Dynamic traffic guidance method and device based on multi-source data fusion and storage medium

Abstract

The invention belongs to the technical field of dynamic traffic, and particularly discloses a multi-source data fusion traffic dynamic induction method, which comprises the following steps: S1, traffic information is acquired from multi-source heterogeneous data, and the data comprises real-time traffic information, navigation map data, a road network topological structure and a public traffic timetable; s2, carrying out data fusion processing on the traffic information obtained in the step S1, and generating fused traffic state estimation; s3, on the basis of travel historical records and preferences of the user, constructing a personalized travel guide model, and providing an optimal travel suggestion; and S4, monitoring the traffic condition in real time, performing dynamic adjustment based on user feedback and emergencies, and optimizing the travel path. By integrating real-time information from various data sources, comprehensive perception of road traffic conditions is ensured, traffic changes in space and time can be processed at the same time, the weight is dynamically adjusted according to the reliability of the data sources, and the influence of data islands and information loss is reduced.

Description

Technical Field

[0001] The present invention relates to the field of dynamic traffic technology, and in particular to a traffic dynamic induction method, device and storage medium for multi-source data fusion. Background Art

[0002] With the acceleration of global urbanization, urban transportation systems are facing increasingly severe challenges. Traffic congestion, emergencies (such as traffic accidents and construction) and asymmetric traffic information have led to uneven distribution of road resources, greatly reducing the efficiency of urban transportation systems. At the same time, as people's requirements for travel experience continue to increase, how to provide users with real-time and personalized traffic guidance solutions to reduce travel time and costs has become an urgent problem to be solved.

[0003] Traditional traffic guidance systems usually rely on a single data source (such as traffic monitoring cameras or navigation map data). Due to the lack of multi-source data integration, it is difficult to fully and accurately perceive traffic conditions, resulting in poor traffic guidance effects. In addition, existing systems are often unable to flexibly respond to emergencies such as traffic flow fluctuations and accidents, resulting in insufficient dynamic adjustment capabilities of the system and poor user travel experience. Especially when users' personalized needs are becoming increasingly strong, existing traffic guidance methods are difficult to provide travel suggestions that meet individual preferences and needs, and travel plans lack flexibility and pertinence.

[0004] On the other hand, with the advancement of technology, the diversification of data sources provides opportunities for the optimization of intelligent transportation systems. Traffic information can come from multiple heterogeneous data sources, including traffic monitoring cameras, GPS data, road sensors, navigation maps, road network topology, public transportation timetables, etc. However, due to the different data sources and heterogeneous formats, how to effectively integrate these multi-source data has become the key to improving the traffic induction effect. At the same time, how to adjust travel plans in real time and dynamically guide users through intelligent algorithms based on the fusion of multi-source data is also a major problem faced by existing technologies. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a traffic dynamic induction method, device and storage medium with multi-source data fusion to solve the problem that the prior art lacks multi-source data fusion, making it difficult to fully and accurately perceive traffic conditions, resulting in poor traffic induction effect.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0007] A traffic dynamic induction method based on multi-source data fusion includes the following steps:

[0008] S1. Acquire traffic information from multi-source heterogeneous data, wherein the data includes real-time traffic information, navigation map data, road network topology and public transportation timetable;

[0009] S2, performing data fusion processing on the traffic information obtained by S1 to generate a fused traffic state estimation;

[0010] S3: Build a personalized travel guidance model based on the user's travel history and preferences to provide the best travel suggestions;

[0011] S4. Monitor traffic conditions in real time, make dynamic adjustments based on user feedback and emergencies, and optimize travel routes;

[0012] S5. Integrate the data and services of transportation operators, navigation map providers, and smart travel service providers, build a unified data sharing standard, and achieve data interoperability and information exchange between different platforms and applications.

[0013] The data fusion in S2 is calculated using a spatiotemporal graph convolutional network and a Bayesian data fusion model, respectively, wherein the spatiotemporal graph convolutional network includes the following steps:

[0014] Obtain traffic surveillance cameras, vehicle GPS, navigation map data, and road network topology data;

[0015] The above data are processed using a spatiotemporal graph convolutional network to construct spatial and temporal dependencies. The traffic state estimation of the spatiotemporal graph convolutional network is calculated by the following formula:

[0016]

[0017] in, represents the feature of node v at time t+1, N(v) represents the neighborhood set of node v, and W (t) is the convolution weight and σ is the activation function.

