Bluetooth positioning methods, systems, devices, storage media, and software products

By acquiring environmental interference data and optimizing the ranging parameter combination using parameter optimization algorithms, the problem of insufficient accuracy of Bluetooth positioning in complex environments is solved, achieving higher precision and more stable Bluetooth positioning results.

CN123124086APending Publication Date: 2026-11-17CHONGQING CHANGAN AUTOMOBILE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611540989.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-30
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing Bluetooth positioning methods suffer from low positioning accuracy in some scenarios, especially in complex environments where Bluetooth signals are easily affected by multipath reflections, blockages, and electromagnetic interference, leading to inaccurate ranging results.

Method used

By acquiring environmental interference data of the vehicle, the environmental complexity index is determined, and the initial ranging parameter combination is optimized using a parameter optimization algorithm to obtain the target ranging parameter combination, thereby realizing the adaptive adjustment of the ranging parameters and dynamically adapting to the real-time environmental interference state.

Benefits of technology

It significantly improves the accuracy and stability of Bluetooth positioning, reduces the adverse effects of multipath interference, occlusion interference and electromagnetic interference on ranging results, and improves the adaptability and accuracy of positioning in complex scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123124086A_ABST
    Figure CN123124086A_ABST
Patent Text Reader

Abstract

This invention provides a Bluetooth positioning method, system, device, storage medium, and program product, relating to the field of Bluetooth positioning technology. After establishing a Bluetooth communication connection between the user device and the vehicle, the Bluetooth positioning method collects various types of interference information in the vehicle's current environment, constructs an environmental complexity index, and adaptively optimizes multiple initial ranging parameter combinations based on this index to determine the ranging parameter combination that matches the current environment before performing Bluetooth positioning on the user device. This solution can improve the accuracy, stability, and environmental adaptability of digital key positioning in complex scenarios, reducing problems such as mis-locking and locking delays caused by environmental fluctuations and terminal differences.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of Bluetooth positioning technology, and in particular to a Bluetooth positioning method, system, device, storage medium, and program product. Background Technology

[0002] With the rapid development of intelligent vehicle technology, digital keys are gradually replacing traditional mechanical keys and remote keys, becoming the core solution for keyless entry systems. Digital keys use Bluetooth technology to enable communication between the vehicle and the user's device, supporting seamless interaction functions (e.g., automatically unlocking the doors, activating welcome lights and ambient lighting when the user approaches the vehicle, or automatically locking the doors when the user moves away). This function relies on high-precision positioning technology to ensure accurate distance determination between the user and the vehicle in complex environments (such as underground parking lots, open-air parking lots, and open outdoor areas).

[0003] Currently, in-vehicle digital keys typically employ Bluetooth 6.0 Channel Sounding (CS) ranging, relying on ranging observations from multiple vehicle anchor points to achieve Bluetooth positioning. CS ranging primarily utilizes Phase-Based Ranging (PBR) technology for distance measurement. Specifically, firstly, the user device initiating the ranging sequentially selects multiple different carrier frequencies and sends detection signals to the in-vehicle reflective device acting as the reflector. Upon receiving a detection signal at the same carrier frequency, the in-vehicle reflective device sends a signal back to the user device, completing the signal exchange at the current carrier frequency before switching to the next carrier frequency and repeating the detection process. Secondly, based on the phase difference between the transmitted and received signals at multiple carrier frequencies, the user device or the in-vehicle reflective device eliminates phase integer ambiguity and calculates the straight-line distance between the user device and a single in-vehicle reflective device. If multiple in-vehicle reflective devices are deployed, the spatial coordinates of the user device are calculated by combining multiple sets of distance observations to achieve Bluetooth positioning. Finally, the vehicle digital key controller executes corresponding vehicle control commands based on the automaker's preset welcome / departure distance thresholds or spatial determination rules.

[0004] However, the inventors discovered that the above Bluetooth positioning method has a problem of low positioning accuracy in some scenarios. Summary of the Invention

[0005] This invention provides a Bluetooth positioning method, system, device, storage medium, and program product to solve the problem that the Bluetooth positioning method using CS ranging and relying on the ranging observation values ​​of multiple vehicle anchor points has low positioning accuracy in some scenarios.

[0006] In a first aspect, the present invention provides a Bluetooth positioning method, comprising: responding to a user device loaded with a vehicle digital key establishing a Bluetooth communication connection with a vehicle, acquiring current environmental interference data and multiple sets of initial ranging parameter combinations of the vehicle, wherein the current environmental interference data includes interference data corresponding to at least one of multipath interference, occlusion interference, and electromagnetic interference; determining a current environmental complexity index based on the current environmental interference data; performing parameter optimization processing on the multiple sets of initial ranging parameter combinations based on the current environmental complexity index to obtain a target ranging parameter combination; and performing Bluetooth positioning of the user device based on the target ranging parameter combination.

[0007] In one possible implementation, determining the current environmental complexity index based on the current environmental interference data includes: identifying the environmental interference index corresponding to the current environmental interference data, wherein the environmental interference index includes at least one of multipath interference index, electromagnetic interference index, and occlusion interference index; and calculating the current environmental complexity index based on the environmental interference index.

[0008] In one possible implementation, the current environmental complexity index is calculated based on the environmental interference index, including: weighting the multipath interference index, electromagnetic interference index and occlusion interference index based on preset weighting coefficients to obtain the current environmental complexity index.

[0009] In one possible implementation, based on the current environmental complexity index, parameter optimization is performed on multiple initial ranging parameter combinations to obtain the target ranging parameter combination. This includes: combining the current environmental complexity index with multiple initial ranging parameter combinations to generate multiple ranging parameter combinations to be evaluated; inputting each of the multiple ranging parameter combinations to be evaluated into a parameter selection model for prediction, obtaining the prediction results output by the parameter selection model, the prediction results including the predicted ranging error corresponding to each ranging parameter combination to be evaluated; determining multiple candidate ranging parameter combinations based on the predicted ranging error corresponding to each ranging parameter combination to be evaluated; and performing parameter optimization on the multiple candidate ranging parameter combinations to obtain the target ranging parameter combination.

[0010] In one possible implementation, parameter optimization is performed on multiple sets of candidate ranging parameter combinations to obtain the target ranging parameter combination. This includes: using an optimization algorithm to iteratively optimize multiple sets of candidate ranging parameter combinations to obtain the target ranging parameter combination; wherein the optimization algorithm completes the iterative optimization according to the optimization rules corresponding to the environmental complexity level, and the environmental complexity level is positively correlated with the environmental complexity index.

[0011] In one possible implementation, the optimization rule adaptively adjusts the ranging parameters according to the level of environmental complexity. The ranging parameters include at least one of the following: single-round ranging duration, total ranging duration, connection interval, number of ranging frequency points, channel state information sampling interval, and antenna configuration index.

[0012] The environmental complexity level includes at least one of Level 1, Level 2, and Level 3 complexity environments. The optimization rule for Level 1 complexity environments prioritizes reducing terminal power consumption and reduces the configuration of ranging parameters. The optimization rule for Level 2 complexity environments prioritizes balancing terminal power consumption and ranging accuracy and configures ranging parameters to a moderate level. The optimization rule for Level 3 complexity environments prioritizes ensuring ranging accuracy and enhances the configuration of ranging parameters.

[0013] In one possible implementation, the prediction result also includes the parameter importance weights corresponding to each ranging parameter in the candidate ranging parameter combination. The parameter importance weights are used to reflect the importance of the ranging parameter's influence on the ranging error.

[0014] An optimization algorithm is used to iteratively optimize multiple sets of candidate ranging parameter combinations to obtain the target ranging parameter combination. This includes: selecting multiple target ranging parameters from all ranging parameters corresponding to the candidate ranging parameter combinations according to the parameter importance weights of each ranging parameter; constructing a set of parameter points to be evaluated based on the multiple target ranging parameters; and using an optimization algorithm to iteratively optimize the set of parameter points to be evaluated to obtain the target ranging parameter combination.

[0015] In one possible implementation, a set of parameter points to be evaluated is constructed based on multiple target ranging parameters, including:

[0016] For each target ranging parameter among multiple target ranging parameters, the range is expanded according to the upper and lower limits of the parameter values ​​of the target ranging parameters in each candidate ranging parameter combination to obtain the parameter sampling space corresponding to the target ranging parameter. Uniform sampling is performed within the parameter sampling space corresponding to each target ranging parameter to obtain multiple parameter sampling values ​​corresponding to each target ranging parameter. These multiple parameter sampling values ​​are combined to generate multidimensional sampling points. For the other ranging parameters in the candidate ranging parameter combinations besides the target ranging parameters, the known ranging parameter combination with the smallest Euclidean distance to the multidimensional sampling points is found among all candidate ranging parameter combinations, and the known parameter values ​​corresponding to the remaining ranging parameters in the known ranging parameter combinations are extracted. The known parameter values ​​corresponding to the remaining ranging parameters are combined with the parameter sampling values ​​corresponding to the target ranging parameters in the multidimensional sampling points to obtain the parameter point set to be evaluated.

[0017] In one possible implementation, the optimization algorithm is a Bayesian optimization algorithm. The surrogate model of the Bayesian optimization algorithm is constructed using a Gaussian kernel function that incorporates the environmental complexity index. The surrogate model is used to output the predicted mean and predicted variance of each parameter point in the set of parameters to be evaluated during the iterative optimization process.

[0018] In one possible implementation, an optimization algorithm is used to iteratively optimize the set of parameter points to be evaluated to obtain the target ranging parameter combination, including: calculating the predicted mean and predicted variance corresponding to each parameter point to be evaluated according to the surrogate model; using a sliding window of a preset size combined with the expected boost function, iteratively sampling the set of parameter points to be evaluated according to the predicted mean and predicted variance until the preset iteration termination condition is met to obtain the target parameter point set to be evaluated;

[0019] In the set of target parameter points to be evaluated, the range parameter combination with the smallest predicted range error is selected as the target range parameter combination.

[0020] In one possible implementation, the parameter selection model is a random forest model, which is trained as follows: a training dataset is obtained, which contains multiple sets of training samples, each set of training samples including a ranging parameter vector, an environment complexity index, and a true value of the ranging error; based on the training dataset, the pre-trained random forest model with configured hyperparameters is iteratively trained to obtain the trained random forest model.

[0021] In one possible implementation, the pre-trained random forest model includes multiple decision trees. Based on the training dataset, the pre-trained random forest model with configured hyperparameters is iteratively trained to obtain a trained random forest model. This includes: for each decision tree in the pre-trained random forest model, performing the following independent training: sampling data from the training dataset to obtain single-tree training samples corresponding to the current decision tree; using the random subspace method to sample the ranging parameter vector of the single-tree training samples to obtain sampled features; forming a feature vector based on the sampled features and the environmental complexity index; and performing node splitting training on the current decision tree based on the feature vector to obtain a trained single decision tree.

[0022] In one possible implementation, the current decision tree is trained by node splitting based on the feature vector to obtain a trained single decision tree. This includes: traversing all candidate split points corresponding to all features in the feature vector, using the minimum overall root mean square error after node splitting as the splitting criterion, and recursively completing node splitting layer by layer; stopping node splitting and generating leaf nodes when the node depth reaches a preset maximum depth or the number of samples in the node is less than or equal to a preset threshold; the leaf node outputs the average ranging error of the corresponding training samples, and uses the average ranging error as the predicted ranging error value of the current decision tree.

[0023] In one possible implementation, during the node splitting training of the current decision tree based on the feature vector, the method further includes: calculating the sum of the root mean square error reductions of each feature across all split nodes, and using the sum of the root mean square error reductions as the parameter importance weight of the feature in the current decision tree; and for all decision trees in the random forest model, using the average of the parameter importance weights of the same feature in each decision tree as the parameter importance weight of the feature in the random forest model.

[0024] In one possible implementation, the hyperparameters of the pre-trained random forest model are determined as follows: based on Latin hypercube sampling, multiple sets of hyperparameter combinations are generated within a first preset range of the number of decision trees and a second preset range of the depth of a single tree, and corresponding multiple sets of candidate random forest models are constructed; the out-of-bag error of each candidate random forest model is calculated, and the hyperparameters corresponding to the candidate random forest model with the smallest out-of-bag error are selected as the hyperparameters of the pre-trained random forest model.

[0025] In one possible implementation, it also includes: scheduling and executing the user device local training mode, incrementally training the random forest model based on the user's personalized training data, and obtaining a personalized random forest model adapted to the user.

[0026] Alternatively, the vehicle-mounted system can be scheduled to perform training, and the random forest model can be incrementally trained based on the user's personalized training data to obtain a personalized random forest model adapted to the user.

[0027] In a second aspect, the present invention provides a Bluetooth positioning device, comprising:

[0028] The acquisition module is used to acquire the vehicle's current environmental interference data and multiple sets of initial ranging parameters in response to the establishment of a Bluetooth communication connection between the user device loaded with the vehicle digital key and the vehicle. The current environmental interference data includes interference data corresponding to at least one of multipath interference, occlusion interference, and electromagnetic interference.

[0029] The determination module is used to determine the current environmental complexity index based on the current environmental interference data;

[0030] The parameter optimization module is used to optimize multiple sets of initial ranging parameter combinations based on the current environmental complexity index to obtain the target ranging parameter combination.

[0031] The Bluetooth positioning module is used to locate the user device via Bluetooth based on a combination of target ranging parameters.

[0032] In one possible implementation, the determining module is specifically used to: identify environmental interference indicators corresponding to the current environmental interference data based on the current environmental interference data, wherein the environmental interference indicators include at least one of multipath interference indicators, electromagnetic interference indicators, and occlusion interference indicators; and calculate the current environmental complexity index based on the environmental interference indicators.

[0033] In one possible implementation, the determining module is also used to: weight the multipath interference index, electromagnetic interference index and occlusion interference index based on preset weighting coefficients to obtain the current environmental complexity index.

[0034] In one possible implementation, the parameter optimization module is specifically used to: combine the current environmental complexity index with multiple initial ranging parameter combinations to generate multiple ranging parameter combinations to be evaluated; input the multiple ranging parameter combinations to be evaluated into the parameter selection model for prediction to obtain the prediction results output by the parameter selection model, the prediction results including the predicted ranging error corresponding to each ranging parameter combination to be evaluated; determine multiple candidate ranging parameter combinations based on the predicted ranging error corresponding to each ranging parameter combination to be evaluated; and perform parameter optimization processing on the multiple candidate ranging parameter combinations to obtain the target ranging parameter combination.

[0035] In one possible implementation, the parameter optimization module is further used to: employ an optimization algorithm to iteratively optimize multiple sets of candidate ranging parameter combinations to obtain the target ranging parameter combination; wherein, the optimization algorithm completes the iterative optimization according to the optimization rules corresponding to the environmental complexity level, and the environmental complexity level is positively correlated with the environmental complexity index.

[0036] In one possible implementation, the optimization rule adaptively adjusts the ranging parameters according to the level of environmental complexity. The ranging parameters include at least one of the following: single-round ranging duration, total ranging duration, connection interval, number of ranging frequency points, channel state information sampling interval, and antenna configuration index.

[0037] The environmental complexity level includes at least one of Level 1, Level 2, and Level 3 complexity environments. The optimization rule for Level 1 complexity environments prioritizes reducing terminal power consumption and reduces the configuration of ranging parameters. The optimization rule for Level 2 complexity environments prioritizes balancing terminal power consumption and ranging accuracy and configures ranging parameters to a moderate level. The optimization rule for Level 3 complexity environments prioritizes ensuring ranging accuracy and enhances the configuration of ranging parameters.

[0038] In one possible implementation, the prediction result also includes the parameter importance weights corresponding to each ranging parameter in the candidate ranging parameter combination. The parameter importance weights are used to reflect the importance of the ranging parameter to the ranging error. The parameter optimization module is also used to: select multiple target ranging parameters from all ranging parameters corresponding to the candidate ranging parameter combination according to the parameter importance weights corresponding to each ranging parameter; construct a set of parameter points to be evaluated based on the multiple target ranging parameters; and use an optimization algorithm to iteratively optimize the set of parameter points to be evaluated to obtain the target ranging parameter combination.

[0039] In one possible implementation, the parameter optimization module is further configured to: for each target ranging parameter among multiple target ranging parameters, expand the interval according to the upper and lower limits of the parameter values ​​of the target ranging parameters in each set of candidate ranging parameter combinations to obtain the parameter sampling space corresponding to the target ranging parameter; uniformly sample within the parameter sampling space corresponding to each target ranging parameter to obtain multiple parameter sampling values ​​corresponding to each target ranging parameter; combine the multiple parameter sampling values ​​corresponding to each target ranging parameter to generate multidimensional sampling points; for other ranging parameters in the candidate ranging parameter combinations besides the target ranging parameters, find the known ranging parameter combination with the smallest Euclidean distance to the multidimensional sampling points among all candidate ranging parameter combinations, and extract the known parameter values ​​corresponding to the remaining ranging parameters in the known ranging parameter combinations; combine the known parameter values ​​corresponding to the remaining ranging parameters with the parameter sampling values ​​corresponding to the target ranging parameters in the multidimensional sampling points to obtain the parameter point set to be evaluated.

[0040] In one possible implementation, the optimization algorithm is a Bayesian optimization algorithm. The surrogate model of the Bayesian optimization algorithm is constructed using a Gaussian kernel function that incorporates the environmental complexity index. The surrogate model is used to output the predicted mean and predicted variance of each parameter point in the set of parameters to be evaluated during the iterative optimization process.

[0041] In one possible implementation, the parameter optimization module is further used to: calculate the predicted mean and predicted variance corresponding to each parameter point to be evaluated based on the surrogate model; use a sliding window of a preset size combined with the expected lifting function to iteratively sample the set of parameter points to be evaluated based on the predicted mean and predicted variance until the preset iteration termination condition is met to obtain the target set of parameter points to be evaluated; and select the ranging parameter combination with the smallest predicted ranging error as the target ranging parameter combination from the target set of parameter points to be evaluated.

[0042] In one possible implementation, the parameter selection model is a random forest model, which is trained as follows: a training dataset is obtained, which contains multiple sets of training samples, each set of training samples including a ranging parameter vector, an environment complexity index, and a true value of the ranging error; based on the training dataset, the pre-trained random forest model with configured hyperparameters is iteratively trained to obtain the trained random forest model.

