Intelligent navigation path dynamic planning method based on multi-source data fusion

Abstract

The invention relates to the technical field of satellite navigation and positioning, in particular to an intelligent navigation path dynamic planning method based on multi-source data fusion, and the method comprises the following steps: S1, collecting and preprocessing Beidou positioning, sensor and traffic information data, and generating standardized input through Kalman filtering noise reduction and space-time alignment; according to the method, the three-dimensional road network is constructed through the Beidou elevation data, the multi-target reinforcement learning is fused to dynamically optimize the path weight, the personalized path planning of safety, efficiency and energy consumption is realized in combination with the user portrait, and the road condition data is compressed and transmitted by using the Beidou short message; obstacle avoidance area marking and rapid path generation are realized through AR gesture interaction and Monte Carlo tree search, dynamic closed-loop optimization of a user and a system is formed in combination with multi-modal feedback, and navigation real-time performance and user experience are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of satellite navigation and positioning technology, and in particular to an intelligent navigation path dynamic planning method based on multi-source data fusion. Background Art

[0002] Intelligent navigation is a multi-disciplinary technology system that integrates environmental perception, path planning, dynamic adjustment, and intelligent processing of human-computer interaction to achieve efficient, safe, and personalized navigation services. Its core lies in the use of advanced sensors, communications, computing, and artificial intelligence algorithms to overcome the limitations of traditional navigation, adapt to complex and dynamic environments, and meet the diverse needs of users. Existing navigation systems mainly rely on single satellite positioning and static electronic maps, facing the following core challenges:

[0003] Insufficient positioning reliability in complex environments: In scenarios such as urban canyons, tunnels, and mountainous areas, satellite signals are easily affected by obstruction or multipath effects, resulting in reduced positioning accuracy. Furthermore, the system lacks the ability to fuse multi-source data, making it impossible to achieve continuous and reliable positioning. Traditional data fusion methods do not consider the probability of data source conflicts. When sensor data is inconsistent, it is easy to cause inaccurate environmental perception models.

[0004] The path planning strategy is simple and lacks dynamic adaptability: the use of a single objective optimization with fixed weights cannot balance multi-dimensional requirements such as safety, efficiency, and energy consumption. In particular, it cannot optimize energy consumption or circumvent load restrictions for special types of users such as electric vehicles and freight vehicles. Real-time road condition updates are delayed, and there is a lack of predictive ability for future road conditions. Path planning results often fall into local optimality and cannot cope with dynamic traffic changes.

[0005] The user interaction model is rigid and lacks personalization: it is mainly based on "command input-route output", with a single means of interaction, unable to support real-time intervention in complex scenarios, and no user behavior analysis model has been established. It is impossible to dynamically adjust path preferences based on driving habits, resulting in a "one-size-fits-all" navigation experience.

[0006] Based on this, the present invention provides an intelligent navigation path dynamic planning method based on multi-source data fusion to solve the above-mentioned technical problems. Summary of the Invention

[0007] The purpose of the present invention is to provide an intelligent navigation path dynamic planning method based on multi-source data fusion. The present invention realizes high-precision positioning and anti-occlusion continuous positioning by fusing Beidou RTK and multi-source data, dynamically allocates data weights in combination with evidence theory to construct a real-time dynamic environment perception model, improves positioning reliability in complex environments, and constructs a three-dimensional road network based on Beidou elevation data and integrates multi-objective reinforcement learning to dynamically optimize path weights. It realizes personalized path planning with safety, efficiency and energy consumption by combining user portraits, utilizes Beidou short message compression to transmit road condition data, realizes obstacle avoidance area marking and fast path generation through AR gesture interaction and Monte Carlo tree search, and forms dynamic closed-loop optimization of users and systems by combining multimodal feedback, which significantly improves navigation real-time performance and user experience.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] The present invention provides an intelligent navigation path dynamic planning method based on multi-source data fusion, comprising the following steps:

[0010] S1: Collect and preprocess BeiDou positioning, sensor, and traffic information data, and generate standardized inputs through Kalman filtering, noise reduction, and spatiotemporal alignment;

[0011] S2: Based on BeiDou RTK high-precision positioning and multi-source data fusion, it adopts conflict data adaptive fusion and multi-modal failure compensation strategies to build a dynamic environment perception model;

[0012] S3: Utilizing BeiDou elevation data and real-time traffic flow, a multi-objective reinforcement learning framework dynamically optimizes path weights. User profile analysis adjusts safety, efficiency, and energy consumption decision preferences in real time based on driving behavior characteristics.

