Anomaly perception and analysis method of multi-source data in the Internet of Things based on AI

Abstract

The present invention discloses an AI-based method for anomaly perception and analysis of multi-source data in the Internet of Things, which relates to the field of data anomaly perception and analysis. The main scheme is: by analyzing and calculating communication quality data, the dynamic entropy value of the signal-to-noise ratio and the beamforming distortion of the physical layer channel of the communication system are obtained; by analyzing and calculating perception parameter data, the Doppler profile similarity and point cloud structure entropy of the base station signal data are obtained; by calculating the hardware anomaly index of the base station equipment based on the equipment status data; by analyzing and calculating the environmental interference data, the multipath interference factor of the channel is obtained; and further analysis is performed to obtain the comprehensive feature fusion coefficient and the comprehensive anomaly probability; by analyzing historical fault data and the comprehensive anomaly probability, the fault type is judged, which is conducive to improving the multi-source data fusion capability of the analysis method, enhancing the dynamic adaptability of the evaluation threshold and improving the diagnostic efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of data anomaly perception and analysis, and specifically to an AI-based multi-source data anomaly perception and analysis method for the Internet of Things. Background Art

[0002] With the rapid development of Internet of Things technology, the scale of equipment and data dimensions in scenarios such as smart industry and smart cities are growing exponentially. Real-time anomaly perception of multi-source heterogeneous data has become a core requirement for ensuring system reliability.

[0003] Traditional anomaly detection methods often focus on a single data dimension, such as device status or communication quality, and struggle to cope with the dynamic coupling of multiple data sources in complex environments. For example, in the Industrial Internet of Things, device hardware anomalies can compound with environmental interference, leading to fluctuations in communication links and distortion of sensory data. Using a single threshold for a single dimension can easily lead to misjudgments or missed detections.

[0004] Therefore, current technologies suffer from the following flaws: First, they lack the ability to fuse multi-source data. Existing methods typically analyze communication quality or device operating parameters independently, lacking modeling of the correlation between sensory data and environmental interference, resulting in one-sided extraction of abnormal features. Second, they suffer from poor dynamic adaptability. Most solutions rely on static thresholds or fixed rules, unable to effectively capture nonlinear changes such as channel dynamic entropy and beam distortion, resulting in high false alarm rates in complex electromagnetic environments or mobile scenarios. Third, diagnostic logic lags. Existing technologies often make decisions based on isolated real-time data and lack correlation analysis with historical fault databases, making it difficult to intelligently infer fault types and locate root causes. Summary of the Invention

[0005] (1) Technical problems solved

[0006] In response to the shortcomings of the existing technology, the present invention provides an AI-based IoT multi-source data anomaly perception and analysis method, which at least solves the problems of insufficient multi-source data fusion capability, poor dynamic adaptability, and lagging diagnostic logic in the existing technology.

[0007] (2) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: an AI-based IoT multi-source data anomaly perception and analysis method, comprising:

[0009] Step 1: Obtain IoT communication quality data, perception parameter data, device status data, and environmental interference data, and perform preprocessing;

[0010] Step 2: By analyzing and calculating the communication quality data, the dynamic entropy value of the signal-to-noise ratio and the beamforming distortion of the communication system's physical layer channel are obtained. By analyzing and calculating the perception parameter data, the Doppler profile similarity and point cloud structure entropy of the base station signal data are obtained. The hardware anomaly index of the base station equipment is calculated based on the equipment status data. The multipath interference factor of the channel is obtained by analyzing and calculating the environmental interference data.

[0011] Step 3: Obtain the comprehensive feature fusion coefficient by comprehensively analyzing the Doppler profile similarity, point cloud structure entropy, and multipath interference factor; and calculate the comprehensive anomaly probability based on the comprehensive feature fusion coefficient;

[0012] Step 4: Set a comprehensive evaluation threshold, analyze and evaluate the comprehensive abnormality probability based on the comprehensive evaluation threshold, and then determine whether to initiate the diagnostic instruction;

[0013] Step 5: Determine the fault type by analyzing historical fault data and comprehensive abnormality probability.