[0043] In one possible implementation, the pre-trained random forest model includes multiple decision trees. Based on the training dataset, the pre-trained random forest model with configured hyperparameters is iteratively trained to obtain a trained random forest model. This includes: for each decision tree in the pre-trained random forest model, performing the following independent training: sampling data from the training dataset to obtain single-tree training samples corresponding to the current decision tree; using the random subspace method to sample the ranging parameter vector of the single-tree training samples to obtain sampled features; forming a feature vector based on the sampled features and the environmental complexity index; and performing node splitting training on the current decision tree based on the feature vector to obtain a trained single decision tree.

[0044] In one possible implementation, the current decision tree is trained by node splitting based on the feature vector to obtain a trained single decision tree. This includes: traversing all candidate split points corresponding to all features in the feature vector, using the minimum overall root mean square error after node splitting as the splitting criterion, and recursively completing node splitting layer by layer; stopping node splitting and generating leaf nodes when the node depth reaches a preset maximum depth or the number of samples in the node is less than or equal to a preset threshold; the leaf node outputs the average ranging error of the corresponding training samples, and uses the average ranging error as the predicted ranging error value of the current decision tree.

[0045] In one possible implementation, during the node splitting training of the current decision tree based on the feature vector, the method further includes: calculating the sum of the root mean square error reductions of each feature across all split nodes, and using the sum of the root mean square error reductions as the parameter importance weight of the feature in the current decision tree; and for all decision trees in the random forest model, using the average of the parameter importance weights of the same feature in each decision tree as the parameter importance weight of the feature in the random forest model.

[0046] In one possible implementation, the hyperparameters of the pre-trained random forest model are determined as follows: based on Latin hypercube sampling, multiple sets of hyperparameter combinations are generated within a first preset range of the number of decision trees and a second preset range of the depth of a single tree, and corresponding multiple sets of candidate random forest models are constructed; the out-of-bag error of each candidate random forest model is calculated, and the hyperparameters corresponding to the candidate random forest model with the smallest out-of-bag error are selected as the hyperparameters of the pre-trained random forest model.

[0047] In one possible implementation, the Bluetooth positioning device further includes a scheduling execution module (not shown), which is used to: schedule and execute the user equipment local training mode, incrementally train the random forest model based on the user's personalized training data, and obtain a personalized random forest model adapted to the user.

[0048] Alternatively, the vehicle-mounted system can be scheduled to perform training, and the random forest model can be incrementally trained based on the user's personalized training data to obtain a personalized random forest model adapted to the user.

[0049] Thirdly, the present invention provides a Bluetooth positioning system, comprising: a cloud service platform, a vehicle terminal, and a user equipment terminal; the vehicle terminal includes a Bluetooth controller and a vehicle infotainment system, and the user equipment terminal is loaded with a vehicle digital key application program (APP).

[0050] The cloud service platform is used to respond to terminal model download requests by pushing the trained parameter selection model to the vehicle terminal or to the user device terminal.

[0051] A Bluetooth controller is used to execute the Bluetooth positioning method provided in the first aspect above in response to the establishment of a Bluetooth communication connection between the user equipment terminal and the vehicle terminal.

[0052] In one possible implementation, the cloud service platform includes a digital key service module, a remote control module, a data module, and an algorithm module;

[0053] The digital key service module communicates with the vehicle digital key APP, and is used to interact with the vehicle digital key APP for training data and model data, and provides a data transmission interface.

[0054] The remote control module communicates with the vehicle's infotainment system and is used to respond to the vehicle's model download request by pushing the trained parameter selection model to the vehicle's infotainment system.

[0055] The data module is used to organize and standardize the format of the collected data uploaded by users, and generate a training dataset that is suitable for model training.

[0056] The algorithm module is used to perform offline training and hyperparameter optimization of the parameter selection model based on the training dataset, resulting in a trained parameter selection model.

[0057] In one possible implementation, the vehicle digital key app includes a training module;

[0058] The training module is used to download the trained parameter selection model from the digital key service module and incrementally train the parameter selection model based on the user's personalized training data to obtain a personalized parameter selection model adapted to the user.

[0059] In one possible implementation, the vehicle's infotainment system downloads the trained parameter selection model from the remote control module and incrementally trains the parameter selection model based on the user's personalized training data to obtain a personalized parameter selection model adapted to the user.

[0060] Fourthly, the present invention provides a Bluetooth positioning device, comprising: a processor, and a memory communicatively connected to the processor;

[0061] The memory stores the instructions that the computer executes;

[0062] The processor executes computer execution instructions stored in memory to implement the Bluetooth positioning method as described in the first aspect above.

[0063] Fifthly, the present invention provides a vehicle comprising: at least one processor, and a memory communicatively connected to the processor;

[0064] The memory stores the instructions that the computer executes;

[0065] The processor executes computer execution instructions stored in memory to implement the Bluetooth positioning method as described in the first aspect above.

[0066] In a sixth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the Bluetooth positioning method provided in the first aspect above.

[0067] In a seventh aspect, the present invention provides a computer program product, comprising: a computer program that, when executed by a processor, implements the Bluetooth positioning method provided in the first aspect above.

[0068] This invention provides a Bluetooth positioning method, system, device, storage medium, and program product. The Bluetooth positioning method responds to a user device equipped with a vehicle digital key establishing a Bluetooth communication connection with the vehicle, acquiring the vehicle's current environmental interference data and multiple sets of initial ranging parameter combinations. The current environmental interference data includes interference data corresponding to at least one of multipath interference, obstruction interference, and electromagnetic interference. Based on the current environmental interference data, a current environmental complexity index is determined. Further, based on the current environmental complexity index, parameter optimization processing is performed on the multiple sets of initial ranging parameter combinations to obtain a target ranging parameter combination. Finally, based on the target ranging parameter combination, Bluetooth positioning of the user device is performed. This invention determines the current environmental complexity index based on current environmental interference data, and optimizes multiple initial ranging parameter combinations according to the current environmental complexity index to obtain the target ranging parameter combination. This enables adaptive matching of ranging parameters under different environmental interference scenarios, allowing the Bluetooth ranging and positioning process to dynamically adapt to the real-time environmental interference state. This effectively reduces the adverse effects of environmental factors such as multipath interference, occlusion interference, and electromagnetic interference on the ranging results, and significantly improves the Bluetooth positioning accuracy, operational stability, and adaptability to complex scenarios of user equipment relative to vehicles. Attached Figure Description

[0069] Figure 1 This is a schematic diagram of the structure of a Bluetooth positioning system provided in an embodiment of the present invention;

[0070] Figure 2 This is a schematic diagram of the structure of a Bluetooth positioning system provided in another embodiment of the present invention;

[0071] Figure 3 This is a schematic diagram of the structure of a Bluetooth positioning system provided in another embodiment of the present invention;

[0072] Figure 4 This is a flowchart illustrating a Bluetooth positioning method provided in an embodiment of the present invention;

[0073] Figure 5 A flowchart illustrating a Bluetooth positioning method provided in another embodiment of the present invention;

[0074] Figure 6 A flowchart illustrating a Bluetooth positioning method provided in another embodiment of the present invention;

[0075] Figure 7 This is a flowchart illustrating a Bluetooth positioning method provided in another embodiment of the present invention.

[0076] Figure 8 This is a flowchart illustrating the training method of the random forest model provided in an embodiment of the present invention;

[0077] Figure 9 This is a schematic diagram of the structure of the Bluetooth positioning device provided in an embodiment of the present invention;

[0078] Figure 10 This is a schematic diagram of the structure of a Bluetooth positioning device provided in an embodiment of the present invention.

[0079] The accompanying drawings have illustrated specific embodiments of the invention, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the invention in any way, but rather to illustrate the concept of the invention to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0080] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0081] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0082] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0083] The terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, product, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, product, or apparatus. Without further limitation, the presence of additional identical or equivalent elements in the process, method, product, or apparatus that includes elements is not excluded. For example, the use of terms such as "first," "second," etc., to indicate names does not imply any particular order.

[0084] In related technologies, vehicle digital keys typically employ Bluetooth 6.0 CS ranging and rely on ranging observations from multiple vehicle anchor points to achieve Bluetooth positioning. CS ranging primarily utilizes PBR (Progressive Backpropagation) technology for distance measurement. Specifically, firstly, the user device initiating the ranging sequentially selects multiple different carrier frequencies and sends detection signals to the vehicle-mounted reflective device acting as the reflector. Upon receiving a detection signal at the same carrier frequency, the vehicle-mounted reflective device sends a signal back to the user device, completing the signal exchange at the current carrier frequency before switching to the next carrier frequency and repeating the detection process. Secondly, based on the phase difference between the transmitted and received signals corresponding to multiple carrier frequencies, the user device or the vehicle-mounted reflective device eliminates phase integer ambiguity and calculates the straight-line distance between the user device and a single vehicle-mounted reflective device. If multiple vehicle-mounted reflective devices are deployed, the spatial coordinates of the user device are calculated by combining multiple sets of distance observations to achieve Bluetooth positioning. Finally, the vehicle digital key controller executes corresponding vehicle control commands according to the vehicle manufacturer's preset welcome / departure distance threshold or spatial determination rules.

[0085] However, the aforementioned Bluetooth positioning methods cannot dynamically adjust ranging parameters according to the user's actual environment, resulting in inaccurate positioning when using the same ranging parameters in different scenarios. For example, in environments such as underground parking garages, building perimeters, and outdoor parking lots, Bluetooth signals are easily affected by multipath reflections, vehicle or wall obstructions, and electromagnetic interference from surrounding electronic devices, causing significant fluctuations in ranging results. The inability to dynamically adjust fixed ranging parameters in sync with environmental changes leads to poor environmental adaptability, insufficient positioning stability, and low overall positioning accuracy for these Bluetooth positioning methods.

[0086] Based on the technical problems existing in the above-mentioned related technologies, the embodiments of the present invention introduce an environmental interference perception and adaptive adjustment mechanism for ranging parameters. The current environmental complexity index is determined according to the current environmental interference data, and multiple initial ranging parameter combinations are optimized according to the current environmental complexity index to obtain the target ranging parameter combination. This enables adaptive matching of ranging parameters under different environmental interference scenarios, allowing the Bluetooth ranging and positioning process to dynamically adapt to the real-time environmental interference state. This effectively reduces the adverse effects of environmental factors such as multipath interference, occlusion interference, and electromagnetic interference on the ranging results, and significantly improves the Bluetooth positioning accuracy, operational stability, and adaptability to complex scenarios of the user equipment relative to the vehicle.

[0087] The application scenarios of the embodiments of the present invention will be described below first.

[0088] The Bluetooth positioning method provided in this invention is applicable to the digital key system of smart cars, and enables seamless interaction between the vehicle and user equipment (such as automatic unlocking, locking, and welcoming) through Bluetooth positioning technology.

[0089] Figure 1This is a schematic diagram of the structure of a Bluetooth positioning system provided in an embodiment of the present invention. Figure 1 As shown, the Bluetooth positioning system includes a cloud service platform, a vehicle terminal, and a user equipment terminal. The vehicle terminal includes a Bluetooth controller and a vehicle infotainment system, while the user equipment terminal is equipped with a vehicle digital key app.

[0090] For example, a Bluetooth controller can also be described as a digital key controller. In this embodiment of the invention, it is uniformly described as a Bluetooth controller.

[0091] like Figure 1 As shown, the Bluetooth controller and the vehicle infotainment system (V2X) can establish a communication connection through a gateway (V2X gateway).

[0092] For example, user equipment terminals include, but are not limited to, portable mobile terminals, such as user mobile phones, user smart wearable devices, etc.

[0093] Among them, the cloud service platform is used to respond to terminal model download requests by pushing the trained parameter selection model to the vehicle terminal or to the user equipment terminal.

[0094] The Bluetooth controller is used to execute the Bluetooth positioning method provided above in response to the establishment of a Bluetooth communication connection between the user equipment terminal and the vehicle terminal.

[0095] For example, such as Figure 1 As shown, the vehicle digital key APP includes a ranging module and a training module that are interconnected. The ranging module is used to interact with the Bluetooth controller to transmit ranging parameters.

[0096] It is understood that in the Bluetooth positioning system provided in this embodiment of the invention, by setting up a collaborative architecture of cloud service platform, vehicle terminal and user equipment terminal, the trained parameter selection model can be distributed to the vehicle terminal or user equipment terminal as needed, making the model deployment method more flexible and adaptable to different computing power configurations and system architectures. After the Bluetooth controller establishes a Bluetooth communication connection between the user equipment terminal and the vehicle terminal, it executes the Bluetooth positioning method. It can select a matching combination of ranging parameters based on the complexity of the current environment, thereby reducing the impact of multipath reflection, obstruction and electromagnetic interference on the positioning results. This makes the vehicle's judgment of the user's approach, stay or leave more accurate, thereby improving the stability and real-time performance of digital key unlocking, locking and welcoming interaction, and reducing false triggering and missed triggering in complex scenarios.

[0097] Alternatively, in one possible implementation, the cloud service platform includes a digital key service module, a remote control module, a data module, and an algorithm module.

[0098] The digital key service module communicates with the vehicle digital key APP, and is used to interact with the vehicle digital key APP for training data and model data, and provides a data transmission interface.

[0099] For example, the digital key service module receives Bluetooth ranging data collected by the vehicle digital key APP in different parking environments and under different terminal conditions, and sends the trained parameter selection model data to the APP side interface, thereby establishing a stable data closed loop between the user device terminal and the cloud service platform.

[0100] The remote control module communicates with the vehicle's infotainment system and is used to respond to the vehicle's model download request by pushing the trained parameter selection model to the vehicle's infotainment system.

[0101] Understandably, when the remote control module receives a model download request from the vehicle's infotainment system, it pushes the trained parameter selection model to the system, enabling the vehicle terminal to load the optimized model in a timely manner and participate in Bluetooth positioning decisions. This improves the accuracy, stability, and scene adaptability of digital key positioning in complex environments, and reduces the probability of mis-locking, locking delays, and location misjudgments.

[0102] The data module is used to organize and standardize the collected data uploaded by users, and generate a training dataset suitable for model training.

[0103] For example, the data module can clean, unify fields, and standardize formats of the raw collected data, so that data from multiple sources, multiple models, and multiple scenarios can be used uniformly for model training.

[0104] The algorithm module is used to perform offline training and hyperparameter optimization of the parameter selection model based on the training dataset, resulting in a trained parameter selection model.

[0105] For example, the algorithm module performs offline training and hyperparameter optimization based on the training dataset, thereby obtaining a parameter selection model with stronger generalization ability to environmental complexity and parameter combination relationships.

[0106] Optionally, such as Figure 1 As shown, the vehicle digital key APP includes a training module, which is used to download the trained parameter selection model from the digital key service module and incrementally train the parameter selection model based on the user's personalized training data to obtain a personalized parameter selection model adapted to the user.

[0107] For example, the training module supports local model self-training.

[0108] In one specific implementation, the training module downloads the parameter selection model that has been trained offline from the digital key service module of the cloud service platform, and performs incremental training on the parameter selection model based on the personalized training data collected locally on the user's device to obtain a personalized parameter selection model adapted to the user.

[0109] For example, the vehicle digital key APP may also include a data module (not shown in the figure). In the local training mode of the user device, the data module is used to collect the ranging parameters and positioning results generated during the model self-training process, and report the collected data to the digital key service module, or for local personalized model training. At the same time, the trained personalized random forest model is sent to the Bluetooth controller of the vehicle terminal for storage and use via Bluetooth communication.

[0110] For example, the vehicle digital key APP may also include a positioning module (not shown in the figure), which is used to complete the Bluetooth positioning between the user device and the vehicle, and send the positioning result and positioning-related parameters to the data module of the vehicle digital key APP.

[0111] Figure 2 This is a schematic diagram of a Bluetooth positioning system provided in another embodiment of the present invention. After receiving the parameter selection model that has completed offline training from the digital key service module on the user device side, the training module can continue to train by combining the personalized training data formed by the user in actual vehicle use. This allows the model to further learn the user's corresponding device model, carrying location, usage habits, and common parking environment characteristics while retaining the ability to adapt to the general environment in the cloud. As a result, a more personalized parameter selection model that fits the actual positioning scenario can be formed for different individual users.

[0112] Understandably, through this incremental training method, the model can more accurately correct parameter selection under Bluetooth signal fluctuations, multipath reflections, obstructions, and electromagnetic interference. This allows the vehicle digital key APP to output a more suitable combination of ranging parameters after establishing a Bluetooth communication connection with the vehicle, thereby improving the positioning accuracy and stability of unlocking, locking, and welcome triggering, reducing the misjudgment rate in complex environments, and improving the seamless interaction experience.

[0113] Alternatively, in one possible implementation, the vehicle-mounted system is used to download the trained parameter selection model from the remote control module and incrementally train the parameter selection model based on the user's personalized training data to obtain a personalized parameter selection model adapted to the user.

[0114] For example, the parameter selection model downloaded by the vehicle's infotainment system from the remote control module can be transmitted to the Bluetooth controller via a gateway.

[0115] Figure 3This is a schematic diagram of the Bluetooth positioning system provided in another embodiment of the present invention. After the vehicle's infotainment system downloads the trained parameter selection model from the remote control module, it can first obtain the general parameter selection capability formed by the cloud based on multi-scenario training data, and then perform incremental training by combining the user's personalized training data accumulated during the long-term operation of the vehicle. This allows the model to retain its basic generalization ability while further learning the Bluetooth signal characteristics corresponding to the current user's device model, carrying location, proximity habits, and common parking environments.

[0116] Understandably, this incremental training method enables the personalized parameter selection model formed on the vehicle side to more accurately match ranging parameter combinations for different environmental complexities, thereby reducing positioning fluctuations in complex interference scenarios such as underground garages, building perimeters, or outdoor areas. This results in more timely unlocking responses when the user approaches the vehicle and more stable locking judgments when leaving the vehicle, thereby reducing false triggers and missed triggers and improving the accuracy, stability, and user adaptability of digital key Bluetooth positioning.

[0117] The following describes in detail the specific implementation of the Bluetooth positioning method provided by the present invention, taking the Bluetooth controller in the above-mentioned Bluetooth positioning system as the execution subject and in conjunction with specific embodiments.

