the system needs to process these videos in real time to extract vehicle information and traffic events. Social media data is obtained through open API interfaces. The system extracts text or image data related to traffic conditions from social platforms and stores these data for analysis.

[0081] Through the above data acquisition method, the present invention can obtain comprehensive and diverse traffic data, providing a basis for subsequent data processing and analysis.

[0082] S2. Preprocess multimodal data and extract feature vectors

[0083] After acquiring multimodal data, the system needs to preprocess the data to ensure the uniformity and processability between different types of data. Preprocessing of multimodal data is crucial to improving the accuracy of subsequent data fusion and analysis. This step includes the process of anomaly detection, cleaning, feature extraction, and conversion of structured, semi-structured, and unstructured data into feature vectors. The generation of feature vectors is the basis for subsequent data fusion. They can represent the core features of each data type and are used for further model processing.

[0084] In this embodiment, the step of preprocessing the multimodal data and extracting the feature vector includes:

[0085] 1. Preprocessing of structured data:

[0086] In this embodiment, the structured data mainly includes traffic flow, vehicle speed, lane occupancy, etc. These data are subjected to anomaly detection and data cleaning through Gaussian Mixture Model (GMM). This method identifies possible abnormal data points by fitting multi-peak distribution.

[0087] The probability density function of the GMM model can be expressed as:

[0088]

[0089] in, For the The weights of a Gaussian distribution, and are the mean and covariance matrix, respectively. Outliers are removed by detecting points that deviate from the main distribution to ensure the validity and consistency of the data.

[0090] The cleaned structured data is further standardized and feature vectors are extracted ,This feature vector contains the standardized values ​​of various traffic parameters, which will be used for data fusion and time series analysis later.

[0091] 2. Preprocessing of semi-structured data:

[0092] In this embodiment, the semi-structured data mainly includes the vehicle's GPS data and user feedback data. GPS data is the geographic location information periodically sent by the vehicle during driving, and the format is usually ,in and Indicates the latitude and longitude of the vehicle, Indicates the sampling time.

[0093] In the preprocessing step, the GPS data is first smoothed by an interpolation algorithm to solve the missing data caused by network fluctuations. The interpolation algorithm fills in the missing location information in the middle by the distance and time relationship between two points:

[0094]

[0095]

[0096] in, , and , Respectively represent the longitude and latitude data of the previous and next moments, , For the corresponding time.

[0097] User feedback data is input as text and labels, and needs to be cleaned to remove noise information. Keywords are extracted through the natural language processing (NLP) model and further converted into vector form for subsequent analysis.

[0098] Finally, GPS data and smoothed user feedback data are extracted as feature vectors and , respectively represent the spatiotemporal distribution characteristics of location and user feedback.

[0099] 3. Preprocessing of real-time video data:

[0100] In this embodiment, real-time video data is processed by a convolutional neural network (CNN). First, the real-time video stream collected by the traffic camera is decomposed into continuous image frames. ,Each frame of the image is input into CNN as a matrix for object detection, and key traffic elements such as vehicles and pedestrians are extracted.

[0101] The process of CNN for object detection can be represented as a series of convolution operations:

[0102]

[0103] in, and are the convolution kernel and bias, is the activation function, represents the convolution operation, is the feature vector extracted by the convolutional layer.

[0104] The feature map processed by the convolution layer is pooled and extracted into a low-dimensional feature vector through the fully connected layer. , which represents the traffic information extracted from the current video frame.

[0105] 4. Preprocessing of text data:

[0106] In this example, the text data from social media is processed by a natural language processing (NLP) model to extract semantic information. First, the text is segmented and tagged, and then encoded by a pre-trained BERT model to convert it into a word vector. The output of the BERT model is a word embedding vector. ,This vector retains the semantic information in social media feedback such as events such as road congestion or accidents.

[0107] Text data processed by NLP is compressed into feature vectors ,This vector represents the semantic information of the current traffic event, and will be subsequently combined with other modal data in the data fusion phase.

[0108] In this embodiment, the multimodal data processed by the above steps are converted into feature vectors, which are:

[0109] Feature vectors for structured data ;

[0110] Feature vector of GPS trajectory and the feature vector of user feedback ;

[0111] Feature vector of video data ;

[0112] Feature vectors for social media text data .

[0113] These feature vectors will be used in subsequent data fusion steps and provide basic information for time series modeling and intelligent decision-making.

