A bluetooth path loss model parameter self-adaptive calibration method and system

Abstract

The application belongs to the technical field of indoor positioning, and discloses a Bluetooth path loss model parameter self-adaptive calibration method and system. Through fusing a historical learning mechanism and a hybrid optimization strategy, the application realizes dynamic updating of path loss model parameters, and takes into account global search capability and local optimization precision. Compared with existing calibration methods of fixed parameters or single optimization algorithms, the application can automatically correct model parameters according to environmental changes, and solves the problem of precision reduction of traditional models under the conditions of multipath effect, shielding interference and environmental mutation. By introducing a hybrid optimization framework of a genetic algorithm and a particle swarm algorithm, the parameter optimization process obtains high-quality initial values in the global search stage, and quickly converges in the local fine adjustment stage, thereby significantly improving calibration efficiency and precision. Meanwhile, the addition of the historical learning mechanism enables the model to have time memory characteristics, can fuse historical parameter information to smooth mutation, and improves stability and robustness of the method in a complex dynamic environment.

Description

Technical Field

[0001] This invention belongs to the field of indoor positioning technology and relates to an adaptive calibration method and system for Bluetooth path loss model parameters. Background Technology

[0002] Bluetooth Low Energy (BLE) positioning technology is widely used in smart buildings, asset management, and indoor navigation due to its low power consumption, low cost, and ease of deployment. Its core principle is to establish a path loss model based on the Received Signal Strength Indicator (RSSI), estimate the distance between transmitting and receiving nodes based on signal strength, and calculate the location using geometric or least squares methods. However, in complex indoor environments, signal propagation is often affected by multipath effects, wall obstruction, metal reflection, and human activity, causing significant fluctuations in RSSI values ​​and unstable parameters in the path loss model. A path loss model with fixed parameters cannot dynamically reflect environmental changes, leading to a distortion in the mapping relationship between signal strength and distance. This results in accumulated ranging errors, significantly reducing the system's positioning accuracy and robustness.

[0003] Commonly used path loss model parameter calibration methods include static parameter estimation methods based on empirical fitting, local modeling methods based on regional segmentation, and global search methods based on optimization algorithms. Among these, static parameter estimation methods based on empirical fitting rely on offline data, making them difficult to adapt to dynamic environments; local modeling methods based on regional segmentation are limited by regional divisions, leading to unstable accuracy; and while single optimization algorithms have some adaptability, they are prone to getting trapped in local optima or parameter oscillations, making efficient and stable real-time calibration difficult. Furthermore, existing methods generally ignore the temporal correlation of channel parameters and fail to fully utilize historical calibration results for dynamic correction, resulting in a lack of buffering capacity for sudden environmental disturbances.

[0004] In summary, existing path loss model parameter calibration methods have the following drawbacks: 1. Fixed path loss model parameters, unable to dynamically reflect environmental changes: Existing path loss models generally use static empirical parameters. When the indoor environment changes, the parameters cannot be adjusted in real time, leading to distortion of the mapping relationship between RSSI and distance, and a significant increase in ranging error. 2. Limitations of single optimization algorithms: Traditional calibration methods based on single optimization algorithms suffer from slow global convergence or premature local convergence when searching for path loss parameters, making it difficult to balance search accuracy and computational efficiency. 3. Excessive reliance on single optimization results: Existing path loss model parameter calibration methods lack historical information fusion mechanisms and fail to utilize the temporal correlation of the RSSI channel to smoothly correct parameters, resulting in sensitivity to environmental changes, severe parameter oscillations, and reduced system stability. Summary of the Invention

[0005] The purpose of this invention is to propose an adaptive calibration method for Bluetooth path loss model parameters. This method can integrate historical learning mechanisms and hybrid optimization strategies to achieve dynamic updating and robust correction of Bluetooth path loss model parameters, thereby enabling the Bluetooth path loss model to maintain high-precision positioning performance in complex and ever-changing environments.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] An adaptive calibration method for Bluetooth path loss model parameters includes the following steps:

[0008] Step 1. Deploy Bluetooth beacons within the indoor space to be calibrated, place receiving units at different known distances, and collect a set of data using the standard sampling method. The signal and its corresponding distance are estimated using a path loss model to obtain initial parameters.

[0009] Step 2. Keeping the Bluetooth beacon position unchanged, continuously collect data from the Bluetooth beacon at a fixed distance. Signal, and will collect continuous The sequences are arranged in chronological order to construct a sliding window, with a sliding window length of [length missing]. The step size is 1. It is a natural number;

[0010] Step 3. Starting with the first sliding window, the parameter calibration process within each sliding window is as follows:

[0011] Step 3.1. Based on the initial parameters and the current sliding window... indivual The signal, combined with the path loss model, is used to fit the current sliding window. A reference distance is used to form the initial model parameter set with the initial parameters;

[0012] Step 3.2. First, perform global joint optimization on the initial model parameter set to obtain a set of globally optimal parameter sets; then, perform local search in the neighborhood of the obtained globally optimal parameter set and output a set of locally optimal parameter sets as the parameter optimization result.

