An environmental learning threshold-based internet of things node security positioning method and system

Abstract

The application discloses an Internet of Things node security positioning method and system based on an environment learning threshold, belongs to the technical field of Internet of Things security positioning and wireless ranging, and through the use of a reference node with a known position and a calibration node for multi-round sparse sampling in an application environment or a similar environment in advance, the method acquires a measured distance, calculates a distance, and obtains a normalized difference value, inputs an artificial intelligence model to learn a trusted threshold value in the current environment; in actual positioning, the measured distance between a node to be positioned and an anchor node is acquired, positioning calculation is performed, a normalized difference value statistic is calculated, the normalized difference value statistic is compared with the trusted threshold value learned in advance; if the threshold is exceeded, it is determined that the anchor node set is abnormal, and after malicious anchor nodes are excluded through node-by-node verification or clustering positioning analysis, the node is repositioned. The application can adapt to ranging noise time variation and anchor node distribution change, so that online calculation overhead is reduced, and the security and reliability of Internet of Things node positioning are improved.

Description

Technical Field

[0001] This invention belongs to the field of IoT security positioning and wireless ranging technology, specifically relating to an IoT node security positioning method and system based on environmental learning threshold. Background Technology

[0002] In IoT systems, numerous sensor nodes, mobile terminals, industrial equipment, and smart devices rely on location information for resource scheduling, anomaly alerts, data association, and security control. Common positioning methods typically utilize wireless ranging information between anchor nodes with known locations and the node to be located for location calculation. For example, distances or angles between nodes can be obtained based on received signal strength indicators, time of arrival, time difference of arrival, angle of arrival, or channel state information.

[0003] In real-world applications, wireless signals are affected by factors such as obstruction, multipath propagation, non-line-of-sight propagation, electromagnetic interference, personnel movement, and changes in equipment posture. This leads to discrepancies between measured and actual geometric distances, and these discrepancies are typically environmentally relevant and time-varying. Therefore, even in the absence of malicious nodes, the positioning results and the calculated distances between anchor nodes may not perfectly match the measured distances.

[0004] On the other hand, IoT positioning scenarios may be vulnerable to security threats such as malicious anchor node attacks, pseudo-coordinate attacks, pseudo-range attacks, replay attacks, and cloned node attacks. Malicious anchor nodes can report incorrect locations or generate abnormal ranging information, causing the positioning results of the node to be located to deviate. Existing methods often use fixed error thresholds or empirical thresholds to determine the reliability of anchor nodes, but fixed thresholds are difficult to adapt to different building structures, equipment models, channel conditions, and anchor node distributions; setting the threshold too small will cause false alarms, while setting it too large will cause missed alarms; especially in scenarios where the anchor node distribution changes, ranging noise varies over time, and only partial sampling point data can be obtained, it is difficult to directly calculate the reliability threshold using a strict closed mathematical formula.

[0005] Therefore, existing technologies still lack a secure positioning mechanism that can be reused in similar or identical environments and can quickly adjust the reliability threshold based on the distribution of anchor nodes and measurement noise. Summary of the Invention

[0006] This invention aims to provide a secure positioning method and system for IoT nodes based on environmental learning thresholds. By performing sparse sampling in the application environment or similar environments, it learns a reasonable range of normalized differences between measured distance and positioning calculation distance. During the usage phase, it adaptively outputs a reliable threshold value based on the current anchor node ranging results, thereby identifying and eliminating malicious anchor nodes and improving positioning security and reliability.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A secure location method for IoT nodes based on environmental learning thresholds includes the following steps:

[0009] S1. In the wireless positioning environment to be applied, obtain the ranging information between the node to be positioned and multiple anchor nodes with known locations, and determine the measured distance between the node to be positioned and each anchor node based on the ranging information; perform positioning calculation on the node to be positioned based on the positions of the multiple anchor nodes and the measured distance between the node to be positioned and each anchor node to obtain the estimated position of the node to be positioned.

[0010] S2. Based on the estimated position of the node to be located and the positions of each anchor node, calculate the calculated distance between the node to be located and each anchor node; based on the measured distance and calculated distance between the node to be located and each anchor node, calculate the normalized difference or normalized difference statistic corresponding to the node to be located.

[0011] S3. Input the measured distance, calculated distance, normalized difference, reference node location information, anchor node location information, and / or normalized difference statistics corresponding to the reference node in the current environment into a pre-trained confidence threshold prediction model, and output the confidence threshold value in the current environment; the confidence threshold prediction model is a model trained using multiple rounds of calibration samples formed by reference nodes and calibration nodes with known locations in the application environment or similar environments; compare the normalized difference or normalized difference statistics corresponding to the node to be located with the confidence threshold value; when the normalized difference or normalized difference statistics do not exceed the confidence threshold value, determine that the set of anchor nodes participating in the positioning is reliable, and output the estimated position of the node to be located; when the normalized difference or normalized difference statistics exceed the confidence threshold value, determine that the set of anchor nodes participating in the positioning is abnormal;

[0012] S4. When the set of anchor nodes participating in the positioning is determined to be abnormal, perform node-by-node verification or cluster positioning analysis on the anchor nodes participating in the positioning, exclude, replace or mark suspected abnormal anchor nodes, and re-position based on the processed set of anchor nodes.

