Calibration application method for signal intensity positioning of Internet of Things equipment
By establishing a path loss model related to air density and humidity in IoT devices, the problem of inaccurate measurements caused by neglected environmental factors in the prior art is solved, and high accuracy and cost savings in the measurement of distance between devices are achieved.
Patent Information
- Application Number
- CN202510530187.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-25
- Publication Date
- 2025-05-30
AI Technical Summary
When measuring distance between IoT devices, the prior art ignores the influence of environmental factors such as air density and humidity, resulting in inaccurate measurement results, high equipment costs and limited applicability.
A calibration application method for signal strength positioning of IoT devices is adopted. A path loss model considering air density and humidity is established before leaving the factory. After installation, the actual distance between devices is determined by detecting environmental factors and using the path loss model for accurate calculations.
It significantly improves measurement accuracy, is suitable for complex environmental conditions, reduces equipment hardware cost and calculation complexity, and realizes efficient calibration and model verification.
Smart Images
Figure CN120074696A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical fields of communication data processing and Internet of Things (IoT) positioning, and particularly relates to a calibration application method for IoT device signal strength positioning. Background Art
[0002] Civil and general commercial IoT systems and their participating devices have the following technical characteristics: 1. The installation layout spacing between devices in the system is small, usually with a maximum interval of dozens of meters, and the installation interval is less than or equal to the effective communication distance of their own communication modules. 2. The communication frequency and communication transmission power of IoT devices generally do not change easily and will be a fixed value after the wireless communication module and application scenario used are determined. 3. The receiving and transmitting gains of the antennas used by IoT devices are fixed, and the antenna parameters of the devices basically do not change after leaving the factory. 4. The wireless communication capabilities, storage capabilities, and program operation capabilities of all IoT devices in the same system are basically the same, and these functional parameters are much smaller than those of base station devices. Some calibration, debugging, and positioning methods applicable to base stations are not suitable for these civil and general IoT systems.
[0003] For example, communication base stations use the TDOA algorithm for calculating the time difference of arrival of communication signals or the AOA algorithm for the angle of arrival of signals. These two technologies have high hardware computing power requirements for the base stations for signal transmission and reception, and also require dedicated hardware for accurately measuring the signal arrival time or a dedicated array antenna for the arrival angle. The time synchronization accuracy requirements for the time synchronization between the base stations to be ranged are also very high, and precise time synchronization is required. It is applicable to the positioning applications of large communication base stations and the IoT devices supported by the surrounding hardware functions. The equipment costs of the above two methods are very high and are not suitable for popularization in the general civil field. Moreover, when using these methods in buildings, they are easily affected by the irregular placement of large objects in the building, and their reflected signals interfere with the communication arrival time or the arrival angle. For example, when the parking lot is not full, the random parking of vehicles is not easy to summarize, so the data parameters of the signals are not easy to count and calculate.
[0004] In civil and commercial IoT device networking systems, dedicated positioning base stations for measuring the arrival time of wireless signals are usually not established. The clock synchronization performance of ordinary devices in the IoT system is poor and cannot meet the basic requirements of the TDOA algorithm. Moreover, the application scenarios of these IoT devices usually do not require particularly high-precision positioning, and positioning with a distance control at the decimeter level is sufficient for use.
[0005] If there are too many IoT devices participating in the system, in order to achieve better management and usage effects, it is usually hoped to use algorithms such as machine learning to determine the position relationship between the devices in the system. It is more hoped to apply a simple, time-saving, labor-saving, and cost-saving construction and debugging method.
[0006] In summary, the signal strength, i.e., RSSI positioning technology, has relatively low overall device hardware costs and computational complexity. It is suitable for being extended to scenarios where there are a large number of devices in the system that need to determine their relative positions, and signal strength positioning will not be affected by environmental objects placed irregularly.
[0007] During the path from the transmission to the reception of IoT devices, in addition to the attenuation caused by distance, mainly air density and humidity affect the signal strength. When the air humidity increases, the water molecule content increases. Water molecules have an absorption effect on the electromagnetic waves of wireless communication and will also increase the scattering of electromagnetic waves, reducing the finally received signal strength. A higher air density will also lead to a more significant attenuation of the signal strength. Traditional methods for measuring the distance between devices often ignore the influence of environmental factors such as air density and humidity on signal propagation, resulting in inaccurate measurement and calculation results.