[0114] S3. The extracted feature vectors are fused through the self-attention mechanism to generate a unified fusion feature

[0115] After preprocessing and extracting the feature vectors of multimodal data, these feature vectors need to be effectively fused. Since the impact of different modal data on traffic conditions may vary, a single fusion strategy is difficult to fully capture the complexity of multimodal information. In order to solve this problem, the present invention adopts a self-attention mechanism to achieve efficient feature fusion by dynamically adjusting the weights of different data source features, so that the system can flexibly select the most important data source information according to the current traffic conditions. This step is the key to ensuring the accuracy of subsequent time series modeling and prediction.

[0116] In this embodiment, the step of fusing the extracted feature vectors through the self-attention mechanism includes:

[0117] Initialization of input feature vector:

[0118] In this embodiment, the multimodal data feature vector after processing in step S2 includes: the feature vector of the structured data , the feature vector of real-time video data , and the feature vector of social media text data These feature vectors represent the traffic information in structured, unstructured and semi-structured data respectively.

[0119] 2. Generation of query, key and value vectors:

[0120] The core of the self-attention mechanism is to assign appropriate weights to each data source to ensure that the most representative information is fully utilized. In this embodiment, a query vector is first generated for each feature vector , key vector , and value vector ,in The feature vectors represent structured data, video data, and text data respectively.

[0121] These vectors are generated as follows:

[0122]

[0123] in, , , are the weight matrices corresponding to queries, keys, and values, respectively. is the original feature vector.

[0124] 3. Calculation of attention weight:

[0125] In order to dynamically adjust the importance of each feature vector, in this embodiment, the dot product attention mechanism is used to calculate the correlation between features. is a query vector based on the feature vector and key vector The attention weight is calculated based on the similarity between them. The attention weight calculation formula is as follows:

[0126]

[0127] in, is the key vector The dimension of This process ensures that data from different modalities receive different attention weights according to their relevance to the current task.

[0128] 4. Weighted fusion:

[0129] After calculating the attention weight of each feature vector Afterwards, this embodiment performs a vectorization of the value of each feature vector Perform weighted summation to generate the final unified fusion feature vector The calculation formula is as follows:

[0130]

[0131] The fused feature vector combines the effective information in different modal data and dynamically adjusts the weights according to their relevance, ensuring that the system can adaptively select the most important feature information under different traffic conditions.

[0132] In this embodiment, a multi-head self-attention mechanism can also be introduced, that is, the above self-attention calculation process is repeated multiple times to generate multiple different attention weight sets. The output results of each head will eventually be spliced ​​and generated through a fully connected layer to generate richer fusion features. The use of this multi-head mechanism can improve the expressiveness of the model, allowing the system to focus on different dimensions and features of traffic data.

[0133] In this embodiment, the process of feature fusion through the self-attention mechanism can effectively solve the problem of information importance differences in multimodal data. The attention mechanism dynamically adjusts the weights of different features, allowing the system to more flexibly select and fuse effective information when facing complex traffic environments. This fusion feature provides a unified feature input for subsequent time series modeling and prediction, ensuring the effectiveness and robustness of the model.

[0134] S4. Use the time series model to model the fusion features and generate real-time predictions of traffic flow

[0135] After acquiring and fusing the multimodal feature vectors, these fused features need to be modeled in time series. Traffic flow has obvious time dependence, and the current traffic conditions are often closely related to previous traffic data. Therefore, the present invention uses time series models such as long short-term memory networks (LSTM) to capture this temporal correlation to generate accurate traffic flow predictions. This process ensures that the system can dynamically track traffic changes and predict future traffic trends based on historical data.

[0136] In this embodiment, the steps of modeling the fusion features using the time series model and generating a real-time prediction of traffic flow include:

[0137] 1. Input fusion feature vector:

[0138] In this embodiment, the time step from step S3 The unified fusion feature vector It is used as the input of the time series model. The vector integrates the key information of structured, semi-structured and unstructured data, and dynamically adjusts the weights of different modal features. Contains global traffic information at the current moment.

[0139] 2. Choice of timing modeling:

[0140] In this embodiment, a long short-term memory network (LSTM) is used as a time series model. The special structure of LSTM can effectively handle long-term dependencies and capture the changing pattern of traffic flow over time. LSTM has an input gate, a forget gate, and an output gate, which are used to determine how the information of the current time step is updated and passed to the next step.