[0013] Furthermore, the reliable threshold prediction model is trained through the following steps:

[0014] S31. In the wireless positioning environment to be applied or a wireless positioning environment similar to the environment to be applied, select at least one reference node with a known location and determine multiple calibration nodes with known locations; the reference node is a reference signal receiver or an equivalent signal receiver node; the calibration node includes a physical anchor node with a known location, and / or a virtual anchor node formed by a mobile calibration node at multiple sampling points.

[0015] S32. Obtain the distance measurement information between the reference node and multiple calibration nodes, and determine the measurement distance from the reference node to each calibration node based on the distance measurement information;

[0016] S33. Treat the reference node as the node to be located. Based on the positions of the multiple calibration nodes and the measured distances between the reference node and each calibration node, perform positioning calculations on the reference node to obtain the estimated position of the reference node. Calculate the calculated distances between the reference node and each calibration node based on the estimated position of the reference node and the positions of each calibration node.

[0017] S34. Based on the measured distance and calculated distance between the reference node and each calibration node, calculate the normalized difference or normalized difference statistic corresponding to each calibration node.

[0018] S35. The measured distance, calculated distance, normalized difference, reference node position information, calibration node position information, number of sampling points, and / or normalized difference statistics formed by multiple rounds of sampling are used as feature inputs of the artificial intelligence model; the basic confidence threshold label is determined based on the normalized difference statistics and sampling point number coefficient corresponding to each round of sampling; then, the basic confidence threshold label is adjusted in combination with the actual distance between the reference node and each calibration node, the relative topological symmetry level between the reference node and multiple calibration nodes, and the system application requirements to determine the target confidence threshold value of the sampling round, which serves as the supervision label for the sampling round.

[0019] The samples from all rounds are used to form a training sample set, which is then divided into a training set, a validation set, and a test set according to a preset ratio. The artificial intelligence model is iteratively trained and optimized based on the training set and the validation set. Finally, the model that performs best on the validation set and passes the evaluation on the test set is selected as the reliable threshold prediction model for outputting reliable threshold values ​​in the current environment.

[0020] Furthermore, the ranging information includes one or more of Received Signal Strength Indication (RSSI), Time of Arrival (TOA), Time Difference of Arrival (TDOA), Angle of Arrival (AOA), and Channel State Information (CSI); when using Received Signal Strength Indication for ranging, the Received Signal Strength Indication is converted into the measured distance based on the path loss model.

[0021] Furthermore, the normalized difference statistic includes one or more of the mean, variance, maximum, minimum, median, and quantile of the normalized difference; when the normalized difference statistic is greater than the confidence threshold, it is determined that there are abnormal or malicious anchor nodes in the corresponding anchor node set or anchor node group.

[0022] Furthermore, the node-by-node verification includes: removing one anchor node from the set of anchor nodes that have been judged as abnormal, and recalculating the estimated position and normalized difference statistic of the node to be located using the remaining anchor nodes; if the normalized difference statistic decreases to within the confidence threshold after removing an anchor node, or the estimated position of the node to be located tends to stabilize, then the anchor node is identified as a suspected abnormal anchor node or a malicious anchor node.

[0023] Furthermore, the clustered localization analysis includes: dividing the anchor nodes participating in the localization into multiple anchor node clusters, calculating the estimated position of the node to be located based on different anchor node clusters, and calculating the normalized difference statistic corresponding to each anchor node cluster; selecting anchor node clusters whose normalized difference statistic is less than the confidence threshold as reliable anchor node clusters; when the localization results of multiple anchor node clusters conflict with each other and the normalized difference statistic does not exceed the confidence threshold, rejecting the current localization result or indicating that the current localization environment is abnormal.

[0024] Preferably, the sampling point quantity coefficient is a coefficient that is preset manually, obtained by looking up a table, or calculated by a function based on the number of sampling points in each round; the smaller the number of sampling points, the larger the sampling point quantity coefficient is, or the coefficient is not less than that when the number of sampling points is large, so as to compensate for the uncertainty caused by small sample sampling.

[0025] Preferably, when the calibration node includes a virtual anchor node formed by a mobile calibration node at multiple sampling points, the mobile calibration node moves within the inscribed rectangular or inscribed square region of the reference node's communication coverage area according to a preset route, a random route, or a combination of both, and transmits positioning signals and reports its own position at multiple sampling points; the sampling points are sparse sampling points, and it is not required to cover all grid points or pixels within the inscribed rectangular or inscribed square region.

[0026] Furthermore, the relative topology symmetry level is used to characterize the degree of spatial distribution balance of multiple calibration nodes around the reference node; the relative topology symmetry level is not used to directly calculate the confidence threshold value according to a deterministic mathematical formula, but is used as one of the influencing factors for setting the target confidence threshold value supervision label;

[0027] The classification of relative topology symmetry levels takes into account the following factors: the directional distribution characteristics and distance distribution characteristics of each calibration node relative to the reference node, the geometric enclosure degree of multiple calibration nodes around the reference node, and application requirements, thereby dividing the relative topology into several levels. The higher the directional distribution of multiple calibration nodes around the reference node and the higher the enclosure degree of the reference node, the higher the relative topology symmetry level. The lower the relative topology symmetry level, the more concentrated the multiple calibration nodes are on one side of the reference node or the more uneven the directional distribution. When the relative topology symmetry level is lower than a preset threshold, it indicates that the spatial layout of the current calibration nodes is sensitive to ranging noise. Normal ranging noise can easily cause large positioning deviations or fluctuations in normalized difference. An incremental increase should be performed based on the basic trust threshold label. The increase range is manually set according to the tolerance of the application scenario for positioning error, false alarm rate, and false negative rate.