[0013] Step 3.3. Construct a length of Historical parameter queue It updates historical parameters using a first-in, first-out (FIFO) principle; and assigns parameters sequentially using a time-decay weighted strategy. Different weights for each historical parameter It is a natural number;

[0014] The parameter optimization results of the current sliding window are used to perform a weighted average with each historical parameter in the historical parameter queue to obtain the parameter calibration fusion result of the current sliding window, which is then stored as a historical parameter. ;

[0015] Step 4. Shift by one step size and repeat step 3; after all sliding windows have been solved, use the parameter calibration fusion result of the last sliding window as the final path loss model parameter to complete the adaptive calibration process.

[0016] Furthermore, based on the aforementioned adaptive calibration method for Bluetooth path loss model parameters, this invention also proposes a corresponding adaptive calibration system for Bluetooth path loss model parameters, which adopts the following technical solution:

[0017] An adaptive calibration system for Bluetooth path loss model parameters includes a Bluetooth beacon, a receiving unit, and a computer device; wherein the receiving unit is connected to the computer device; the computer device includes a memory and a processor;

[0018] A computer program is stored in the memory; when the processor executes the computer program, it implements the steps of an adaptive calibration method for Bluetooth path loss model parameters as described above.

[0019] The present invention has the following advantages:

[0020] As described above, this invention constructs an adaptively adjustable low-power Bluetooth path loss model parameter calibration method by integrating a historical learning mechanism and a hybrid optimization strategy. This method achieves dynamic updating of path loss model parameters while simultaneously considering both global search capability and local optimization accuracy, exhibiting significant technical advantages. Compared to existing calibration methods with fixed parameters or single optimization algorithms, this invention can automatically correct model parameters according to environmental changes, effectively solving the problem of accuracy degradation in traditional models under multipath effects, occlusion interference, and sudden environmental changes. By introducing a multi-level hybrid optimization framework combining genetic algorithms and particle swarm optimization, the parameter optimization process obtains high-quality initial values ​​in the global search phase and converges rapidly in the local fine-tuning phase, thereby significantly improving calibration efficiency and accuracy. Simultaneously, the inclusion of a historical learning mechanism endows the model with temporal memory characteristics, enabling the integration of historical parameter information to smoothly handle sudden changes, thus enhancing the stability and robustness of the method in complex dynamic environments. Attached Figure Description

[0021] Figure 1 This is a flowchart of the adaptive calibration method for Bluetooth path loss model parameters in an embodiment of the present invention;

[0022] Figure 2 This is a flowchart illustrating the processing of the combination of the GA algorithm and the PSO algorithm in an embodiment of the present invention.

[0023] Figure 3 This is a schematic diagram showing the arrangement of the Bluetooth beacon and the receiving unit in an embodiment of the present invention;

[0024] Figure 4 This is a structural block diagram of the Bluetooth path loss model parameter adaptive calibration system in an embodiment of the present invention;

[0025] Figure 5 This is a schematic diagram of the arrangement of the Bluetooth beacon and the receiving unit in a specific embodiment of the present invention. Detailed Implementation

[0026] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0027] This embodiment addresses the problems of existing Bluetooth Low Energy (BLE) path loss models in complex indoor environments, such as fixed parameters, poor dynamic adaptability, insufficient optimization accuracy, and slow response to environmental changes. It provides an adaptive calibration method for Bluetooth path loss model parameters that integrates a historical learning mechanism and a hybrid optimization strategy. By establishing a multi-level hybrid optimization framework, a genetic algorithm (GA) is used for global parameter search, followed by particle swarm optimization (PSO) for local fine-tuning, achieving high-precision calibration of core parameters such as path loss exponent, reference signal strength, and reference distance. Simultaneously, a time-attenuation-weighted historical learning mechanism is introduced to dynamically fuse historical parameter information, enabling the model to maintain sensitive response while possessing good smoothness and anti-interference capabilities. Through this design, real-time updates and adaptive adjustments of path loss model parameters can be achieved, maintaining high positioning accuracy and system robustness in different indoor scenarios. This fundamentally improves the performance degradation problem of traditional methods under environmental changes, providing an effective technical approach for high-precision calibration of BLE indoor positioning systems.