[0018] The Bayesian data fusion model includes the following calculation steps:

[0019] Jointly analyze traffic information from different data sources and use a Bayesian data fusion model to dynamically adjust the weights of each data source based on its reliability;

[0020] The fused traffic state is estimated by the following formula:

[0021]

[0022] Among them, X is the traffic status, D i is the observed data from different data sources, P(X) is the prior probability, P(D1,D2,…,D n |X) is the likelihood function.

[0023] The generation of the guidance model in S3 includes the following steps:

[0024] Build user profiles through adaptive models based on users’ historical travel records and preferences;

[0025] The user's future travel behavior is predicted by a bidirectional long short-term memory network, and the prediction formula is:

[0026]

[0027] Among them, W forward and W backward are the weight matrices of the forward and backward networks respectively, and are the forward and backward hidden layer states, respectively.

[0028] The personalized travel recommendations in S3 are generated through the following steps:

[0029] Combining historical data of users and real-time traffic conditions, the scheme is evaluated through a multi-objective optimization algorithm, taking into account time, cost and travel comfort;

[0030] The multi-dimensional optimal path solution is generated by the Pareto optimal solution search algorithm. The multi-objective optimization model is:

[0031]

[0032] Among them, f1(r) represents the time cost, f2(r) represents the expense cost, and f3(r) represents the travel comfort.

[0033] The dynamic adjustment of the plan in step S4 is performed by the following steps:

[0034] The Q-learning reinforcement learning algorithm is used to dynamically adjust the path selection strategy according to the real-time traffic conditions and user feedback. The Q value is updated by the following formula:

[0035] Q(s,a)=Q(s,a)+α(r+γmaxQ(s′,a′)-Q(s,a))

[0036] Among them, Q(s,a) represents the Q value of selecting action a in state s, α is the learning rate, γ is the discount factor, and r is the reward value.

[0037] The dynamic adjustment in S4 includes the following steps:

[0038] Real-time monitoring of external traffic conditions, including traffic flow, emergencies, and weather changes;

[0039] A Bayesian reinforcement learning model is used to adjust the user's path selection according to real-time feedback. The weight update of the path adjustment is calculated by the following formula:

[0040]

[0041] in, is the updated weight, η is the learning rate, r t is the current feedback reward, The reward predicted by the model.

[0042] The travel path planning in step S4 includes the following steps:

[0043] Perform global path planning to determine the user's main driving route;

[0044] Perform local path optimization and refine the specific driving path through a hierarchical dynamic programming model. The optimization model of the global path planning is:

[0045]

[0046] Among them, f1(r) represents the time cost and f2(r) represents the fee cost.

[0047] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a traffic dynamic induction method of multi-source data fusion is implemented.

[0048] A readable storage medium stores a computer program, and when the computer program is executed by a processor, a traffic dynamic induction method of multi-source data fusion is implemented.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. The present invention ensures a comprehensive perception of road traffic conditions by integrating real-time traffic information, navigation map data, road network topology and public transportation timetable from multiple data sources. The spatiotemporal graph convolutional network and Bayesian data fusion model are used to simultaneously process traffic changes in space and time, and dynamically adjust weights according to the reliability of the data source. Compared with a single data source, the system can more accurately reflect the actual traffic status and reduce the impact of data silos and information loss.

[0051] 2. The present invention analyzes the user's historical travel data and preferences, combines the bidirectional long short-term memory network to predict user behavior, and generates a personalized travel guidance strategy. Each user has different travel needs. The system can provide personalized travel suggestions based on personal historical behavior, current traffic conditions and preferences to meet users' diverse needs for the shortest time, best route, lowest cost, etc., and improve user experience.

[0052] 3. The present invention monitors the changes in road traffic in real time, such as traffic congestion, accidents or road closures, and other emergencies. The system can adjust the travel plan in time, and use the Bayesian reinforcement learning model to dynamically adjust the path selection in combination with user feedback and real-time traffic data. When responding to sudden traffic events, the system can adaptively adjust the travel strategy to ensure that users obtain the best dynamic travel plan, greatly improving the flexibility and accuracy of travel.