[0028] Preferably, the artificial intelligence model includes one or more of the following: multilayer perceptron neural network, convolutional neural network, recurrent neural network, Transformer model, graph neural network, support vector regression model, random forest regression model, gradient boosting regression model, or ensemble learning model.

[0029] Preferably, when the number of existing anchor nodes with known locations in the deployment environment reaches a preset number, there is no need to deploy additional reference signal receivers and mobile calibration nodes. Instead, one of the anchor nodes with known locations is selected as a reference node, and the remaining anchor nodes with known locations are used as calibration nodes. Distance and positioning calculations are performed to obtain the measured distance, calculated distance, and normalized difference corresponding to the reference node. The corresponding data is then used as input to the artificial intelligence model to learn or update the confidence threshold value in the current environment.

[0030] This invention also protects a secure positioning system for IoT nodes based on an environment learning threshold, comprising:

[0031] The communication ranging module is used to acquire ranging information between the node to be located and multiple anchor nodes with known locations, and to acquire ranging information between the reference node and multiple anchor nodes with known locations; and to obtain the corresponding measurement distance based on the ranging information.

[0032] A reference sampling module is used to perform positioning calculations on the reference node based on the positions of the plurality of anchor nodes and the ranging information, so as to obtain the estimated position of the reference node;

[0033] The positioning calculation module is used to perform positioning calculations on the node to be positioned based on the positions of multiple anchor nodes and their corresponding measured distances, so as to obtain the estimated position of the node to be positioned.

[0034] The difference construction module is used to calculate the calculated distance from the reference node to each anchor node based on the estimated position of the reference node and the position of each anchor node, and to calculate the normalized difference or normalized difference statistic based on the distance measurement information and the corresponding measured distance.

[0035] The threshold processing module is used to train and infer reliable threshold values ​​based on an artificial intelligence model. During the training phase, the measured distance, calculated distance, normalized difference, reference node position information, anchor node position information, and / or normalized difference statistics corresponding to the reference node are input into the artificial intelligence model for training to obtain a reliable threshold value prediction model. During the application phase, the similar features corresponding to the reference node in the current environment are input into the reliable threshold value prediction model to output the reliable threshold value in the current environment.

[0036] The safety determination module is used to calculate the calculated distance from the node to be located to each anchor node based on the estimated position of the node to be located, and to calculate the normalized difference or normalized difference statistic based on the measured distance and calculated distance between the node to be located and each anchor node. The normalized difference or its statistic is compared with the confidence threshold value to determine whether the set of anchor nodes participating in the positioning is abnormal.

[0037] The abnormal anchor node processing module is used to perform individual exclusion or clustered positioning analysis on the anchor nodes involved in the positioning when the security determination module determines that the set of anchor nodes involved in the positioning is abnormal. If the normalized difference or normalized difference statistic decreases after excluding or replacing a certain anchor node, the anchor node is marked as a suspected abnormal anchor node. If clustered positioning analysis is used, positioning calculations are performed based on different anchor node clusters, and anchor node clusters with normalized difference or normalized difference statistic less than a preset value are selected as reliable anchor node clusters to obtain the final positioning result.

[0038] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0039] This invention learns a reliable threshold value by using sparse sampling data from the application environment or similar environments, rather than using a fixed empirical threshold, and can adapt to different building structures, wireless propagation conditions, anchor node distributions and ranging noise.

[0040] This invention normalizes and compares the measured distance with the calculated distance derived from the positioning result, avoiding the phenomenon that the larger the distance, the larger the absolute error, which directly affects the judgment result, and making the residuals between different distances and different anchor nodes more comparable.

[0041] The training phase of this invention does not require sampling points to cover all grid points or pixels. Training samples can be constructed using only some random or designed sampling points, thus reducing on-site sampling costs.

[0042] This invention compensates for time-varying noise and sample size differences through multiple rounds of measurement and sampling point number coefficients, enabling the threshold value to be adjusted according to sample size and environmental fluctuations.

[0043] This invention can complete threshold updates by utilizing the mutual communication between anchor nodes when there are a sufficient number of anchor nodes, without the need for additional moving calibration equipment, thus facilitating rapid maintenance after deployment. Attached Figure Description

[0044] Figure 1 This is a flowchart of the IoT node secure positioning method based on environment learning threshold of the present invention;

[0045] Figure 2 This is a schematic diagram of the sampling scenario during the training phase of the present invention;

[0046] Figure 3 This is a flowchart illustrating the training data construction and trusted threshold learning process of this invention.

[0047] Figure 4 This is a schematic diagram illustrating the threshold update method for the present invention when the number of anchor nodes is sufficiently large, eliminating the need for additional equipment.

[0048] Figure 5 This is a graph showing the change in the abnormal anchor node detection rate with the distance measurement bias of malicious nodes in the simulation experiment of this invention.

[0049] Figure 6 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0050] To make the technical solution of the present invention clearer, the technical solution of the present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.