[0141] 3. Calculation of LSTM unit:

[0142] LSTM network receives fusion features As a time step Input, update its hidden state and cell status The calculation formula of the timing model is:

[0143]

[0144] in, is the time step The hidden state of is the time step The unified fusion feature vector.

[0145] The specific calculation steps of LSTM are as follows:

[0146] Input Gate Determine the current input The importance of is calculated as:

[0147]

[0148] in, is the hidden state at the previous time step, is the weight matrix, is the bias term, is the activation function (Sigmoid).

[0149] Forget Gate It is used to decide which historical information needs to be retained or forgotten. The calculation formula is:

[0150]

[0151] in, and are the weight matrix and bias term of the forget gate respectively.

[0152] Cell state It is the core of LSTM and is used to store long-term information. The current cell state is updated through the forget gate and the input gate. The calculation formula is:

[0153]

[0154] in, is the cell state at the previous time step, and is the weight matrix and bias term updated for the cell state, is the hyperbolic tangent function.

[0155] Output Gate Determine the hidden state at the current time step How to update, the formula is:

[0156]

[0157] Updated hidden state for:

[0158]

[0159] in, and are the weight and bias of the output gate.

[0160] 4. Prediction output:

[0161] In this embodiment, through continuous time step operations, the hidden state of LSTM Is updated with the latest time step information, and finally generates the traffic flow prediction result. It can capture information from multiple time steps in the past and use it to predict traffic flow at the current moment.

[0162] 5. Rolling forecast and real-time update:

[0163] In this embodiment, the system adopts a rolling prediction method, that is, while predicting at each time step, the state of the LSTM model is continuously updated with the latest input data to ensure the real-time performance of traffic flow prediction. Whenever new traffic data arrives, the system can update the input vector To generate new prediction results, thus achieving dynamic prediction.

[0164] In this embodiment, the LSTM-based time series modeling effectively solves the time dependency problem in traffic flow. By using the gating mechanism, LSTM can flexibly handle short-term and long-term traffic changes and generate accurate real-time traffic flow prediction results. This prediction result provides a reliable input basis for subsequent intelligent decision-making, ensuring that the traffic management system can make timely responses and adjustments based on future traffic flows.

[0165] S5. Based on the traffic flow prediction results, intelligent decision-making algorithms are used to optimize traffic signal control and recommend routes.

[0166] After completing the real-time prediction of traffic flow, the system needs to make intelligent decisions based on the prediction results. The core goal of traffic management is to reduce congestion by optimizing the duration and phase configuration of traffic lights, while providing the best path recommendation for vehicles. In order to achieve these goals, the present invention uses a deep reinforcement learning (DRL) algorithm to optimize traffic signal control and combines it with a path recommendation algorithm to achieve dynamic guidance of vehicles. These optimization schemes are based on traffic flow prediction results and can be adaptively adjusted in complex and dynamic traffic environments to improve the operating efficiency of the entire transportation system.

[0167] In this embodiment, based on the traffic flow prediction results, the steps of using an intelligent decision-making algorithm to optimize traffic signal control and recommend routes include:

[0168] 1. Optimization of traffic signal control:

[0169] In this embodiment, a deep reinforcement learning (DRL) algorithm is used to optimize and adjust traffic lights. Based on the traffic flow prediction results generated in the previous step, the system can identify potential congestion in advance and improve traffic flow by dynamically adjusting the duration and phase of the traffic lights.

[0170] In the DRL framework, the system first defines the state space, action space, and reward function.

[0171] State Space : Includes traffic flow prediction value, current vehicle queue length, lane occupancy rate and other information, which describes the current traffic intersection conditions.

[0172] Action Space : Contains possible traffic light duration and phase configurations (such as green and red light duration adjustments and traffic priority adjustments in different directions).

[0173] Reward Function : It is used to measure the effect of each action. The design goal of the reward function in this embodiment is to minimize the waiting time of the vehicle and maximize the traffic efficiency. Its form is:

[0174]

[0175] in, is the weight factor, For the The waiting time of a vehicle, is the vehicle's speed, It is the number of vehicles observed simultaneously in the current traffic signal control area, that is, the total number of all vehicles queuing or passing in the signal light control area under the current traffic conditions.