[0118] Figure 4 This is a flowchart illustrating a Bluetooth positioning method according to an embodiment of the present invention. Figure 4 As shown, a specific implementation of this Bluetooth positioning method may include the following steps:

[0119] S401, in response to the establishment of a Bluetooth communication connection between a user device loaded with a vehicle digital key and the vehicle, the current environmental interference data of the vehicle and multiple sets of initial ranging parameter combinations are obtained. The current environmental interference data includes interference data corresponding to at least one of multipath interference, obstruction interference, and electromagnetic interference.

[0120] The vehicle digital key is a vehicle access control credential loaded in the user device, used to enable the user device to participate in the Bluetooth communication and positioning process on the vehicle side after authentication; the user device is a mobile terminal that has the vehicle digital key APP installed or stores the digital key credential, which in this invention can be a smartphone, smart wearable device, etc.; the vehicle is an in-vehicle object that performs environmental perception, ranging parameter processing and Bluetooth positioning; the Bluetooth communication connection is a communication link established between the user device and the vehicle, used to carry the interaction of positioning-related parameters.

[0121] For example, the current environmental interference data can be interference characterization data collected by the vehicle at the current moment, reflecting the Bluetooth propagation environment between the vehicle and the user device, including interference data corresponding to at least one of multipath interference, obstruction interference, and electromagnetic interference. Multipath interference refers to the interference type corresponding to phase shift, intensity fluctuation, or waveform distortion caused by the Bluetooth signal reaching the vehicle receiver through multiple propagation paths. It is used to characterize the interference characteristics during Bluetooth CS ranging, where the signal forms multiple propagation paths through reflection and refraction, thereby causing phase jitter and delay spread at the receiver and leading to ranging errors. Electromagnetic interference refers to the interference type that affects the Bluetooth channel quality due to surrounding wireless devices, power equipment, or complex electromagnetic environments. It is used to characterize the level of environmental signal-to-noise ratio degradation and reflect the degree of electromagnetic noise interference in the environment. Obstruction interference refers to the interference type corresponding to attenuation, diffraction, or additional reflection caused by people, walls, vehicle bodies, pillars, or other obstacles obstructing the direct Bluetooth path. It is used to characterize the degree of power attenuation of the direct or reflected Bluetooth signal, reflecting the attenuation level caused by physical obstruction of the direct signal.

[0122] For example, the interference data corresponding to multipath interference includes, but is not limited to, the standard deviation of the measured multipath phase deviation and the standard deviation of the phase of the pre-calibrated medium multipath level. Among them, the standard deviation of the multipath phase deviation is used to characterize the intensity of phase fluctuation introduced by multipath reflection. The larger the standard deviation of the multipath phase deviation, the stronger the multipath interference, the more severe the phase fluctuation, and the higher the environmental complexity.

[0123] The standard deviation of the multipath phase bias can be obtained as follows: Obtain the Channel State Information (CSI) output by Bluetooth 6.0 CS, and extract a set of subcarrier phases from the CSI (e.g., ...). (N is typically 20 to 40 subcarriers); based on the extracted set of subcarrier phases, the phase difference between adjacent subcarriers is calculated to eliminate the common phase components caused by the fixed phase offset between the transmitter and receiver and the carrier frequency offset; further, the standard deviation of the multipath phase deviation is calculated based on the phase difference sequence.

[0124] For example, the pre-calibrated phase standard deviation of the intermediate multipath level value can be 0.5 rad. This embodiment of the invention does not limit the value of the phase standard deviation of the intermediate multipath level value; it can be determined according to the actual application requirements.

[0125] For example, the interference data corresponding to electromagnetic interference includes, but is not limited to, the measured average signal-to-noise ratio (SNR) and the maximum SNR under ideal conditions in an anechoic chamber. Among them, the SNR can effectively reflect the electromagnetic noise level of the Bluetooth channel. The lower the measured average SNR, the more severe the environmental electromagnetic interference, and the higher the corresponding environmental complexity.

[0126] The measured average signal-to-noise ratio can be obtained in the following way: obtain the CSI output of Bluetooth 6.0 CS, read the signal-to-noise ratio value of each subcarrier from the CSI, and calculate the arithmetic average of the signal-to-noise ratios of all subcarriers to obtain the measured average signal-to-noise ratio under the current environment.

[0127] For example, the maximum signal-to-noise ratio under ideal conditions in an anechoic chamber can be 30 dB. This embodiment of the invention does not limit the value of the maximum signal-to-noise ratio under ideal conditions in an anechoic chamber; it can be determined according to actual application requirements.

[0128] For example, the interference data corresponding to obstruction interference includes, but is not limited to, the signal power of the direct path and the total signal power of the reflected path. Among them, the degree of attenuation of the signal power of the direct path can intuitively represent the strength of physical obstruction. The lower the proportion of the signal power of the direct path and the higher the proportion of the signal power of the reflected path, the more severe the obstruction of the current Bluetooth transmission link and the higher the corresponding environmental complexity.

[0129] The direct path signal power and the total reflected path signal power can be obtained as follows: Perform an inverse Fourier transform on the frequency domain CSI complex sequence output by Bluetooth 6.0 CS to obtain the time domain channel impulse response (CIR); identify the first main peak in the time domain channel impulse response, take it as the direct propagation path, and extract the corresponding peak power as the direct path signal power; sum the powers of the remaining multipath components with a delay greater than a preset delay threshold to obtain the total reflected path signal power.

[0130] For example, in multiple sets of initial ranging parameter combinations, the ranging parameters corresponding to each set of initial ranging parameter combinations include, but are not limited to, single-round ranging duration, total ranging duration, connection interval, number of ranging frequency points, channel state information sampling interval, and antenna configuration index. Specifically, the single-round ranging duration is the duration occupied by a single complete CS detection interaction process; the total ranging duration is the maximum allowed total time from the start to the end of a single ranging task; the connection interval is the time interval between adjacent connection events between Bluetooth master and slave devices; the number of ranging frequency points is the number of carrier frequency points used in a single round of CS ranging; the channel state information sampling interval is the time interval between two adjacent CSI acquisitions; and the antenna configuration index is a number used to select a preset antenna transceiver combination scheme, corresponding to different transceiver antenna configurations. It should be noted that all ranging parameters involved in this invention can include any one or more dimensions of the single-round ranging duration, total ranging duration, connection interval, number of ranging frequency points, channel state information sampling interval, and antenna configuration index.

[0131] For example, in one possible implementation, multiple sets of initial ranging parameter combinations can be obtained as follows: Based on the preset value range and preset sampling step size of each ranging parameter, Latin hypercube sampling is used to generate multiple sets of initial ranging parameter combinations. Specifically, for the six-dimensional ranging parameters (single-round ranging duration, total ranging duration, connection interval, number of ranging frequency points, channel state information sampling interval, and antenna configuration index), upper and lower limits for the corresponding values ​​of each ranging parameter and the sampling step size are set respectively; Latin hypercube sampling is used to perform layered sampling of the six-dimensional ranging parameter space to ensure that the sampling points of each dimension of the ranging parameter are evenly distributed within their respective value ranges, avoiding local clustering of sampling points, and achieving better coverage of the overall ranging parameter space with a limited number of samples, resulting in multiple sets of sampled ranging parameter vectors, each set of sampled ranging parameter vectors constituting a set of initial ranging parameter combinations.

[0132] For example, the total number of sampling samples can be set to generate a corresponding number of initial ranging parameter combinations. This embodiment of the invention does not specifically limit the number of sampling samples or the value range of each ranging parameter; it can be flexibly configured based on hardware computing power, real-time ranging requirements, and model training needs.

[0133] S402, Determine the current environmental complexity index based on the current environmental interference data.

[0134] Among them, the current environmental complexity index is a quantitative value that comprehensively represents the overall interference level of the environment in which Bluetooth positioning is located. It is used to transform the current environmental interference state into a quantitative input that can participate in the decision of ranging parameters.

[0135] For example, in one possible implementation, after acquiring the current environmental interference data, the data can be preprocessed to mitigate the adverse effects of inconsistencies in units, sampling frequencies, and timestamps between data from different sources. Then, based on the preprocessed current environmental interference data, a preset processing method is used to calculate the current environmental complexity index. In practical applications, the calculation process for the current environmental complexity index can be adaptively adjusted based on data integrity and dynamic environmental changes to further improve the stability of the output results.

[0136] For example, the current environmental complexity index can not only be output as a single continuous value, but can also be further discretized into complexity levels, such as three levels: low environmental complexity, medium environmental complexity, and high environmental complexity, or subdivided into multiple complexity ranges. The vehicle determines the complexity level of the current environment by comparing the environmental complexity score with a preset threshold range, and uses this complexity level as the basis for selecting subsequent parameter optimization strategies.

[0137] Understandably, in this step, by uniformly converting the current environmental interference data into the current environmental complexity index, the vehicle can perform parameter filtering and optimization based on a single quantitative entry point in subsequent processing. This reduces the processing dispersion caused by making decisions separately for different interference situations, and ensures that the parameter adjustment basis is consistent with the current scenario, thereby providing a foundation for stable positioning output in complex environments.

[0138] S403, based on the current environmental complexity index, performs parameter optimization on multiple initial ranging parameter combinations to obtain the target ranging parameter combination.

[0139] Among them, parameter optimization refers to adapting multiple initial ranging parameter combinations based on the current environmental complexity index in order to determine the target ranging parameter combination that is suitable for the current environment.

[0140] For example, the target ranging parameter combination is the final ranging parameter configuration actually used for Bluetooth positioning calculation after environmental adaptation.

[0141] For example, in one possible implementation, multiple sets of initial ranging parameter combinations are compared, filtered, or adapted based on the current environmental complexity index, so that the output target ranging parameter set is suitable for the current Bluetooth propagation environment. For instance, in scenarios with strong electromagnetic interference, the environmental complexity index value is high. In such cases, an initial ranging parameter combination with more ranging frequency points or a longer ranging duration can be selected as the target ranging parameter combination to increase the amount of sampling information and suppress ranging errors introduced by electromagnetic interference.

[0142] For example, in specific implementations, when optimizing multiple initial ranging parameter combinations based on the current environmental complexity index, the system dynamically selects the combination with the smallest ranging error as the target ranging parameter combination by comparing the predicted ranging errors of different initial ranging parameter combinations under the same environment. For instance, in an underground parking garage scenario where the environmental complexity index is high, the system prioritizes a ranging parameter combination with a ranging frequency (Nf) of 64 and a single-wheel ranging duration (Tr) of 300ms. Compared to the fixed parameters preset by the vehicle manufacturer (such as Nf=48, Tr=200ms), the ranging error is reduced by 15%. This process achieves adaptive optimization in complex environments through parameter optimization.

[0143] For example, environmental complexity indicators include multipath interference indicators, electromagnetic interference indicators, and occlusion interference indicators. The following explanation will focus on a scenario where a single type of interference is dominant.

[0144] For example, in scenarios with significant obstruction interference (e.g., the Bluetooth direct path is blocked by a vehicle body or metal obstacles) and moderate multipath interference and electromagnetic interference, the initial ranging parameter combination can be: single-wheel ranging time 300ms, total ranging time 90s, connection interval 30ms, ranging frequency points 30, channel state information sampling interval 20ms, and antenna configuration index 3. In this scenario, the multipath interference index is 0.6, the electromagnetic interference index is 0.3, and the obstruction interference index is 0.65. The current environmental complexity index is obtained by weighted fusion calculation of the three interference indices. For example, the multipath interference weight is 0.5, the electromagnetic interference weight is 0.3, and the obstruction interference weight is 0.2, resulting in a current environmental complexity index of 0.52.

[0145] In this scenario, the optimization objective is to achieve the optimal comprehensive evaluation index. Ranging-related parameters are adjusted based on the current environmental complexity index. Specifically, if the current environmental complexity index exceeds a preset threshold, the ranging time is extended to collect more ranging samples and obtain stable ranging data. Simultaneously, antenna configurations are switched to enable more antenna paths and avoid direct sunlight obstruction. Ranging-related parameters are iteratively adjusted, and the optimization iteration stops and the target ranging parameter combination is output when the optimization evaluation index meets a preset convergence condition. The comprehensive optimization evaluation index considers ranging error, ranging stability, and ranging resource overhead.

[0146] Understandably, the core challenge in scenarios with significant obstruction is that the signal power of the direct path is attenuated by obstacles. Therefore, it is preferable to use multiple receiving antenna paths for diversity reception. In this scenario, multipath interference and electromagnetic interference are not prominent, and the number of ranging frequency points does not need to be configured to the maximum value to avoid unnecessary RF power consumption.

[0147] For example, in scenarios with significant electromagnetic interference (e.g., numerous 2.4G WiFi and Bluetooth devices nearby) and moderate multipath interference and obstruction interference, the initial ranging parameter combination can be: single-round ranging duration 500ms, total ranging duration 30s, connection interval 30ms, ranging frequency points 60, channel state information sampling interval 20ms, and antenna configuration index 1. In this scenario, the multipath interference index is 0.7, the electromagnetic interference index is 0.82, and the obstruction interference index is 0.3. The current environmental complexity index is obtained by weighted fusion calculation of the three interference indices. For example, the multipath interference weight is 0.5, the electromagnetic interference weight is 0.3, and the obstruction interference weight is 0.2, resulting in a current environmental complexity index of 0.656.

[0148] In this scenario, the optimization objective is to achieve the optimal comprehensive evaluation index. Ranging-related parameters are adjusted based on the current environmental complexity index. Specifically, if the current environmental complexity index exceeds a preset threshold, the ranging duration per round is increased to collect more ranging samples; simultaneously, the number of ranging frequency points is increased, and electromagnetic interference is suppressed through multi-frequency diversity. Ranging-related parameters are iteratively adjusted, and when the optimization evaluation index meets the preset convergence condition, the optimization iteration stops and the target ranging parameter combination is output.

[0149] Understandably, strong electromagnetic noise reduces the signal-to-noise ratio (SNR) and introduces significant phase measurement noise. Therefore, it is advisable to increase the number of ranging frequencies and use multi-frequency ranging measurements to avoid electromagnetic noise with adjacent frequencies.

[0150] For example, in scenarios with significant multipath interference (e.g., underground parking garages, metal vehicle body reflections) and moderate electromagnetic and occlusion interference, the initial ranging parameter combination can be: single-wheel ranging duration 300ms, total ranging duration 30s, connection interval 30ms, ranging frequency points 60, channel state information sampling interval 20ms, and antenna configuration index 4. In this scenario, the multipath interference index is 1.7, the electromagnetic interference index is 0.2, and the occlusion interference index is 0.4. The current environmental complexity index is obtained by weighted fusion calculation of the three interference indices. For example, the multipath interference weight is 0.5, the electromagnetic interference weight is 0.3, and the occlusion interference weight is 0.2, resulting in a current environmental complexity index of 0.99.

[0151] In this scenario, the optimization objective is to achieve the optimal comprehensive evaluation index. Ranging-related parameters are adjusted based on the current environmental complexity index. Specifically, if the current environmental complexity index exceeds a preset complexity threshold, the number of ranging frequency points is increased, ranging sampling is completed on more subcarrier frequencies, and interference deviations at single frequencies are reduced. Simultaneously, available antenna paths are expanded to collect multi-dimensional spatial measurement data, providing sufficient sample support for backend multipath phase separation calculations. Ranging-related parameters are iteratively adjusted, and when the optimization evaluation index meets preset convergence conditions, the optimization iteration stops and the target ranging parameter combination is output. It is understandable that when multipath interference is dominant, the channel superimposed with complex reflection components, resulting in severe phase distortion. This necessitates relying on more subcarrier frequency points and more spatial phase samples to improve the resolution of the multipath separation algorithm, thereby effectively suppressing ranging jitter and offset caused by multipath interference.

[0152] For example, in an open and ideal scenario with low environmental complexity (no obvious obstacles and few surrounding interference sources), the multipath interference index is 0.3, the electromagnetic interference index is 0.15, and the occlusion interference index is 0.1. The current environmental complexity index is obtained by weighted fusion calculation of the three interference indices. For example, the weight of multipath interference is 0.5, the weight of electromagnetic interference is 0.3, and the weight of occlusion interference is 0.2, and the calculated current environmental complexity index is 0.22.

[0153] The environment in this scenario has low interference levels and good ranging channel conditions. There's no need to increase ranging duration, the number of ranging frequencies, or switch complex antenna configurations. The corresponding target ranging parameter combination can be: single-round ranging duration 300ms, total ranging duration 30s, connection interval 30ms, number of ranging frequencies 30, channel state information sampling interval 20ms, and antenna configuration index 1. Specifically, in an open and ideal scenario, the direct path signal quality is high, and multipath, electromagnetic, and obstruction interference are all at low levels. Using the basic ranging parameter combination directly ensures ranging accuracy while minimizing power consumption overhead from RF ranging. It's understandable that this step, by incorporating the current environmental complexity into the ranging parameter processing, allows the ranging parameter combination to be adapted to different environmental scenarios, reducing the ranging error shift caused by the same set of fixed ranging parameters in different environments.

[0154] S404 performs Bluetooth positioning of user equipment based on the combination of target ranging parameters.

[0155] Bluetooth positioning refers to the process by which a vehicle calculates the distance, orientation, or relative area relationship between a user device and the vehicle based on a combination of target ranging parameters, thereby determining the position of the user device relative to the vehicle and using this positional state in digital key control scenarios.

[0156] For example, in one possible implementation, after acquiring the target ranging parameter combination, the Bluetooth controller controls the Bluetooth communication module to continuously conduct communication sampling with the user equipment. The sampled data includes at least the received signal strength sequence, measurement results of each channel, communication timestamps, and necessary retry statistics. The Bluetooth controller preprocesses the raw sampled data based on the configured ranging parameters included in the target ranging parameter combination and performs signal feature calculation based on the target ranging parameter combination to obtain an estimated distance between the user equipment and the vehicle. When the vehicle is equipped with multiple Bluetooth antennas, multiple door measurement points, or multiple receiving channels, the multi-channel ranging measurement results can be fused to solve for the comprehensive distance or area score of the user equipment relative to the vehicle. After completing the distance estimation, the Bluetooth controller further performs position determination based on the ranging results. For example, the Bluetooth controller can compare the comprehensive distance with a preset area boundary threshold to determine whether the user equipment is in the far field outside the vehicle, the near field outside the vehicle, the driver's side, the passenger side, or the interior area; it can also identify the approach direction of the user equipment relative to the vehicle by combining the signal differences, temporal continuity, and area score change trends of different measurement points. If the system only needs to implement the functions of contactless unlocking and locking, the Bluetooth positioning output can be simplified to: whether the user device has entered the preset range and whether it has stayed in the range for a preset duration threshold; if the system needs to support functions such as welcome light illumination, trunk unlocking, and personalized welcome interaction, the vehicle outputs fine-grained location tags and sends them to the body controller to execute the corresponding interactive actions.