[0051] like Figure 1 As shown, the present invention provides a secure positioning method for IoT nodes based on environmental learning thresholds, comprising the following steps:

[0052] S1. In the wireless positioning environment to be applied, obtain the ranging information between the node to be positioned and multiple anchor nodes with known locations, and determine the measured distance between the node to be positioned and each anchor node based on the ranging information; perform positioning calculation on the node to be positioned based on the positions of the multiple anchor nodes and the measured distance between the node to be positioned and each anchor node to obtain the estimated position of the node to be positioned.

[0053] Specifically, the ranging information includes one or more of the following: received signal strength indication, time of arrival, time difference of arrival (TDOA), angle of arrival, and channel state information; the node to be located is represented as... The measured distances from the node to be located to each anchor node are:

[0054] ;

[0055] in, Indicates the node to be located to the first... The measured distance between each anchor node;

[0056] Based on the anchor node location and the measured distance, the estimated location of the node to be located is calculated using a positioning algorithm, denoted as . ;

[0057] S2. Based on the estimated position of the node to be located. Based on the positions of each anchor node, calculate the distance between the node to be located and each anchor node, expressed as: ,

[0058] In the formula, Indicates the first The location of each anchor node Indicates the node to be located to the first... Calculated distance between anchor nodes;

[0059] Then, based on the measured distance and calculated distance between the node to be located and each anchor node, the normalized difference corresponding to the node to be located is calculated, expressed as:

[0060] ,

[0061] In the formula, Indicates the first Normalized difference corresponding to each anchor node The minimum effective distance is preset to be close to 0; the normalized difference corresponding to each anchor node is statistically analyzed to obtain the normalized difference statistics of the current positioning process of the node to be located;

[0062] S3. Input the measured distance, calculated distance, normalized difference, reference node location information, anchor node location information, and / or normalized difference statistics corresponding to the reference node in the current environment into the pre-trained confidence threshold prediction model, and output the confidence threshold value in the current environment; compare the normalized difference or normalized difference statistics corresponding to the node to be located with the confidence threshold value; when the normalized difference or normalized difference statistics do not exceed the confidence threshold value, determine that the set of anchor nodes participating in the positioning is reliable, and output the estimated position of the node to be located; when the normalized difference or normalized difference statistics exceed the confidence threshold value, determine that there are abnormal anchor nodes or malicious anchor nodes in the set of anchor nodes participating in the positioning.

[0063] S4. When the set of anchor nodes participating in the positioning is determined to contain abnormal or malicious anchor nodes, an elimination method can be used for analysis. Specifically, the system eliminates one anchor node from the set of anchor nodes participating in the positioning in turn, and recalculates the position of the node to be located using the remaining anchor nodes. At the same time, it recalculates the normalized difference statistic between the measured distance and the calculated distance. If the estimated position of the node to be located tends to stabilize after eliminating a certain anchor node, and the normalized difference statistic decreases significantly, it indicates that the anchor node has an abnormal impact on the positioning results, and the anchor node can be marked as a suspected malicious or abnormal anchor node.

[0064] When there are many abnormal anchor nodes, or when the method of eliminating them one by one cannot obtain stable positioning results, a clustered positioning method is adopted. The system divides the anchor nodes involved in the positioning into multiple anchor node clusters, calculates the estimated position of the node to be located based on different anchor node clusters, and calculates the normalized difference statistic corresponding to each anchor node cluster. The anchor node cluster with the normalized difference statistic less than the confidence threshold is selected as the confidence anchor node cluster, and the position of the node to be located is output based on the confidence anchor node cluster. If the positioning results of multiple anchor node clusters conflict with each other and none of them exceed the confidence threshold, the positioning result can be rejected, or a serious anomaly in the current environment can be indicated.

[0065] Specifically, the reliable threshold value is used to characterize the reasonable deviation range allowed between the measured distance and the calculated distance under the current communication environment. Because wireless signals are affected by factors such as obstruction, multipath propagation, noise, equipment differences, and the spatial distribution of anchor nodes in different application environments, a fixed artificial threshold is difficult to adapt to different scenarios. Therefore, this invention uses reference nodes with known locations to simulate the node to be located, obtains training samples through multiple rounds of ranging and positioning calculations, and uses an artificial intelligence model to learn the reliable threshold value under the current environment. Figures 2 to 3 As shown, the specific steps are as follows:

[0066] S31. In the wireless positioning environment to be applied, or a wireless positioning environment similar to the one to be applied, a reference node R is selected, whose actual location is known; the locations of the surrounding anchor nodes participating in the ranging are denoted as... , ;

[0067] S32. Obtain the distance measurement information between the reference node and multiple anchor nodes with known positions, and determine the measured distance between the reference node and each anchor node based on the distance measurement information, expressed as:

[0068] ,

[0069] in, Indicates the reference node and the first The measured distance between anchor nodes is obtained by RSSI, TOA, TDOA, AOA, or a combination thereof;

[0070] S33. Treat the reference node as the node to be located. Based on the positions of the multiple anchor nodes and the measured distances from the reference node to each anchor node, perform positioning calculations on the reference node to obtain its estimated position, denoted as . Based on the estimated position of the reference node Based on the positions of each anchor node, calculate the distance from the reference node to each anchor node, expressed as:

[0071] ,

[0072] in, This indicates the distance from the reference node obtained by back-calculation from the positioning result to the first... Calculated distance between anchor nodes For the first The location of each anchor node;