[0013] S4: Based on the Beidou short message protocol, a differential road condition compression transmission mechanism is used to implement closed-loop interaction between user obstacle avoidance area marking and system alternative path generation in the AR interface, and parameter optimization is achieved through tactile and voice multimodal feedback.

[0014] S5: Complete the hardware and software integration deployment, and verify the system performance and reliability through simulation and field testing.

[0015] In S1, Beidou positioning, sensor, and traffic information data are collected and preprocessed, and standardized input is generated through Kalman filtering, noise reduction, and spatiotemporal alignment. The specific steps are as follows:

[0016] S1.1: Collect raw positioning data through Beidou multi-frequency receiver;

[0017] S1.2: Synchronously acquire vehicle-mounted sensor data, including acceleration / angular velocity from the MEMS inertial measurement unit, lidar point cloud, and visual camera images;

[0018] S1.3: Access to real-time traffic conditions, accidents, and construction information from third-party traffic platforms;

[0019] S1.4: Use extended Kalman filtering to eliminate multipath errors on BeiDou data and perform dynamic trajectory smoothing in combination with IMU data.

[0020] S1.5: Align multi-source data using a timestamp synchronization algorithm and convert BeiDou WGS-84 coordinates to a local plane coordinate system using UTM projection transformation.

[0021] The BeiDou multi-frequency receiver supports B1I / B2I / B3I multi-band signal reception, meeting the signal strength range of -130dBm to -100dBm.

[0022] In S2, based on BeiDou RTK high-precision positioning and multi-source data fusion, conflict data adaptive fusion and multi-modal failure compensation strategies are adopted to build a dynamic environment perception model. The specific steps are as follows:

[0023] S2.1: Use ground-based augmentation stations to obtain differential corrections to achieve centimeter-level real-time positioning with a lateral accuracy of ≤0.1m and an elevation accuracy of ≤0.2m;

[0024] S2.2: In areas where satellite signals are blocked, the BeiDou / INS tight integration algorithm is used to maintain positioning continuity by fusing inertial data through the federated Kalman filter.

[0025] S2.3: Use the evidence theory weighted fusion algorithm to model the conflict probability between Beidou positioning and lidar obstacle detection results, and automatically assign data credibility weights;

[0026] S2.4: Based on the fused positioning and sensor data, construct a dynamic grid map that includes road geometry and real-time status.

[0027] In S2.3, the weighted fusion algorithm based on evidence theory is used to model the conflict probability between Beidou positioning and lidar obstacle detection results, and automatically assign data credibility weights. The specific steps are as follows:

[0028] S2.3.1: Generate a position probability distribution m1(A) for BeiDou positioning data and an obstacle presence probability distribution m2(B) for LiDAR obstacle detection results;

[0029] S2.3.2: Calculate the probability of conflict between two data sources

[0030] S2.3.3: Dynamically adjust the credibility weight according to the conflict probability k Where w1+w2=1.

[0031] S3 utilizes BeiDou elevation data and real-time traffic flow to dynamically optimize path weights through a multi-objective reinforcement learning framework. User profile analysis adjusts safety, efficiency, and energy consumption decision preferences in real time based on driving behavior characteristics. The specific steps are as follows:

[0032] S3.1: Use BeiDou elevation data to construct a three-dimensional road network including slope and altitude, with nodes as intersections and energy consumption coefficients added to edge attributes;

[0033] S3.2: Establish an optimization function J = αT + βE + γS, where T is the travel time after adjusting for real-time traffic conditions, E is the energy consumption calculated based on the vehicle dynamics model F = m·g·sinθ, and S is the safety risk value weighted by historical accident rates.