[0176] At each time step , the system selects the best action through the Deep Q-Network (DQN) algorithm , that is, the adjustment strategy of the traffic light. The specific steps are as follows:

[0177] 1) At each time step , according to the time step The state space , select the time step Action Space Adjustment strategies as signal lights;

[0178] 2) After executing the action, get the time step The reward function , and observe the new time step The state space , update the value of depth Q according to the Bellman equation.

[0179] DQN updates the value function of the deep Q through the Bellman equation to evaluate the value of each action. The update formula of the value function of the deep Q is:

[0180]

[0181] in, is the learning rate, is the discount factor, For the current reward, is the current state, For the current action, is the time step The state space and action space The value of depth Q.

[0182] As the system continues to operate and learn, DRL can gradually optimize the control strategy of traffic lights and achieve adaptive adjustment of traffic lights, thereby reducing the average waiting time of vehicles and improving traffic efficiency.

[0183] 2. Optimization of path recommendation:

[0184] In addition to optimizing traffic lights, this embodiment also recommends routes for vehicles to further reduce congestion and improve the driving efficiency of individual vehicles. The route recommendation algorithm is based on traffic flow prediction results and dynamically calculates the optimal route to guide vehicles to avoid congested sections.

[0185] In this embodiment, the route recommendation adopts a multi-objective optimization method, and the optimization objectives include minimizing the travel time and avoiding highly congested sections. The multi-objective optimization function is:

[0186]

[0187] in, Is the path The total driving time, is the congestion level of the path, is a weighting factor used to balance travel time and congestion avoidance.

[0188] In addition, the route recommendation system uses federated learning to collaboratively train the route selection model. Federated learning allows vehicles in multiple regions to collaboratively update the route recommendation model, improving the prediction accuracy of the overall model while protecting user privacy. Specifically, vehicles in each region train the route model locally and update the global model using the following formula:

[0189]

[0190] in, It is The model weights of each client, It is The number of samples in each region, is the total number of samples.

[0191] 3. Comprehensive intelligent decision-making system:

[0192] In this embodiment, the optimization of traffic signal control and route recommendation cooperate with each other. The signal control system improves local traffic flow by optimizing the duration of traffic lights; the route recommendation system provides vehicles with the best route to guide them to avoid congested sections, thereby alleviating global traffic pressure.

[0193] The system will continuously adjust traffic light strategies and recommended routes based on real-time traffic flow forecasts and vehicle feedback to ensure that the system can maintain optimal operating efficiency in a dynamic traffic environment.

[0194] In this embodiment, through the combination of deep reinforcement learning algorithm and path recommendation algorithm, the system can dynamically adjust traffic lights and provide the best path recommendation solution based on traffic flow prediction results. This intelligent decision-making system can effectively alleviate traffic congestion, improve vehicle traffic efficiency, and further enhance the personalization and global optimization capabilities of path recommendation through federated learning.

[0195] S6. Applying the traffic signal control and route recommendation results to real-time traffic management

[0196] After completing traffic flow prediction and optimizing traffic signal control and route recommendations, the system needs to apply these optimization results to actual traffic management. This step not only includes sending the optimized signal light configuration and route recommendations to traffic infrastructure and vehicle terminals, but also includes continuous monitoring of traffic conditions, feedback data, and further optimization of the decision-making process through adaptive adjustments of the system. Through real-time application and feedback mechanisms, the system can effectively respond to dynamic changes in the traffic environment and ensure that the decision results produce the expected effects in actual scenarios.

[0197] In this embodiment, the step of applying the traffic signal control and route recommendation results to real-time traffic management includes:

[0198] 1. Real-time application of traffic signal control:

[0199] In this embodiment, the system sends control instructions to the traffic light control system at each traffic intersection in real time based on the traffic light duration and phase adjustment strategy optimized by the aforementioned deep reinforcement learning algorithm (DRL). Specifically, the system updates the configuration parameters of the traffic lights (such as the duration and sequence of green and red lights) every other time step to ensure that they can be adjusted according to the latest traffic flow forecast.

[0200] The signal light duration at each traffic intersection and the priority of traffic in each direction are determined by the optimization results, with the goal of reducing the average waiting time of vehicles and improving the traffic efficiency of each lane. The signal light adjustment instructions are sent to the intersection control equipment through the traffic management center to ensure its timely response.