[0008] IoT devices usually use wireless communication technology for data transmission, and the signal transmission distance is affected by various environmental factors. In the existing technology, most distance measurement methods only consider the free space path loss and ignore the influence of environmental factors such as air density and humidity on signal propagation, resulting in certain errors in the measurement results. Summary of the Invention
[0009] The present invention provides a calibration application method for signal strength positioning of IoT devices. Its main purpose is to comprehensively consider factors such as air density and humidity that affect path loss, generate a path loss function model during the factory calibration process, and use the function model to accurately calculate the distance between IoT devices after the IoT devices are installed and applied. The relative distance calculation and measurement accuracy between devices after the deployment of the IoT system are improved.
[0010] As Figure 1 shown, the method of the present invention includes the following steps: S1: Before leaving the factory, perform the device calibration process to establish a path loss model L(d, ρ, h) considering air density and humidity; S2: When installing the device, detect the air density and humidity of the current environment; S3: After the device installation is completed, each device can act as a transmitter to send a broadcast signal including the device address, radio frequency transmission intensity, and transmitter antenna gain data; each device can act as a receiver to receive broadcast signals within the communication range, record the received device address and the received signal strength value reflected at the receiver correspondingly; each device retrieves its own receiver antenna gain value and calculates the signal strength change value according to the empirical formula. S4: Substitute the above data into the empirical formula containing the path loss model to inversely calculate the actual distance d between the devices.
[0011] Generally speaking, the signal strength weakens as the transmission distance increases. The influencing factors of environmental factors: air density and air humidity, will also affect the signal strength, and the influence increases as the transmission distance increases, which also belongs to the loss model related to the distance d. Then the aforementioned "distance loss ± environmental factor influence" can be unified and expressed in a function: L(d,ρ,h) is the path loss, f is the wireless communication frequency, d is the distance between wireless communication devices, c is the speed of light, α(ρ) is the influence coefficient of air density on path loss, β(h) is the influence coefficient of humidity on path loss, γ(ρ,h) is the interaction influence coefficient of air density and humidity, ρ is the air density, and h is the humidity.
[0012] The communication frequencies of the wireless communication modules of the same product devices in mass production are the same, and in most cases, the communication frequency is not allowed to be changed after leaving the factory. Then the device manufacturer can sample some products from this production batch during factory production to obtain the path loss calibration function of the signal strength of the products in this batch affected by air density and humidity.
[0013] The signal strength relationship between devices can be expressed by an empirical formula: Received Signal Strength = Radio Frequency Transmit Strength + Transmitter Antenna Gain – Distance Loss ± Environmental Factor Influence + Receiver Antenna Gain. Preferably, according to the foregoing content, the empirical formula is: L(d,ρ,h) = Radio Frequency Transmit Strength + Transmitter Antenna Gain + Receiver Antenna Gain – Received Signal Strength, and this formula is used for the calculation in step S4.
[0014] Optionally, α(ρ) is expressed as α(ρ) = a 1 ρ + a 2 exp(a 3 ρ), where a 1 , a 2 , a 3 are parameters to be determined during the calibration process. The linear term a 1 ρ represents the basic linear influence of air density on signal attenuation. The exponential term a 2 exp(a 3 ρ) represents that when the air density increases to a certain extent, a non-linear rapid growth effect may occur. This is based on the atmospheric attenuation theory that when the air density increases, signal scattering and absorption increase exponentially.
[0015] Optionally, β(h) is expressed as β(h) = b 1 h + b 2 ln(1 + b 3 h), where b 1 , b 2, b 3 is a parameter to be determined during the calibration process. The linear term b 1 h represents the basic linear effect of humidity on signal attenuation. The logarithmic term b 2 ln(1 + b 3 h) represents that there may be a saturation effect in the humidity influence. This is based on the water vapor absorption theory. When the humidity reaches a certain level, the increasing influence will slow down.
[0016] For the air density and humidity interaction influence coefficient γ(ρ, h) mentioned above, during calibration, only the basic interaction, or only the high-order interaction effect, or a combination of both the basic interaction and the high-order interaction effect can be adopted.
[0017] Optionally, γ(ρ, h) is expressed as , c 1 ρh represents the basic interaction between air density and humidity. The basic interaction refers to the phenomenon that the difference in the response amount between different levels of one factor changes with different levels of other factors, and is used to study the non-independent effects of several factors simultaneously. The basic interaction can measure the degree to which the effect change of different levels of one factor depends on another or several factors. c 2 ρ²h½ and c 3 ρ½h² represent the high-order interaction effects between air density and humidity. This is based on the complex atmospheric propagation model, and complex non-linear interactions will occur between air density and humidity.