[0034] S3.3: Use a deep reinforcement learning framework to input user profiles and real-time environment status and output dynamic weight coefficients;

[0035] S3.4: Based on the improved A* algorithm, a time window expansion mechanism is introduced to predict the traffic conditions in the next 5 minutes and generate the global optimal path.

[0036] S3.3 uses a deep reinforcement learning framework to input user profiles and real-time environment status and output dynamic weight coefficients. The specific steps are as follows:

[0037] S3.3.1: Construct the state space S, which contains the user profile features u=[u1,u2,…,u n ] and the real-time environment state e=[e1,e2,…,e m ];

[0038] S3.3.2: Define the action space A as the weight coefficient vector [α, β, γ], satisfying α + β + γ = 1;

[0039] S3.3.3: Design a reward function R = -λ1T - λ2E + λ3S, where λ1, λ2, and λ3 are preset constants.

[0040] S3.3.4: Fit the policy function through a deep neural network and update the network parameters using experience replay and gradient descent algorithms.

[0041] In S4, based on the Beidou short message protocol, a differential road condition compression transmission mechanism is used to perform closed-loop interaction between user obstacle avoidance area marking and system alternative path generation in the AR interface, and parameters are optimized through tactile and voice multimodal feedback. The specific steps are as follows:

[0042] S4.1: Using the CS-SLP compressed sensing algorithm, key information such as the coordinates of the congested area and the type of accident is encoded into a binary sequence and transmitted to terminals in remote areas via Beidou short messages;

[0043] S4.2: The user uses gestures in the AR interface to mark the avoidance area, and the system generates an alternative path within 1.5 seconds based on Monte Carlo tree search, with a path deviation of ≤5%;

[0044] S4.3: Synchronize path correction results in real time through voice prompts, tactile vibrations, and AR real-scene arrow overlays, and collect user operation data to update model parameters.

[0045] S4.1 uses the CS-SLP compressed sensing algorithm to encode key information such as the coordinates of the congested area and the accident type into a binary sequence, which is then transmitted to terminals in remote areas via Beidou short messages. The specific steps are as follows:

[0046] S4.1.1: Represent the congestion area coordinates as a two-dimensional grid matrix The non-zero elements represent congested areas;

[0047] S4.1.2: Design the measurement matrix Perform linear projection on the matrix X to generate the measurement vector y = φX;

[0048] S4.1.3: Reconstructing sparse signals using the orthogonal matching pursuit algorithm satisfies min||X||0s·t·y=φX;

[0049] S4.1.4: Reconstruct the signal The accident type is encoded as a binary symbol sequence B, which is further compressed by Huffman coding.

[0050] In S5, hardware and software integration deployment is completed, and system performance and reliability are verified through simulation and field testing. The specific steps are as follows:

[0051] S5.1: Integrates a Qualcomm Snapdragon 820A chip, BeiDou / GPS multi-mode receiver, 9-axis IMU, and lidar interface, and supports 5G / short message dual communication modes;

[0052] S5.2: Deploy edge computing modules to process real-time data, and cloud servers to run reinforcement learning model training;

[0053] S5.3: Simulate five scenarios in urban and mountainous areas on the CARLA platform to verify path planning response time and replanning accuracy.

[0054] S5.4: A total of 2,000 kilometers of road testing, covering extreme environments such as heavy rain and tunnels, and recording the mean time between failures.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] The present invention realizes high-precision positioning and anti-occlusion continuous positioning through the fusion of Beidou RTK and multi-source data, dynamically allocates data weights in combination with evidence theory to build a real-time dynamic environment perception model, improves positioning reliability in complex environments, and constructs a three-dimensional road network based on Beidou elevation data and integrates multi-objective reinforcement learning to dynamically optimize path weights. It realizes personalized path planning with safety, efficiency and energy consumption in combination with user portraits, utilizes Beidou short message compression to transmit road condition data, realizes obstacle avoidance area marking and fast path generation through AR gesture interaction and Monte Carlo tree search, and forms dynamic closed-loop optimization of users and systems in combination with multimodal feedback, which significantly improves navigation real-time performance and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a flow chart of an intelligent navigation path dynamic planning method based on multi-source data fusion according to the present invention.