[0028] like Figure 1 As shown, the adaptive calibration method for Bluetooth path loss model parameters in this embodiment includes the following steps:

[0029] Step 1. Deploy Bluetooth beacon AP1 in the indoor space to be calibrated, and place receiving units at different known distances, such as... Figure 3 As shown, a set of samples was collected using the standard sampling method. The signal and its corresponding distance are estimated using a path loss model to obtain initial parameters. The obtained initial parameters are the path loss exponent. Signal strength at reference distance .

[0030] Specifically, a low-power Bluetooth beacon is deployed as a fixed signal transmitter in the indoor space to be calibrated, and a Bluetooth gateway is configured as a receiving unit to receive signals in real time, which is the received signal strength.

[0031] The broadcast power of the Bluetooth Low Energy beacon is set to 0dBm, and the broadcast interval is 200ms.

[0032] Receiver units are placed at different known distances, and the received signal strength is collected using a standard sampling method. and corresponding distance To form a sample set : ; This represents the number of samples.

[0033] The principle of the Standard Sampling Method is as follows:

[0034] In the process of calibrating the path loss model, in order to obtain the model parameters and An experimental method for accurately estimating RSSI (or received power) data at a set of known standard distance points and establishing a "power-distance" mapping relationship.

[0035] Based on the collected sample data, a path loss model is established:

[0036] .

[0037] in This represents the actual distance between the receiving unit and the Bluetooth beacon location. for The received signal strength at that location For the signal strength at the reference distance, This is the path loss index. For reference distance, This is the noise term.

[0038] The linear form of the path loss model is shown below:

[0039] .

[0040] in, ;by Establish a linear equation:

[0041] .

[0042] in, The coefficients of the linear equation are obtained using the least squares method:

[0043] .

[0044] in Indicates all The average value, Indicates all The average value;

[0045] The initial parameters estimated using the path loss model are: .

[0046] Step 1 is the traditional Bluetooth calibration method, and the traditional method only uses this one step. and Signal-to-distance conversion is performed, and the calibration method of this invention is based on this initial estimate. and Further, more precise adaptive calibration will be performed.

[0047] Step 2. Keeping the Bluetooth beacon position unchanged, continuously collect data from the Bluetooth beacon at a fixed distance. Signal, and will collect continuous The sequences are arranged in chronological order to construct a sliding window, with a sliding window length of [length missing]. The step size is 1. It is a natural number.

[0048] In the same experimental environment as in step 1, keeping the spatial position of the Bluetooth beacon device unchanged, the receiver is used as the core data acquisition node to continuously receive the RSSI signal from the Bluetooth beacon at a fixed distance, with the reference distance... unknown.

[0049] Continuous The sequence is divided into multiple sliding windows in chronological order, and the length of each sliding window is... For example, setting it to 20 means that each sliding window contains 20 consecutive RSSI data points.

[0050] For example, the first sliding window, which contains elements from continuous... The second sliding window starts from the first sampling point to the 20th sampling point of the sequence. The sampling points from the 2nd to the 21st of the sequence are then repeated.

[0051] The overlap rate between any two adjacent sliding windows is 95%, meaning that a new sliding window is generated each time a sampling point is slid.

[0052] Each sliding window undergoes an independent parameter calibration process (the parameter calibration process for each window will be described in detail in step 3 below) to capture the channel's time-varying characteristics. The specific rules for constructing the sliding window are as follows:

[0053] In the formula, Indicates the window number. For window length, Total data volume; This indicates the RSSI values ​​in different orders within the window; It is an index variable.

[0054] Step 3. Starting with the first sliding window, perform parameter calibration within each sliding window.

[0055] It can be roughly divided into two levels of processing:

[0056] The first stage employs the GA algorithm to globally and jointly optimize the path loss exponent, signal strength at the reference distance, and the reference distance. Then, the PSO algorithm is used to perform a fine local search near the optimal solution of GA to obtain the current optimization result.

[0057] The second stage introduces a historical learning mechanism. By maintaining a queue of historical parameters and combining it with a time decay weighting strategy, historical parameters are integrated with the current optimization results to achieve dynamic parameter calibration, suppress parameter fluctuations caused by sudden environmental changes, and improve system robustness.

[0058] The parameter calibration process within each sliding window is described in detail below.

[0059] Step 3.1. Based on the initial parameters and the current sliding window... indivual The signal, combined with the path loss model, is used to fit the current sliding window. A reference distance is used to form the initial model parameter set with the initial parameters.

[0060] Within each sliding window, a logarithmic path loss model is established as shown above.

[0061] The initial parameters obtained in step 1 and As a baseline reference value, and based on the current sliding window indivual The signal, through the path loss model, fits the current sliding window. Each reference distance is defined as follows: , .

[0062] Initial model parameter set As a search variable in the global joint optimization algorithm.