[0018] Under different air density and humidity conditions, measure the signal strength at a known distance, and use regression analysis to determine the path loss model parameters.
[0019] Such as Figure 2 , the calibration process includes: the operation process of collecting calibration data of Internet of Things devices; the function modeling process of calculating the path loss model L(d, ρ, h) with the calibration data; parameter estimation and verification of the modeled function L(d, ρ, h), and those meeting the conditions are uniformly distributed to the Internet of Things devices of this batch for storage and used for calculation after installation.
[0020] The operation process of collecting calibration data of Internet of Things devices includes: The calibration device inputs the fixed frequency value of the Internet of Things device; several Internet of Things devices form a queue and are arranged in sequence, and the device interval is set as required; the signal strength values received by each device in this queue are collected by the method of controlling variables; The control variable method includes: keeping the air density constant, calibrating the β(h) coefficient, with each constant air density as a group, and multiple groups of calibrations can be carried out by adjusting the air density; keeping the humidity constant, calibrating the α(ρ), with each constant humidity as a group, and multiple groups of calibrations can be carried out by adjusting the humidity; The collected values and the device interval data are entered into the computer, and subsequently, the non-linear path loss models under the individual influences of air density and humidity are calculated through the function modeling process; Relying on multiple groups of collected values and device interval data, the path loss model affected by the interaction of air density and humidity is calculated through computer simulation; The complete path loss model function is stored in each Internet of Things device produced in this batch, and this function is called during the positioning and debugging between devices after actual installation, and the later calculation is carried out through the air density and humidity parameters at the installation site.
[0021] The function modeling process adopts the LMA fitting algorithm, including the following calculation steps: Initial parameter estimation: Using multiple linear regression analysis on the collected values to obtain the initial estimated parameter values; Optimization objective: Using the function of the estimated parameter values, calculating the minimized predicted value under the condition of the actual measured value, and calculating the mean square error between the predicted value and the actual measured value; Using LMA to iteratively update the parameter values until the mean square error converges; Cross-validation: Using k-fold cross-validation, such as k = 10, to evaluate the model performance.
[0022] The calibration process includes the following parameter estimation and verification steps: Residual analysis: Checking the normality and homoscedasticity of the residuals, and using the Durbin-Watson test to evaluate the autocorrelation of the residuals; Parameter significance test: Using the t-test to evaluate the statistical significance of each parameter; Model comparison: Using the Akaike information criterion or the Bayesian information criterion to compare models of different complexities, and selecting the optimal L(d,ρ,h) from them. When the data sample size in the calibration process is too large, preferably, the Bayesian information criterion is adopted.
[0023] The beneficial effects of the present invention are mainly reflected in the following aspects: First, by introducing a path loss model related to air density and humidity, it is possible to more accurately correct the intensity loss during signal propagation, thereby significantly improving the measurement accuracy. The adopted path loss model not only considers the individual effects of air density and humidity, but also takes into account their interactive effects to further improve the measurement accuracy. This enables the communication intensity positioning to better adapt to environmental changes, maintain the measurement accuracy, and be applicable to various complex environmental conditions. Second, efficient calibration: The calibration process adopted in the present invention combines the method of controlling variables and regression analysis, which can efficiently determine the model parameters. This method not only saves time but also improves the accuracy of the calibration results. Third, model verification and optimization: Through steps such as residual analysis, parameter significance test, and model comparison, the present invention can strictly verify and optimize the model to ensure the reliability and effectiveness of the model. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 is the calibration application method flow of the signal strength positioning of the Internet of Things device Figure 1 .
[0025] Figure 2 is the calibration application method flow of the signal strength positioning of the Internet of Things device Figure 2 .
[0026] Figure 3 is an implementation manner of the calibration operation process of the present invention. DETAILED DESCRIPTION OF THE INVENTION
[0027] The method of the present invention has the best application effect for devices with a uniform distribution of antenna polarization directions. The transceiver gain performance of such antennas for wireless communication intensity does not change with the propagation direction, and the gain is consistent among different devices without difference. The numerical values of the antenna transmit and receive gains can be obtained from the manufacturer. A typical implementation case is UWB communication, whose antenna polarization is circular.