[0058] Figure 2 This is a multi-source data fusion flow chart of the intelligent navigation path dynamic planning method based on multi-source data fusion in the present invention.

[0059] Figure 3 This is a reinforcement learning decision flow chart for an intelligent navigation path dynamic planning method based on multi-source data fusion in the present invention. DETAILED DESCRIPTION

[0060] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described 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 making any creative efforts shall fall within the scope of protection of the present invention.

[0061] Example:

[0062] like Figure 1-Figure 3As shown, this embodiment provides an intelligent navigation path dynamic planning method based on multi-source data fusion, including the following steps: S1: Collect and pre-process Beidou positioning, sensor and traffic information data, and generate standardized input through Kalman filtering noise reduction and spatiotemporal alignment; S2: Based on Beidou RTK high-precision positioning and multi-source data fusion, conflict data adaptive fusion and multi-modal failure compensation strategy are adopted to build a dynamic environment perception model; S3: Utilize Beidou elevation data and real-time traffic flow to dynamically optimize path weights through a multi-objective reinforcement learning framework, wherein user portrait analysis adjusts decision preferences for safety, efficiency and energy consumption in real time according to driving behavior characteristics; S4: Based on the differential road condition compression transmission mechanism of the Beidou short message protocol, a closed-loop interaction between user obstacle avoidance area marking and system alternative path generation is performed in the AR interface, and parameters are optimized through tactile and voice multi-modal feedback; S5: Complete the hardware and software integration deployment, and verify the system performance and reliability through simulation and field testing.

[0063] In this embodiment, it should also be noted that S1 collects and preprocesses Beidou positioning, sensor, and traffic information data, and generates standardized input through Kalman filtering for noise reduction and spatiotemporal alignment. The specific steps are as follows: S1.1: Collect raw positioning data through a Beidou multi-frequency receiver; the Beidou multi-frequency receiver supports B1I / B2I / B3I multi-band signal reception, meeting the signal strength range of -130dBm to -100dBm. S1.2: Synchronously acquire on-board sensor data, including acceleration / angular velocity from the MEMS inertial measurement unit, lidar point cloud, and visual camera images; S1.3: Access real-time dynamic information on road conditions, accidents, and construction from a third-party traffic platform; S1.4: Apply an extended Kalman filter to the Beidou data to eliminate multipath effect errors, and combine it with IMU data for dynamic trajectory smoothing; S1.5: Align multi-source data through a timestamp synchronization algorithm, and convert Beidou WGS-84 coordinates into a local plane coordinate system using the UTM projection transformation.

[0064] Furthermore, it should be noted that the B1I band (1561.098MHz) is used for conventional navigation and provides meter-level positioning accuracy; the B2I band (1207.14MHz) and the B3I band (1268.52MHz) support high-precision measurements, eliminating ionospheric delay errors through multi-frequency combination observations, and improving accuracy to decimeter levels. When the received signal strength is in the range of -130dBm to -100dBm, it indicates that the receiver is in normal working condition (Note: the typical strength of the satellite signal reaching the ground is about -130dBm, and it may be attenuated to -100dBm due to obstruction in urban environments). The original positioning data consists of information such as pseudorange, carrier phase observations, satellite ephemeris, timestamp and signal-to-noise ratio, which provides the basis for subsequent high-precision positioning solutions. MEMS inertial measurement unit: The data output is three-axis acceleration (m / s 2), angular velocity (° / s); update frequency is 100Hz; accuracy specification: acceleration noise ≤100μg / √Hz. LiDAR: Data output is a 3D point cloud (X, Y, Z coordinates); update frequency is 10Hz; accuracy specification: distance accuracy ≤2cm (within 100 meters). Vision camera: Data output is an RGB image (1920×1080 pixels); update frequency is 30fps; accuracy specification: lane line detection error ≤10cm. The time synchronization algorithm uses the NTPv4 protocol, calibrating the on-board terminal time using the BeiDou receiver's built-in high-precision clock (error ≤100ns). The synchronization period is 1 second, ensuring that the timestamp deviation of multi-source data is ≤10ms.