[0157] Alternatively, in one possible implementation, to improve the stability of the positioning output, the positioning results can be subjected to time-series smoothing processing within multiple consecutive sampling periods; the final position determination result is only updated when the positioning status of several consecutive sampling periods meets the consistency determination condition, thereby suppressing the problems of false triggering and missed triggering caused by single ranging fluctuations.

[0158] For example, in one possible implementation, after completing the Bluetooth positioning, the measurement quality information during the positioning process, the final target ranging parameter combination used, and the positioning output result can be written into the session record for subsequent parameter library updates or complexity model calibration. When the vehicle detects that the user equipment has left the predetermined range, or the Bluetooth communication connection is lost, the positioning session based on the target ranging parameter combination ends. If the current environmental interference data changes significantly during the session, the vehicle can re-trigger steps S401 to S403 to update the target ranging parameter combination and continue executing this step.

[0159] It is understandable that after this step is completed, the vehicle obtains the effective position result of the user equipment relative to the vehicle, and can use this position result for door unlocking, locking, welcoming function triggering, or function suppression. Based on the above analysis, it can be seen that this invention directly uses the combination of target ranging parameters that have undergone environmental adaptation and optimization for Bluetooth positioning, so that the positioning calculation no longer relies on a single fixed ranging model, but maintains a more stable distance estimation and position determination output when the actual communication environment changes.

[0160] In this embodiment of the invention, in response to a user device loaded with a vehicle digital key establishing a Bluetooth communication connection with the vehicle, the system acquires the vehicle's current environmental interference data and multiple sets of initial ranging parameter combinations. Based on the current environmental interference data, it determines the current environmental complexity index. Furthermore, based on the current environmental complexity index, it performs parameter optimization processing on the multiple sets of initial ranging parameter combinations to obtain a target ranging parameter combination. Then, based on the target ranging parameter combination, it performs Bluetooth positioning of the user device.

[0161] In this embodiment of the invention, after the vehicle establishes a Bluetooth communication connection with the user equipment, it first collects interference data corresponding to multipath interference, occlusion interference, and electromagnetic interference in the current scene, and converts it into a current environment complexity index. Then, based on the current environment complexity index, it performs optimization processing on multiple sets of initial ranging parameter combinations to finally obtain a target ranging parameter combination that adapts to the current environment. The user equipment's position relative to the vehicle is determined using this target ranging parameter combination. This enables the ranging parameters to be adjusted synchronously with environmental changes during Bluetooth positioning, limiting the fluctuation of ranging error in complex environments to the current scene adaptation range. Consequently, the unlocking, locking, and welcoming trigger determination in the digital key scenario maintains higher consistency with the user's actual position, significantly improving the Bluetooth positioning accuracy, operational stability, and adaptability of the user equipment relative to the vehicle.

[0162] The following is combined Figure 5 A detailed explanation is provided on a specific implementation method for step S402, which determines the current environmental complexity index based on the current environmental interference data.

[0163] Figure 5 This is a flowchart illustrating a Bluetooth positioning method according to another embodiment of the present invention. Figure 5 As shown, a specific implementation of this Bluetooth positioning method, based on current environmental interference data, to determine the current environmental complexity index may include the following steps:

[0164] S501, based on the current environmental interference data, identify the environmental interference index corresponding to the current environmental interference data, the environmental interference index including at least one of multipath interference index, electromagnetic interference index and obstruction interference index.

[0165] For example, in one possible implementation, it is identified whether the current environmental interference data includes interference data corresponding to multipath interference, electromagnetic interference, and occlusion interference. Based on the interference data corresponding to multipath interference, a multipath interference index is calculated; based on the interference data corresponding to electromagnetic interference, a corresponding electromagnetic interference index is calculated; and based on the interference data corresponding to occlusion interference, a corresponding occlusion interference index is calculated. The interference data corresponding to multipath interference, electromagnetic interference, and occlusion interference are similar to those described above and will not be repeated here.

[0166] For example, the multipath interference index can be expressed by the following formula:

[0167]

[0168] in, Indicates multipath interference index, This represents the standard deviation of the measured multipath phase deviation. This represents the phase standard deviation of the pre-calibrated intermediate multipath level value.

[0169] For example, with a pre-calibrated standard deviation of the phase of a moderate multipath level value of 0.5 rad, in an ideal open environment (no multipath interference), the standard deviation of the measured multipath phase deviation can be 0 to 0.25 rad, corresponding to a multipath interference index of 0 to 0.5; in a normal indoor environment (weak multipath interference), the standard deviation of the measured multipath phase deviation can be 0.25 rad to 0.5 rad, corresponding to a multipath interference index of 0.5 to 1; in an underground garage / metal environment (strong multipath interference), the standard deviation of the measured multipath phase deviation can be 0.5 rad to 1 rad, corresponding to a multipath interference index of 1 to 2, etc.

[0170] For example, the electromagnetic interference index can be expressed by the following formula:

[0171]

[0172] in, Indicates electromagnetic interference index, This represents the measured average signal-to-noise ratio. This represents the maximum signal-to-noise ratio under ideal conditions in a darkroom.

[0173] For example, the maximum signal-to-noise ratio (SNR) in an ideal darkroom is 30 dB. In an ideal open environment, the measured average SNR can range from 25 dB to 30 dB, corresponding to an electromagnetic interference (EMI) index of 0 to 0.17. In a typical indoor environment, the measured average SNR can range from 20 dB to 25 dB, corresponding to an EMI index of 0.17 to 0.33. In an underground parking garage / metal environment, the measured average SNR can range from 0 dB to 20 dB, corresponding to an EMI index of 0.33 to 1, etc.

[0174] For example, the occlusion interference index can be expressed by the following formula:

[0175]

[0176] in, Indicates the occlusion and interference index. Indicates the signal power along the reflected path. This represents the total power of the signal along the direct path.

[0177] For example, in an ideal open environment, the value of the shading interference index can be 0 to 0.2; in a normal indoor environment, the value of the shading interference index can be 0.2 to 0.5; and in an underground garage / metal environment, the value of the shading interference index can be 0.5 to 0.7, etc.

[0178] S502, Calculate the current environmental complexity index based on the environmental interference index.

[0179] For example, in one possible implementation, the environmental interference indicators are comprehensively calculated based on the correlation between them, or the environmental interference indicators are directly weighted and summed according to preset weighting coefficients to obtain the current environmental complexity index. The preset weighting coefficients can be obtained through experimental calibration or by training different interference items using weighted labeling on scenario samples. The contribution ratios of multipath interference, electromagnetic interference, and occlusion interference indicators to the current environmental complexity index can be configured according to the actual application scenario of the Bluetooth positioning system. This establishes a mapping relationship between sample labeling results and environmental complexity in different scenarios, completing the labeling training of the environmental complexity tag. When any environmental interference indicator increases, the current environmental complexity index increases accordingly, reflecting the degree of adverse effect of the current environment on the ranging results.

[0180] For example, the current environmental complexity index can be used as the input result for a single positioning cycle and refreshed synchronously after the current environmental interference data is updated.

[0181] Optionally, in one possible implementation, the multipath interference index, electromagnetic interference index, and occlusion interference index are weighted based on preset weighting coefficients to obtain the current environmental complexity index.

[0182] For example, the preset weighting coefficients can be obtained through pre-test calibration. In one possible implementation, environmental complexity index values ​​corresponding to various typical scenarios are predefined (e.g., the environmental complexity index value of an ideal darkroom is 0, the environmental complexity index value of an outdoor parking lot is 1, and the environmental complexity index value of a completely interfering environment is 2). Multipath interference index, electromagnetic interference index, and occlusion interference index are collected and calculated multiple times under each typical scenario. Combining the environmental complexity index weighting calculation formula, the preset weighting coefficients corresponding to the multipath interference index, electromagnetic interference index, and occlusion interference index are solved by least squares fitting.

[0183] For example, the preset weighting coefficients for the multipath interference index, electromagnetic interference index, and occlusion interference index can be 0.4, 0.4, and 0.2, respectively. These preset weighting coefficients can be obtained through experimental calibration and are used to comprehensively characterize the influence of the current environment on the ranging error. This embodiment of the invention does not limit the specific values ​​of the preset weighting coefficients and can be flexibly configured according to the actual application scenario.

[0184] Understandably, the preset weighting coefficients (such as 0.4 for multipath interference, 0.4 for electromagnetic interference, and 0.2 for occlusion interference) are determined through experimental fitting to ensure the applicability of the environmental complexity index in different scenarios. For example, in an open-air parking lot scenario, the weight of the electromagnetic interference index can be reduced to 0.2, while the weight of the occlusion interference index can be increased to 0.4 to adapt to the signal propagation characteristics in an open environment. This weighting mechanism allows the environmental complexity index to flexibly adapt to the needs of multiple scenarios.

[0185] For example, the weighted calculation formula for the environmental complexity index can be expressed as:

[0186]

[0187] in, Indicators representing environmental complexity This represents the preset weighting coefficients corresponding to the multipath interference index. This represents the preset weighting coefficients corresponding to the electromagnetic interference index. This represents the preset weighting coefficient corresponding to the occlusion interference index.

[0188] It is understood that in the above embodiments, by converting the dispersed multipath interference, occlusion interference and electromagnetic interference information into calculable environmental interference indicators, and then using preset weighting coefficients to complete normalized synthesis, the current environmental complexity index is output. This allows multiple environmental interferences to be uniformly represented as the current environmental complexity index, thereby establishing a direct correspondence between the selection of positioning parameters and the current environmental state, improving the adaptability of the Bluetooth positioning process to complex scenarios, and allowing ranging fluctuations caused by environmental changes to be more accurately incorporated into the parameter selection criteria.

[0189] This invention, through its embodiments, identifies environmental interference indicators corresponding to current environmental interference data and calculates a current environmental complexity index based on these indicators. This allows for the unification of multiple environmental interferences into a single current environmental complexity representation, outputting a current environmental complexity result that matches the current environment. This enables subsequent parameter optimization to be directly based on this current environmental complexity result. Since the current environmental complexity index reflects the combined impact of multipath interference, electromagnetic interference, and occlusion interference on Bluetooth ranging, the system's response to environmental changes is more consistent in underground parking garages, building perimeters, or densely populated vehicle areas, significantly improving the stability and consistency of the Bluetooth positioning process.

[0190] It should be noted that the environmental complexity index constructed in this embodiment of the invention can also be extended to positioning algorithms based on Bluetooth signal strength (Received Signal Strength Indication, or RSSI). Specifically, the calculated environmental complexity metric can be used to classify scene interference, select a calibration dataset matching the interference classification, optimize the RSSI positioning model or distance mapping curve, and thus improve positioning accuracy in complex scenes. The following section will combine... Figure 6 The following is a detailed explanation of a specific implementation method for step S403, which optimizes multiple sets of initial ranging parameter combinations based on the current environmental complexity index to obtain the target ranging parameter combination.

[0191] Figure 6 This is a flowchart illustrating a Bluetooth positioning method provided in another embodiment of the present invention. Figure 6 As shown, in this Bluetooth positioning method, based on the current environmental complexity index, multiple initial ranging parameter combinations are optimized to obtain the target ranging parameter combination. A specific implementation may include the following steps:

[0192] S601 combines the current environmental complexity index with multiple sets of initial ranging parameters to generate multiple sets of ranging parameter combinations to be evaluated.

[0193] For example, the current environmental complexity index is combined with multiple sets of initial ranging parameters to generate multiple sets of ranging parameters to be evaluated.

[0194] It is understandable that each set of ranging parameters to be evaluated includes multidimensional ranging parameters and a current environmental complexity index. The multidimensional ranging parameters are similar to those described above and will not be repeated here.

[0195] S602, input multiple sets of ranging parameter combinations to be evaluated into the parameter selection model for prediction, and obtain the prediction results output by the parameter selection model. The prediction results include the prediction ranging error corresponding to each set of ranging parameter combinations to be evaluated.

[0196] For example, the parameter selection model can be constructed as a model containing a feature encoding layer and an error regression layer. In one possible implementation, each set of ranging parameters to be evaluated is input into the parameter selection model, and a joint feature vector is generated through feature encoding. Then, the corresponding predicted ranging error is output through the error regression layer. If the parameter selection model adopts a neural network model, the network parameters can be trained based on historical environmental samples and measured ranging errors, making the model adaptable to various application environments such as underground garages, building perimeters, and parking lots.

[0197] Alternatively, in another possible implementation, the parameter selection model can be a random forest model, which comprises multiple decision trees, and the maximum depth of each decision tree is determined. Specifically, each set of ranging parameters to be evaluated is input into each decision tree in the random forest model, and the single-tree ranging error of each set of ranging parameters to be evaluated is obtained from the output of each decision tree. The average of the single-tree ranging errors output by all decision trees is taken as the predicted ranging error corresponding to each set of ranging parameters to be evaluated output by the random forest model.

[0198] S603, based on the predicted ranging error corresponding to each set of ranging parameter combinations to be evaluated, determines multiple sets of candidate ranging parameter combinations.

[0199] For example, in one possible implementation, the predicted ranging errors corresponding to each set of ranging parameter combinations to be evaluated are arranged in ascending order, and the corresponding number of sets of ranging parameter combinations to be evaluated are selected as multiple candidate ranging parameter combinations based on a preset number of candidates. In another possible implementation, the ranging parameter combinations to be evaluated with predicted ranging errors less than a preset ranging error threshold are determined as candidate ranging parameter combinations. In this embodiment of the invention, the specific values ​​of the preset number of candidates and the preset ranging error threshold are not limited; they can be determined according to actual application requirements.

[0200] S604 performs parameter optimization on multiple candidate ranging parameter combinations to obtain the target ranging parameter combination.

[0201] For example, in one possible implementation, based on the ranging error corresponding to each set of candidate ranging parameter combinations, the set of candidate ranging parameter combinations with the smallest ranging error is selected as the target ranging parameter combination from among multiple sets of candidate ranging parameter combinations.

[0202] Alternatively, in another possible implementation, an optimization algorithm is used to iteratively optimize multiple sets of candidate ranging parameter combinations to obtain the target ranging parameter combination.

[0203] The optimization algorithm performs iterative optimization based on the optimization rules corresponding to the environmental complexity level, which is positively correlated with the environmental complexity index.

[0204] For example, the environmental complexity level is positively correlated with the environmental complexity index, that is, the higher the value of the environmental complexity index, the higher the corresponding environmental complexity level.

[0205] For example, the environmental complexity levels include Level 1 complexity environment (low complexity environment), Level 2 complexity environment (medium complexity environment), and Level 3 complexity environment (high complexity environment). Specifically, when the environmental complexity index is between [0, 0.6), the corresponding environmental complexity level is Level 1 complexity environment; when the environmental complexity index is between [0.6, 1.2), the corresponding environmental complexity level is Level 2 complexity environment; and when the environmental complexity index is between [1.2, 2.0], the corresponding environmental complexity level is Level 3 complexity environment.

[0206] It should be noted that in the embodiments of the present invention, different optimization rules are used for different levels of environmental complexity.

[0207] Optionally, in one possible implementation, the optimization rule adaptively adjusts the ranging parameters according to the environmental complexity level. These ranging parameters include at least one of the following: single-round ranging duration, total ranging duration, connection interval, number of ranging frequency points, channel state information sampling interval, and antenna configuration index.

[0208] The environmental complexity level includes at least one of Level 1, Level 2, and Level 3 complexity environments. The optimization rule for Level 1 complexity environments prioritizes reducing terminal power consumption and reduces the configuration of ranging parameters. The optimization rule for Level 2 complexity environments prioritizes balancing terminal power consumption and ranging accuracy and configures ranging parameters to a moderate level. The optimization rule for Level 3 complexity environments prioritizes ensuring ranging accuracy and enhances the configuration of ranging parameters.

[0209] For example, a level 1 complexity environment corresponds to a low-interference scenario. In this scenario, the optimization rule prioritizes reducing terminal power consumption by reducing the configuration of ranging parameters. This reduction can include decreasing the number of active antennas, shortening the single-round ranging time, reducing the number of ranging frequency points, and increasing the channel state information sampling interval. While maintaining ranging accuracy, this reduces RF operating time and lowers the RF power consumption of the digital key terminal.

[0210] For example, in some preferred embodiments, the target ranging parameter combination obtained after iterative optimization of multiple candidate ranging parameter combinations in a first-level complexity environment by the above optimization rules can be: single-round ranging duration 300ms, total ranging duration 30s, connection interval 30ms, number of ranging frequency points 30, channel state information sampling interval 20ms, and antenna configuration index 1.

[0211] Understandably, in low-interference scenarios corresponding to Level 1 complexity environments, the levels of multipath interference, obstruction interference, and electromagnetic interference are low, resulting in favorable ranging conditions. This invention reduces the configuration of ranging parameters through optimization rules, decreasing the number of active antennas, reducing the number of ranging frequency points, and lengthening the channel state information sampling interval. While ensuring that preset ranging accuracy targets are met, this shortens the operating time of the RF circuit and reduces the energy consumption of the RF module and antenna array. This combination of ranging parameters does not require a large amount of RF resources for redundant measurements, significantly reducing the standby power consumption and ranging power consumption of the digital key terminal, extending the terminal's battery life, and without causing significant loss of ranging accuracy.

[0212] For example, a level-two complexity environment corresponds to a medium interference scenario. In this scenario, the optimization rule prioritizes balancing terminal power consumption and ranging accuracy, and the ranging parameters are configured to a medium level. This medium level configuration can involve moderately increasing the number of active antennas, the duration of a single ranging round, and the sampling frequency, while also considering ranging accuracy, ranging response latency, and terminal power consumption.