[0028] The full name of UWB is Ultra Wide Band, which is an ultra-wideband wireless carrier communication technology. Its core is to use non-sinusoidal narrow pulses in the nanosecond level to transmit data. Its characteristics are: strong anti-interference ability, strong multipath resolution ability; able to maintain stable communication in complex environments; high positioning accuracy, can be used for precise positioning, with an accuracy of up to centimeter level; simple device structure and low cost.
[0029] Based on this empirical formula: L(d,ρ,h) = radio frequency transmission intensity + transmit end antenna gain + receive end antenna gain – received signal intensity, where the numerical values of the transmit end antenna gain and the receive end antenna gain are fixed and do not change with the device installation and the relative direction of signal transmission.
[0030] If all devices in the Internet of Things system adopt the same wireless communication module and antenna, then according to the reciprocity principle of the antenna: the basic characteristic parameters of the same antenna as a transmitter or receiver are the same, and the empirical formula can be further simplified. That is: the transmitting antenna gain = the receiving antenna gain.
[0031] When multiple communication modules or antennas are mixed in the Internet of Things system, the transmitting antenna gain and the receiving antenna gain need to be calculated separately.
[0032] In other application scenarios, such as when the antenna polarization direction is elliptical or pointer-shaped, and the polarization direction can be accurately identified and the gain for each propagation direction can be calculated, the method of the present invention can also be used to obtain a more accurate calculated value of the distance between devices, but corresponding calculations need to be made based on the function model and numerical values of the planned direction.
[0033] As Figure 1 shown, the method of the present invention includes the following steps: S1: Perform a device calibration process before leaving the factory to establish a path loss model L(d, ρ, h) considering air density and humidity; S2: When installing the device, detect the air density and humidity of the current environment; S3: After the device installation is completed, each device can act as a transmitter to send a broadcast signal including the device address, radio frequency transmission intensity, and transmitting antenna gain data; each device can act as a receiver to receive broadcast signals within the communication range, record the received device address and the received signal strength value reflected at the receiving end; each device retrieves its own receiving antenna gain value and calculates the signal strength change value according to the empirical formula. S4: Substitute the above content data into the empirical formula containing the path loss model to inversely calculate the actual distance d between devices.
[0034] The following implementation situations may occur in step S2: 1. In a certain project, due to the phased completion of the building, the Internet of Things devices in the building also need to follow the construction cycle of the building. For example, the building is divided into 1 to 4 phases and is completed in the four seasons of spring, summer, autumn, and winter respectively. Then the air density and humidity in each season are different, and the Internet of Things devices installed in phases 1 to 4 need to be measured once during each installation. 2. For the deployment of medium and large-scale Internet of Things projects, it is impossible to complete the device installation in one day. If there are sudden changes in temperature or changes between sunny and rainy days during the construction period, the air density and humidity need to be re-detected at least on the day of the change. 3. If the Internet of Things device is a street lamp and there are obvious altitude changes or changes in installation environment types such as cities or wetlands during the street lamp installation process, the instruments for detecting air density and humidity can follow the construction vehicle to collect data in real time and correct the calculation of the relative distance.
[0035] Regarding the content described in step S3, there are the following implementation manners: The completion of installation can be when the IoT device automatically starts running the embedded program after being powered on, which is applied to the scenario of machine self-learning without debugging. In this scenario, air density and humidity sensors can be evenly distributed in the IoT system and automatically complete data collection, so that the calculated result of the distance between devices is more accurate. It can also be that after all IoT devices on a floor or in a region are installed, an installation completion instruction is sent by the debugging device, corresponding to the scenario of manual debugging. Then, it only requires manual operation to collect air density and humidity once using the debugging device to complete this step.
[0036] During the calibration process, select the IoT device to be calibrated, and enter its operating frequency f in MHz, radio frequency transmission intensity P_t in dBm, and the antenna gain values of the receiving and transmitting ends in dBm into the calibration device. Under different air densities ρ in kg / m³ and relative humidities h in %, collect measurement data.
[0037] As Figure 3 shown, the data that also needs to be entered into the calibration device includes: the known distance between IoT devices in m, and the received signal strength P_r in dBm measured at each interval distance. Collect enough data points to cover the expected working environment range. The received signal strength corresponds to the device address of the broadcast signal emitted, which is convenient for identifying which two specific devices are performing signal strength data interaction. The device spacing can be arranged in the same interval in sequence as shown in the figure, so that the distances between devices include d, 2d, 3d, etc., saving operation space. The minimum device spacing d should be less than the maximum effective range of device wireless communication. Of course, the device intervals can be arranged in other intervals or arrangements according to needs, and this part belongs to well-known technologies and will not be elaborated further.