[0065] In this embodiment, it should also be noted that, in S2, based on Beidou RTK high-precision positioning and multi-source data fusion, conflict data adaptive fusion and multi-modal failure compensation strategies are adopted to construct a dynamic environment perception model. The specific steps are as follows: S2.1: Use ground-based augmentation stations to obtain differential corrections to achieve centimeter-level real-time positioning, with lateral accuracy ≤0.1m and elevation ≤0.2m; S2.2: In satellite signal blocking areas, the Beidou / INS tight combination algorithm is enabled, and the inertial data is fused through the federal Kalman filter to maintain positioning continuity; S2.3: The evidence theory weighted fusion algorithm is used to model the conflict probability between Beidou positioning and lidar obstacle detection results, and automatically assign data credibility weights; the specific steps are as follows: S2.3.1: Generate a position probability distribution m1 (A) for Beidou positioning data, and generate an obstacle existence probability distribution m2 (B) for the lidar obstacle detection results; S2.3.2: Calculate the conflict probability of the two data sources S2.3.3: Dynamically adjust the credibility weight according to the conflict probability k Where w1 + w2 = 1. S2.4: Based on the fused positioning and sensor data, construct a dynamic grid map including road geometry and real-time status.

[0066] Furthermore, it is important to note that positioning accuracy is guaranteed by: ① Error elimination: Carrier phase differential technology eliminates systematic errors in satellite signal propagation, theoretically achieving lateral accuracy of ±0.05m and elevation accuracy of ±0.1m. ② Real-time design: The differential correction frequency is updated at 1Hz, ensuring real-time positioning results in dynamic scenarios.

[0067] In this embodiment, it should also be noted that, in S3, Beidou elevation data and real-time traffic flow are used to dynamically optimize path weights through a multi-objective reinforcement learning framework, wherein user portrait analysis adjusts decision preferences for safety, efficiency, and energy consumption in real time according to driving behavior characteristics. The specific steps are as follows: S3.1: Use Beidou elevation data to construct a three-dimensional road network including slope and altitude, with nodes as intersections and edge attributes with additional energy consumption coefficients; S3.2: Establish an optimization function J = αT + βE + γS, wherein: T is the travel time after real-time road condition correction, E is the energy consumption calculated based on the vehicle dynamics model F = m·g·sinθ, and S is the safety risk value weighted by the historical accident rate; S3.3: Use a deep reinforcement learning framework, input user portraits and real-time environmental status, and output dynamic weight coefficients; the specific steps are as follows: S3.3.1: Construct a state space S, which includes user portrait features u = [u1, u2,…, u n ] and the real-time environment state e=[e1,e2,·,e m ]; S3.3.2: Define the action space A as the weight coefficient vector [α, β, γ], satisfying α + β + γ = 1; S3.3.3: Design the reward function R = -λ1T - λ2E + λ3S, where λ1, λ2, and λ3 are preset constants; S3.3.4: Fit the policy function using a deep neural network and update the network parameters using experience replay and gradient descent. S3.4: Based on the improved A* algorithm, introduce a time window expansion mechanism to predict traffic conditions in the next 5 minutes and generate the global optimal path.

[0068] Furthermore, it should be noted that the elevation data has a resolution of 5 meters x 5 meters, with a vertical accuracy of ≤0.5 meters. By mapping elevation values ​​to road network nodes and combining them with road vector data, a three-dimensional topological network is constructed that includes altitude and slope. User profile features include rapid acceleration frequency, average vehicle speed, and historical route preferences; environmental conditions include real-time road conditions, weather, and three-dimensional road network slope. The state vector is a 12-dimensional vector. A dual actor-critic network architecture is employed: the actor network (two fully connected layers, each with 64 neurons) outputs the weighting strategy, while the critic network evaluates the state value. The time window expansion mechanism uses an LSTM neural network to predict road conditions for the next five minutes. Input data includes: historical 15-minute road section speeds; time characteristics such as weekdays / weekends, morning and evening rush hours; and weather warning information (for factors affecting traffic efficiency, such as rain and snow). The prediction error is kept within 10%, providing forward-looking data for route planning.