[0201] 2. Real-time application of route recommendation:

[0202] For vehicle terminals, in this embodiment, the system pushes the route recommendation results to users through mobile applications or navigation devices. The route recommendation optimization scheme is based on the current traffic flow forecast and can generate a personalized optimal route for each vehicle. The specific recommended route is calculated by the multi-objective optimization formula that minimizes travel time and congestion level:

[0203]

[0204] in, For path The driving time, is the congestion coefficient, To adjust the parameters.

[0205] The user's vehicle's navigation device will adjust the driving route based on the system's recommended path, avoid congested sections, and dynamically update navigation instructions based on real-time traffic conditions.

[0206] 3. Real-time monitoring and feedback of traffic conditions:

[0207] In this embodiment, after applying the signal control and route recommendation solutions, the system will continue to monitor the actual traffic conditions and obtain the latest sensor data, GPS trajectory data and video surveillance information through the data acquisition module. Through the feedback of these data, the system can evaluate the effectiveness of the current decision in real time.

[0208] For the control effect of traffic lights, the system monitors the waiting time and speed of vehicles at each intersection, and adjusts the next signal light strategy based on feedback information. For the effect of route recommendation, the system monitors the matching degree between the actual driving time of vehicles and the recommended route, and further optimizes the route selection through this information.

[0209] 4. Real-time adjustment and adaptive optimization:

[0210] In this embodiment, the system has an adaptive optimization function, which can dynamically adjust the signal control and route recommendation strategy based on the real-time feedback of traffic data. For example, if it is detected that the length of the vehicle queue at a certain intersection has increased significantly, the system will immediately recalculate the signal light duration of the intersection to ensure that traffic can resume flow as soon as possible.

[0211] Adaptive optimization ensures that the system is flexible enough to deal with emergencies or traffic anomalies through continuous learning and adjustment. For example, when an accident or sudden weather changes occur, the system will prioritize these abnormal situations, adjust the timing of traffic lights, and guide vehicles to avoid the affected sections through route recommendations.

[0212] 5. System overall feedback mechanism:

[0213] In this embodiment, the system continuously optimizes traffic signal control and route recommendation through a closed-loop feedback mechanism. The results of each signal light adjustment and route recommendation will be used as input for the next time step to participate in the recalculation of traffic flow prediction and signal optimization.

[0214] This feedback mechanism relies on continuous monitoring by the data collection and analysis module to ensure that the system can adapt to complex traffic flow changes. Driven by real-time data input, the system continuously learns and optimizes, making the decision results of traffic light control and route recommendation gradually approach the optimal state.

[0215] In this embodiment, by applying the optimized traffic signal control and route recommendation scheme to the real-time traffic management system, the system can dynamically adjust the signal light duration and vehicle driving route, thereby effectively alleviating traffic congestion. The real-time monitoring and feedback mechanism ensures that the system remains flexible in the ever-changing traffic environment, and further improves the system's response speed and decision-making effect through adaptive optimization functions.

[0216] In some embodiments, the method of the present invention adopts an architecture that combines edge computing with cloud computing, where the edge nodes are responsible for preprocessing local data and the cloud nodes are responsible for complex calculations and model training of global traffic data.

[0217] In this hybrid architecture, edge nodes and cloud nodes perform their respective functions and coordinate their work to give full play to their respective advantages, ensuring that the system can balance local rapid response and global complex computing needs while processing large amounts of real-time traffic data.

[0218] 1. Edge computing nodes

[0219] Local data preprocessing: Edge nodes are deployed at key locations in the traffic network (such as traffic lights, cameras, sensors, etc.) and are responsible for real-time preprocessing of multimodal data collected locally. These data include:

[0220] Structured data (traffic flow, vehicle speed, lane occupancy, etc.) is collected directly by sensors.

[0221] Unstructured data (video stream data, user feedback, etc.) is obtained through cameras and user devices.

[0222] Real-time and low-latency processing: Edge nodes have strong real-time processing capabilities, can quickly complete data cleaning, feature extraction and other operations, reduce the delay of data transmission to the cloud, and perform preliminary intelligent analysis locally. For example, the real-time video stream collected by the camera can be processed by the convolutional neural network through the edge node to extract traffic information of vehicles and pedestrians and reduce the redundancy of video data.

[0223] Sharing computing pressure: Edge nodes can perform preliminary traffic flow analysis and short-term signal control optimization locally, especially in scenarios that require low-latency response (such as real-time response to sudden accidents or congestion). This processing method can effectively reduce the system's dependence on cloud computing resources and improve processing speed.