[0073] S34. Based on the measured distance and calculated distance between the reference node and each anchor node, calculate the normalized difference corresponding to each anchor node, expressed as:

[0074] ,

[0075] in, Indicates the first Normalized difference corresponding to each anchor node This is a preset minimum effective distance close to 0, used to avoid abnormal amplification of the normalization result or division by zero errors caused by a measurement distance that is too small;

[0076] Statistical analysis is performed on multiple normalized differences obtained from the same round of sampling to obtain the difference statistics for that round of sampling. These normalized difference statistics include one or more of the following: mean, variance, maximum, minimum, median, and quantiles of the normalized differences. The following is the formula for calculating the average of the normalized differences from that round:

[0077] ,

[0078] In the formula, This represents the average of the normalized differences in this round of sampling; the normalized difference statistic is used to describe the degree of consistency between the measured distance and the calculated distance in the current environment;

[0079] S35. The measured distance, calculated distance, normalized difference, reference node position information, calibration node position information, number of sampling points, and / or normalized difference statistics formed by multiple rounds of sampling are used as feature inputs of the artificial intelligence model; a basic confidence threshold label is determined based on the normalized difference statistics and sampling point number coefficient corresponding to each round of sampling; then, the basic confidence threshold label is adjusted in combination with the actual distance between the reference node and each calibration node, the relative topological symmetry level between the reference node and multiple calibration nodes, and system application requirements to determine the target confidence threshold value of the sampling round, which serves as the supervision label for the sampling round.

[0080] Specifically, when the actual distance between the reference node and each calibration node is relatively large, the impact of ranging noise on the positioning result increases, which can increase the target confidence threshold value. The reference node position information and calibration node position information are used to characterize the relative topology between the reference node and the calibration node. When the directional distribution of multiple calibration nodes relative to the reference node is more uneven, the degree of concentration on one side of the reference node is higher, or the degree of enclosure of the reference node is lower, the symmetry level of the relative topology is lower. Under the same label setting conditions, if the symmetry level of the relative topology is lower than the preset threshold, an incremental increase should be performed on the basis of the basic confidence threshold label. The increase range is manually set according to the tolerance of the application scenario for positioning error, false alarm rate and false alarm rate, so as to obtain the final target confidence threshold value supervision label.

[0081] The sampling point quantity coefficient is a coefficient that is preset manually, obtained by looking up a table, or calculated by a function based on the number of sampling points in each round. The smaller the number of sampling points, the larger the sampling point quantity coefficient is, or the coefficient is not less than that when the number of sampling points is large, so as to compensate for the uncertainty caused by small sample sampling.

[0082] The samples from all rounds are used to form a training sample set, which is then divided into a training set, a validation set, and a test set according to a preset ratio. The artificial intelligence model is iteratively trained and optimized based on the training set and the validation set. Finally, the model that performs best on the validation set and passes the evaluation on the test set is selected as the reliable threshold prediction model for outputting reliable threshold values ​​in the current environment.

[0083] In one specific implementation, to improve the adaptability of the reliability threshold to time-varying noise, samples are repeatedly collected at different times, different sampling points, and different combinations of anchor nodes. When an intelligent vehicle is used as the mobile carrier, the intelligent vehicle moves within a rectangular or square area covered by the reference node according to a preset path, a random path, or a semi-random path, and transmits or receives signals at multiple sampling locations. The sampling points are sparse sampling points; it is not necessary to cover all pixel positions in the entire area, but only to cover a few typical positions that can reflect the current environmental signal propagation characteristics.

[0084] In one specific implementation, taking into account the directional distribution characteristics and distance distribution characteristics of each calibration node relative to the reference node, the geometric enclosure degree of multiple calibration nodes to the reference node, and application requirements, the relative topology is divided into several symmetry levels. For example, it can be divided into two levels: high symmetry level and low symmetry level, or into three levels: high, medium, and low symmetry levels.

[0085] When the relative topology symmetry level is low, it indicates that the spatial layout of the current calibration node is sensitive to ranging noise. Equal ranging noise may be geometrically amplified during the solution process, leading to increased fluctuations in the calculated position and normalized difference statistics. To avoid misjudging such normal noise fluctuations as abnormal, the confidence threshold label should be appropriately increased. The artificial intelligence model learns the nonlinear mapping relationship between topology, ranging features, normalized difference statistics, number of sampling points, application requirements, and confidence threshold values ​​through different relative topologies and their corresponding supervision labels.

[0086] In a preferred embodiment, the artificial intelligence model employs a regression prediction model. The model input includes one or more of the following: measured distance, calculated distance, normalized difference, normalized difference statistic, anchor node position, reference node position, number of anchor nodes, and number of sampling points; the model output is a confidence threshold value under the current environment.

[0087] In a preferred embodiment, the artificial intelligence model employs a multilayer perceptron regression network. This network includes an input layer, several hidden layers, and an output layer. The input layer receives a feature vector composed of measured distance, calculated distance, normalized difference, and their statistics; the hidden layers are used to extract the correlation between ranging error, positioning error, and node distribution characteristics in the current environment; the output layer outputs a non-negative scalar as a confidence threshold value in the current environment. During model training, the model parameters can be optimized using a mean squared error loss function, a mean absolute error loss function, or a Huber loss function.