[0038] Establish a non-linear path loss model, L(d,ρ,h) = 20log 10 (4πfd / c) + α(ρ)d + β(h)d + γ(ρ,h)d. Where: L(d,ρ,h) is the path loss, c is the speed of light (m / s), α(ρ) is the influence coefficient of air density on path loss, β(h) is the influence coefficient of humidity on path loss, and γ(ρ,h) is the interaction influence coefficient of air density and humidity.
[0039] The design of the L(d,ρ,h) function comprehensively considers linear and non-linear effects, and can more accurately describe the complex influence of air density and humidity on signal propagation. By introducing exponential, logarithmic, and power functions, the saturation effect, rapid growth effect, and various complex interactions existing in the signal propagation process can be captured.
[0040] Based on the collected measurement acquisition data, determine the specific functional forms and their coefficients of α(ρ), β(h), and γ(ρ,h). Using multiple nonlinear regression analysis, the specific regression analysis calculation process can be computed by computer software, and the content listed here is only for those skilled in the art to understand the core method ideas and facilitate implementation.
[0041] After collecting the data, each function uses multiple nonlinear regression analysis respectively, and substitutes the acquired data into the following function models to calculate the parameters: where, a 1 , a 2 , a 3 ; b 1 , b 2 , b 3 ; c 1 , c 2 , c 3 are undetermined coefficients during calibration, which can be specific numerical values, including 0. The parameters can also be designed as function expression forms according to needs. In specific applications, the function expression should be able to calculate a definite numerical value based on other relevant parameters or the specific numerical values of air density, humidity, and propagation distance.
[0042] Preferably, α(ρ) = a 1 ρ + a 2 exp(a 3 ρ).
[0043] Preferably, β(h) = b 1 h + b 2 ln(1 + b 3 h).
[0044] Preferably, γ(ρ,h) = c 1 ρh + c 2 ρ²h½ + c 3 ρ½h².
[0045] In the formula of γ(ρ,h), c 1 ρh. According to the basic principles in atmospheric physics based on the interaction between air density ρ and humidity h, air density affects the refractive index of electromagnetic waves, while humidity affects the water vapor content in the air. Both jointly affect signal propagation, and this interaction has a linear relationship. The coefficient c 1 allows adjusting the degree of this basic interaction.
[0046] In addition to the basic interaction between air density and humidity, there are also complex model relationships with interactions in a specific order. Higher-order interaction effects c 2 ρ²h½ and c 3It is expressed as ρ½h². According to the theory of atmospheric radio wave propagation, the effects of air density and humidity are not just a simple linear superposition. Nonlinear effects may stem from phenomena such as multiple scattering, molecular resonance absorption, and complex atmospheric turbulence.
[0047] In c 2 The term ρ²h½: The ρ² term reflects that air density may have a stronger effect. This is based on the fact that when air density increases, the intermolecular interaction enhances, which may lead to an increase in the nonlinearity of signal scattering and absorption. The h½ term indicates that the effect of humidity may tend to saturate in some cases. Considering that after the water vapor content increases to a certain extent, its additional effect may weaken.
[0048] In c 3 The term ρ½h²: The ρ½ term takes into account that the effect of air density may tend to level off within certain ranges. The h² term reflects that humidity may have a more significant nonlinear effect in some cases. This is based on the fact that in a high-humidity environment, the aggregation of water vapor molecules leads to a more complex signal attenuation effect.
[0049] Considering the overall design, γ(ρ,h) adopts a polynomial form: The choice of the polynomial form is to balance the complexity of the model. It allows for nonlinear effects while maintaining relative computational simplicity. Among them, the ρh term captures the basic linear interaction. The ρ²h½ and ρ½h² terms capture possible asymmetric nonlinear effects, allowing the model to exhibit different parameter expressions under different combinations of air density and humidity, which can be represented by piecewise functions.
[0050] The choice of exponents: Using simple exponents such as 1 / 2 and 2 is to avoid overfitting while still capturing the main nonlinear features. This choice also takes into account computational efficiency, making the model more practical in actual applications.