[0063] Step 3.2. First, perform global joint optimization on the initial model parameter set to obtain a set of globally optimal parameter sets; then, perform local search in the neighborhood of the obtained globally optimal parameter set to output a set of locally optimal parameter sets as the parameter optimization results.

[0064] First, an initial population for the Genetic Algorithm (GA) is constructed using the initial model parameter set. Then, the optimal solution is searched in the global parameter space through selection, crossover, and mutation genetic operations, outputting a set of globally optimal parameter sets. .

[0065] in This represents the signal strength, path loss index, and reference distance at the optimized reference distance.

[0066] The globally optimal parameter set It can provide high-quality initial solutions for local optimization.

[0067] like Figure 2 As shown, the specific processing procedure of the genetic algorithm is as follows:

[0068] First, set the initial model parameters. Treating each individual as an individual, an initial population is constructed using real-number encoding; the population size is set to 100, the crossover probability is 0.7, the mutation probability is 0.2, and the maximum number of iterations is 60 generations.

[0069] The formula for constructing the initial population is as follows:

[0070] ;

[0071] ;

[0072] ;

[0073] in The values ​​are the signal strength, path loss exponent, and reference distance at the reference distance to be optimized.

[0074] In the formula The initial estimated parameters are the signal strength at the reference distance, the path loss exponent, and the reference distance, respectively.

[0075] The random disturbance factor follows an interval... The uniform distribution of the initial parameters is used to generate random individuals within ±10% of the initial parameters, thereby constructing a diverse initial population.

[0076] Where MSE represents the mean square error between the model-calculated value and the observed RSSI.

[0077] Then, high-quality individuals are selected through a tournament selection operator, parameters are recombined using an arithmetic crossover operator, and population diversity is maintained through a Gaussian mutation operator; as the iteration proceeds, the population gradually evolves and approaches the optimal solution;

[0078] After the genetic algorithm converges, it outputs the globally optimal parameter set. This globally optimal parameter set can accurately characterize the channel characteristics within the current time window and provide a high-quality initial solution for subsequent local optimization.

[0079] in This represents the signal strength, path loss index, and reference distance at the optimized reference distance.

[0080] Although GA has global search capabilities, its convergence speed is slow near the optimal region.

[0081] To further improve the accuracy of parameter solving, this invention introduces the Particle Swarm Optimization (PSO) algorithm for local fine-tuning within the neighborhood of the GA output result, thereby accelerating convergence and eliminating residual errors.

[0082] Specifically, with the globally optimal parameter set as the center, the particle swarm optimization (PSO) algorithm is used to construct a particle swarm search space in its neighborhood and randomly generate multiple particles. The velocity and position are iteratively updated, and a set of locally optimal parameter sets is output.

[0083] The specific processing steps of the particle swarm optimization algorithm are as follows:

[0084] The globally optimal parameter set output by the genetic algorithm Centered on the particle swarm search space, a particle swarm search space is constructed within its neighborhood.

[0085] Define the possible parameter boundaries for each particle's position update as follows:

[0086] ;

[0087] ;

[0088] In the formula, , Upper and lower bounds for updating particle positions; and These represent the path loss exponent optimized by the genetic algorithm and the signal strength at the reference distance, respectively. and This is a scaling factor used to limit the local search range.

[0089] The constants 2.3 and 3.7 represent the physically reasonable range of the path loss exponent. 2.3 indicates the lower limit of decay close to free space, and 3.7 indicates the upper limit in multipath and occlusion environments, used to avoid the algorithm searching for non-physical solutions.

[0090] Based on the statistical characteristics of Bluetooth signal propagation indoors, the path loss index is typically between 2.0 and 4.0.

[0091] The constants –45 and –75 represent empirical boundary values ​​for signal strength. The Bluetooth received signal strength at the reference distance typically varies between –45 dBm and –75 dBm; –45 dBm corresponds to a strong signal scenario, and –75 dBm corresponds to a weak signal scenario.

[0092] Limiting this range ensures that the model fitting results conform to the actual power decay law.

[0093] Parameters 0.2 and 3 represent the local perturbation values ​​of the path loss exponent and signal strength, respectively, indicating the search step size of the particle swarm near the optimal solution of the genetic algorithm. By setting perturbations of ±0.2 and ±3 dBm, a fine search of the local optimum can be achieved while maintaining the physical meaning of the parameters.

[0094] Forty particles are randomly generated within the particle swarm search space, with a maximum of 80 iterations; the initial inertia weight ω is set to 0.9 and linearly decreased to 0.4; the learning factor... Random factor It takes values ​​within the interval [0,1].

[0095] Each particle represents a set of candidate parameter vectors. .

[0096] in These represent the signal strength at the candidate reference distance, the path loss index, and the reference distance, respectively.

[0097] In each iteration, the particle swarm optimization algorithm updates its search direction based on the historical best and global best of the particles.