[0069] In this embodiment, it should also be noted that the differential road condition compression transmission mechanism based on the Beidou short message protocol in S4 performs closed-loop interaction between user obstacle avoidance area marking and system alternative path generation in the AR interface, and optimizes parameters through tactile and voice multimodal feedback. The specific steps are as follows: S4.1: Using the CS-SLP compressed sensing algorithm, the key information of the congested area coordinates and accident type is encoded into a binary sequence, and transmitted to the remote area terminal through the Beidou short message; the specific steps are as follows: S4.1.1: Represent the congested area coordinates as a two-dimensional grid matrix Non-zero elements represent congested areas; S4.1.2: Design measurement matrix Perform linear projection on the matrix X to generate the measurement vector y = φX; S4.1.3: Reconstruct the sparse signal using the orthogonal matching pursuit algorithm Satisfy min||X||0s·t·y=φX; S4.1.4: reconstruct the signal The accident type is encoded as a binary symbol sequence B, which is further compressed using Huffman coding. S4.2: The user uses gestures on the AR interface to mark the avoidance area. The system generates an alternative path within 1.5 seconds using a Monte Carlo tree search, with a path deviation of ≤5%. S4.3: Path correction results are synchronized in real time through voice prompts, haptic vibrations, and AR real-world arrow overlays. User operation data is collected to update model parameters.

[0070] Furthermore, it should be noted that the search strategy is as follows: ① Selection phase: Select high-potential nodes based on the confidence upper bound formula: Where Q(s,a) is the action value, N(s) is the number of node visits, N(s,a) is the number of action executions, and c is the exploration coefficient. ② Expansion and Simulation Phase: Randomly simulate paths to the endpoint and evaluate the cumulative cost. ③ Backpropagation: Update the node values ​​along the path and prioritize the optimal branch. ④ Performance Guarantee: Through a pruning strategy, such as setting a maximum search depth of 10 layers, we ensure path generation within 1.5 seconds and a path deviation of ≤5%. Voice Prompt: Using Text-to-Speech technology, the announcement includes: ① the reason for the path correction (e.g., "Accident ahead, detour recommended"); ② the estimated time savings of the new path. Haptic Feedback: Steering wheel vibration pattern encoding: ① high-frequency short vibration: impending turn signal; ② low-frequency long vibration: emergency obstacle avoidance warning. AR Reality Enhancement: Overlays a semi-transparent arrow and distance label, with arrow width positively correlated with priority (arrow width increases by 50% for emergency paths).

[0071] In this embodiment, it should also be noted that the hardware and software integration deployment is completed in S5, and the system performance and reliability are verified through simulation and field tests. The specific steps are as follows: S5.1: Integrate Qualcomm Snapdragon 820A chip, Beidou / GPS multi-mode receiver, 9-axis IMU, and lidar interface, and support 5G / short message dual communication modes; S5.2: Deploy edge computing modules to process real-time data, and run reinforcement learning model training on cloud servers; S5.3: Simulate five types of scenarios in cities and mountainous areas on the CARLA platform to verify the path planning response time and re-planning accuracy; S5.4: Accumulate 2,000 kilometers of road tests, covering extreme environments such as heavy rain and tunnels, and record the average time between failures.

[0072] Furthermore, it should be noted that the Qualcomm Snapdragon 820A chip: adopts a 14nm process, integrates a quad-core Kryo CPU (main frequency 2.2GHz) and an Adreno 530GPU, supports parallel processing of high-load tasks such as lidar point clouds and visual images, and reduces power consumption by 30% compared to the previous generation; built-in DSP coprocessor for real-time data preprocessing (such as sensor signal filtering). Beidou / GPS multi-mode receiver: supports BDS B1I / B2I / B3I and GPS L1 / L2 frequency bands, cold start positioning time ≤30 seconds, compatible with NMEA0183 protocol and Beidou short message protocol, and communicates with the main control chip through the SPI interface. 9-axis IMU: integrates a three-axis accelerometer (range ±16g, resolution 16 bits), a three-axis gyroscope (range ±2000° / s), and a three-axis magnetometer, providing attitude solution data, an update frequency of 100Hz, and zero bias stability ≤50μ g . 5 types of scenarios: urban CBD, mountain roads, rural roads, expressway sections, and complex intersections.