[0053] 4. The present invention provides efficient traffic flow perception and prediction capabilities for traffic management departments. By integrating multi-source data and using machine learning and deep learning technologies to analyze and predict traffic conditions in real time, management departments can better plan signal light timing, adjust traffic strategies, and optimize road network design. Traffic managers can use the traffic induction data provided by the present invention to understand road conditions in a timely manner and make corresponding management decisions.

[0054] 5. Through intelligent route planning and dynamic adjustment of real-time traffic conditions, the system can effectively reduce the time users spend on congested roads, avoid ineffective waiting and detours, and optimize travel routes. At the same time, the system combines a multi-objective optimization algorithm to weigh multiple dimensions such as time, cost, and comfort, helping users choose the travel plan that best meets their needs and reducing unnecessary expenses. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 The traffic induction flow chart of multi-source data fusion of the present invention;

[0056] Figure 2 It is a schematic diagram of the traffic dynamic induction device of the present invention;

[0057] Figure 3 It is a schematic diagram of a readable storage medium of the present invention. DETAILED DESCRIPTION

[0058] The technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings 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.

[0059] Example 1

[0060] Reference Figure 1-Figure 3 In a first embodiment of the present invention, the present invention provides a traffic dynamic induction method based on multi-source data fusion, comprising the following steps:

[0061] S1. Obtain traffic information from multi-source heterogeneous data, including real-time traffic information, navigation map data, road network topology and public transportation timetable;

[0062] The multi-source data fusion method of the present invention relies on the effective collection and cleaning of multiple types of data. To achieve this goal, data collection covers multiple data sources such as real-time traffic flow, vehicle location, and road condition information in navigation maps. Based on these data, standardization and cleaning operations are performed to ensure that the data can smoothly enter the next stage of fusion processing.

[0063] In this embodiment, the multi-source data collection process includes the following main steps:

[0064] Collection of real-time traffic information:

[0065] The system obtains real-time data about the current traffic conditions of the road through traffic monitoring cameras, road sensors (such as ground sensors), vehicle GPS and other equipment. These data include vehicle speed, traffic flow, accidents, vehicle location, etc. In order to ensure the timeliness of the data, the present invention uses a stream processing framework (such as Apache Flink) to collect and process real-time data streams to ensure low data latency and timely feedback.

[0066] Obtaining navigation map data:

[0067] By calling public API interfaces such as Amap or Baidu Map, you can obtain road condition information in real-time navigation maps. Such data includes road length, road classification (such as expressways, trunk roads, etc.), road closure information, road congestion level and speed limit, etc. This information can not only provide real-time road traffic conditions, but also provide basic data support for future route planning.

[0068] Obtaining the road network topology:

[0069] The system also obtains the latest road network structure data from the city's traffic management department, covering the detailed information of the main roads, secondary roads, branches and traffic infrastructure (such as bridges and tunnels) in the road network. These data are stored in the form of vector diagrams or topological diagrams and serve as an important basis for route planning and traffic guidance.

[0070] Collection of public transport timetables:

[0071] The system connects with the city's public transportation system to obtain data including bus and subway departure times, station information, transfer information, and line changes. This information provides support for multi-modal travel plans and helps users comprehensively choose a mixed travel plan of public transportation and private cars.

[0072] In this embodiment, the collected data comes from a wide range of sources and is heterogeneous. Therefore, after the data collection is completed, the data from different sources need to be standardized and preprocessed to ensure the uniformity of the data format and the alignment of timestamps for subsequent data fusion processing.

[0073] Data standardization: The system normalizes data in different formats, for example, converting traffic flow information obtained from different devices and data sources into unified units (such as vehicle speed units and time units). This standardization operation ensures that different data sources can be effectively integrated.

[0074] Timestamp alignment: Since information from different data sources has different collection frequencies and time delays, the system uses a sliding time window algorithm to align data timestamps. In each time window, ensure that the collection times of different data sources match, so that synchronized data is used in the same time period. The time span of the sliding window is set by the specific business scenario, generally in seconds or minutes.