[0213] For example, in some preferred embodiments, the target ranging parameter combination obtained after iterative optimization of multiple candidate ranging parameter combinations in a second-level complexity environment by the above optimization rules can be: single-round ranging duration 500ms, total ranging duration 30s, connection interval 30ms, number of ranging frequency points 60, channel state information sampling interval 20ms, and antenna configuration index 2.

[0214] Understandably, in medium-interference scenarios corresponding to level 2 complexity environments, where there is moderate multipath interference, obstruction interference, or electromagnetic interference, simply reducing ranging parameters will lead to a decrease in ranging accuracy. However, fully opening RF resources will result in unnecessary power consumption. This invention employs an optimization rule to adopt a moderate configuration for ranging parameters, appropriately increasing the number of antennas, single-round ranging duration, and sampling frequency. This increases the amount of ranging sampling information while slightly increasing terminal power consumption. This combination of ranging parameters can balance ranging accuracy, ranging response latency, and terminal power consumption, avoiding excessive simplification in low-interference scenarios or full resource opening in high-interference scenarios, thus achieving a balance between Bluetooth ranging performance and terminal power consumption.

[0215] For example, a level-three complexity environment corresponds to a high-interference scenario. In this scenario, the optimization rule prioritizes ensuring ranging accuracy and enhances the ranging parameters. These enhancements include activating the maximum number of available antennas, increasing the single-round ranging duration, improving the ranging and CSI sampling rates, increasing the number of ranging frequency points, and suppressing ranging errors introduced by multipath interference, obstruction, and electromagnetic noise through multiple measurements and statistical averaging, while also tolerating a moderate increase in terminal power consumption.

[0216] For example, in some preferred embodiments, the target ranging parameter combination obtained after iterative optimization of multiple candidate ranging parameter combinations in a level-three complexity environment using the above optimization rules can be: single-round ranging duration 500ms, total ranging duration 60s, connection interval 30ms, number of ranging frequency points 60, channel state information sampling interval 20ms, and antenna configuration index 4.

[0217] Understandably, in high-interference scenarios corresponding to level-three complexity environments, multipath reflections, object obstruction, and electromagnetic noise interference are strong, easily leading to significant deviations in single Bluetooth ranging results. This invention employs an enhanced configuration optimization strategy, activating the maximum available antenna array to extend the total ranging time, and relying on multiple measurements and statistical averaging to mitigate ranging errors caused by various environmental interferences. This ranging parameter combination prioritizes ranging accuracy, increasing the amount of observation information by adding radio frequency sampling resources. Although terminal radio frequency power consumption may increase, it ensures the reliability of digital key Bluetooth positioning in harsh, high-interference environments, solving the problems of ranging failure and positioning drift in complex scenarios.

[0218] It should be noted that the embodiments of the present invention only exemplarily divide environmental complexity into three levels: Level 1, Level 2, and Level 3. In practical applications, more refined level classification methods can also be used for environmental complexity. Any technical solution that configures corresponding optimization rules according to different environmental complexity levels and completes adaptive optimization of positioning parameters based on the optimization rules is within the protection scope of the present invention.

[0219] For example, in one possible implementation, intelligent optimization algorithms such as particle swarm optimization, genetic algorithm, and simulated annealing algorithm can be used to perform iterative update operations such as perturbation, crossover, and resampling on at least some of the ranging parameters in each group of candidate ranging parameter combinations. Combined with the new predicted ranging error results output by the parameter selection model, the parameter search direction and search range are dynamically updated, and the optimization is continuously iterated until the preset iteration convergence condition or the maximum number of iterations condition is met. Finally, the optimal ranging parameter combination is output as the target ranging parameter combination.

[0220] It is understood that the above-mentioned optimization algorithms are all general optional solutions, and the embodiments of the present invention do not limit the specific type of optimization algorithm. In a preferred embodiment, a Bayesian optimization algorithm that integrates environmental complexity indicators can be used to carry out refined parameter optimization. Compared with traditional intelligent algorithms, it can effectively reduce the number of iterations in the high-dimensional parameter space and improve the accuracy and efficiency of ranging parameter matching in complex interference scenarios.

[0221] Optionally, the prediction results also include the parameter importance weights corresponding to each ranging parameter in the candidate ranging parameter combination. These parameter importance weights are used to reflect the importance of the ranging parameter's influence on the ranging error.

[0222] In this embodiment of the invention, the current environmental complexity index is combined with multiple initial ranging parameter combinations to generate multiple sets of ranging parameter combinations to be evaluated. These multiple sets of ranging parameter combinations to be evaluated are then input into a parameter selection model for prediction, and the prediction results output by the parameter selection model are obtained. Further, based on the predicted ranging error corresponding to each set of ranging parameter combinations to be evaluated, multiple sets of candidate ranging parameter combinations are determined. Then, an optimization algorithm is used to iteratively optimize the multiple sets of candidate ranging parameter combinations to obtain the target ranging parameter combination. This allows the current environmental complexity to directly participate in the selection of ranging parameters. The parameter selection model first completes the prediction of ranging error, and then the optimization algorithm performs secondary optimization on the better parameter set to obtain the target ranging parameter combination that matches the current environment, thereby improving the accuracy of the target ranging parameter combination. As a result, when performing Bluetooth positioning based on this target ranging parameter combination, the ranging parameters can be adaptively adjusted with changes in the environment, and the positioning results in complex environments can remain stable, thus improving positioning accuracy.

[0223] For example, in one possible implementation, when the parameter selection model is a random forest model, each set of ranging parameters to be evaluated is input into each decision tree in the random forest model to obtain the single-tree importance weights corresponding to each ranging parameter in each set of ranging parameters to be evaluated output by each decision tree, and the average of the single-tree importance weights output by all decision trees is used as the parameter importance weights corresponding to each ranging parameter output by the random forest model.

[0224] For example, the parameter importance weights corresponding to each ranging parameter are normalized weights, that is, the sum of the parameter importance weights corresponding to each ranging parameter in each set of ranging parameters to be evaluated is 1.

[0225] The following is combined Figure 7 This paper provides a detailed explanation of a specific implementation method that uses an optimization algorithm to iteratively optimize multiple sets of candidate ranging parameter combinations to obtain the target ranging parameter combination.

[0226] Figure 7 This is a flowchart illustrating a Bluetooth positioning method provided in another embodiment of the present invention. Figure 7 As shown, in this Bluetooth positioning method, an optimization algorithm is used to iteratively optimize multiple sets of candidate ranging parameter combinations to obtain the target ranging parameter combination. A specific implementation may include the following steps:

[0227] S701, based on the parameter importance weights corresponding to each ranging parameter, selects multiple target ranging parameters from all ranging parameters corresponding to the candidate ranging parameter combinations.

[0228] For example, based on the parameter importance weights corresponding to each ranging parameter, multiple target ranging parameter dimensions are selected from all ranging parameter dimensions as target ranging parameters.

[0229] For example, in one possible implementation, the importance weights of each ranging parameter are matched with a preset importance weight threshold, and ranging parameters with importance weights higher than the preset importance weight threshold are selected as target ranging parameters. In another possible implementation, the importance weights of each ranging parameter are matched with a weight ranking rule, and several ranging parameters with high importance weight rankings are selected as target ranging parameters. The number of target ranging parameters can be set according to the complexity of the current environment and the dimensions of the candidate ranging parameter combinations; this embodiment of the invention does not specifically limit this.

[0230] For example, in specific implementations, parameter importance weights are normalized (e.g., Nf=0.35, Tr=0.25, ACI=0.20) to guide the optimization algorithm to prioritize adjusting high-weight parameters. For instance, in electromagnetic interference scenarios, the system prioritizes optimizing Nf and Tr, while reducing the priority of adjusting Icsi, thereby minimizing computational resource waste. This mechanism improves parameter optimization efficiency through weight-guided optimization.

[0231] S702 constructs a set of parameter points to be evaluated based on multiple target ranging parameters.

[0232] For example, one possible implementation is to discretely sample the range of values ​​for the target ranging parameters to form multiple parameter points, and combine these points with the baseline values ​​of the remaining unselected ranging parameters to form a set of parameter points to be evaluated. This set of parameter points to be evaluated serves as the input space for an optimization algorithm, which can employ one or more of the following algorithms: genetic algorithm, particle swarm optimization, Bayesian optimization, or hill climbing search algorithm. The algorithm iteratively updates the set of parameter points to be evaluated, thereby outputting a combination of target ranging parameters with a smaller predicted ranging error. The baseline values ​​can be selected as the mean, median, or boundary values ​​within the candidate ranging parameter combinations. The sampling interval, search boundary, and iteration termination condition in the set of parameter points to be evaluated can be pre-set based on the stability of the vehicle-side Bluetooth link and the complexity of the current environment. In practical applications, this algorithm can also choose other optimization models, which are not limited in this embodiment of the invention.

[0233] Optionally, another possible implementation of this step is as follows: For each target ranging parameter among multiple target ranging parameters, expand the interval according to the upper and lower limits of the parameter values ​​of the target ranging parameters in each group of candidate ranging parameter combinations to obtain the parameter sampling space corresponding to the target ranging parameter; uniformly sample within the parameter sampling space corresponding to each target ranging parameter to obtain multiple parameter sampling values ​​corresponding to each target ranging parameter; combine the multiple parameter sampling values ​​corresponding to each target ranging parameter to generate multidimensional sampling points; for the other ranging parameters in the candidate ranging parameter combinations besides the target ranging parameters, find the known ranging parameter combination with the smallest Euclidean distance to the multidimensional sampling points among all candidate ranging parameter combinations, and extract the known parameter values ​​corresponding to the remaining ranging parameters in the known ranging parameter combinations; combine the known parameter values ​​corresponding to the remaining ranging parameters with the parameter sampling values ​​corresponding to the target ranging parameters in the multidimensional sampling points to obtain the parameter point set to be evaluated.

[0234] Among them, the target ranging parameter is the parameter used for key optimization from the candidate ranging parameter combination. Its value range is determined by the upper and lower limits in the candidate ranging parameter combination, and the parameter sampling space is formed by expanding outward based on this.

[0235] The parameter sampling space is used to define the subsequent sampling boundary. The sampling extension can be set uniformly according to the preset interval, or it can be set separately according to the value span of different ranging parameters, so that the sampling range covers the neighborhood changes of the target ranging parameters.

[0236] The parameter sample value is a discrete value that is uniformly obtained at fixed intervals within the parameter sampling space. Multiple parameter sample values ​​are used to characterize the distribution of the target ranging parameters within the sampling range.

[0237] Multidimensional sampling points are composed of parameter sampling values ​​of multiple target ranging parameters, and are used to represent candidate positions in the target ranging parameter space.

[0238] Other ranging parameters are ranging parameters other than the target ranging parameter in the candidate ranging parameter combination. Known ranging parameter combinations are ranging parameter combinations that already exist in the candidate set. Their corresponding known parameter values ​​can be used as supplementary information and spliced ​​with multi-dimensional sampling points to form complete parameter points to be evaluated.

[0239] For example, in one specific implementation, a parameter index table is first established based on candidate ranging parameter combinations, and the upper and lower limits of each ranging parameter value, sampling interval, and parameter vector of known combinations are recorded. For each target ranging parameter, a preset sampling margin can be extended to both sides according to its upper and lower limits, and then multiple parameter sampling values ​​are generated at equal intervals within the extended interval. Subsequently, Cartesian combinations are performed on the parameter sampling values ​​of multiple target ranging parameters to obtain a set of sampling points covering multidimensional space. For each sampling point, the parameter sampling value corresponding to the target ranging parameter is used as the distance comparison benchmark. The Euclidean distance between the sampling point and each known ranging parameter combination in the target dimension is calculated among all known ranging parameter combinations, and the known ranging parameter combination with the smallest distance is selected. When multiple known ranging parameter combinations have the same distance, one can be determined according to the recording order or preset priority. Then, the known parameter values ​​corresponding to the remaining ranging parameters are extracted from the known ranging parameter combination, and these known parameter values ​​are concatenated with the target ranging parameter sampling values ​​in the current sampling point to obtain the corresponding set of parameter points to be evaluated. In practical applications, the sampling interval, expansion margin, and Euclidean distance calculation dimension can all be configured according to the number of ranging parameters, candidate combination density, and computing resources. This embodiment of the invention does not limit these settings.

[0240] For example, in specific implementations, for sampling points of target ranging parameters (such as Nf), non-target ranging parameters (such as Icsi=20ms) are supplemented by finding the known combination with the smallest Euclidean distance to the multidimensional sampling points (such as Nf=64, Tr=300ms), thereby reducing the number of discretization samplings in the parameter space. For instance, among the sampling points of Nf=64, the known combination closest to the combination of Tr=300ms (Tr=300ms, Icsi=20ms) is selected, avoiding full-range sampling of Icsi and reducing computational overhead. Before calculating the Euclidean distance, the target ranging parameters (such as Nf, Tr, ACI) can be normalized (e.g., 0-1 standardization) to avoid the influence of dimensional differences on distance calculation. For example, Nf=64 is normalized to 0.8, and Tr=300ms is normalized to 0.6, ensuring that the Euclidean distance accurately reflects the similarity of the ranging parameter combinations.

[0241] Understandably, this construction method first expands the target ranging parameters by interval and uniformly samples them, and then uses the nearest neighbor known combination to supplement the remaining ranging parameters. This ensures that the set of parameter points to be evaluated not only covers the range of variation of the target ranging parameters, but also maintains a continuous association with the known candidate space. Without changing the structure of the known candidate ranging parameters, it can generate a more representative set of parameter points to be evaluated, making the parameter optimization coverage more complete, the sampling distribution of the target ranging parameters more uniform, and reducing the evaluation bias caused by insufficient discretization of the parameter space.

[0242] S703 employs an optimization algorithm to iteratively optimize the set of parameter points to be evaluated, thereby obtaining the target ranging parameter combination.

[0243] For example, one possible implementation is to use the set of parameter points to be evaluated as the input space of the optimization algorithm, iteratively update the set of parameter points to be evaluated, and thus output the target ranging parameter combination with smaller predicted ranging error. The optimization algorithm can be one or more of the following: genetic algorithm, particle swarm optimization algorithm, Bayesian optimization algorithm, or hill-climbing search algorithm.

[0244] Optionally, the optimization algorithm is a Bayesian optimization algorithm, and the surrogate model of this Bayesian optimization algorithm is constructed using a Gaussian kernel function that incorporates the complexity index of the current environment. This surrogate model is used to output the predicted mean and predicted variance of each parameter point in the set of parameters to be evaluated during the iterative optimization process. The predicted variance is the predicted covariance.

[0245] In this embodiment of the invention, the Bayesian optimization algorithm refers to a probabilistic optimization method used for iterative optimization of the set of parameter points to be evaluated. It establishes a surrogate model based on historical sampling results and selects the next round of evaluation points in conjunction with a sampling function. The surrogate model refers to the statistical prediction model in the Bayesian optimization algorithm, which outputs the predicted mean and predicted variance for each parameter point to be evaluated, characterizing the expected prediction ranging error and uncertainty of that parameter point. The Gaussian kernel function incorporating the current environmental complexity index refers to the kernel function form after incorporating the current environmental complexity index into the kernel function correlation measurement, enabling the surrogate model to differentiate the parameter point relationships under different environmental conditions.

[0246] Understandably, Gaussian processes can simultaneously output predicted values ​​and corresponding uncertainties, making them suitable as surrogate models. Therefore, Gaussian processes were selected as the surrogate model for the Bayesian optimization algorithm. A Gaussian process is primarily determined by the mean function and the covariance kernel function, with the covariance mainly calculated through the kernel function. To better evaluate the impact of the environment on the surrogate model, an environmental complexity index is introduced into the kernel function, resulting in a Gaussian kernel function based on environmental complexity.

[0247] For example, the Gaussian kernel function based on environment complexity can be expressed by the following formula:

[0248]

[0249] in, This represents the output value of the Gaussian kernel function based on environment complexity. Let represent the ranging parameter vector corresponding to the i-th parameter point in the set of parameter points to be evaluated, which is composed of target ranging parameters. Let represent the ranging parameter vector corresponding to the j-th parameter point in the set of parameter points to be evaluated, which is composed of target ranging parameters. This represents the Euclidean distance between two sets of distance measurement parameter vectors. This represents the environmental complexity index corresponding to the i-th parameter point to be evaluated. This represents the environmental complexity index corresponding to the j-th parameter point to be evaluated. This represents the absolute value of the difference between the environmental complexity index of the i-th and j-th parameter points to be evaluated. This represents the natural exponential function.

[0250] For example, in specific implementations, the Gaussian kernel function adjusts the kernel similarity calculation by introducing the absolute value of the difference in environmental complexity indicators. For instance, in occlusion interference scenarios, increasing the weight of the difference in environmental complexity indicators makes the surrogate model pay more attention to parameter combinations in high-complexity scenarios, thereby improving the scenario adaptability of Bayesian optimization.

[0251] In practice, the Gaussian kernel function calculates the similarity between parameter combinations by combining the environmental complexity index (S) with parameter vectors (such as Nf, Tr, ACI). For example, the combination with S=0.6 has a high similarity to the combination with S=0.5, while the combination with S=0.6 has a low similarity to the combination with S=1.0, ensuring that the surrogate model can distinguish the parameter requirements of different interference scenarios.

[0252] For example, the covariance matrix constructed based on the output values ​​of the Gaussian kernel function described above can be represented by the following formula:

[0253]

[0254] Where K represents the covariance matrix, This represents the cocorrelation value between the ranging parameter vector consisting of target ranging parameters corresponding to the first parameter point to be evaluated and the ranging parameter vector consisting of target ranging parameters corresponding to the second parameter point to be evaluated, and so on for other matrix elements.

[0255] Alternatively, another possible implementation of this step is as follows: Calculate the predicted mean and predicted variance corresponding to each parameter point to be evaluated based on the surrogate model; use a sliding window of a preset size combined with the expected lifting function to iteratively sample the parameter point set to be evaluated based on the predicted mean and predicted variance until the preset iteration termination condition is met to obtain the target parameter point set to be evaluated; in the target parameter point set to be evaluated, select the ranging parameter combination with the smallest predicted ranging error as the target ranging parameter combination.