[0088] In another preferred embodiment, such as Figure 4 As shown, when the number of existing known anchor nodes in the deployment environment reaches a preset number, there is no need to deploy additional reference signal receivers and mobile calibration nodes. Instead, one of the known anchor nodes is selected as a reference node, and the remaining known anchor nodes are used as calibration nodes. Distance and positioning calculations are performed to obtain the measured distance, calculated distance, and normalized difference corresponding to the reference node. The corresponding data is then used as the input to the reliable threshold prediction model, and the reliable threshold value in the current environment is output.

[0089] To verify the effectiveness of the method of this invention, the following simulation experiment was conducted: 15 anchor nodes were deployed in a 200m×200m simulation area, including 3 malicious nodes; a random forest regression model was used, with 150 decision trees and a maximum depth of 14, and 900 sets of training samples were used, with 60 calibration nodes (sampling points) set in each set. The experimental results are as follows. Figure 5 As shown, the ranging bias caused by malicious nodes gradually increases from 2m to 20m. With the increase in the ranging bias of malicious nodes, the detection rate of abnormal anchor nodes gradually improves. When the ranging bias exceeds 10m, the detection rate stabilizes at over 90%, proving that the present invention can effectively identify and eliminate malicious anchor nodes, ensuring the accuracy and robustness of the positioning results. The above random forest model and its parameters are only an example. In practical applications, neural networks or other artificial intelligence models with regression prediction capabilities can also be used, and the model structure and parameters can be adjusted according to the scale of training data, application environment, and positioning requirements.

[0090] like Figure 6 As shown, the present invention provides an IoT node security positioning system based on environment learning thresholds, comprising:

[0091] The communication ranging module is used to acquire ranging information between the node to be located and multiple anchor nodes with known locations, and to acquire ranging information between the reference node and multiple anchor nodes with known locations; and to obtain the corresponding measurement distance based on the ranging information.

[0092] A reference sampling module is used to perform positioning calculations on the reference node based on the positions of the plurality of anchor nodes and the ranging information, so as to obtain the estimated position of the reference node;

[0093] The positioning calculation module is used to perform positioning calculations on the node to be positioned based on the positions of multiple anchor nodes and their corresponding measured distances, so as to obtain the estimated position of the node to be positioned.

[0094] The difference construction module is used to calculate the calculated distance from the reference node to each anchor node based on the estimated position of the reference node and the position of each anchor node, and to calculate the normalized difference or normalized difference statistic based on the measured distance corresponding to the ranging information and the calculated distance.

[0095] The threshold processing module is used to train and infer reliable threshold values ​​based on an artificial intelligence model. During the training phase, the measured distance, calculated distance, normalized difference, reference node position information, anchor node position information, and / or normalized difference statistics corresponding to the reference node are input into the artificial intelligence model for training to obtain a reliable threshold value prediction model. During the application phase, the similar features corresponding to the reference node in the current environment are input into the reliable threshold value prediction model to output the reliable threshold value in the current environment.

[0096] The safety determination module is used to calculate the calculated distance from the node to be located to each anchor node based on the estimated position of the node to be located, and to calculate the normalized difference based on the measured distance and calculated distance between the node to be located and each anchor node. The normalized difference or its statistic is compared with the confidence threshold value to determine whether the set of anchor nodes participating in the positioning is abnormal.

[0097] The abnormal anchor node processing module is used to perform individual exclusion or clustered positioning analysis on the anchor nodes involved in the positioning when the security determination module determines that the set of anchor nodes involved in the positioning is abnormal. If the normalized difference or normalized difference statistic decreases after excluding or replacing a certain anchor node, the anchor node is marked as a suspected abnormal anchor node. If clustered positioning analysis is used, positioning calculations are performed based on different anchor node clusters, and the anchor node cluster with the smaller normalized difference or normalized difference statistic is selected as the reliable anchor node cluster to obtain the final positioning result.

[0098] Specifically, the system provided by this invention can be deployed in IoT gateways, edge computing devices, host computers, servers, mobile terminals, or embedded development boards. Anchor nodes can be implemented using wireless communication modules based on ZigBee, Wi-Fi, Bluetooth, UWB, LoRa, or other wireless communication protocols. Reference nodes can be fixed signal receivers or installed on intelligent vehicles, mobile robots, drones, or manually operated mobile sampling devices. The system software may include data acquisition programs, ranging calculation programs, positioning calculation programs, artificial intelligence model training programs, reliable threshold output programs, and abnormal anchor node detection programs.

[0099] In practical applications, training samples can be collected using reference nodes and surrounding anchor nodes in a typical application environment to train a reliable threshold prediction model. In subsequent similar environments, the trained reliable threshold prediction model can be directly used to quickly output reliable threshold values ​​based on a small amount of reference ranging data, without the need for extensive manual recalibration. Even if the anchor node distribution changes, the system can quickly obtain reliable threshold values ​​adapted to the current anchor node distribution by re-inputting the current anchor node position, measured distance, calculated distance, and normalized difference statistics.

[0100] This invention combines environmental sampling, location calculation, normalized difference construction, and artificial intelligence threshold learning. It can adaptively generate reliable threshold values ​​under different wireless propagation environments and different anchor node distributions, and identify abnormal anchor nodes based on these reliable threshold values, thereby improving the accuracy, security, and robustness of IoT node positioning results.