[0075] Data cleaning and exception handling: In actual data collection, it is inevitable to encounter noise data, missing values ​​or abnormal data. The system uses the triple standard deviation method to detect abnormal values. If the data deviates from the mean value by more than three times the standard deviation, it is judged as abnormal data. In addition, for missing data, the present invention uses linear interpolation to complete the data. The interpolation formula is as follows:

[0076]

[0077] Among them, x(t1) and x(t2) are adjacent known values, t1 and t2 are corresponding time points, and x(t) is the estimated value of the missing time point. Noise filtering: Since the sensor may be interfered by the environment, there is noise in the collected traffic data.

[0078] To this end, the present invention uses a Kalman filter to filter out noise from the data during the preprocessing process. The Kalman filter process is as follows:

[0079]

[0080] in, is the optimal estimate at time k, z k is the observed value, K kis the Kalman gain coefficient, and H is the measurement matrix. The Kalman filter continuously adjusts the state estimation, filters out noise, and improves the accuracy of the data.

[0081] In this embodiment, after data cleaning and standardization are completed, the data will enter the subsequent multi-source data fusion step, and the specific fusion method will be described in detail in the subsequent steps.

[0082] Through the above steps, the system can ensure that the heterogeneous data obtained from different data sources have a unified format, synchronized timestamps, and have high accuracy and reliability after cleaning and noise processing.

[0083] S2, performing data fusion processing on the traffic information obtained by S1 to generate a fused traffic state estimation;

[0084] For step S2, the present invention performs fusion processing based on the multi-source data collected and preprocessed in the aforementioned step S1 to generate a comprehensive and accurate traffic state estimation. The purpose of data fusion is to effectively integrate heterogeneous data of different types and sources so that the system can accurately perceive and predict traffic conditions. This process combines the spatiotemporal graph convolutional network (ST-GCN) and the Bayesian data fusion model to process data from three dimensions: space, time, and reliability of different data sources to ensure the accuracy and reliability of traffic state estimation.

[0085] In this embodiment, data fusion technology is the core step of the intelligent traffic guidance system, ensuring that multi-dimensional information from different data sources can be processed collaboratively to provide accurate basic data for subsequent personalized travel guidance and dynamic adjustment.

[0086] Specific description

[0087] In this embodiment, the fusion processing of multi-source data adopts the following two core technologies: spatiotemporal graph convolutional network (ST-GCN) and Bayesian data fusion model, which respectively process traffic flow analysis in the spatiotemporal dimension and reliability integration of multiple data sources.

[0088] Applications of Spatiotemporal Graph Convolutional Networks:

[0089] In this embodiment, the spatiotemporal graph convolutional network is used to process the spatiotemporal characteristics of road traffic data. The urban road network is modeled as a directed graph, in which traffic infrastructure such as road sections and intersections are regarded as nodes in the graph, and the connections between nodes represent the topological structure of the road. In this way, the system can simultaneously capture the changes in the road spatial structure and traffic status over time.

[0090] Convolution operation: The system first uses graph convolutional networks to extract spatial features, and then propagates local information on the road topology map through convolution operations, so that the traffic status of each road section can be affected by its adjacent road sections. Then, the system combines time convolution to extract time series features of traffic status at different times and model the changes in traffic flow in the time dimension. The specific spatiotemporal convolution formula is as follows:

[0091]

[0092] in, represents the feature of node v at time t+1, N(v) represents the set of nodes adjacent to node v, and W (t) is the weight matrix of the convolutional layer, ∫ is the activation function, b (t) is the bias term.

[0093] Through this operation, the system is able to generate time-space related traffic status estimates. Spatial neighborhood influence modeling: In a graph convolutional network, the characteristics of each node depend not only on its own traffic conditions, but also on the traffic conditions in its area. Therefore, the system considers the flow, speed and other information of adjacent road sections during the convolution process, thereby making a more comprehensive estimate of the node's traffic status.

[0094] Applications of Bayesian Data Fusion Models:

[0095] In this embodiment, the Bayesian data fusion model is used to integrate traffic information from different data sources. The reliability, timeliness and coverage of different data sources vary, so the influence weights of different data sources on the final traffic state estimation need to be dynamically adjusted during the fusion process. Bayesian inference provides effective theoretical support for this.