[0014] In the preferred embodiment of the above-mentioned AI-based IoT multi-source data anomaly perception and analysis method, the dynamic entropy value of the signal-to-noise ratio of the subcarrier is obtained by analyzing and calculating the communication quality data, and the formula based on it is as follows:

[0015] ;

[0016] in, SND t Indicates the dynamic entropy value of the signal-to-noise ratio of the subcarrier; SN c Indicates the c The instantaneous signal-to-noise ratio of the subcarriers; c Indicates the subcarrier number, N Indicates the number of subcarriers; SN avg Represents the average instantaneous signal-to-noise ratio of all subcarriers;

[0017] The beamforming distortion is calculated based on the following formula:

[0018] ;

[0019] Among them, BFD t Indicates the beamforming distortion degree; W id,ab represents the theoretical beam weight matrix, W re,ab represents the actual beam weight matrix; a and b denote the number of matrix rows and columns respectively, m and nRepresents the maximum number of matrix rows and columns respectively.

[0020] In the preferred embodiment of the above-mentioned AI-based IoT multi-source data anomaly perception and analysis method, the Doppler profile similarity is obtained by analyzing the perception parameter data, and the calculation formula is as follows:

[0021] ;

[0022] Among them: DPS t represents the Doppler profile similarity; Dref(v) represents the Doppler velocity distribution reference spectrum generated by historical normal data training; Dobs(v) represents the Doppler velocity distribution function observed in real time; v represents the moving speed of the target object;

[0023] The formula for calculating the point cloud structural entropy is as follows:

[0024] ;

[0025] Among them, PSE t represents the point cloud structure entropy; P k Represents the probability density of the three-dimensional point cloud in the spatial bin k, k represents the sequence number of the spatial bin, and K represents the total number of spatial bins.

[0026] In the preferred embodiment of the above-mentioned AI-based IoT multi-source data anomaly perception and analysis method, the hardware anomaly index is obtained by analyzing the device status data, and the formula based on it is as follows:

[0027] ;

[0028] Among them, HAL t Indicates the hardware abnormality index; T J Indicates the junction temperature of the device chip; T th Indicates the temperature threshold of the chip; T mac Indicates the maximum temperature allowed by the device chip under normal working conditions; M u Indicates memory usage; M to represents the total memory capacity; ɑ represents the weight coefficient of the temperature state; β represents the weight coefficient of the memory state, and ɑ+β=1.

[0029] In the preferred embodiment of the above-mentioned AI-based IoT multi-source data anomaly perception and analysis method, the multipath interference factor is obtained by analyzing the environmental interference data, and the formula based on it is as follows:

[0030] ;

[0031] Among them, MIF t Represents the multipath interference factor; SYl Indicates the l The time delay of the multipath component; h l Indicates the l The channel coefficient of the multipath; L represents the number of multipaths.

[0032] In the preferred embodiment of the above-mentioned AI-based IoT multi-source data anomaly perception and analysis method, the dynamic entropy value of the signal-to-noise ratio, beamforming distortion, Doppler profile similarity, and point cloud structure entropy are extracted through a convolutional neural network to obtain a feature vector 1. The hardware anomaly index and multipath interference factor are then processed through a recurrent neural network to obtain a feature vector 2. The two feature vectors are then fused to obtain a comprehensive feature fusion coefficient. The formula is as follows:

[0033] ;

[0034] Among them, Conv1d represents one-dimensional convolution, which means the convolution operation used to process one-dimensional sequence data; GRU is a type of recurrent neural network RNN, which is used to process time series data; represents the continuous state of the hardware anomaly index and multipath interference factor within a time series segment. tg:t represents a time series segment, represents the time period from time tg to time t, and g represents the size of the time window. ⊕ represents the feature fusion operation, which is used to concatenate or add two feature vectors element by element to generate a joint feature vector.