[0098] Speed ​​updated to:

[0099] .

[0100] Location updated to:

[0101] .

[0102] In the formula: For particles In iteration The velocity vector at time; For particle i in iteration Position vector at time; Inertial weight; , For learning factors; , It is a random number in the interval [0,1].

[0103] Represents particles The historical best position; This indicates the group's historical best position.

[0104] By iteratively calculating the velocity and position update equations, particles dynamically adjust their positions within the parameter space and gradually approach the global optimum based on the fitness function. As the particle swarm optimization algorithm converges, the swarm's historical optimal position is determined. Selected as the final output parameter, where and These correspond to the signal strength at the reference distance and the path loss exponent, respectively, in the path loss model. This is the reference distance. This parameter set represents the combination that minimizes the model's prediction error within the current time window.

[0105] This invention takes the globally optimal parameter set as input, constructs a particle swarm search space in its neighborhood, performs iterative updates of velocity and position, outputs a locally optimal parameter set as the parameter optimization result, and passes it to the historical learning mechanism for fusion.

[0106] This invention is based on a two-stage optimization structure of GA and PSO. GA is responsible for global search of the optimal solution space, while PSO performs local fine search in its neighborhood, effectively balancing the breadth of the search and the convergence speed, and improving the accuracy and stability of parameter optimization.

[0107] Step 3.3. Construct a length of Historical parameter queue It updates historical parameters using a first-in, first-out (FIFO) principle; and assigns parameters sequentially using a time-decay weighted strategy. Different weights for each historical parameter It is a natural number.

[0108] This invention utilizes the temporal correlation characteristics of RSSI propagation to construct a dynamically evolving parameter queue. During the sliding window processing, not only are the path loss parameters of the current window optimized, but a queue with a capacity of... (For example, 10) The model employs an exponential decay weighting strategy to assign higher weights to recent historical parameters. This time decay weighting strategy aims to enable the model to respond quickly to the latest environmental changes while smoothing out historical noise, thereby achieving more accurate parameter calibration. In other words, it uses a weighted average of the current optimal solution and past experience, with higher weights for more recent data and lower weights for older data.

[0109] Establish a length of First-In-First-Out (FIFO) parameter history queue :

[0110] .

[0111] Historical parameter queue Before storage The optimization results for each time window.

[0112] Each time a new sliding window completes its parameter calculation, the optimization result of the current sliding window is shared. Push it to the end of the queue; when the queue length exceeds When the time comes, the oldest record is automatically deleted to keep the queue length constant.

[0113] Historical parameter queue Used to store historical parameters, i.e., the parameter calibration and fusion results for each sliding window, and when the historical parameter queue... Historical parameters reached Afterwards, each time a new historical parameter is added, it will be deleted. The earliest historical parameters in China.

[0114] The local optimal parameter set of the first sliding window is used as the parameter calibration fusion result of the first sliding window.

[0115] Starting from the second sliding window, the parameter calibration fusion result of the current sliding window is calculated using the historical parameter fusion mechanism; starting from the second window, the previous... The parameter results of each window are included in the fusion calculation.

[0116] If historical parameter queue The number of historical parameter records is less than Then utilize The existing historical parameters are weighted and fused; when the number of historical records reaches the length... Afterwards, each fusion uses The latest historical parameters are weighted and fused together.

[0117] When the number of historical records reaches the preset queue length Afterward, the system enters a stable self-learning phase. This design ensures that the algorithm responds quickly during the startup phase and maintains temporal smoothness and parameter stability in the later stages of operation.

[0118] The parameter optimization results of the current sliding window are used to perform a weighted average with each historical parameter in the historical parameter queue to obtain the parameter calibration fusion result of the current sliding window, which is then stored as a historical parameter. .

[0119] The specific process of the historical parameter fusion mechanism is as follows:

[0120] When the system is in the first sliding window (number k=1), due to the historical parameter queue If empty, there is no historical data available for fusion. The current window optimization result is directly output as the final parameter calibration fusion result, i.e.:

[0121] .

[0122] Starting from the second window (k=2), the system enables the historical learning fusion mechanism.

[0123] If the number of historical records is less than Then utilize the existing A weighted fusion of historical samples is performed; when the number of historical records reaches... After that, the algorithm enters a stable running phase, and the fusion process maintains a constant length.

[0124] Define a time decay coefficient α∈(0,1) and a fusion scaling coefficient λ∈(0,1); where α controls the weight of historical parameters decreasing over time, and λ controls the relative importance of the current sliding window and the results of the historical sliding windows; the weight calculation formula is:

[0125] .

[0126] in Indicates the weight of historical parameters; express The number of current historical parameters.

[0127] The fusion calculation formula is as follows:

[0128] ;

[0129] ;

[0130] .