[0224] 2. Responsibilities of cloud computing nodes

[0225] Complex calculation of global data: Cloud nodes are mainly responsible for large-scale historical traffic data analysis and model training tasks. Compared with edge nodes, the cloud has stronger computing power and storage resources and can handle complex computing tasks. Including:

[0226] Global traffic flow prediction: Based on the pre-processed data uploaded by the edge nodes, the cloud nodes can conduct in-depth analysis of traffic data on a global scale and capture the long-term dependent traffic flow change trends in the city through time series models.

[0227] Training and optimization of complex models: The cloud is responsible for training deep reinforcement learning models for optimizing traffic light configuration and global route recommendation. By processing a large amount of historical data and feedback, the cloud can continuously optimize the traffic light control strategy and route recommendation algorithm to ensure that they adapt to complex traffic environments.

[0228] Model distribution and collaborative optimization: After completing complex model training in the cloud, the trained model will be distributed to each edge node for local rapid inference and application. At the same time, the cloud and edge nodes are optimized through dynamic resource allocation and collaborative scheduling, so that computing tasks can be flexibly switched between the cloud and the edge according to real-time traffic conditions.

[0229] Through this architecture, the system can not only ensure the timeliness of local real-time data processing, but also perform global complex calculations and model training in the cloud, so that the overall traffic management system has strong real-time, scalability and intelligent analysis capabilities.

[0230] In general, the present invention provides a method for intelligent real-time processing and analysis of big data, which is mainly used in traffic management systems. The method obtains multimodal traffic data, including structured, semi-structured and unstructured data, performs preprocessing and extracts feature vectors, and then uses the self-attention mechanism to fuse these features to generate unified fusion features. The fused features are modeled in time series through a long short-term memory network to predict changes in traffic flow, and based on this, a deep reinforcement learning algorithm is used to optimize traffic light control, and a path recommendation algorithm is combined to provide the optimal driving route for vehicles. Through real-time application, monitoring and feedback mechanisms, the system can adaptively adjust signal control and path recommendation strategies, effectively alleviate traffic congestion, and improve overall traffic management efficiency.

[0231] The system for intelligent real-time processing and analysis of big data described below and the method for intelligent real-time processing and analysis of big data described above can be referenced to each other.

[0232] Please refer to the attached Figure 3 The present invention also provides a system for intelligent real-time processing and analysis of big data, comprising:

[0233] The data acquisition module 100 is used to acquire structured data, semi-structured data and unstructured data;

[0234] The data processing module 200 is used to clean and extract features from the acquired multimodal data;

[0235] A data fusion module 300, for fusing multimodal data through a self-attention mechanism;

[0236] A time series prediction module 400, for performing time series modeling and prediction of traffic flow based on fusion features;

[0237] A decision module 500, for intelligent traffic signal control and route recommendation based on traffic flow prediction;

[0238] The resource scheduling module 600 is used to dynamically allocate resources between edge nodes and cloud nodes to ensure optimal efficiency of real-time processing.

[0239] The system of this embodiment can be used to execute the above method embodiment, and its principle and technical effect are similar, which will not be described in detail here. Although the embodiments of the present invention have been shown and described, it can be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention.

Claims

1. A method for intelligent real-time processing and analysis of big data, characterized in that: The following steps are involved: S1, based on traffic sensors, vehicle GPS, Internet devices and traffic cameras, real-time acquisition of multimodal traffic data, wherein the multimodal traffic data includes structured data, semi-structured data and unstructured data; S2, preprocessing the multimodal traffic data to extract feature vectors; S3, the extracted feature vectors are fused through the self-attention mechanism to generate a unified fused feature vector; S4, using the time series model to model the unified fusion feature vector and generate real-time traffic flow prediction results; S5, based on the traffic flow prediction result, using an intelligent decision-making algorithm to optimize traffic signal control and recommend routes, and obtain traffic signal control and route recommendation results; S6, applying the traffic signal control and route recommendation results to real-time traffic management; Among them, in step S5, the step of using an intelligent decision-making algorithm to optimize traffic signal control and recommend routes based on the traffic flow prediction result includes: S51, define basic elements of deep reinforcement learning, wherein the basic elements include state space , Action Space And the reward function ; The state space : Traffic condition parameters including traffic flow, lane occupancy rate, and vehicle queue length at the current traffic intersection; The action space : Including the phase and duration of traffic lights, as well as the switching duration of green, red and yellow lights; The reward function for: in, is the weight factor, For the The waiting time of a vehicle, is the vehicle's speed, is the number of vehicles observed simultaneously in the current traffic signal control area, that is, the total number of all vehicles queuing or passing in the signal light control area under the current traffic conditions; S52, based on the basic elements, obtaining the duration and phase of the traffic light, and optimizing the duration and phase of the traffic light using a deep Q network algorithm: S521, at each time step , according to the time step The state space , select the time step Action Space Adjustment strategies as signal lights; S522, after executing the action, obtain the time step The reward function , and observe the new time step The state space , update the value of depth Q according to the Bellman equation, which is: in, is the learning rate, is the discount factor, is the time step The state space and action space The value of depth Q; S523, during the continuous training process, by continuously optimizing the value of depth Q, the optimal traffic light duration and phase strategy is gradually learned to reduce the average waiting time of vehicles and improve traffic efficiency.