[0256] The surrogate model is used to characterize the mapping relationship between the combination of ranging parameters and the predicted ranging error. The surrogate model can be trained based on historical known samples, so that the surrogate model can output the predicted mean and predicted variance of each parameter point to be evaluated.

[0257] The predicted mean is used to reflect the expected predicted distance measurement error corresponding to the parameter point to be evaluated, and the predicted variance is used to reflect the uncertainty of the parameter point to be evaluated. Together, they provide a basis for subsequent sampling.

[0258] For example, a sliding window is used to limit the range of local parameter points participating in each evaluation. Its preset size can be set to a fixed number of samples to continuously update historical and current sampling results. For example, in a Bayesian optimization algorithm, the size of the sliding window can be 5 to maintain the sampling points obtained in the last 5 iterations. This sliding window controls computational overhead by limiting the range of local parameter points participating in the evaluation and achieves continuous local sampling in conjunction with the expected improvement function. This embodiment of the invention does not limit the specific values ​​of the preset size and the sliding window; they can be determined according to the actual application requirements.

[0259] The expected improvement function measures the potential for improvement of the current parameter point relative to the known optimal parameter point, and combines the predicted mean and predicted variance to give the sampling priority.

[0260] For example, the expected improvement function can be expressed by the following formula:

[0261]

[0262] in, This represents the expected improvement value, used to assess the potential of finding a better (smaller ranging error) parameter for the parameter point x under evaluation. The parameter point with the largest value will be used as the sampling point for the next round. This represents the mean of the predictions made by the Gaussian surrogate model for the parameter point x to be evaluated; This represents the standard deviation (uncertainty) of the predictions made by the Gaussian surrogate model for the parameter point x to be evaluated. This represents the minimum distance measurement error among all observed sampling points within the sliding window; The perturbation constant (hyperparameter) represents the balance between exploration and utilization, used to control whether to search more of the unknown parameter space; its value can be 0.1. This represents the cumulative distribution function of the standard normal distribution. Z represents the probability density function of the standard normal distribution; Z represents the normalized intermediate variable.

[0263] For example, normalized intermediate variables can be represented by the following formula:

[0264]

[0265] For example, based on the surrogate model, the predicted mean and predicted variance for each parameter point to be evaluated are calculated, and the corresponding Gaussian posterior formula can be expressed by the following formula:

[0266]

[0267]

[0268]

[0269] in, This indicates the parameter point to be evaluated. This represents the predicted mean of the current parameter points to be evaluated (corresponding to the predicted distance measurement error). Let represent the prediction variance of the current parameter point to be evaluated, and y represent the vector of ranging errors corresponding to all known observation sampling points within the sliding window. The covariance vector represents the current parameter point to be evaluated. and all known sampling points within the sliding window kernel similarity, Indicates the current parameter point to be evaluated. Its own kernel function value.

[0270] For example, the size of the sliding window can be 5. This sliding window is used to maintain the sampling points obtained from the most recent 5 iterations, and the covariance matrix is ​​constructed using only the sampling points within the sliding window, thus controlling the computational cost.

[0271] For example, in one specific implementation, multiple sets of candidate ranging parameter combinations and their corresponding current environmental complexity indices can be input into a surrogate model. The surrogate model outputs the predicted mean and predicted variance of each parameter point to be evaluated. Then, the expected improvement value is calculated within the local range covered by the sliding window, and the next round of sampling points is selected accordingly, so that the sampling process simultaneously considers the local optimal region and the potential improvement region. As the number of sampling rounds increases, the model parameters and candidate set are continuously updated. When the termination condition is met, the set of target parameter points to be evaluated is used as the final screening space. Then, the ranging parameter combination with the smallest predicted ranging error is selected from this set of target parameter points to be evaluated as the target ranging parameter combination. If the surrogate model used adopts a Gaussian process model, its kernel function can integrate the current environmental complexity index to enhance the fitting ability of the predicted ranging error distribution under complex environments. In practical applications, other models can also be selected, and this embodiment of the invention does not limit this.

[0272] Understandably, this approach first uses the predicted mean and predicted variance to characterize the predicted ranging error level and uncertainty of each parameter point to be evaluated during iterative optimization. Then, it achieves local continuous sampling through a sliding window and the expected boost function. Finally, it selects the optimal ranging parameter combination from the converged set of target parameter points to be evaluated. This allows the target ranging parameter combination to come from a fully screened candidate set and match the complexity of the current environment, thereby improving the stability of parameter optimization and the consistency of positioning results.

[0273] In this embodiment of the invention, multiple target ranging parameters are selected from all ranging parameters corresponding to candidate ranging parameter combinations based on the parameter importance weights corresponding to each ranging parameter. A set of parameter points to be evaluated is constructed based on the multiple target ranging parameters. Furthermore, an optimization algorithm is used to iteratively optimize the set of parameter points to be evaluated to obtain the target ranging parameter combination. Since the iterative optimization process focuses on the parameter dimensions that have a greater impact on the predicted ranging error, the interference of irrelevant parameters is reduced, making the target ranging parameter combination more consistent with the Bluetooth ranging characteristics in the current environment. This improves the convergence efficiency of the parameter combination and the adaptability of the target ranging parameter combination.

[0274] Optionally, the parameter selection model provided in this embodiment of the invention can be a random forest model. The following is in conjunction with... Figure 8 The training method of the random forest model provided in the embodiments of the present invention will be described in detail.

[0275] Figure 8 This is a schematic diagram illustrating the training process of the random forest model provided in an embodiment of the present invention. Figure 8 As shown, the training method for this random forest model may include the following steps:

[0276] S801, Obtain the training dataset, which contains multiple sets of training samples. Each set of training samples includes a ranging parameter vector, an environment complexity index, and the true value of the ranging error.

[0277] For example, the ranging parameter vector can be a vector consisting of the single-round ranging duration, the total ranging duration, the connection interval, the number of ranging frequency points, the channel state information sampling interval, and the antenna configuration index.

[0278] For example, in one possible implementation, under various environmental complexity scenarios (e.g., low, medium, and high environmental complexity scenarios), the preset value range and sampling constraints corresponding to each ranging parameter are determined, and parameter sampling is performed using Latin hypercube sampling (LHS). 200 training samples are collected in each environmental complexity scenario, resulting in a total of 600 training samples, improving the model's generalization performance. The collected training samples are labeled with ground truth ranging errors. Specifically, for each training sample, a laser rangefinder is used to obtain the true distance, and the absolute value of the difference between the measured distance obtained based on Bluetooth ranging and the true distance is calculated. This absolute value is used as the ground truth ranging error for that sample group, constructing a training dataset containing ranging parameter vectors, environmental complexity indicators, and ground truth ranging errors.

[0279] In its implementation, Latin hypercube sampling divides the environmental complexity index (S) into several intervals (e.g., 0 to 0.4, 0.4 to 0.8, 0.8 to 1.2) and generates samples evenly within each interval, ensuring a balanced distribution of samples across low, medium, and high environmental complexity scenarios. For example, 200 samples are generated within the interval S=0.6, covering typical combinations of multipath, occlusion, and electromagnetic interference, thus avoiding sample concentration in low-complexity scenarios.

[0280] In practice, the distribution of the environmental complexity index (S) in the 600 training samples generated by Latin hypercube sampling corresponds to the preset weighting coefficients (0.4, 0.4, 0.2). For example, samples with a high proportion of multipath interference index (e.g., S>0.8) account for about 40%, samples with a high proportion of electromagnetic interference index (e.g., S<0.4) account for about 40%, and samples with a high proportion of occlusion interference index (e.g., 0.4≤S≤0.8) account for about 20%, which is consistent with the weighting coefficients.

[0281] For example, the preset range of single-wheel distance measurement duration (Tr) can be (0, 500ms], and the sampling constraint, i.e. the sampling step size, can be 100ms. The corresponding values ​​of single-wheel distance measurement duration can be 100ms, 200ms, 300ms, 400ms, and 500ms.

[0282] For example, the preset range of the total ranging time (Tc) can be (0, 90s], and the sampling constraint, i.e. the sampling step size, can be 10s. The corresponding values ​​of the total ranging time can be 10s, 20s, 30s, 40s, 50s, 60s, 70s, 80s, and 90s.

[0283] For example, the preset value range corresponding to the connection interval (Ic) can be (0, 90ms], the sampling constraint, i.e. the sampling step size, can be 30ms, and the corresponding connection interval values ​​can be 30ms, 60ms, and 90ms.

[0284] For example, the preset range of the number of ranging frequency points (Nf) can be [24, 72], and the sampling constraint, i.e. the sampling step size, can be 8, corresponding to the values ​​of the number of ranging frequency points being 24, 32, 40, 48, 56, 64, and 72.

[0285] For example, the preset value range corresponding to the channel state information sampling interval (Icsi) can be (0, 50ms], the sampling constraint, i.e. the sampling step size, can be 10ms, and the corresponding connection interval values ​​can be 10ms, 20ms, 30ms, 40ms, and 50ms.

[0286] For example, the preset value range corresponding to the antenna configuration index (ACI) can be [0, 7], the sampling constraint, i.e. the sampling step size, can be 1, and the corresponding value of the antenna configuration index can be 0, 1, 2, 3, 4, 5, 6, 7.

[0287] For example, the preset value range for the environmental complexity index (S) can be [0, 2].

[0288] For example, the i-th training sample in the training dataset can be represented as:

[0289]

[0290] Where D represents the i-th training sample. Represents the distance measurement parameter vector. Indicators representing environmental complexity in different scenarios. This represents the true value of the ranging error.

[0291] For example, the ranging parameter vector can be represented as:

[0292]

[0293] in, Indicates the duration of single-wheel distance measurement. Indicates the total distance measurement time. Indicates the connection interval. Indicates the number of ranging frequency points. Indicates the sampling interval for channel state information. This indicates the antenna configuration index.

[0294] S802, based on the training dataset, iteratively train the pre-trained random forest model with configured hyperparameters to obtain the trained random forest model.

[0295] The hyperparameters configured in the pre-trained random forest model may include the number of decision trees, the maximum depth of a single decision tree, the minimum number of samples per leaf node, and the feature random sampling ratio. These hyperparameters are configured before training.

[0296] For example, in one possible implementation, the pre-trained random forest model first establishes an initial decision tree structure based on the training dataset, and then inputs each set of training samples into each decision tree for fitting. Next, based on the sample partitioning results, the partitioning thresholds and splitting directions of the decision tree nodes are updated, allowing the pre-trained random forest to gradually approximate the mapping relationship between the ranging parameter vector, the environmental complexity index, and the true value of the ranging error. After iterative training, the output of the trained random forest model can be used to predict the ranging error of candidate ranging parameter combinations and provide a basis for subsequent parameter optimization.

[0297] For example, in practical applications, the random forest model can also work in conjunction with the parameter optimization module. When the current environmental complexity index and multiple sets of ranging parameters to be evaluated are input, the random forest model predicts the ranging error corresponding to each combination and outputs the results used to evaluate the merits of the parameter combinations.

[0298] Optionally, the pre-trained random forest model includes multiple decision trees. In another possible implementation of this step, for each decision tree in the pre-trained random forest model, the following independent training is performed: data sampling is performed in the training dataset to obtain single-tree training samples corresponding to the current decision tree; the ranging parameter vector of the single-tree training samples is sampled using the random subspace method to obtain sampled features; based on the sampled features and the environmental complexity index, a feature vector is formed, and based on the feature vector, the current decision tree is trained by node splitting to obtain a trained single decision tree.

[0299] For example, a specific implementation method for independently training each decision tree in a pre-trained random forest model can be as follows: First, the training dataset is sampled with replacement to obtain single-tree training samples corresponding to each decision tree, so that the sample distribution received by different decision trees remains different; then, for the ranging parameter vector in the single-tree training samples, a portion of the ranging parameters are randomly selected as sampling features according to a preset feature sampling ratio; after feature sampling, the sampling features are concatenated with the environmental complexity index corresponding to the current sample to form a feature vector for node splitting judgment, and the feature vector is input into the current decision tree to perform node splitting training, so that the node is recursively divided according to the splitting gain or the decrease in ranging error until the maximum depth, minimum number of samples or minimum gain threshold is met, and then leaf nodes are generated, thus obtaining a single decision tree that has been trained.

[0300] For example, in practical applications, the decision tree can be a tree model suitable for regression prediction, and the embodiments of the present invention do not limit this.

[0301] It is understood that in this embodiment of the invention, the random subspace method is used to sample the ranging parameters and then combined with the environmental complexity index for training, which can keep the learning direction of different decision trees differentiated. Thus, after each decision tree completes independent training, the results are merged into the output of the random forest model, forming a trained random forest model, which is then used to predict ranging errors.

[0302] In this embodiment of the invention, by sampling, dividing, and training each decision tree separately, the correlation between decision trees in the random forest model is reduced, and environmental complexity information is explicitly introduced into the node splitting process. This allows the random forest model to simultaneously learn the correspondence between ranging parameters and environmental changes. Consequently, the trained random forest model exhibits more stable ranging error prediction in complex environments under Bluetooth positioning scenarios, and the output parameter selection results match the current environment more closely.

[0303] In this embodiment of the invention, a training dataset containing multiple sets of training samples is obtained, each set including a ranging parameter vector, an environmental complexity index, and the true value of the ranging error. Based on the training dataset, a pre-trained random forest model with configured hyperparameters is iteratively trained to obtain a trained random forest model. This allows the trained random forest model to more stably fit and predict ranging errors in complex environments, thereby providing reliable support for parameter selection in Bluetooth positioning and improving the reliability of Bluetooth positioning.

[0304] Optionally, in the above embodiments, a possible implementation of training the current decision tree by splitting nodes based on the feature vector to obtain a trained single decision tree can be as follows: traverse all candidate split points corresponding to all features in the feature vector, and recursively complete the node splitting layer by layer with the minimum overall root mean square error after node splitting as the splitting criterion; when the node depth reaches a preset maximum depth or the number of samples in the node is less than or equal to a preset threshold, stop the node splitting and generate a leaf node; the leaf node outputs the average ranging error of the corresponding training samples, and uses the average ranging error as the predicted ranging error value of the current decision tree.

[0305] For example, in one specific implementation, the current decision tree receives feature vectors sampled from a random subspace and generates candidate split points for each feature dimension based on the midpoints of adjacent samples after discrete or continuous sorting. During training, the ranging error statistics of the sample sets of the left and right child nodes are calculated for each candidate split point, and the overall root mean square error (RMSE) after splitting is obtained accordingly. The candidate split point that minimizes the RMSE is selected as the optimal split position of the current node, and then the samples are divided into child nodes and recursively splitting continues. When the depth of the current node reaches a preset maximum depth, or the number of samples in the node is less than or equal to a preset sample number threshold, the current node stops splitting and directly generates a leaf node. The average of the true ranging error values ​​of the training samples contained in the leaf node is calculated to obtain the output value of the leaf node, which is used as the predicted ranging error value corresponding to the leaf node. The preset maximum depth and preset sample number threshold can be pre-configured according to the training sample size and tree model complexity. For example, the maximum depth can be set to 8 to 16 layers, and the preset sample number threshold can be set to 5 to 20. This embodiment of the invention does not limit these settings.

[0306] For example, in practical applications, this decision tree can be used as a single base learner in a random forest to regress and predict the ranging error of different ranging parameter combinations in Bluetooth positioning scenarios. Since the splitting criterion uses the overall root mean square error minimization and stops splitting when the depth or sample threshold is reached, the decision tree can form a constraint between fitting ability and structural complexity. The average ranging error output by the leaf nodes is directly used as the prediction result of the current decision tree for the ranging error of the current input feature.

[0307] Understandably, by traversing all candidate split points corresponding to all features in the feature vector, and using the minimum overall root mean square error after node splitting as the splitting criterion, node splitting is completed recursively layer by layer. When the node depth reaches the preset maximum depth or the number of samples in the node is less than or equal to the preset threshold, node splitting stops and leaf nodes are generated. The leaf node outputs the average ranging error of the corresponding training samples, and the average ranging error is used as the ranging error prediction value of the current decision tree. This enables a single decision tree to complete recursive regression modeling for ranging error under the drive of feature vectors, and uses the average value of the leaf nodes as a stable output. This makes the decision tree have a more consistent representation ability for ranging error fluctuations in complex environments, and thus provides a usable basis for ranging error prediction for subsequent ranging parameter optimization.

[0308] Compared to related technologies that cannot assess the importance of ranging parameters based on actual scenarios, resulting in a uniform adjustment mode for ranging parameters across different scenarios, digital key ranging and positioning involves multiple adjustable parameters, such as ranging duration, environmental interference, connection duration, and number of antenna paths based on the CS positioning method. These parameters have varying importance in different scenarios, and without targeted adjustment of the correct parameters for different scenarios, good positioning results cannot be obtained. In the Bluetooth positioning method provided in this embodiment of the invention, the importance of each ranging parameter is evaluated simultaneously during the training process of the above-mentioned algorithm model, enabling accurate adjustment of ranging parameters under different environments, thereby improving the positioning effect.

[0309] Optionally, during the node splitting training of the current decision tree based on the feature vector, the method further includes: calculating the sum of the root mean square error reductions of each feature at all split nodes, and using the sum of the root mean square error reductions as the parameter importance weight of the feature in the current decision tree; for all decision trees in the random forest model, using the average of the parameter importance weights of the same feature in each decision tree as the parameter importance weight of the feature in the random forest model.

[0310] For example, in one specific implementation, when the current decision tree performs node splitting training layer by layer based on feature vectors, the root mean square error reduction brought by each candidate feature is calculated for each splitting node. After the optimal splitting feature is determined at that node, the root mean square error reduction of that feature at subsequent splitting nodes continues to accumulate until the entire decision tree training is completed. After training, the root mean square error reduction of the same feature at all splitting nodes is summed to obtain the parameter importance weight of that feature in the current decision tree. For multiple decision trees in a random forest, the parameter importance weight of the same feature in each decision tree is obtained, and then the average of the parameter importance weights of the feature in all decision trees is taken as the final parameter importance weight of the feature in the random forest model. This parameter importance weight can be directly output as a feature scoring result and used for subsequent filtering or sorting of ranging parameters. In practical applications, this model can also adopt other tree numbers and node splitting strategies, which are not limited in this embodiment of the invention.