[0101] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A secure positioning method for IoT nodes based on environmental learning thresholds, characterized in that, Includes the following steps: S1. In the wireless positioning environment to be applied, obtain the ranging information between the node to be positioned and multiple anchor nodes with known locations, and determine the measured distance between the node to be positioned and each anchor node based on the ranging information. Based on the positions of the multiple anchor nodes and the measured distances between the node to be located and each anchor node, the node to be located is located by calculation to obtain the estimated position of the node to be located. S2. Based on the estimated position of the node to be located and the positions of each anchor node, calculate the calculated distance between the node to be located and each anchor node; Based on the measured distance and calculated distance between the node to be located and each anchor node, calculate the normalized difference or normalized difference statistic corresponding to the node to be located; S3. Input the measured distance, calculated distance, normalized difference, reference node location information, anchor node location information, and / or normalized difference statistics corresponding to the reference node in the current environment into the pre-trained confidence threshold prediction model, and output the confidence threshold value in the current environment; the confidence threshold prediction model is a model trained using multiple rounds of calibration samples formed by reference nodes and calibration nodes with known locations in the application environment or similar environments; compare the normalized difference or normalized difference statistics corresponding to the node to be located with the confidence threshold value; When the normalized difference or normalized difference statistic does not exceed the confidence threshold, the set of anchor nodes participating in the positioning is determined to be reliable, and the estimated position of the node to be positioned is output. When the normalized difference or normalized difference statistic exceeds the confidence threshold, the set of anchor nodes participating in the positioning is determined to be abnormal. S4. When the set of anchor nodes participating in the positioning is determined to be abnormal, perform node-by-node verification or cluster positioning analysis on the anchor nodes participating in the positioning, exclude, replace or mark suspected abnormal anchor nodes, and re-position based on the processed set of anchor nodes.

2. The IoT node secure positioning method based on environment learning threshold according to claim 1, characterized in that, The reliable threshold prediction model is trained through the following steps: S31. In the wireless positioning environment to be applied or a wireless positioning environment similar to the environment to be applied, select at least one reference node with a known location and determine multiple calibration nodes with known locations; the reference node is a reference signal receiver or an equivalent signal receiver node; the calibration node includes a physical anchor node with a known location, and / or a virtual anchor node formed by a mobile calibration node at multiple sampling points. S32. Obtain the distance measurement information between the reference node and multiple calibration nodes, and determine the measurement distance from the reference node to each calibration node based on the distance measurement information; S33. Treat the reference node as the node to be located. Based on the positions of the multiple calibration nodes and the measured distances between the reference node and each calibration node, perform positioning calculations on the reference node to obtain the estimated position of the reference node. Calculate the calculated distances between the reference node and each calibration node based on the estimated position of the reference node and the positions of each calibration node. S34. Based on the measured distance and calculated distance between the reference node and each calibration node, calculate the normalized difference or normalized difference statistic corresponding to each calibration node. S35. The measured distance, calculated distance, normalized difference, reference node position information, calibration node position information, number of sampling points and / or normalized difference statistics formed by multiple rounds of sampling are used as feature inputs of the artificial intelligence model. The basic confidence threshold label is determined based on the normalized difference statistic and the number of sampling points corresponding to each round of sampling. Then, the basic confidence threshold label is adjusted by combining the actual distance between the reference node and each calibration node, the relative topological symmetry level between the reference node and multiple calibration nodes, and the system application requirements to determine the target confidence threshold value of the sampling round, which serves as the supervision label for the sample of that round. The samples from all rounds are used to form a training sample set, which is then divided into a training set, a validation set, and a test set according to a preset ratio. The artificial intelligence model is iteratively trained and optimized based on the training set and the validation set. Finally, the model that performs best on the validation set and passes the evaluation on the test set is selected as the reliable threshold prediction model for outputting reliable threshold values ​​in the current environment.

3. A secure positioning method for IoT nodes based on environmental learning thresholds according to claim 1 or 2, characterized in that, The ranging information includes one or more of the following: received signal strength indication, time of arrival, time difference of arrival, angle of arrival, and channel state information; when using received signal strength indication for ranging, the received signal strength indication is converted into the measured distance based on the path loss model.

4. The IoT node secure positioning method based on environment learning threshold according to claim 1, characterized in that, The normalized difference statistic includes one or more of the mean, variance, maximum, minimum, median, and quantile of the normalized difference; when the normalized difference statistic is greater than the confidence threshold, it is determined that there are abnormal or malicious anchor nodes in the corresponding anchor node set or anchor node group.

5. The IoT node secure positioning method based on environment learning threshold according to claim 1, characterized in that, The node-by-node verification includes: removing one anchor node from the set of anchor nodes that have been judged to be abnormal, and recalculating the estimated position and normalized difference statistic of the node to be located using the remaining anchor nodes; if the normalized difference statistic decreases to within the confidence threshold after removing an anchor node, or the estimated position of the node to be located tends to stabilize, then the anchor node is identified as a suspected abnormal anchor node or a malicious anchor node.

6. The IoT node secure positioning method based on environment learning threshold according to claim 1, characterized in that, The clustered localization analysis includes: dividing the anchor nodes involved in the localization into multiple anchor node clusters, calculating the estimated position of the node to be located based on different anchor node clusters, and calculating the normalized difference statistic corresponding to each anchor node cluster; selecting anchor node clusters whose normalized difference statistic is less than the confidence threshold as reliable anchor node clusters; when the localization results of multiple anchor node clusters conflict with each other and the normalized difference statistic does not exceed the confidence threshold, rejecting the current localization result or indicating that the current localization environment is abnormal.