[0051] By adjusting the coefficients c 1 , c 2 , c 3 , the model can adapt to different environmental conditions and device characteristics. If some interaction effects are not obvious, the corresponding coefficients can be close to or equal to zero, and the model will simplify automatically.
[0052] The calibration process of α(ρ) and β(h) is relatively simple. In cases where high calculation result accuracy is required, it can be achieved by simply collecting more data under different controlled variable conditions.
[0053] The data collection and calculation for γ(ρ,h) are relatively complex, and calibration can be achieved through the following methods. The following method steps can also be applied to the data collection and simulation calculation in the calibration process of α(ρ) and β(h) when needed.
[0054] For data acquisition design, the factorial design method is adopted to ensure that the air density ρ and humidity h are evenly distributed within the expected range. For example, at least 5x5 factor levels are designed, that is, 5 different levels are selected for both air density and humidity. During testing, the device operates at a fixed transmission power and frequency.
[0055] Measure the parameters at each factor level. For each queue, measure the signal strength at multiple known distance points. Record the precise air density, relative humidity, distance, and received signal strength. Repeat the measurement multiple times for each queue under each factor condition to reduce random errors.
[0056] Use an environmental chamber to simulate different air density and humidity conditions.
[0057] Among them, the air density can be achieved by adjusting the temperature and pressure according to the formula ρ=(P*M) / (R*T). In the formula, P is the pressure of the gas, M is the molar mass of air, R is the ideal gas constant, and T is the temperature of air. The molar mass M of air is approximately 0.0289 kg / mol, and the ideal gas constant R is 8.3145 J / (mol·K).
[0058] Preprocess the data. For outlier detection: Use the Grubbs test or Z-score method to identify and handle outliers.
[0059] When simulating and calculating the γ(ρ,h) model, the Levenberg-Marquardt algorithm, namely LMA, is used for fitting and parameter estimation. LMA combines the advantages of the gradient descent method and the Gauss-Newton method. For nonlinear least squares problems, LMA is usually more stable and efficient than other methods. LMA can handle the correlation between parameters well and is suitable for the model of this method. In the case where the initial parameter estimation is inaccurate, LMA can still converge to the correct solution.
[0060] Such as Figure 2 , the function modeling process adopts the LMA fitting algorithm, including the following fitting calculation steps: Initial parameter estimation: Use multiple linear regression analysis for the collected values to obtain the initial estimated parameter values; Optimization objective: Use the function of the estimated parameter values to calculate the minimized predicted values under the actual measurement conditions, and calculate the mean square error between the predicted values and the actual measurement values; Use LMA to iteratively update the parameter values until the mean square error converges; Cross-validation: Use k-fold cross-validation, such as k = 10, to evaluate the performance of the model. k-fold cross-validation can help select the model and parameter combination with the best performance, provide a quantitative index for the reliability of the model, and enable a reasonable expectation of the model's effect in practical applications. The process is automatically calculated by a computer program. If it does not meet the expectation, the LMA fitting algorithm is re-run. If the k-fold cross-validation fails multiple times, the calibration data acquisition operation process is redesigned and data is re-acquired.
[0061] For example Figure 2 , the calibration process includes the following parameter estimation and verification steps: Residual analysis: Check the normality and homoscedasticity of the residuals, and use the Durbin-Watson test to evaluate the autocorrelation of the residuals; Parameter significance test: Use the t-test to evaluate the statistical significance of each parameter; Model comparison: Use the Akaike information criterion or the Bayesian information criterion to compare models of different complexities and select the optimal L(d,ρ,h). When the data sample size in the calibration process is too large, preferably, the Bayesian information criterion is adopted.
[0062] The Durbin-Watson test, abbreviated as the DW test, is a statistical method used to detect whether there is first-order autocorrelation in the residual terms in regression analysis. It is mainly used for the analysis of time series data to determine whether the residuals are independent of each other. The Durbin-Watson test is implemented in a variety of statistical software, such as R and Python. In R, the durbin.watson function in the tseries package can be used; in Python, the statsmodels library provides a direct implementation.
[0063] In this patented technology, the DW test is used to detect whether there is first-order autocorrelation in the residual terms in regression analysis. If there is autocorrelation in the residuals, it indicates that the model may have omitted important variables or the model form is set incorrectly, which will affect the accuracy and reliability of the model. Through the DW test, it can be judged whether the regression result based on the path loss model is reliable, ensuring that the model can accurately reflect the true relationship between variables. It helps to judge whether the model is reasonable. If there is autocorrelation in the residuals, the model can be adjusted in time, such as adding omitted variables or changing the model form, so as to improve the prediction accuracy and reliability of the model and make the distance between devices calculated based on this model more accurate.