[0035] In the preferred embodiment of the above-mentioned AI-based IoT multi-source data anomaly perception and analysis method, the comprehensive anomaly probability is calculated based on the comprehensive feature fusion coefficient, and the formula is as follows:

[0036] ;

[0037] Among them, Pan t represents the comprehensive abnormal probability; σ represents the Sigmoid function, which is used to map the output value to the range of 0 to 1; λ r The attention weights of the feature vectors r of different dimensions of the comprehensive feature fusion coefficient are calculated by the softmax function, referring to the formula: , where wa is the attention weight matrix; r represents the sequence number of the feature vectors of different dimensions of the comprehensive feature fusion coefficient, which takes values ​​of 1, 2, 3 or 4, corresponding to the communication dimension, perception dimension, device dimension and environment dimension respectively; MLP r It is a frontal neural network with layer perceptrons corresponding to feature vectors r of different dimensions, which is used to process information of corresponding feature dimensions.

[0038] In the preferred embodiment of the above-mentioned AI-based IoT multi-source data anomaly perception and analysis method, the comprehensive evaluation threshold is calculated based on the comprehensive anomaly probability of the historical time period, and the formula is as follows:

[0039] ;

[0040] Among them, TH t represents the comprehensive evaluation threshold; μ t-1 Represents the moving average of the comprehensive abnormal probability in the previous time period; τ t-1 It represents the standard deviation of the comprehensive abnormal probability in the previous time period, ρ is a constant, and τ t-1 Multiplication indicates multiples of the standard deviation;

[0041] Based on the comprehensive evaluation threshold, the comprehensive abnormal probability is analyzed and evaluated, and the method for determining whether to start the diagnostic instruction is as follows: t >TH t When the diagnostic process is started, t ≤TH t The diagnostic process is not started.

[0042] In the preferred embodiment of the above-mentioned AI-based IoT multi-source data anomaly perception and analysis method, the posterior probabilities of different fault causes are obtained by analyzing the comprehensive anomaly probability, based on the following formula:

[0043] ;

[0044] Among them, P (X i |Pan t ) represents the observed comprehensive abnormal probability Pan t In the case of failure reason X i The posterior probability of occurrence; P (Pan t-z |X i ) indicates that the fault reason is X i In the event of a fault, the historical comprehensive abnormal probability Pan observed in the historical fault log t-z The probability of P(X i ) indicates the cause of the fault X i Prior probability of occurrence; i The serial number representing the fault cause, which is a positive integer.

[0045] In the preferred solution of the above-mentioned AI-based IoT multi-source data anomaly perception and analysis method, the method for determining the fault type is: traversing P (X i |Pan t ) under different fault causes Xi, P (Xi |Pan t ) are prioritized from large to small. At this time, P(X i |Pan t ) The fault cause X corresponding to the maximum value i The first priority fault cause is used as the fault type for this judgment. The priorities of the other fault causes are based on P (X i |Pan t ) values, decreasing from large to small;

[0046] If the first priority fault cause is eliminated, the remaining fault causes will be checked in order of priority.

[0047] (3) Beneficial effects

[0048] The present invention provides an AI-based method for analyzing anomaly perception of multi-source data in the Internet of Things, which has the following beneficial effects:

[0049] (1) By preprocessing, communication quality data, perception parameters, device status, and environmental interference data are mapped to a unified numerical range, eliminating the interference of dimensional differences on the fusion model, and achieving the "computability" of multi-source data, laying the foundation for subsequent cross-modal feature fusion, avoiding feature deviations caused by data heterogeneity, and solving the problem that the existing technology lacks a unified preprocessing and alignment mechanism for cross-modal data (communication, perception, equipment, environment), resulting in data heterogeneity that is difficult to eliminate, and thus causing feature fusion failure or misjudgment.

[0050] (2) By analyzing and calculating various types of data, we can obtain features such as the dynamic entropy value of the signal-to-noise ratio of the physical layer channel of the communication system, the beamforming distortion degree, the Doppler profile similarity and point cloud structure entropy of the base station signal data, the hardware anomaly index of the base station equipment, and the multipath interference factor of the channel. Extracting these targeted features can accurately characterize the status and potential anomalies of different aspects of the IoT system. The comprehensive application of these features makes the anomaly perception analysis more in-depth and detailed, and can effectively capture various potential anomalies, thereby improving the accuracy and sensitivity of anomaly perception.