7. The IoT node secure positioning method based on environment learning threshold according to claim 2, characterized in that, When the calibration node includes a virtual anchor node formed by a mobile calibration node at multiple sampling points, the mobile calibration node moves within the inscribed rectangular or square region of the reference node's communication coverage area according to a preset route, a random route, or a combination of both, and transmits positioning signals and reports its own position at multiple sampling points; the sampling points are sparse sampling points, and it is not required to cover all grid points or pixels within the inscribed rectangular or square region.

8. The IoT node secure positioning method based on environment learning threshold according to claim 2, characterized in that, The sampling point quantity coefficient is a coefficient that is preset manually, obtained by looking up a table, or calculated by a function based on the number of sampling points in each round. The smaller the number of sampling points, the larger the sampling point quantity coefficient is, or the coefficient is not less than that when the number of sampling points is large.

9. The IoT node secure positioning method based on environment learning threshold according to claim 2, characterized in that, The relative topology symmetry level is manually divided into at least two levels according to preset labeling rules, based on the directional distribution characteristics, distance distribution characteristics, and the degree of enclosure of each calibration node relative to the reference node. The higher the relative topology symmetry level, the more evenly the directional distribution of multiple calibration nodes around the reference node and the higher the degree of enclosure of the reference node. Conversely, the lower the relative topology symmetry level, the more concentrated the multiple calibration nodes are on one side of the reference node or the more uneven the directional distribution. Under the same label setting conditions, if the relative topology symmetry level is lower than the preset threshold, it is used as one of the judgment conditions for raising the target confidence threshold value supervision label. The increase range is manually set according to the tolerance of the application scenario for positioning error, false alarm rate, and false negative rate.

10. A secure positioning method for IoT nodes based on environmental learning thresholds according to claim 2, characterized in that, The system application requirements correspond to the following requirement parameters: tolerance for one or more of the following: positioning error, false alarm rate, and false negative rate.

11. A secure positioning method for IoT nodes based on environmental learning thresholds according to claim 2, characterized in that, The artificial intelligence model includes one or more of the following: multilayer perceptron, convolutional neural network, recurrent neural network, Transformer model, graph neural network, support vector regression model, random forest regression model, gradient boosting regression model, or ensemble learning model.

12. The IoT node secure positioning method based on environment learning threshold according to claim 2, characterized in that, When the number of known anchor nodes in the deployment environment reaches a preset number, there is no need to deploy additional reference signal receivers and mobile calibration nodes. Instead, one of the known anchor nodes is selected as a reference node, and the remaining known anchor nodes are used as calibration nodes. Distance and positioning calculations are performed to obtain the measured distance, calculated distance, and normalized difference corresponding to the reference node. The corresponding data is then used as input to the artificial intelligence model to learn or update the confidence threshold value in the current environment.

13. A secure positioning system for IoT nodes based on environmental learning thresholds, characterized in that, The method for secure positioning of IoT nodes based on environment learning thresholds as described in any one of claims 1 to 12 includes: The communication ranging module is used to acquire ranging information between the node to be located and multiple anchor nodes with known locations, and to acquire ranging information between the reference node and multiple anchor nodes with known locations; and to obtain the corresponding measurement distance based on the ranging information. A reference sampling module is used to perform positioning calculations on the reference node based on the positions of the plurality of anchor nodes and the ranging information, so as to obtain the estimated position of the reference node; The positioning calculation module is used to perform positioning calculations on the node to be positioned based on the positions of multiple anchor nodes and their corresponding measured distances, so as to obtain the estimated position of the node to be positioned. The difference construction module is used to calculate the calculated distance from the reference node to each anchor node based on the estimated position of the reference node and the position of each anchor node, and to calculate the normalized difference or normalized difference statistic based on the distance measurement information and the corresponding measured distance. The threshold processing module is used to train and infer reliable threshold values ​​based on an artificial intelligence model. During the training phase, the measured distance, calculated distance, normalized difference, reference node position information, anchor node position information, and / or normalized difference statistics corresponding to the reference node are input into the artificial intelligence model for training to obtain a reliable threshold value prediction model. During the application phase, the similar features corresponding to the reference node in the current environment are input into the reliable threshold value prediction model to output the reliable threshold value in the current environment. The safety determination module is used to calculate the calculated distance from the node to be located to each anchor node based on the estimated position of the node to be located, and to calculate the normalized difference or normalized difference statistic based on the measured distance and calculated distance between the node to be located and each anchor node. The normalized difference or its statistic is compared with the confidence threshold value to determine whether the set of anchor nodes participating in the positioning is abnormal. The abnormal anchor node processing module is used to perform individual exclusion or clustered positioning analysis on the anchor nodes involved in the positioning when the security determination module determines that the set of anchor nodes involved in the positioning is abnormal. If the normalized difference or normalized difference statistic decreases after excluding or replacing a certain anchor node, the anchor node is marked as a suspected abnormal anchor node. If clustered positioning analysis is used, positioning calculations are performed based on different anchor node clusters, and anchor node clusters with normalized difference or normalized difference statistic less than a preset value are selected as reliable anchor node clusters to obtain the final positioning result.

Images