[0131] in[ ]express The Middle One historical parameter, ; These represent the signal strength at the reference distance, the path loss index, and the reference distance, respectively.

[0132] In this embodiment, the time decay coefficient α = 0.9; the fusion ratio coefficient λ = 0.7.

[0133] The time-weighted fusion parameters were obtained through calculation. And use it as the final output of the current window; This represents the signal strength, path loss index, and reference distance at the reference distance after the current sliding window fusion.

[0134] Step 4. Shift by one step size and repeat step 3 to achieve rolling updates and dynamic calibration of the path loss model parameters.

[0135] After all sliding window calculations are completed, the parameter calibration and fusion result of the last sliding window is used as the final path loss model parameter, thus completing the adaptive calibration process of the Bluetooth path loss model parameter.

[0136] After obtaining the final path loss model parameters, substitute them into the path loss model, and the actual distance can be obtained based on the collected RSSI signal. This significantly improves indoor positioning accuracy. At the same time, it fully demonstrates the crucial role of utilizing historical data for learning and modeling in enhancing parameter calibration accuracy and algorithm generalization ability.

[0137] In addition, to verify the effectiveness of the method of the present invention, an experimental verification scheme for an adaptive calibration system for Bluetooth path loss model parameters is provided to illustrate the implementation process of the method in a real-world scenario.

[0138] In this embodiment, four Xinchida Low Power Bluetooth beacons are used as the Bluetooth signal transmitter. The broadcast interval of the beacons is set to 200 ms, and the transmission power is 0 dBm. The system is equipped with a Bluetooth receiver gateway as the receiving unit, which is connected to a computer device for data acquisition and processing, such as... Figure 5 As shown. The experimental environment was a closed indoor hall with dimensions of 6.28 m × 4.7 m. The walls were made of concrete and contained metal components, representing a typical multipath propagation environment.

[0139] Four Bluetooth beacons are fixed at the four corners of the hall, with their planar coordinates as follows: AP1 (0.00, 0.00), AP2 (0.00, 4.17), AP3 (6.28, 0.00), and AP4 (6.28, 4.17), to achieve uniform signal coverage.

[0140] To determine the initial reference values ​​for the path loss model, a Bluetooth beacon was deployed at the geometric center of the hall as a signal source, and sampling points were set at 0.5 m intervals within a range of 0–5 m from the beacon.

[0141] Multiple sets of RSSI signal data were collected at each sampling point, and linear fitting was performed using the least squares method to obtain the initial parameters of the path loss model, where the path loss exponent was approximately 3.09 and the path loss at the reference distance (1 m) was -47.13 dB.

[0142] In the above experimental environment, the coordinates of the receiving unit were set to (3.0, 2.4), and it continuously received RSSI signals from four beacons as the core data acquisition node, obtaining a total of 4560 raw data entries.

[0143] To improve data stability, a sliding window method was used to construct the data sequence. The sliding window length was set to 20 sampling points, the step size was 1, and the overlap rate of adjacent windows was about 98%, resulting in 4480 effective windows.

[0144] Outliers are removed from the signal sequence within each window using the interquartile range (IQR) algorithm to reduce measurement noise.

[0145] Subsequently, the adaptive calibration of the Bluetooth path loss model parameters is performed within the computer device, including the following steps:

[0146] I. A genetic algorithm is used for global optimization, with an initial population size of 100 individuals and a maximum number of iterations of 60 generations. Tournament selection, mixed crossover operator (probability 0.7), and Gaussian mutation operator (probability 0.2) are used to maintain population diversity.

[0147] II. Based on the optimal solution output by the genetic algorithm, a local fine-grained search is performed using the particle swarm optimization algorithm. The number of particles is set to 40, the maximum number of iterations is 80, the learning factor is 0.5, and the inertia weight is linearly decreased from 0.9 to 0.4.

[0148] III. Based on the optimization results, a historical parameter fusion mechanism is introduced. A decay factor of 0.9 is set to update the historical parameters with weights, and the fusion ratio is 70% of the current optimal parameters and 30% of the historical parameters, so as to achieve smooth adaptive adjustment of the model parameters.

[0149] This experimental process enables automatic calibration and dynamic correction of path loss model parameters. After multiple iterations, the system obtains stable path loss exponent and reference signal strength parameters, providing reliable input for subsequent Bluetooth positioning algorithms.

[0150] To further verify the adaptability of the method in different scenarios, the above experimental process can be repeated in different indoor and outdoor environments (such as dormitories, open spaces, stairwells, underground parking garages, etc.) to collect RSSI data and perform parameter calibration. The system completes the adaptive adjustment of the model through the same algorithm process to achieve stable estimation of path loss characteristics in complex multipath environments.