[0096] Bayesian inference formula: The system uses the Bayesian formula to jointly infer data from different data sources. Suppose we have multiple data sources {D1, D2, …, D n}After obtaining the observation of the traffic state X of the same road section, the Bayesian inference model calculates the fused traffic state estimate through the following formula:

[0097]

[0098] Among them, P(X|D1,D2,…,D n ) is the posterior probability of traffic state P(D1,D2,…,D n |X) is the likelihood function of different data sources, P(X) is the prior probability of the traffic state P(D1,D2,…,D n ) is the marginal probability of the observed data, and the reliability of the data source is dynamically adjusted; in the Bayesian model, the reliability of the data source should be adjusted according to its contribution.

[0099] For more reliable or newer data sources, their weights are higher in the fusion process, while for historical data or low-confidence data sources, their weights are lower. In this way, the system can automatically adjust the influence of data according to the changes in the data source, ensuring that the fused traffic state estimation has a high degree of credibility.

[0100] Perception fusion and multimodal data integration:

[0101] In order to further improve the accuracy of traffic state estimation, this embodiment also combines data from different types of sensors (such as cameras, ground sensing coils, and vehicle GPS) through multimodal data fusion. Multimodal fusion technology combines video data with time series data to achieve comprehensive perception of complex traffic scenes.

[0102] Application of multimodal fusion network: The data from traffic cameras is processed through convolutional neural networks (CNN) to extract visual features such as traffic flow and accidents in the images. At the same time, long short-term memory networks (LSTM) are used to process time series data (such as vehicle speed and location) to predict future traffic changes. Finally, these different modal data are integrated through the fusion network to form a more accurate estimate of traffic status.

[0103] S3: Build a personalized travel guidance model based on the user's travel history and preferences to provide the best travel suggestions;

[0104] For step S3, the formulation of the personalized travel guidance strategy of the present invention is based on the multi-source data fusion results in the aforementioned step S2, as well as the user's historical travel data and preferences. By building a personalized travel model, the system can generate the best travel suggestions based on each user's travel behavior and preferences. In order to achieve personalized route recommendations, the present invention uses adaptive personalized models and deep learning technologies, such as bidirectional long short-term memory networks (BiLSTM) to predict user behavior, and generates personalized optimal travel strategies through multi-objective optimization algorithms.

[0105] Specific description

[0106] In this embodiment, the process of formulating personalized travel guidance strategies includes multi-objective optimization processing of user portrait construction, user travel behavior prediction, and personalized recommendation. This process combines the user's historical travel records and preferences, and predicts future travel needs through deep learning models to ensure the accuracy of personalized recommendations.

[0107] User portrait construction and historical travel data analysis:

[0108] In this embodiment, the system generates a user profile for each user by analyzing the user's historical travel data, such as common travel time, preferred routes, and frequently visited destinations. User profiles are mainly based on historical data and user behavior habits. Through these portraits, the user's travel mode can be captured to provide a reference for subsequent travel strategy formulation.

[0109] The system uses an adaptive personalized model to classify user travel patterns based on the analysis of user preferences. For example, for commuting users, the system identifies their frequently used routes and travel times; for users who travel occasionally, the system focuses on their past destinations and travel time preferences. This classification information helps provide personalized travel strategies for different users.

[0110] User travel behavior prediction:

[0111] In this embodiment, in order to predict the user's future travel needs, the system uses a bidirectional long short-term memory network (BiLSTM). The network processes the user's historical travel records, captures the forward and backward dependencies between travel time periods and travel modes, and thus predicts the user's future travel behavior.

[0112] The working principle of BiLSTM is to pass the input data forward and backward, and the model can capture complex time dependencies based on the context information. The specific formula is:

[0113]

[0114] Among them, W forward and W backward are the weight matrices of the forward and backward networks respectively, and Represent the forward and backward hidden layer states respectively, σ is the activation function, Represents the prediction of the user's future travel behavior at time t. Through this bidirectional LSTM network, the system can effectively predict the user's departure time, possible destination, and preferred route, providing data support for personalized travel guidance.