[0064] One implementation is as follows: After estimating the path loss model parameters, the residual data of the model is input into a statistical software that supports the DW test for verification. Run the DW test function in the software, and the function will calculate the DW statistic based on the residual data. Generally, the value range of the DW statistic is between 0 and 4. If the statistic is close to 2, it indicates that there is no autocorrelation in the residuals; if the statistic is far from 2, the closer it is to 0, the stronger the positive autocorrelation, and the closer it is to 4, the stronger the negative autocorrelation.
[0065] The t-test is used to evaluate the statistical significance of each parameter. In the path loss model, the t-test can be used to determine whether 1 , a 2 , a 3 ; b 1 , b 2 , b 3 ; c 1 , c 2 , c 3 these coefficients are significantly non-zero. If a certain coefficient is not significant, it indicates that the influence of this variable on the path loss may not be obvious and it is not necessary in the model. The parameter can be assigned a value of 0, which helps to optimize the model structure.
[0066] For each parameter in the path loss model, the t-statistic is calculated using the sample data. In different statistical software, such as the t.test function in R language and the relevant functions in the scipy.stats module in Python, the corresponding sample data and the parameters of the hypothesis test, such as assuming the coefficient is 0, are input for calculation. Determine the importance of each parameter in the model, eliminate the insignificant parameters, simplify the model, avoid overfitting of the model, and at the same time improve the interpretability of the model, so that the model can more accurately reflect the influence of factors such as air density and humidity on the path loss, thereby improving the accuracy of the calculation of the distance between devices.
[0067] The Akaike Information Criterion, abbreviated as AIC in English, is a statistic used for model selection. It is based on information theory and aims to balance the goodness of fit and complexity of the model to help select the model that can best explain the data with the fewest parameters. AIC is used to compare the fitting effects of different models. Usually, the model with the smallest AIC value is selected to avoid overfitting of the model.
[0068] It is used to compare path loss models with different complexities, balance the goodness of fit and complexity of the model, and select the model that can best explain the data with the fewest parameters. In this patent, there may be multiple different forms of path loss models, and AIC can help determine the optimal model, avoiding overfitting caused by the model being too complex or being too simple to accurately describe the data.
[0069] For each path loss model to be compared, calculate the AIC value according to the likelihood function value and the number of parameters of the model. In practical applications, calculations can be implemented with the help of statistical software such as relevant libraries in R language and Python.
[0070] The calculation formula of the AIC value is AIC = -2ln(s)+2k, where s is the maximum likelihood estimate of the path loss model and k is the number of parameters in the model. After the calculation is completed, compare the AIC values of different models. Select the optimal model to improve the generalization ability of the model, so that the model can stably and accurately calculate the distance between IoT devices in different environments, and enhance the practicality and reliability of the model.
[0071] The Bayesian Information Criterion, abbreviated as BIC. BIC is an extension of AIC, especially suitable for Bayesian environments and has better performance when the sample size is large. In the present invention, BIC is also used for model selection. Similar to AIC, by weighing the goodness of fit and complexity of the model, select the most suitable path loss model to provide more reliable model support for accurately calculating the distance between devices. Similar to AIC, calculate the BIC value according to the likelihood function value and the number of parameters of the model, and implement the calculation in statistical software such as relevant libraries in R language and Python.
[0072] Taking Python as an example, after fitting different path loss models, calculate the BIC value of each model according to the calculation formula of BIC: BIC = -2ln(s)+kln(n), where n is the sample size, and then compare the BIC values of different models. When the sample size is large, the optimal model can be selected more accurately, avoiding model overfitting, improving the model's ability to interpret data and prediction accuracy, so as to calculate the distance between IoT devices more precisely and adapt to the complex and changeable IoT application environment.