[0051] (3) By comprehensively analyzing the Doppler profile similarity, point cloud structure entropy, and multipath interference factor, the comprehensive feature fusion coefficient is obtained, and the comprehensive anomaly probability is calculated based on this. This comprehensive feature fusion method fully utilizes the advantages of each feature, organically integrates information from different aspects, and can more comprehensively reflect the overall operating status of the Internet of Things system. The calculation of the comprehensive anomaly probability quantifies the influence of multiple features into a specific probability value, intuitively indicating the possibility of system anomalies, providing a direct basis for subsequent evaluation and decision-making, and helping to achieve early warning and timely handling of anomalies in the Internet of Things system, thereby enhancing the reliability and stability of the system.

[0052] (4) Based on the comprehensive evaluation threshold, the comprehensive abnormality probability is analyzed and evaluated, and then a decision is made whether to initiate the diagnostic instruction. Setting a reasonable comprehensive evaluation threshold can effectively control false positives and missed positives, ensuring that the diagnostic instruction is initiated only when the abnormality probability reaches a certain level, thus avoiding unnecessary waste of resources and system interference. In this way, the automated and intelligent evaluation and decision-making of abnormal conditions in the IoT system are achieved, the operation and maintenance efficiency and management level of the system are improved, and potential failure risks can be responded to in a timely manner, reducing the probability and impact of failures.

[0053] (5) By analyzing historical fault data and comprehensive abnormality probabilities, the fault type can be determined. By utilizing the experience and knowledge in historical fault data and combining it with the current comprehensive abnormality probability, the specific fault type can be identified more accurately. This helps operation and maintenance personnel prepare corresponding maintenance plans and resources in advance, quickly locate and resolve fault problems, reduce fault repair time and system downtime, and improve the availability and user experience of the IoT system. At the same time, through statistics and analysis of fault types, it can also provide a strong basis for system optimization and improvement, further improving the performance and reliability of the IoT system. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a schematic diagram of the steps of the AI-based IoT multi-source data anomaly perception and analysis method of the present invention. DETAILED DESCRIPTION

[0055] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0056] Example 1

[0057] See also Figure 1The present invention provides an AI-based method for analyzing anomaly perception of multi-source data in the Internet of Things, including:

[0058] Step 1: Obtain IoT communication quality data, perception parameter data, device status data, and environmental interference data, and perform preprocessing.

[0059] It should be noted that after obtaining the communication quality data, perception parameter data, device status data and environmental interference data, the obtained data are normalized and preprocessed to eliminate the dimensional differences of different data, and then subsequent formula calculations are performed to facilitate the fusion calculation of multiple data.

[0060] Step 2: By analyzing and calculating the communication quality data, the dynamic entropy value of the signal-to-noise ratio and the beamforming distortion of the physical layer channel of the communication system are obtained; by analyzing and calculating the perception parameter data, the Doppler profile similarity and point cloud structure entropy of the base station signal data are obtained; by calculating the hardware anomaly index of the base station equipment based on the equipment status data; by analyzing and calculating the environmental interference data, the multipath interference factor of the channel is obtained.

[0061] Step 201: The communication quality data includes obtaining the instantaneous signal-to-noise ratio of each subcarrier from the network base station or access point, aggregating the instantaneous signal-to-noise ratio data of all subcarriers collected into a data processing unit, counting the number of subcarriers, summing the aggregated instantaneous signal-to-noise ratio data, and then dividing it by the number of subcarriers to obtain the average instantaneous signal-to-noise ratio of all subcarriers; then, based on the instantaneous signal-to-noise ratio of the subcarrier and the average instantaneous signal-to-noise ratio of all subcarriers, the dynamic entropy value of the signal-to-noise ratio of the subcarrier is calculated using an entropy value calculation model. SND t , the calculation formula of the value calculation model is as follows:

[0062] ;

[0063] in, SND t Indicates the dynamic entropy value of the signal-to-noise ratio of the subcarrier; SN c Indicates the c The instantaneous signal-to-noise ratio of the subcarriers; c Indicates the subcarrier number, which is a positive integer. N Indicates the number of subcarriers; SN avg Indicates the average instantaneous signal-to-noise ratio of all subcarriers.