[0115] Personalized recommendation and multi-objective optimization:

[0116] In this embodiment, the formulation of personalized travel strategies not only relies on the prediction of user travel behavior, but also combines real-time traffic conditions. In order to meet the user's multiple travel needs (such as shortest time, minimum cost, high comfort, etc.), the system optimizes the travel strategy through a multi-objective optimization algorithm.

[0117] In multi-objective optimization, the system solves the travel needs of different users as different goals and uses the Pareto optimal solution search algorithm to generate multi-dimensional optimal route suggestions. The objective function of multi-objective optimization is:

[0118] The solution search algorithm generates multi-dimensional optimal path suggestions. The objective function of multi-objective optimization is:

[0119]

[0120] Among them, f1(r) represents the travel time cost, f2(r) represents the economic cost (such as fuel costs, tolls, etc.), and f3(r) represents the travel comfort (such as reducing the time spent at traffic lights, avoiding peak hours, etc.).

[0121] Decision-making process of personalized recommendation: In actual applications, the system combines the user's preference weights to weight different optimization goals. For example, if the user attaches more importance to the time factor, the system will increase the weight of the time goal.

[0122] The system uses the Pareto optimization algorithm to find a set of optimal solutions under multi-dimensional objectives, and selects the path plan that best meets user needs.

[0123] Through the personalized travel guidance strategy of this embodiment, the system can provide users with travel suggestions that meet their personalized needs and ensure the best balance between travel time, cost and comfort. In this process, the combination of deep learning models and multi-objective optimization algorithms ensures the intelligence and accuracy of personalized travel strategies.

[0124] S4. Monitor traffic conditions in real time, make dynamic adjustments based on user feedback and emergencies, and optimize travel routes;

[0125] With respect to step S4, the travel plan evaluation and dynamic adjustment of the present invention is based on the personalized travel guidance strategy generated in step S3, combined with real-time monitored traffic conditions and user feedback, to dynamically adjust the travel plan. In order to optimize the user's travel experience, the system evaluates the current travel plan in real time, evaluates it from multiple dimensions such as time, cost, and comfort, and dynamically adjusts the travel plan in combination with real-time data to ensure that the user can obtain the optimal travel path.

[0126] The present invention introduces a Bayesian reinforcement learning model, combines real-time traffic data, user feedback and prediction, and updates and adjusts travel strategies in real time, thereby ensuring the flexibility and dynamism of the solution.

[0127] Specific description

[0128] In this embodiment, the evaluation and dynamic adjustment of travel plans include multi-dimensional evaluation and weight adjustment of travel plans, real-time monitoring and feedback mechanism, and dynamic optimization of paths based on Bayesian reinforcement learning. The system dynamically adjusts the generated personalized plans and optimizes path selection by sensing external factors such as traffic flow, accidents, and weather in real time and combining the user's actual travel situation.

[0129] Travel plan evaluation system:

[0130] In this embodiment, the system establishes a multi-dimensional travel plan evaluation system, which mainly evaluates from three aspects: time cost, cost cost and travel comfort. The needs of each user are different, so the weights of these dimensions are dynamically adjusted.

[0131] The specific model for travel plan evaluation uses a weighted comprehensive evaluation method, which assigns different weights to each evaluation dimension based on the user's personalized needs. The specific weighted scoring formula is as follows:

[0132]

[0133] Among them, S i For each evaluation dimension (such as time, cost, comfort, etc.), w i The system continuously optimizes these weights based on user feedback and historical data to ensure the personalization and accuracy of the evaluation results.

[0134] Specific description of the evaluation dimensions:

[0135] Time dimension: The system scores based on the difference between the actual travel time and the estimated time. The shorter the time and the less delay, the higher the score.

[0136] Cost dimension: The system scores based on economic indicators such as fuel costs and tolls. The lower the cost, the higher the score.

[0137] Comfort dimension: Comfort is scored by reducing the number of stops at traffic lights and avoiding congested roads.

[0138] Real-time monitoring and user feedback mechanism:

[0139] In this embodiment, the system monitors traffic flow, accidents, weather changes and other external factors in real time through various sensors, cameras and traffic management platforms. When congestion, traffic accidents or other emergencies are detected, the system will automatically trigger a dynamic adjustment mechanism.

[0140] The system also receives real-time feedback from users through mobile apps or in-car navigation systems. For example, users can provide feedback on the accuracy and comfort of the current route, and this feedback information will serve as an important basis for subsequent program evaluation and dynamic adjustment.