2. The method for intelligent real-time processing and analysis of big data according to claim 1 is characterized in that: The structured data include traffic flow data, vehicle speed data and lane occupancy data collected by traffic sensors; the semi-structured data include vehicle GPS data and user feedback data collected by vehicle GPS and Internet devices; the unstructured data include real-time video data collected by traffic cameras and social media text data collected in real time by Internet devices.

3. The method for intelligent real-time processing and analysis of big data according to claim 1, characterized in that: The step of preprocessing the multimodal traffic data comprises: Use Gaussian mixture models for anomaly detection and data cleaning on structured data; According to the appearance, size, shape and speed of objects captured by traffic cameras, real-time video data is processed in real time through convolutional neural networks to extract vehicle information and pedestrian information in traffic scenes; Perform sentiment analysis and keyword extraction on social media text data based on natural language processing models.

4. The method for intelligent real-time processing and analysis of big data according to claim 1, characterized in that: The step of fusing the extracted feature vectors through a self-attention mechanism to generate a unified fusion feature includes: Extract feature vectors of structured data, real-time video data, and social media text data respectively, and add tags, wherein the structured data is marked as feature vectors , the real-time video data is a feature vector , the social media text data is marked as a feature vector ; Based on the feature vector , eigenvector and the eigenvector Generate query vector , key vector Sum value vector ,in ; The feature vector is calculated by the following attention weight calculation formula , eigenvector and the eigenvector The attention weights between: in, is the key vector The dimension of , is the feature vector , eigenvector and the eigenvector The attention weight between Based on the feature fusion calculation formula, the feature vector is fused by weighted summation. , eigenvector and the eigenvector Perform feature fusion to generate a unified fusion feature vector , the feature fusion calculation formula is: in, is a unified fusion feature vector, is the value vector of the eigenvector, is the feature vector , eigenvector and the eigenvector The attention weight between .

5. The method for intelligent real-time processing and analysis of big data according to claim 1, characterized in that: The time series model is a long short-term memory network, which is used to capture the time dependency of traffic flow data. The calculation formula of the time series model is: in, is the time step The hidden state of is the time step The hidden state of is the time step The unified fusion feature vector.

6. The method for intelligent real-time processing and analysis of big data according to claim 1, characterized in that: The route recommendation adopts a federated learning method to collaboratively train a route recommendation model through vehicle data in multiple regions and generate personalized route recommendations.

7. The method for intelligent real-time processing and analysis of big data according to claim 1, characterized in that: The method adopts an architecture that combines edge computing with cloud computing, where edge nodes are responsible for preprocessing local data and cloud nodes are responsible for complex calculations and model training of global traffic data.

8. A system for intelligent real-time processing and analysis of big data, applied to the method according to any one of claims 1 to 7, characterized in that: include: Data collection module, used to obtain structured data, semi-structured data and unstructured data; A data processing module is used to clean and extract features from the acquired multimodal data; Data fusion module, used to fuse multimodal data through self-attention mechanism; Time series prediction module, used for time series modeling and prediction of traffic flow based on fusion features; Decision-making module, used for intelligent traffic signal control and route recommendation based on traffic flow prediction.

9. The system for intelligent real-time processing and analysis of big data according to claim 8, characterized in that: The system further comprises: The resource scheduling module is used to dynamically allocate resources between edge nodes and cloud nodes to ensure optimal efficiency of real-time processing.

Images