[0073] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0074] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. An intelligent navigation path dynamic planning method based on multi-source data fusion, characterized in that: The following steps are involved: S1: Collect and preprocess BeiDou positioning, sensor, and traffic information data, and generate standardized inputs through Kalman filtering, noise reduction, and spatiotemporal alignment; S2: Based on BeiDou RTK high-precision positioning and multi-source data fusion, it adopts conflict data adaptive fusion and multi-modal failure compensation strategies to build a dynamic environment perception model; S3: Utilizing BeiDou elevation data and real-time traffic flow, a multi-objective reinforcement learning framework dynamically optimizes path weights. User profile analysis adjusts safety, efficiency, and energy consumption decision preferences in real time based on driving behavior characteristics. S4: Based on the Beidou short message protocol, a differential road condition compression transmission mechanism is used to implement closed-loop interaction between user obstacle avoidance area marking and system alternative path generation in the AR interface, and parameter optimization is achieved through tactile and voice multimodal feedback. S5: Complete the hardware and software integration deployment, and verify the system performance and reliability through simulation and field testing.

2. The intelligent navigation path dynamic planning method based on multi-source data fusion according to claim 1 is characterized in that: In S1, Beidou positioning, sensor, and traffic information data are collected and preprocessed, and standardized input is generated through Kalman filtering, noise reduction, and spatiotemporal alignment. The specific steps are as follows: S1.1: Collect raw positioning data through Beidou multi-frequency receiver; S1.2: Synchronously acquire vehicle-mounted sensor data, including acceleration / angular velocity from the MEMS inertial measurement unit, lidar point cloud, and visual camera images; S1.3: Access to third-party traffic platforms for real-time traffic conditions, accidents, and construction information; S1.4: Use extended Kalman filtering to eliminate multipath errors on BeiDou data and perform dynamic trajectory smoothing in combination with IMU data. S1.5: Align multi-source data using a timestamp synchronization algorithm and convert BeiDou WGS-84 coordinates to a local plane coordinate system using UTM projection transformation.

3. The method for dynamic planning of intelligent navigation paths based on multi-source data fusion according to claim 2, characterized in that: The BeiDou multi-frequency receiver supports B1I / B2I / B3I multi-band signal reception, meeting the signal strength range of -130dBm to -100dBm.

4. The method for dynamic planning of intelligent navigation paths based on multi-source data fusion according to claim 1, characterized in that: In S2, based on BeiDou RTK high-precision positioning and multi-source data fusion, conflict data adaptive fusion and multi-modal failure compensation strategies are adopted to build a dynamic environment perception model. The specific steps are as follows: S2.1: Use ground-based augmentation stations to obtain differential corrections to achieve centimeter-level real-time positioning with a lateral accuracy of ≤0.1m and an elevation accuracy of ≤0.2m; S2.2: In areas where satellite signals are blocked, the BeiDou / INS tight integration algorithm is used to maintain positioning continuity by fusing inertial data through the federated Kalman filter. S2.3: Use the evidence theory weighted fusion algorithm to model the conflict probability between Beidou positioning and lidar obstacle detection results, and automatically assign data credibility weights; S2.4: Based on the fused positioning and sensor data, construct a dynamic grid map that includes road geometry and real-time status.

5. The method for dynamic planning of intelligent navigation paths based on multi-source data fusion according to claim 4, characterized in that: In S2.3, the weighted fusion algorithm based on evidence theory is used to model the conflict probability between Beidou positioning and lidar obstacle detection results, and automatically assign data credibility weights. The specific steps are as follows: S2.3.1: Generate a position probability distribution m1(A) for BeiDou positioning data and an obstacle presence probability distribution m2(B) for LiDAR obstacle detection results; S2.3.2: Calculate the probability of conflict between two data sources S2.3.3: Dynamically adjust the credibility weight according to the conflict probability k Where w1+w2=1.