[0311] For example, in a specific implementation, the out-of-bag error (e.g., 0.15) of the candidate random forest model generated by Latin hypercube sampling (e.g., 10 decision trees with a single tree depth of 8) is lower than that of other combinations (e.g., 5 decision trees with an out-of-bag error of 0.25), verifying the model's predictive stability in complex scenarios (e.g., underground parking garages). This mechanism uses the out-of-bag error to select the optimal model parameters.

[0312] It is understandable that by accumulating the root mean square error reduction of each split node within a single decision tree, and averaging the parameter importance weights of multiple decision trees at the random forest level, the feature importance assessment can simultaneously reflect the local split contribution and the overall integration result, thereby forming a stable parameter importance weight output.

[0313] The implementation of this embodiment ensures that the calculation results of parameter importance weights are consistent with the node splitting contribution, and the influence of training fluctuations of a single decision tree on the results can be suppressed by the mean of multiple decision trees, thus making the parameter importance weights output by the random forest model more suitable as a basis for filtering and ranking ranging parameters.

[0314] Optionally, in one possible implementation, the hyperparameters of the pre-trained random forest model are determined as follows: based on Latin hypercube sampling, multiple sets of hyperparameter combinations are generated within a first preset range of the number of decision trees and a second preset range of the depth of a single tree, and multiple sets of candidate random forest models are constructed accordingly; the out-of-bag error of each candidate random forest model is calculated, and the hyperparameters corresponding to the candidate random forest model with the smallest out-of-bag error are selected as the hyperparameters of the pre-trained random forest model.

[0315] For example, the first preset range for the number of decision trees and the second preset range for the depth of a single tree can be (0, 15). This embodiment of the invention does not limit the specific values ​​of the first and second preset ranges; they can be determined according to actual application requirements.

[0316] For example, in one specific implementation, a first preset range for the number of decision trees and a second preset range for the depth of a single tree can be set based on the number of samples in the training dataset, the range of values ​​for the environmental complexity index, and the fluctuation range of the true value of the ranging error. Then, several sets of discrete sampling points are generated in the two-dimensional range using Latin hypercube sampling. Each set of sampling points corresponds to a set of random forest hyperparameter combinations, and a set of candidate random forest models is constructed accordingly. Subsequently, each candidate random forest model is trained by bootstrapping on the corresponding training samples, and the out-of-bag error is calculated using the samples that are not selected. The out-of-bag errors of all candidate random forest models are compared, and the number of decision trees and the depth of a single tree corresponding to the candidate random forest model with the smallest out-of-bag error are selected as the hyperparameter configuration of the pre-trained random forest model. In practical applications, the values ​​of the number of decision trees and the depth of a single tree can also be set in other ranges, and this embodiment of the invention does not limit this.

[0317] Understandably, by adopting the above hyperparameter determination method, the structure configuration of the pre-trained random forest model can achieve uniform coverage and error filtering within the preset search space, ensuring that the selected hyperparameters match the features of the current training data. Since out-of-bag error is used as the filtering criterion, the determined hyperparameters can more directly reflect the generalization performance of the random forest model on unseen samples, thereby improving the stability and prediction consistency of the subsequent random forest model training.

[0318] Compared to related technologies that cannot allow users to personalize parameter selection and adaptation based on their actual usage scenarios—current Bluetooth positioning methods rely on vehicle manufacturers pre-determining positioning parameters or mapping mechanisms and writing them into the Bluetooth controller, forcing users to use these parameters in a predetermined manner without personalization based on their own usage scenarios—the Bluetooth positioning method provided in this invention can also incrementally train a random forest model by introducing user-specific data on positioning parameters and results related to their own driving scenarios, thereby achieving personalized matching of positioning parameters for different users.

[0319] Optionally, the Bluetooth positioning method provided in this embodiment of the invention further includes: scheduling and executing a local training mode for the user device, incrementally training a random forest model based on the user's personalized training data to obtain a personalized random forest model adapted to the user; or, scheduling and executing a vehicle-mounted system training mode, incrementally training a random forest model based on the user's personalized training data to obtain a personalized random forest model adapted to the user.

[0320] It should be noted that the initial training data collection and model training of the random forest algorithm model are collected by the vehicle manufacturer and completed on the cloud service platform of the aforementioned Bluetooth positioning system. After training is completed, in some embodiments, users can download the random forest model trained on the cloud service platform to their device terminal at any time through the vehicle digital key APP, conduct personalized incremental training based on test data collected in their own driving environment, and then transmit the trained random forest model to the vehicle terminal's Bluetooth controller for update via Bluetooth after training is completed. In other embodiments, after the vehicle digital key establishes a communication connection with the vehicle, the user can also issue a command through the vehicle digital key APP to enable the vehicle's Bluetooth controller to obtain the trained random forest model from the cloud service platform and complete personalized incremental model training locally on the vehicle.

[0321] For example, users can perform personalized incremental training of the random forest model through user device terminals loaded with the vehicle digital key APP, including two modes: local training on the user device terminal and local training in the vehicle. The specific implementation principle is as follows:

[0322] Local incremental training on user device terminals: The user downloads the random forest model trained on the cloud service platform to the vehicle digital key APP; after the model is downloaded, a Bluetooth communication connection is established between the user device terminal and the vehicle; the user triggers the ranging acquisition process at different locations and distances around the vehicle, and the user device terminal performs CS Bluetooth ranging and obtains the Bluetooth calculated distance; the user enters their actual distance relative to the vehicle's Bluetooth main anchor point, and completes the annotation of the ground truth of ranging error for a single set of samples. This annotated data is used as part of the training samples for incremental training of the random forest model, generating personalized parameter combinations adapted to the user scenario; after completing the collection of training samples in multiple scenarios, the user device terminal performs incremental training on the random forest model based on the newly collected personalized training samples to obtain a personalized random forest model adapted to the user's usage scenario; after the incremental training is completed, the user device terminal can send the personalized random forest model to the vehicle Bluetooth controller for storage and use via Bluetooth communication;

[0323] Vehicle Local Incremental Training: After the user device terminal establishes a Bluetooth communication connection with the vehicle, the user triggers the vehicle local training process; the vehicle Bluetooth controller obtains the trained random forest model from the cloud service platform and loads it into the vehicle's infotainment system; multi-scenario training sample collection is completed by referring to the data collection method of the user device terminal's local training; the vehicle's infotainment system completes incremental training of the random forest model based on the collected personalized training samples, generates a personalized random forest model and saves it in the vehicle Bluetooth controller for subsequent CS ranging parameter optimization.

[0324] For example, a specific implementation of incremental training on the user device terminal can be as follows: The user selects to train the positioning algorithm model on their mobile phone within the vehicle's digital key APP page. The APP requests the digital key service module in the cloud service platform to download the trained random forest algorithm model to the mobile phone. The mobile phone and vehicle actively establish a Bluetooth connection. The user clicks "Start Training," and the ranging module in the APP negotiates different ranging parameters with the vehicle sequentially to perform Bluetooth CS ranging. After each ranging measurement, the user fills in the actual distance between the measured location and the actual location. After the user clicks "Stop Ranking," the data module in the APP organizes the collected data and uses it for post-training of the random forest algorithm model. During training, the data module in the APP automatically calculates the error reduction caused by each ranging parameter and normalizes these error reductions as the parameter importance weights of each ranging parameter to the ranging error. The mobile phone sends the trained random forest algorithm model and the parameter importance weights of each ranging parameter to the vehicle's Bluetooth controller via Bluetooth communication.

[0325] In practice, when user-personalized training data is collected through the vehicle digital key APP loaded on the user's device, the system performs three sampling verifications on the "real distance" input by the user. For example, if the user inputs the same distance three times consecutively (e.g., 1.5m), the data is accepted; if the three inputs differ significantly (e.g., 1.5m, 2.0m, 1.8m), the system prompts the user to re-label the data to avoid data noise caused by misoperation.

[0326] For example, a specific implementation of incremental training on the vehicle's local system can be as follows: The user selects to train the positioning algorithm model on the vehicle's digital key APP page. The APP sends a command to the vehicle-side to train the random forest algorithm model to the Bluetooth controller. The Bluetooth controller forwards the request to download the random forest algorithm model to the vehicle-side system via the vehicle gateway. The vehicle-side system requests the remote control module in the cloud service platform to download the random forest algorithm model to the vehicle-side system. The mobile phone and vehicle actively establish a Bluetooth connection. The user clicks "Start Training." The ranging module in the APP negotiates different ranging parameters with the vehicle via the mobile phone to perform Bluetooth CS ranging. After each ranging measurement, the user fills in the actual distance between the measured location and the actual distance. The Bluetooth controller saves each ranging result and the measured distance input by the user via the mobile APP. It then sends the data to the vehicle-side system for post-training of the random forest algorithm model. During training, the vehicle-side system automatically calculates the error reduction caused by each ranging parameter and normalizes these error reductions as the parameter importance weight for each parameter in relation to the ranging error. After the user clicks "End Training," the vehicle-side Bluetooth controller requests the trained random forest algorithm model and the parameter importance weights for each ranging parameter from the vehicle-side system.

[0327] For example, in practical applications, the user device's local training mode or the vehicle-mounted system training mode is scheduled according to the deployment environment, and the random forest model is continuously updated with the user's personalized training data. This ensures that the output of the random forest model is consistent with the user's ranging parameter selection process, thereby making the prediction results of the target ranging parameter combination more closely match the actual usage status of a specific user. Since the training process is based on the user's own samples, the personalized random forest model can form a more suitable parameter mapping relationship for different terminal placement methods, different parking environments, and different usage habits.

[0328] It is understood that, in this embodiment of the invention, the user device local training mode is scheduled to be executed, and the random forest model is incrementally trained based on the user's personalized training data to obtain a personalized random forest model adapted to the user; or, the vehicle-mounted system training mode is scheduled to be executed, and the random forest model is incrementally trained based on the user's personalized training data to obtain a personalized random forest model adapted to the user. This enables the random forest model to complete user-level adaptation without changing the main positioning framework. The model update process can be flexibly switched according to the resources on the terminal side or the vehicle-mounted system side, thereby improving the consistency between ranging parameter prediction and parameter optimization results, and enhancing the adaptability of digital key Bluetooth positioning in scenarios with individual differences.

[0329] In summary, the Bluetooth positioning method provided by the embodiments of the present invention has the following beneficial effects:

[0330] 1) By constructing a Bluetooth positioning environment complexity assessment index that integrates occlusion interference, multipath interference, and electromagnetic interference, the overall interference level of the scene can be quantitatively characterized, and the ranging parameters can be adaptively adapted to the degree of environmental interference, effectively improving the stability of Bluetooth CS ranging under different interference scenarios.

[0331] 2) Achieve quantitative assessment of the importance of Bluetooth positioning parameters, accurately quantify the influence weight of each ranging parameter on the ranging error, support fine-tuning of positioning parameters in different environments, and improve the targeting of parameter optimization;

[0332] 3) Based on the interpretability of the random forest model, the importance weight of each ranging parameter to Bluetooth CS ranging is calculated based on the reduction of ranging error, so as to realize the objective quantification of the parameter contribution, provide a reliable basis for dimensionality reduction to construct the parameter set to be evaluated, and reduce the computational cost of subsequent Bayesian optimization.

[0333] 4) Provides a personalized incremental training mechanism for user equipment terminals, cloud service platforms, and vehicle terminals with multiple selectable deployment modes. It supports model fine-tuning based on real user driving scenarios, overcomes the shortcomings of traditional positioning models that rely solely on offline datasets from the factory and cannot adapt to the actual user environment, and further reduces ranging errors in personalized scenarios.

[0334] 5) By using an improved Gaussian kernel function that incorporates environmental complexity to construct a Bayesian optimization surrogate model, the surrogate model can perceive scene interference features, improve the accuracy of ranging error prediction, and thus more efficiently search for the optimal combination of CS ranging parameters in the current scene.

[0335] The following are embodiments of the apparatus of the present invention, which can be used to execute embodiments of the method of the present invention. For details not disclosed in the embodiments of the apparatus of the present invention, please refer to the embodiments of the method of the present invention.

[0336] Figure 9 This is a schematic diagram of the structure of a Bluetooth positioning device provided in an embodiment of the present invention. Figure 9 As shown, the Bluetooth positioning device 90 includes an acquisition module 910, a determination module 920, a parameter optimization module 930, and a Bluetooth positioning module 940.

[0337] The acquisition module 910 is used to acquire the current environmental interference data and multiple sets of initial ranging parameters of the vehicle in response to the establishment of a Bluetooth communication connection between the user equipment loaded with the vehicle digital key and the vehicle. The current environmental interference data includes interference data corresponding to at least one of multipath interference, occlusion interference and electromagnetic interference.

[0338] The determination module 920 is used to determine the current environmental complexity index based on the current environmental interference data;

[0339] The parameter optimization module 930 is used to perform parameter optimization processing on multiple initial ranging parameter combinations based on the current environmental complexity index to obtain the target ranging parameter combination;

[0340] The Bluetooth positioning module 940 is used to locate user devices via Bluetooth based on a combination of target ranging parameters.

[0341] In one possible implementation, the determining module 920 is specifically used to: identify environmental interference indicators corresponding to the current environmental interference data based on the current environmental interference data, wherein the environmental interference indicators include at least one of multipath interference indicators, electromagnetic interference indicators, and occlusion interference indicators; and calculate the current environmental complexity index based on the environmental interference indicators.

[0342] In one possible implementation, the determining module 920 is further configured to: weight the multipath interference index, electromagnetic interference index and occlusion interference index based on preset weighting coefficients to obtain the current environmental complexity index.

[0343] In one possible implementation, the parameter optimization module 930 is specifically used to: combine the current environmental complexity index with multiple initial ranging parameter combinations to generate multiple ranging parameter combinations to be evaluated; input the multiple ranging parameter combinations to be evaluated into the parameter selection model for prediction to obtain the prediction results output by the parameter selection model, the prediction results including the predicted ranging error corresponding to each ranging parameter combination to be evaluated; determine multiple candidate ranging parameter combinations based on the predicted ranging error corresponding to each ranging parameter combination to be evaluated; and perform parameter optimization processing on the multiple candidate ranging parameter combinations to obtain the target ranging parameter combination.

[0344] In one possible implementation, the parameter optimization module 930 is further configured to: employ an optimization algorithm to iteratively optimize multiple sets of candidate ranging parameter combinations to obtain the target ranging parameter combination; wherein the optimization algorithm completes the iterative optimization according to the optimization rules corresponding to the environmental complexity level, and the environmental complexity level is positively correlated with the environmental complexity index.

[0345] In one possible implementation, the optimization rule adaptively adjusts the ranging parameters according to the level of environmental complexity. The ranging parameters include at least one of the following: single-round ranging duration, total ranging duration, connection interval, number of ranging frequency points, channel state information sampling interval, and antenna configuration index.

[0346] The environmental complexity level includes at least one of three levels: Level 1 complexity, Level 2 complexity, and Level 3 complexity. The optimization rule for Level 1 complexity prioritizes reducing terminal power consumption and reduces the configuration of ranging parameters. The optimization rule for Level 2 complexity prioritizes balancing terminal power consumption and ranging accuracy and configures ranging parameters to a moderate level. The optimization rule for Level 3 complexity prioritizes ensuring ranging accuracy and enhances the configuration of ranging parameters.

[0347] In one possible implementation, the prediction result also includes the parameter importance weights corresponding to each ranging parameter in the candidate ranging parameter combination. The parameter importance weights are used to reflect the importance of the ranging parameter to the ranging error. The parameter optimization module 930 is also used to: select multiple target ranging parameters from all ranging parameters corresponding to the candidate ranging parameter combination according to the parameter importance weights corresponding to each ranging parameter; construct a set of parameter points to be evaluated based on the multiple target ranging parameters; and use an optimization algorithm to iteratively optimize the set of parameter points to be evaluated to obtain the target ranging parameter combination.

[0348] In one possible implementation, the parameter optimization module 930 is further configured to: for each target ranging parameter among multiple target ranging parameters, expand the interval according to the upper and lower limits of the parameter values ​​of the target ranging parameters in each set of candidate ranging parameter combinations to obtain the parameter sampling space corresponding to the target ranging parameter; uniformly sample within the parameter sampling space corresponding to each target ranging parameter to obtain multiple parameter sampling values ​​corresponding to each target ranging parameter; combine the multiple parameter sampling values ​​corresponding to each target ranging parameter to generate multidimensional sampling points; for other ranging parameters in the candidate ranging parameter combinations besides the target ranging parameters, find the known ranging parameter combination with the smallest Euclidean distance to the multidimensional sampling points among all candidate ranging parameter combinations, and extract the known parameter values ​​corresponding to the remaining ranging parameters in the known ranging parameter combinations; combine the known parameter values ​​corresponding to the remaining ranging parameters with the parameter sampling values ​​corresponding to the target ranging parameters in the multidimensional sampling points to obtain the parameter point set to be evaluated.

[0349] In one possible implementation, the optimization algorithm is a Bayesian optimization algorithm. The surrogate model of the Bayesian optimization algorithm is constructed using a Gaussian kernel function that incorporates the environmental complexity index. The surrogate model is used to output the predicted mean and predicted variance of each parameter point in the set of parameters to be evaluated during the iterative optimization process.

[0350] In one possible implementation, the parameter optimization module 930 is further configured to: calculate the predicted mean and predicted variance corresponding to each parameter point to be evaluated based on the surrogate model; use a sliding window of a preset size combined with the expected lifting function to iteratively sample the set of parameter points to be evaluated based on the predicted mean and predicted variance until the preset iteration termination condition is met to obtain the target set of parameter points to be evaluated; and select the ranging parameter combination with the smallest predicted ranging error as the target ranging parameter combination from the target set of parameter points to be evaluated.

[0351] In one possible implementation, the parameter selection model is a random forest model, which is trained as follows: a training dataset is obtained, which contains multiple sets of training samples, each set of training samples including a ranging parameter vector, an environment complexity index, and a true value of the ranging error; based on the training dataset, the pre-trained random forest model with configured hyperparameters is iteratively trained to obtain the trained random forest model.