[0141] Dynamic adjustment based on Bayesian reinforcement learning:

[0142] In this embodiment, in order to adjust the user's travel plan in real time, the system uses the Bayesian reinforcement learning model for path optimization. In reinforcement learning, the system optimizes the user's path selection strategy by constantly interacting with the environment. The Bayesian reinforcement learning model can adaptively adjust the path selection weights in the face of uncertainty or new traffic information, thereby generating a new optimal solution.

[0143] The main steps of the reinforcement learning process include:

[0144] State representation: The system models the current traffic state s as a state vector, including information such as real-time traffic conditions, the user's current location and destination.

[0145] Action selection: The system selects action a according to the current state s, that is, whether to maintain the current path or switch to an alternative path.

[0146] Action selection is performed using the Q-learning algorithm of Bayesian reinforcement learning.

[0147] Q value update: The system updates the Q value of the route based on user feedback and traffic conditions. The update formula is:

[0148] Q(s,a)=Q(s,a)+α(r+γmaxQ(s′,a′)-Q(s,a))

[0149] Among them, Q(s,a) represents the Q value of selecting action a in state s, α is the learning rate, r is the current feedback reward γ is the discount factor, s ′ Indicates the next state.

[0150] Through this process, the system can learn the optimal decision to choose different paths under different traffic conditions. Application of Bayesian inference in path selection: In order to better deal with the uncertainty in the traffic environment, the system dynamically adjusts the selection probability of different paths through Bayesian inference. The Bayesian inference process combines the user's historical feedback and real-time traffic conditions to update the posterior probability of path selection, ensuring that the system can still provide the best path selection solution in a complex traffic environment.

[0151] Application scenarios for solution adjustment:

[0152] Congestion adjustment: When the system detects that the user's current path is congested, the Bayesian reinforcement learning model automatically adjusts the current path weight and prioritizes alternative paths with less congestion.

[0153] Accident detour: When a traffic accident occurs on the road ahead, the system will update the traffic status in real time and suggest users to take a detour.

[0154] User preference adjustment: If the user feedbacks that the current route does not meet their preferences (for example, the route is too long or not comfortable enough), the system will readjust the weight and scheme of route selection based on the user feedback.

[0155] Through this embodiment, the system can flexibly adjust the travel plan based on real-time traffic conditions and user feedback to ensure that the user is always on the optimal path. At the same time, the combination of Bayesian reinforcement learning and real-time evaluation enables the system to adaptively respond to complex traffic environments and provide more personalized and intelligent travel plans.

[0156] S5. Integrate the data and services of transportation operators, navigation map providers, and smart travel service providers, build a unified data sharing standard, and achieve data interoperability and information exchange between different platforms and applications.

[0157] With respect to step S5, the present invention realizes data openness and interoperability by integrating multiple traffic-related data providers, such as traffic operators, navigation map providers, and smart travel service providers, thereby building a unified smart travel service ecosystem.

[0158] The goal of this step is to provide a unified service interface for different types of travel participants, and promote the interconnection and interoperability of different platforms and applications through unified data standards, ensuring smooth data transmission throughout the entire ecosystem.

[0159] The technical solution for this part is based on existing data sharing and platform interoperability technologies, and can achieve seamless integration of data from multiple parties and the implementation of unified standards.

[0160] Example 2

[0161] The second embodiment of the present invention is based on the same inventive concept and proposes a computer-readable storage medium. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, it implements the steps of a traffic dynamic induction method of multi-source data fusion in the above-mentioned embodiment.

[0162] Example 3

[0163] The third embodiment of the present invention is based on the same inventive concept. The present invention proposes a terminal, which includes: a processor and a memory; the processor and the memory communicate with each other; the memory is used to store instructions; the processor is used to execute the instructions in the memory to execute a traffic dynamic induction method of multi-source data fusion of the above embodiment.