6. The method for dynamic planning of intelligent navigation paths based on multi-source data fusion according to claim 1, characterized in that: S3 utilizes BeiDou elevation data and real-time traffic flow to dynamically optimize path weights through a multi-objective reinforcement learning framework. User profile analysis adjusts safety, efficiency, and energy consumption decision preferences in real time based on driving behavior characteristics. The specific steps are as follows: S3.1: Use BeiDou elevation data to construct a three-dimensional road network including slope and altitude, with nodes as intersections and energy consumption coefficients added to edge attributes; S3.2: Establish an optimization function J = αT + βE + γS, where T is the travel time after adjusting for real-time traffic conditions, E is the energy consumption calculated based on the vehicle dynamics model F = m·g·sinθ, and S is the safety risk value weighted by historical accident rates. S3.3: Use a deep reinforcement learning framework to input user profiles and real-time environment status and output dynamic weight coefficients; S3.4: Based on the improved A* algorithm, a time window expansion mechanism is introduced to predict the traffic conditions in the next 5 minutes and generate the global optimal path.

7. The method for dynamic planning of intelligent navigation paths based on multi-source data fusion according to claim 6, characterized in that: S3.3 uses a deep reinforcement learning framework to input user profiles and real-time environment status and output dynamic weight coefficients. The specific steps are as follows: S3.3.1: Construct the state space S, which contains the user profile features u=[u1,u2,…,u n ] and the real-time environment state e=[e1,e2,…,e m ]; S3.3.2: Define the action space A as the weight coefficient vector [α, β, γ], satisfying α + β + γ = 1; S3.3.3: Design a reward function R = -λ1T - λ2E + λ3S, where λ1, λ2, and λ3 are preset constants. S3.3.4: Fit the policy function through a deep neural network and update the network parameters using experience replay and gradient descent algorithms.

8. The method for dynamic planning of intelligent navigation paths based on multi-source data fusion according to claim 1, characterized in that: In S4, based on the Beidou short message protocol, a differential road condition compression transmission mechanism is used to perform closed-loop interaction between user obstacle avoidance area marking and system alternative path generation in the AR interface, and parameters are optimized through tactile and voice multimodal feedback. The specific steps are as follows: S4.1: Using the CS-SLP compressed sensing algorithm, key information such as the coordinates of the congested area and the type of accident is encoded into a binary sequence and transmitted to terminals in remote areas via Beidou short messages; S4.2: The user uses gestures in the AR interface to mark the avoidance area, and the system generates an alternative path within 1.5 seconds based on Monte Carlo tree search, with a path deviation of ≤5%; S4.3: Synchronize path correction results in real time through voice prompts, tactile vibrations, and AR real-scene arrow overlays, and collect user operation data to update model parameters.

9. The intelligent navigation path dynamic planning method based on multi-source data fusion according to claim 8 is characterized in that: S4.1 uses the CS-SLP compressed sensing algorithm to encode key information such as the coordinates of the congested area and the accident type into a binary sequence, which is then transmitted to terminals in remote areas via Beidou short messages. The specific steps are as follows: S4.1.1: Represent the congestion area coordinates as a two-dimensional grid matrix The non-zero elements represent congested areas; S4.1.2: Design the measurement matrix Perform linear projection on the matrix X to generate the measurement vector y = φX; S4.1.3: Reconstructing sparse signals using the orthogonal matching pursuit algorithm satisfies min||X||0s·t·y=φX; S4.1.4: Reconstruct the signal The accident type is encoded as a binary symbol sequence B, which is further compressed by Huffman coding.

10. The intelligent navigation path dynamic planning method based on multi-source data fusion according to claim 1, characterized in that: In S5, hardware and software integration deployment is completed, and system performance and reliability are verified through simulation and field testing. The specific steps are as follows: S5.1: Integrates a Qualcomm Snapdragon 820A chip, BeiDou / GPS multi-mode receiver, 9-axis IMU, and lidar interface, and supports 5G / short message dual communication modes; S5.2: Deploy edge computing modules to process real-time data, and cloud servers to run reinforcement learning model training; S5.3: Simulate five scenarios in urban and mountainous areas on the CARLA platform to verify path planning response time and replanning accuracy. S5.4: A total of 2,000 kilometers of road testing, covering extreme environments such as heavy rain and tunnels, and recording the mean time between failures.

Images