[0352] In one possible implementation, the pre-trained random forest model includes multiple decision trees. Based on the training dataset, the pre-trained random forest model with configured hyperparameters is iteratively trained to obtain a trained random forest model. This includes: for each decision tree in the pre-trained random forest model, performing the following independent training: sampling data from the training dataset to obtain single-tree training samples corresponding to the current decision tree; using the random subspace method to sample the ranging parameter vector of the single-tree training samples to obtain sampled features; forming a feature vector based on the sampled features and the environmental complexity index; and performing node splitting training on the current decision tree based on the feature vector to obtain a trained single decision tree.

[0353] In one possible implementation, the current decision tree is trained by node splitting based on the feature vector to obtain a trained single decision tree. This includes: traversing all candidate split points corresponding to all features in the feature vector, using the minimum overall root mean square error after node splitting as the splitting criterion, and recursively completing node splitting layer by layer; stopping node splitting and generating leaf nodes when the node depth reaches a preset maximum depth or the number of samples in the node is less than or equal to a preset threshold; the leaf node outputs the average ranging error of the corresponding training samples, and uses the average ranging error as the predicted ranging error value of the current decision tree.

[0354] In one possible implementation, during the node splitting training of the current decision tree based on the feature vector, the method further includes: calculating the sum of the root mean square error reductions of each feature across all split nodes, and using the sum of the root mean square error reductions as the parameter importance weight of the feature in the current decision tree; and for all decision trees in the random forest model, using the average of the parameter importance weights of the same feature in each decision tree as the parameter importance weight of the feature in the random forest model.

[0355] In one possible implementation, the hyperparameters of the pre-trained random forest model are determined as follows: based on Latin hypercube sampling, multiple sets of hyperparameter combinations are generated within a first preset range of the number of decision trees and a second preset range of the depth of a single tree, and corresponding multiple sets of candidate random forest models are constructed; the out-of-bag error of each candidate random forest model is calculated, and the hyperparameters corresponding to the candidate random forest model with the smallest out-of-bag error are selected as the hyperparameters of the pre-trained random forest model.

[0356] In one possible implementation, the Bluetooth positioning device further includes a scheduling execution module (not shown), which is used to: schedule and execute the user equipment local training mode, incrementally train the random forest model based on the user's personalized training data, and obtain a personalized random forest model adapted to the user.

[0357] Alternatively, the vehicle-mounted system can be scheduled to perform training, and the random forest model can be incrementally trained based on the user's personalized training data to obtain a personalized random forest model adapted to the user.

[0358] The Bluetooth positioning device provided in this embodiment of the invention can be used to execute the method steps of the above method embodiments. The specific implementation and technical effects are similar, and will not be repeated here.

[0359] Figure 10 This is a schematic diagram of the structure of a Bluetooth positioning device provided in an embodiment of the present invention. Figure 10 As shown, the Bluetooth positioning device 10 includes at least one processor 101 and a memory 102.

[0360] Optionally, the Bluetooth positioning device 10 also includes a communication component 103. The processor 101, memory 102, and communication component 103 are connected via a bus 104.

[0361] In a specific implementation, at least one processor 101 executes computer execution instructions stored in memory 102, causing at least one processor 101 to perform the above-described method.

[0362] The specific implementation process of processor 101 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0363] Understandably, this Bluetooth positioning device, through the collaboration of a processor and memory, stores the instructions corresponding to the Bluetooth positioning method on the device side. This enables the device to directly perform environmental interference perception, environmental complexity characterization, ranging parameter combination optimization, and target parameter matching positioning in vehicle digital key scenarios. After the processor executes the stored instructions, it can dynamically adjust the positioning parameters based on the current multipath interference, occlusion interference, and electromagnetic interference conditions. This avoids the insufficient adaptability of fixed parameters in underground garages, building perimeters, and outdoor parking lots, making the vehicle's judgment of the relative position of the user's device more accurate. Furthermore, it reduces the probability of misjudgments such as approaching an unlocked device, misjudging the lock upon leaving, and locking delays, thus improving Bluetooth positioning accuracy, stability, and scene adaptability.

[0364] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0365] The memory may include Random Access Memory (RAM) and Non-volatile Memory (NVM), such as at least one disk storage device. The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0366] This invention also provides a vehicle, including at least one processor and a memory communicatively connected to the processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory to implement the method provided above.

[0367] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0368] For example, when this computer program product is deployed in a vehicle-side digital key control unit, mobile terminal, or their collaborative processing platform, the Bluetooth positioning method can be embedded in the form of software instructions and directly invoked and executed by the processor. After the processor runs the computer program, it can complete the acquisition of environmental interference information, generation of environmental complexity characterization, optimization of ranging parameter combinations, and target parameter matching based on the established Bluetooth communication connection between the user device and the vehicle, and perform positioning determination accordingly, thereby enabling the Bluetooth positioning capability to be stably reproduced in the form of productized software. This allows for dynamic adaptation to different interference conditions in complex scenarios such as underground garages, building perimeters, and outdoor parking lots, reducing positioning deviations caused by fixed parameters, and thus improving the accuracy, stability, and scene adaptability of the digital key's contactless unlocking, locking, and welcoming interactions.

[0369] This invention also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0370] For example, the computer-readable storage medium can be located in the vehicle-side digital key control unit, mobile terminal, or a processing device that works with it. The stored computer execution instructions, after being invoked by the processor, enable the system to acquire information on multipath interference, obstruction interference, and electromagnetic interference after establishing a Bluetooth communication connection. This information forms an environmental complexity characterization that can participate in positioning decisions. Based on the current environmental state, the system optimizes and matches multiple sets of initial ranging parameter combinations, allowing the ranging parameters to adaptively adjust with environmental changes. Furthermore, Bluetooth positioning can be performed based on the target ranging parameter combination, reducing ranging fluctuations and location misjudgments in complex scenarios such as underground parking garages, building perimeters, and outdoor parking lots. Therefore, it can improve the positioning accuracy, stability, and scene adaptability of the digital key's contactless unlocking, locking, and welcoming interactions.

[0371] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0372] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0373] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the couplings or direct couplings or communication connections shown or discussed may be indirect couplings or communication connections through interfaces, devices, or units, and may be electrical, mechanical, or other forms. Units described as separate components may or may not be physically separate; components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0374] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If the function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the various embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0375] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0376] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A Bluetooth positioning method, characterized in that, include: In response to the establishment of a Bluetooth communication connection between a user device loaded with a vehicle digital key and the vehicle, the current environmental interference data and multiple sets of initial ranging parameter combinations of the vehicle are obtained. The current environmental interference data includes interference data corresponding to at least one of multipath interference, occlusion interference, and electromagnetic interference. Based on the current environmental interference data, determine the current environmental complexity index; Based on the current environmental complexity index, parameter optimization processing is performed on the multiple sets of initial ranging parameter combinations to obtain the target ranging parameter combination; The user equipment is located via Bluetooth based on the target ranging parameter combination.

2. The Bluetooth positioning method according to claim 1, characterized in that, The step of determining the current environmental complexity index based on the current environmental interference data includes: Based on the current environmental interference data, identify the environmental interference index corresponding to the current environmental interference data. The environmental interference index includes at least one of multipath interference index, electromagnetic interference index and occlusion interference index. Calculate the current environmental complexity index based on the environmental interference index.

3. The Bluetooth positioning method according to claim 2, characterized in that, The step of calculating the current environmental complexity index based on the environmental interference index includes: Based on preset weighting coefficients, the multipath interference index, the electromagnetic interference index, and the occlusion interference index are weighted to obtain the current environmental complexity index.

4. The Bluetooth positioning method according to any one of claims 1 to 3, characterized in that, The step of performing parameter optimization processing on the multiple sets of initial ranging parameter combinations based on the current environmental complexity index to obtain the target ranging parameter combination includes: The current environmental complexity index is combined with the multiple sets of initial ranging parameters to generate multiple sets of ranging parameters to be evaluated. The multiple sets of distance measurement parameter combinations to be evaluated are respectively input into the parameter selection model for prediction, and the prediction results output by the parameter selection model are obtained. The prediction results include the prediction distance measurement error corresponding to each set of distance measurement parameter combinations to be evaluated. Based on the predicted ranging error corresponding to each set of ranging parameter combinations to be evaluated, multiple sets of candidate ranging parameter combinations are determined. The multiple sets of candidate ranging parameter combinations are subjected to parameter optimization processing to obtain the target ranging parameter combination.

5. The Bluetooth positioning method according to claim 4, characterized in that, The step of performing parameter optimization processing on the multiple sets of candidate ranging parameter combinations to obtain the target ranging parameter combination includes: An optimization algorithm is used to iteratively optimize the multiple sets of candidate ranging parameter combinations to obtain the target ranging parameter combination. The optimization algorithm performs iterative optimization based on the optimization rules corresponding to the environmental complexity level, and the environmental complexity level is positively correlated with the environmental complexity index.

6. The Bluetooth positioning method according to claim 5, characterized in that, The optimization rule adaptively adjusts the ranging parameters based on the environmental complexity level. The environmental complexity level includes at least one of a level 1 complexity environment, a level 2 complexity environment, and a level 3 complexity environment; The optimization rules corresponding to the first-level complexity environment prioritize reducing terminal power consumption and reduce the configuration of the ranging parameters. The optimization rule corresponding to the second-level complexity environment prioritizes balancing terminal power consumption and ranging accuracy, and configures the ranging parameters to a moderate level. The optimization rules corresponding to the three-level complexity environment prioritize ensuring ranging accuracy and enhance the configuration of the ranging parameters.

7. The Bluetooth positioning method according to claim 5, characterized in that, The prediction results also include the parameter importance weights corresponding to each ranging parameter in the candidate ranging parameter combination. The parameter importance weights are used to reflect the importance of the ranging parameters to the ranging error. The optimization algorithm is used to iteratively optimize the multiple sets of candidate ranging parameter combinations to obtain the target ranging parameter combination, including: Based on the importance weight of each ranging parameter, multiple target ranging parameters are selected from all ranging parameters corresponding to the candidate ranging parameter combination. Based on the multiple target ranging parameters, construct a set of parameter points to be evaluated; The optimization algorithm is used to iteratively optimize the set of parameter points to be evaluated, thereby obtaining the target ranging parameter combination.

8. The Bluetooth positioning method according to claim 7, characterized in that, The step of constructing a set of parameter points to be evaluated based on the multiple target ranging parameters includes: For each target ranging parameter among the multiple target ranging parameters, the range is expanded according to the upper and lower limits of the parameter values ​​of the target ranging parameters in each group of candidate ranging parameter combinations to obtain the parameter sampling space corresponding to the target ranging parameter; Uniform sampling is performed within the parameter sampling space corresponding to each target ranging parameter to obtain multiple parameter sampling values ​​corresponding to each target ranging parameter. Multiple parameter sampling values ​​corresponding to the ranging parameters of each target are combined to generate multi-dimensional sampling points; For the other ranging parameters in the candidate ranging parameter combinations besides the target ranging parameter, find the known ranging parameter combination with the smallest Euclidean distance to the multidimensional sampling point among all the candidate ranging parameter combinations, and extract the known parameter values ​​corresponding to the remaining ranging parameters in the known ranging parameter combination; The known parameter values ​​corresponding to the remaining ranging parameters are combined with the parameter sampling values ​​corresponding to the target ranging parameters in the multidimensional sampling points to obtain the parameter point set to be evaluated.

9. The Bluetooth positioning method according to claim 7, characterized in that, The optimization algorithm is a Bayesian optimization algorithm, and the surrogate model of the Bayesian optimization algorithm is constructed using a Gaussian kernel function that incorporates environmental complexity indicators. The surrogate model is used to output the predicted mean and predicted variance of each parameter point in the set of parameters to be evaluated during the iterative optimization process.

10. The Bluetooth positioning method according to claim 9, characterized in that, The optimization algorithm is used to iteratively optimize the set of parameter points to be evaluated to obtain the target ranging parameter combination, including: Based on the surrogate model, calculate the predicted mean and predicted variance for each parameter point to be evaluated; Using a sliding window of a preset size combined with the expected improvement function, the set of parameter points to be evaluated is iteratively sampled according to the predicted mean and the predicted variance until a preset iteration termination condition is met, thereby obtaining the target set of parameter points to be evaluated. In the set of target parameter points to be evaluated, the range parameter combination with the smallest predicted range error is selected as the target range parameter combination.

11. The Bluetooth positioning method according to claim 4, characterized in that, The parameter selection model is a random forest model, which is trained in the following way: Obtain a training dataset, which contains multiple sets of training samples. Each set of training samples includes a ranging parameter vector, an environment complexity index, and a true value of the ranging error. Based on the training dataset, the pre-trained random forest model with configured hyperparameters is iteratively trained to obtain the trained random forest model.

12. The Bluetooth positioning method according to claim 11, characterized in that, The pre-trained random forest model includes multiple decision trees. The step of iteratively training the pre-trained random forest model with configured hyperparameters based on the training dataset to obtain the trained random forest model includes: For each decision tree in the pre-trained random forest model, perform the following independent training: Data sampling is performed in the training dataset to obtain the single-tree training samples corresponding to the current decision tree; The ranging parameter vector of the single-tree training samples is sampled using the random subspace method to obtain the sampled features; Based on the sampling features and environmental complexity index, a feature vector is formed, and based on the feature vector, the current decision tree is trained by node splitting to obtain a trained single decision tree.

13. The Bluetooth positioning method according to claim 12, characterized in that, The step of performing node splitting training on the current decision tree based on the feature vector to obtain a trained single decision tree includes: Traverse all candidate split points corresponding to all features in the feature vector, and use the minimum overall root mean square error after node splitting as the splitting criterion to recursively complete node splitting layer by layer. When the node depth reaches the preset maximum depth or the number of samples in the node is less than or equal to the preset threshold, node splitting stops and leaf nodes are generated. The leaf node outputs the average ranging error of the corresponding training sample, and uses the average ranging error as the ranging error prediction value of the current decision tree.

14. The Bluetooth positioning method according to claim 12, characterized in that, The process of training the current decision tree by splitting nodes based on the feature vector also includes: Calculate the sum of the root mean square error reductions for each feature across all split nodes, and use the sum of the root mean square error reductions as the parameter importance weight of the feature in the current decision tree; For all decision trees in the random forest model, the average of the parameter importance weights of the same feature in each decision tree is used as the parameter importance weight of the feature in the random forest model.

15. The Bluetooth positioning method according to claim 11, characterized in that, The hyperparameters of the pre-trained random forest model are determined in the following way: Based on Latin hypercube sampling, multiple sets of hyperparameter combinations are generated within a first preset range of the number of decision trees and a second preset range of the depth of a single tree, and corresponding multiple sets of candidate random forest models are constructed. Calculate the out-of-bag error for each candidate random forest model, and select the hyperparameters corresponding to the candidate random forest model with the smallest out-of-bag error as the hyperparameters of the pre-trained random forest model.

16. The Bluetooth positioning method according to claim 11, characterized in that, Also includes: The user equipment local training mode is scheduled and executed, and the random forest model is incrementally trained based on the user's personalized training data to obtain a personalized random forest model adapted to the user. Alternatively, the vehicle-mounted system can be scheduled to perform training, and the random forest model can be incrementally trained based on the user's personalized training data to obtain a personalized random forest model adapted to the user.

17. The Bluetooth positioning method according to any one of claims 1 to 3, characterized in that, The ranging parameters include at least one of the following: single-round ranging duration, total ranging duration, connection interval, number of ranging frequency points, channel state information sampling interval, and antenna configuration index.

18. A Bluetooth positioning device, characterized in that, include: The acquisition module is used to acquire the current environmental interference data and multiple sets of initial ranging parameters of the vehicle in response to the establishment of a Bluetooth communication connection between the user device loaded with the vehicle digital key and the vehicle. The current environmental interference data includes interference data corresponding to at least one of multipath interference, occlusion interference and electromagnetic interference. The determination module is used to determine the current environmental complexity index based on the current environmental interference data; The parameter optimization module is used to perform parameter optimization processing on the multiple sets of initial ranging parameter combinations according to the current environment complexity index, so as to obtain the target ranging parameter combination; The Bluetooth positioning module is used to perform Bluetooth positioning of the user equipment based on the target ranging parameter combination.

19. A Bluetooth positioning system, characterized in that, include: The cloud service platform, vehicle terminal, and user equipment terminal; the vehicle terminal includes a Bluetooth controller and a vehicle infotainment system, and the user equipment terminal is loaded with a vehicle digital key APP. The cloud service platform is used to respond to a terminal model download request by pushing the trained parameter selection model to the vehicle terminal or to the user equipment terminal. The Bluetooth controller is configured to execute the Bluetooth positioning method as described in any one of claims 1 to 17 in response to the establishment of a Bluetooth communication connection between the user equipment terminal and the vehicle terminal.

20. The Bluetooth positioning system according to claim 19, characterized in that, The cloud service platform includes a digital key service module, a remote control module, a data module, and an algorithm module; The digital key service module is communicatively connected to the vehicle digital key APP, used to interact with the vehicle digital key APP on training data and model data, and provides a data transmission interface. The remote control module is connected to the vehicle's infotainment system and is used to push the trained parameter selection model to the vehicle's infotainment system in response to the model download request. The data module is used to organize and standardize the collected data uploaded by users to generate a training dataset suitable for model training. The algorithm module is used to perform offline training and hyperparameter optimization of the parameter selection model based on the training dataset, so as to obtain the trained parameter selection model.

21. The Bluetooth positioning system according to claim 20, characterized in that, The vehicle digital key APP includes a training module; The training module is used to download the trained parameter selection model from the digital key service module, and incrementally train the parameter selection model based on the user's personalized training data to obtain a personalized parameter selection model adapted to the user.

22. The Bluetooth positioning system according to claim 20, characterized in that, The vehicle system is used to download the trained parameter selection model from the remote control module, and perform incremental training on the parameter selection model based on the user's personalized training data to obtain a personalized parameter selection model adapted to the user.

23. A Bluetooth positioning device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the Bluetooth positioning method as described in any one of claims 1 to 17.

24. A vehicle, characterized in that, include: At least one processor and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the Bluetooth positioning method as described in any one of claims 1 to 17.

25. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the Bluetooth positioning method as described in any one of claims 1 to 17.

26. A computer program product, characterized in that, include: A computer program that, when executed by a processor, implements the Bluetooth positioning method according to any one of claims 1 to 17.