[0164] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0165] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A traffic dynamic induction method based on multi-source data fusion, characterized in that: The following steps are involved: S1. Acquire traffic information from multi-source heterogeneous data, wherein the data includes real-time traffic information, navigation map data, road network topology and public transportation timetable; S2, performing data fusion processing on the traffic information obtained by S1 to generate a fused traffic state estimation; S3: Build a personalized travel guidance model based on the user's travel history and preferences to provide the best travel suggestions; S4. Monitor traffic conditions in real time, make dynamic adjustments based on user feedback and emergencies, and optimize travel routes; S5. Integrate the data and services of transportation operators, navigation map providers, and smart travel service providers, build a unified data sharing standard, and achieve data interoperability and information exchange between different platforms and applications.

2. The traffic dynamic induction method based on multi-source data fusion according to claim 1 is characterized in that: The data fusion in S2 is calculated using a spatiotemporal graph convolutional network and a Bayesian data fusion model, respectively, wherein the spatiotemporal graph convolutional network includes the following steps: Obtain traffic surveillance cameras, vehicle GPS, navigation map data, and road network topology data; The above data is processed using a spatiotemporal graph convolutional network to construct spatial and temporal dependencies. The traffic state estimation of the spatiotemporal graph convolutional network is calculated by the following formula: in, represents the feature of node v at time t+1, N(v) represents the neighborhood set of node v, and W (t) is the convolution weight and σ is the activation function.

3. The traffic dynamic induction method based on multi-source data fusion according to claim 2 is characterized in that: The Bayesian data fusion model includes the following calculation steps: Jointly analyze traffic information from different data sources and use a Bayesian data fusion model to dynamically adjust the weights of each data source based on its reliability; The fused traffic state is estimated by the following formula: Among them, X is the traffic status, D i is the observed data from different data sources, P(X) is the prior probability, P(D1,D2,…,D n |X) is the likelihood function.

4. The traffic dynamic induction method based on multi-source data fusion according to claim 1 is characterized in that: The generation of the guidance model in S3 includes the following steps: Build user profiles through adaptive models based on users’ historical travel records and preferences; The user's future travel behavior is predicted by a bidirectional long short-term memory network, and the prediction formula is: Among them, W forward and W backward are the weight matrices of the forward and backward networks respectively, and are the forward and backward hidden layer states, respectively.

5. The traffic dynamic induction method based on multi-source data fusion according to claim 1 is characterized in that: The personalized travel recommendations in S3 are generated through the following steps: Combining historical data of users and real-time traffic conditions, the scheme is evaluated through a multi-objective optimization algorithm, taking into account time, cost and travel comfort; The multi-dimensional optimal path solution is generated by the Pareto optimal solution search algorithm. The multi-objective optimization model is: Among them, f1(r) represents the time cost, f2(r) represents the expense cost, and f3(r) represents the travel comfort.

6. The traffic dynamic induction method based on multi-source data fusion according to claim 1 is characterized in that: The dynamic adjustment in S4 includes the following steps: Real-time monitoring of external traffic conditions, including traffic flow, emergencies, and weather changes; A Bayesian reinforcement learning model is used to adjust the user's path selection according to real-time feedback. The weight update of the path adjustment is calculated by the following formula: in, is the updated weight, η is the learning rate, r t is the current feedback reward, The reward predicted by the model.

7. The traffic dynamic induction method based on multi-source data fusion according to claim 1 is characterized in that: The dynamic adjustment of the plan in S4 is performed through the following steps: The Q-learning reinforcement learning algorithm is used to dynamically adjust the path selection strategy according to the real-time traffic conditions and user feedback. The Q value is updated by the following formula: Q(s,a)=Q(s,a)+α(r+γmaxQ(s′,a′)-Q(s,a)) Among them, Q(s,a) represents the Q value of selecting action a in state s, α is the learning rate, γ is the discount factor, and r is the reward value.

8. The traffic dynamic induction method based on multi-source data fusion according to claim 1 is characterized in that: The travel path planning in S4 includes the following steps: Perform global path planning to determine the user's main driving route; Perform local path optimization and refine the specific driving path through a hierarchical dynamic programming model. The optimization model of the global path planning is: Among them, f1(r) represents the time cost and f2(r) represents the fee cost.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the traffic dynamic induction method of multi-source data fusion as described in any one of claims 1 to 8 is implemented.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for traffic dynamic induction based on multi-source data fusion as described in any one of claims 1 to 8 is implemented.

Images