[0151] This invention introduces a historical learning mechanism and constructs a historical parameter queue and a time decay weighting strategy to achieve smooth updates and dynamic corrections of parameters over time, significantly improving the robustness of the model under sudden environmental changes.

[0152] In addition, during the optimization process, the three key parameters of path loss exponent, reference signal strength and reference distance are solved simultaneously, which breaks through the limitations of the traditional single-parameter fitting model and makes the calibration results more consistent with the physical characteristics of complex indoor channels.

[0153] Example 2

[0154] This embodiment 2 describes an adaptive calibration system for Bluetooth path loss model parameters, including a Bluetooth beacon, a receiving unit, and a computer device; wherein the receiving unit is connected to the computer device, such as... Figure 4 As shown.

[0155] The computer device includes a memory and a processor; a computer program is stored in the memory; when the processor executes the computer program, it is used to implement the Bluetooth path loss model parameter adaptive calibration method as described in Embodiment 1 above.

[0156] Of course, the above description is only a preferred embodiment of the present invention. The present invention is not limited to the above-described embodiments. It should be noted that any equivalent substitutions or obvious modifications made by those skilled in the art under the guidance of this specification fall within the scope of this specification and should be protected by the present invention.

Claims

1. A method for adaptive calibration of Bluetooth path loss model parameters, characterized in that, Includes the following steps: Step 1. Deploy Bluetooth beacons within the indoor space to be calibrated, place receiving units at different known distances, and collect a set of data using the standard sampling method. The signal and its corresponding distance are estimated using a path loss model to obtain initial parameters. Step 2. Keeping the Bluetooth beacon position unchanged, continuously collect data from the Bluetooth beacon at a fixed distance. Signal, and will collect continuous The sequences are arranged in chronological order to construct a sliding window, with a sliding window length of [length missing]. The step size is 1. It is a natural number; Step 3. Starting with the first sliding window, the parameter calibration process within each sliding window is as follows: Step 3.

1. Based on the initial parameters and the current sliding window... indivual The signal, combined with the path loss model, is used to fit the current sliding window. A reference distance is used to form the initial model parameter set with the initial parameters; Step 3.

2. First, perform global joint optimization on the initial model parameter set to obtain a set of globally optimal parameter sets; then, perform local search in the neighborhood of the obtained globally optimal parameter set to output a set of locally optimal parameter sets as the parameter optimization results. Step 3.

3. Construct a length of Historical parameter queue It updates historical parameters using a first-in, first-out (FIFO) principle; and assigns parameters sequentially using a time-decay weighted strategy. Different weights for each historical parameter It is a natural number; The parameter optimization results of the current sliding window are used to perform a weighted average with each historical parameter in the historical parameter queue to obtain the parameter calibration fusion result of the current sliding window, which is then stored as a historical parameter. ; Step 4. Shift by one step size and repeat step 3; after all sliding windows have been solved, use the parameter calibration fusion result of the last sliding window as the final path loss model parameter to complete the adaptive calibration process.

2. The adaptive calibration method for Bluetooth path loss model parameters according to claim 1, characterized in that, The initial model parameter set, the global optimal parameter set, and the local optimal parameter set all include three path loss model parameters: path loss exponent, signal strength at the reference distance, and reference distance. The initial parameters obtained in step 1 are the path loss index and the signal strength at the reference distance.

3. The adaptive calibration method for Bluetooth path loss model parameters according to claim 1, characterized in that, The logarithmic expression of the path loss model is as follows: ； in This represents the actual distance between the receiving unit and the Bluetooth beacon location. for The received signal strength at that location For the signal strength at the reference distance, This is the path loss index. For reference distance, This is the noise term.

4. The adaptive calibration method for Bluetooth path loss model parameters according to claim 1, characterized in that, Step 3.1 specifically involves: The initial parameters obtained in step 1 and As a baseline reference value, and based on the current sliding window indivual The signal, through the path loss model, fits the current sliding window. Each reference distance is defined as follows: , ; Initial model parameter set As a search variable in the global joint optimization algorithm.

5. The adaptive calibration method for Bluetooth path loss model parameters according to claim 1, characterized in that, Step 3.2 specifically involves: First, the initial population of the genetic algorithm is constructed using the initial model parameter set. Then, the optimal solution is searched in the global parameter space through selection, crossover, and mutation genetic operations, and a set of globally optimal parameter sets is output. Then, with the globally optimal parameter set as the center, the particle swarm algorithm is used to construct a particle swarm search space in its neighborhood and randomly generate multiple particles. The velocity and position are iteratively updated, and a set of locally optimal parameter sets is output.