Claims
1. The calibration application method of IoT device signal strength positioning is characterized by , including the following steps: The equipment is calibrated before leaving the factory to establish a path loss model L(d,ρ,h) that takes into account air density and humidity; When installing the equipment, detect the air density and humidity of the current environment; After the equipment is installed, each device can act as a transmitter to send out broadcast signals including device address, RF transmission strength, and transmitter antenna gain data; each device can act as a receiver to receive broadcast signals within the communication range, and record the received device address and its received signal strength value reflected at the receiver; each device retrieves its own receiver antenna gain value and calculates the signal strength change value according to the empirical formula; Substitute the above content data into the empirical formula containing the path loss model to reversely calculate the actual distance d between devices; The path loss model , L(d,ρ,h) is the path loss signal strength value, f is the wireless communication frequency, d is the distance between wireless communication devices, c is the speed of light, α(ρ) is the influence coefficient of air density on path loss, β(h) is the influence coefficient of humidity on path loss, γ(ρ,h) is the interaction coefficient of air density and humidity, ρ is air density, and h is humidity.
2. The calibration application method for signal strength positioning of IoT devices according to claim 1 is characterized in that: The empirical formula is: L(d,ρ,h) = RF transmission strength + transmitting end antenna gain + receiving end antenna gain – received signal strength.
3. The calibration application method for signal strength positioning of IoT devices according to claim 1, characterized in that: The α(ρ) is expressed as α(ρ) = a1ρ + a2exp(a3ρ), where a1, a2, and a3 are parameters that need to be determined during the calibration process.
4. The calibration application method for signal strength positioning of IoT devices according to claim 1, characterized in that: The β(h) is expressed as β(h) = b1h + b2ln(1 + b3h), where b1, b2, and b3 are parameters that need to be determined during the calibration process.
5. The calibration application method for signal strength positioning of IoT devices according to claim 1, characterized in that: The γ(ρ,h) air density and humidity interaction coefficient may be calibrated using only basic interaction, or only high-order interaction, or a combination of basic interaction and high-order interaction.
6. The calibration application method for signal strength positioning of IoT devices according to claim 5, characterized in that: The model combining both basic interaction and higher-order interaction effects when γ(ρ,h) is γ(ρ,h) , c1ρh represents the basic interaction between air density and humidity, c2ρ²h½ and c3ρ½h² represent the higher-order interaction effects between air density and humidity.
7. The calibration application method for positioning the signal strength of an Internet of Things device according to claim 1, characterized in that: The calibration process includes: an operation process of collecting calibration data of IoT devices; a function modeling process of calculating the path loss model L(d, ρ, h) using the calibration data; parameter estimation and verification of the function of modeling L(d, ρ, h), and those that meet the conditions are uniformly sent to the batch of IoT devices as the final path loss model for storage and post-installation calculation.
8. The calibration application method for signal strength positioning of IoT devices according to claim 7, characterized in that: The operation process of collecting IoT device calibration data includes: The calibration device inputs the fixed frequency value of the IoT device; Several IoT devices form a queue and are arranged in sequence, and the device interval is set as needed; The signal strength value received by each device of the team is collected by the control variable method; The control variable method includes: the air density is constant, the β (h) coefficient is calibrated, each constant air density is a group, and multiple groups of calibration can be performed by adjusting the air density; the humidity is constant, the α (ρ) is calibrated, each constant humidity is a group, and multiple groups of calibration can be performed by adjusting the humidity; The collected values and equipment interval data are entered into a computer, and the nonlinear path loss model under the influence of air density and humidity alone is subsequently calculated through a function modeling process; The path loss model of the interaction between air density and humidity is calculated through computer simulation based on multiple sets of collected values and equipment interval data; The complete path loss model function is stored in each IoT device produced in this batch. The function is called during the positioning and debugging between devices after actual installation, and the air density and humidity parameters at the installation site are used for subsequent calculations.
9. The calibration application method for positioning the signal strength of an Internet of Things device according to claim 7 or 8, characterized in that: The function modeling process adopts the LMA fitting algorithm, including the following calculation steps: The collected values were analyzed using multiple linear regression to obtain initial estimated parameter values; Using the function of the estimated parameter value, calculate the minimized predicted value under the condition of the actual measured value, and calculate the mean square error between the predicted value and the actual measured value; Use LMA to iteratively update parameter values until the mean square error converges; Functional model performance was evaluated using k-fold cross validation.
10. The calibration application method for positioning the signal strength of an Internet of Things device according to claim 7, characterized in that: The calibration process includes the following parameter estimation and verification steps: The normality and homoscedasticity of the residuals were checked, and the autocorrelation of the residuals was assessed using the Durbin-Watson test; The statistical significance of each parameter was assessed using t test; Use Akaike Information Criterion or Bayesian Information Criterion to compare models of different complexity and select the optimal L(d,ρ,h) result.