6. The adaptive calibration method for Bluetooth path loss model parameters according to claim 5, characterized in that, In step 3.2, the specific processing procedure of the genetic algorithm is as follows: First, set the initial model parameters. Treating each individual as an individual, an initial population is constructed using real-number encoding; the population size is set to 100, the crossover probability is 0.7, the mutation probability is 0.2, and the maximum number of iterations is 60 generations; The formula for constructing the initial population is as follows: ； ； ； in The signal strength, path loss exponent, and reference distance at the reference distance to be optimized are given. In the formula The initial estimated parameters are the signal strength at the reference distance, the path loss exponent, and the reference distance, respectively. The random disturbance factor follows an interval... The uniform distribution of the initial parameters is used to generate random individuals within ±10% of the initial parameters, thereby constructing a diverse initial population. Then, high-quality individuals are selected through a tournament selection operator, parameters are recombined using an arithmetic crossover operator, and population diversity is maintained through a Gaussian mutation operator; as the iteration proceeds, the population gradually evolves and approaches the optimal solution; After the genetic algorithm converges, it outputs the globally optimal parameter set. ; in This represents the signal strength, path loss index, and reference distance at the optimized reference distance.

7. The adaptive calibration method for Bluetooth path loss model parameters according to claim 5, characterized in that, In step 3.2, the specific processing procedure of the particle swarm algorithm is as follows: The globally optimal parameter set output by the genetic algorithm Centered on the particle swarm search space, a particle swarm search space is constructed within its neighborhood. Define the possible parameter boundaries for each particle's position update as follows: ； ； In the formula, , Upper and lower bounds for updating particle positions; and These represent the path loss exponent optimized by the genetic algorithm and the signal strength at the reference distance, respectively. and This is a scaling factor used to limit the local search range; The constants 2.3 and 3.7 are the lower and upper limits of the physically reasonable range of the path loss exponent; the constants –45 and –75 are the empirical boundary values ​​of the signal strength, respectively; where –45 dBm corresponds to a strong signal scenario and –75 dBm corresponds to a weak signal scenario. Parameters 0.2 and 3 represent the path loss exponent and the local disturbance value of the signal strength, respectively. Forty particles are randomly generated within the particle swarm search space, with a maximum of 80 iterations; the initial inertia weight ω is set to 0.9 and linearly decreased to 0.4; the learning factor... Random factor It takes values ​​within the interval [0,1]. Each particle represents a set of candidate parameter vectors. ; in These represent the signal strength at the candidate reference distance, the path loss exponent, and the reference distance, respectively. In each iteration, the particle swarm optimization algorithm updates its search direction based on the historical and global optima of the particles. As the particle swarm optimization algorithm converges, the population's historical optimal position... Selected as the final output parameter, where and These correspond to the signal strength at the reference distance and the path loss exponent, respectively, in the path loss model. For reference distance.

8. The adaptive calibration method for Bluetooth path loss model parameters according to claim 1, characterized in that, In step 3.3, the historical parameter queue Used to store historical parameters, i.e., the parameter calibration and fusion results for each sliding window, and when the historical parameter queue... Historical parameters reached Afterwards, each time a new historical parameter is added, it will be deleted. The earliest historical parameters in China; The local optimal parameter set of the first sliding window is used as the parameter calibration fusion result of the first sliding window; Starting from the second sliding window, the parameter calibration fusion result of the current sliding window is calculated using the historical parameter fusion mechanism; If historical parameter queue The number of historical parameter records is less than Then utilize The existing historical parameters are weighted and fused; when the number of historical records reaches the length... Afterwards, each fusion uses The latest historical parameters are weighted and fused together.

9. The adaptive calibration method for Bluetooth path loss model parameters according to claim 1, characterized in that, The specific process of the historical parameter fusion mechanism is as follows: Define a time decay coefficient α∈(0,1) and a fusion scaling coefficient λ∈(0,1); where α controls the weight of historical parameters decreasing over time, and λ controls the relative importance of the current sliding window and the results of the historical sliding windows; the weight calculation formula is: ； in Indicates the weight of historical parameters; express The number of current historical parameters; The fusion calculation formula is as follows: ； ； ； in[ ]express The Middle One historical parameter, ; These represent the signal strength at the reference distance, the path loss exponent, and the reference distance, respectively. The time-weighted fusion parameters were obtained through calculation. And use it as the final output of the current window; This represents the signal strength, path loss index, and reference distance at the reference distance after the current sliding window fusion.

10. A Bluetooth path loss model parameter adaptive calibration system, comprising a Bluetooth beacon, a receiving unit, and a computer device; wherein the receiving unit is connected to the computer device; the computer device includes a memory and a processor; The memory stores a computer program; its characteristic is that... When the processor executes the computer program, it implements the steps of an adaptive calibration method for Bluetooth path loss model parameters as described in any one of claims 1 to 9.

Images