[0064] It should be noted that this formula is used to quantify the degree of fluctuation in channel quality. A higher entropy value indicates a more unstable channel state, which is conducive to accurate assessment of the channel state.

[0065] Channel state information for each subcarrier is obtained from the physical layer channel state information of the IoT base station, and beam steering parameters are obtained from the real-time beam steering parameters of the base station antenna array. This information is obtained through the physical layer channel estimation module, a key component of communication systems, used to estimate the characteristics of the signal during channel transmission. The purpose of channel estimation is to predict the channel transmission conditions so that the received signal can be correctly processed. In communication systems, the spectrum is divided into multiple subcarriers, each of which can be independently modulated and demodulated.

[0066] Step 202: The communication quality data also includes a theoretical beam weight matrix extracted from the design document or configuration file of the beamforming algorithm; and an actual channel response matrix calculated by performing channel measurement between the IoT device and the base station, sending a pilot signal and receiving feedback channel state information CSI, and using a channel estimation algorithm such as minimum mean square error (MMSE) or least squares (LS), as the actual beam weight matrix; the beamforming distortion degree (BFD) is obtained by inputting the theoretical beam weight matrix and the interim beam weight matrix into the calculation model of the beamforming distortion degree. t , the calculation formula of the beamforming distortion calculation model is as follows:

[0067] ;

[0068] Among them, BFD t Indicates the beamforming distortion degree; W id,ab represents the theoretical beam weight matrix, W re,ab Represents the actual beam weight matrix, which can be obtained from the base station digital intermediate frequency unit DSP Chip reading; a and b denote the number of matrix rows and columns respectively, m and n Represents the maximum number of matrix rows and columns respectively.

[0069] It should be noted that the numerator of the formula calculates the square root of the sum of the squares of the differences between the theoretical and actual beam weight matrix elements, reflecting the degree of deviation between the actual beamforming and the theoretical design. The denominator is the square root of the sum of the squares of the theoretical beam weight matrix elements. It is used to normalize the numerator, making the distortion result relative, which can more objectively reflect the distortion of the beamforming and avoid comparison deviations caused by the different scales of the theoretical weight matrix itself.

[0070] In IoT communication systems, beamforming technology is crucial for improving the directivity and efficiency of signal transmission. However, actual beamforming performance can deviate from theoretical designs due to various factors, such as hardware variations, environmental interference, and signal reflections. This can lead to problems such as degraded signal transmission quality and unstable communication links. Traditional monitoring methods struggle to accurately quantify this deviation, making it difficult to effectively and promptly detect anomalies in the beamforming process. By calculating the beamforming distortion degree using this formula, we can accurately quantify the difference between actual beamforming and theoretical designs, providing a reliable and intuitive indicator for monitoring beamforming accuracy and stability. During anomaly detection and analysis, if the beamforming distortion degree exceeds the normal range, an early warning can be issued, alerting operators to potential issues with beamforming-related equipment or parameters. This helps identify anomalies that may affect communication quality in advance, allowing appropriate adjustments and optimization measures to ensure the proper functioning of beamforming technology. This, in turn, improves signal transmission quality and link stability in IoT communication systems, reduces communication interruptions and data transmission errors caused by beamforming anomalies, and ensures the efficient and reliable operation of the entire IoT system.

[0071] Step 203: The sensing parameter data includes using a millimeter wave radar signal processing module to collect a large amount of historical normal data, and converting the time domain signal into a frequency domain signal through Fourier transform to extract the Doppler velocity distribution characteristics, thereby generating a Doppler velocity distribution reference spectrum and a real-time Doppler velocity distribution function obtained by real-time acquisition of radar echo signals; and inputting the Doppler velocity distribution reference spectrum generated by historical normal data training and the real-time observed Doppler velocity distribution function into the Doppler profile similarity calculation model to calculate the Doppler profile similarity DPS. t ; The calculation formula of the Doppler profile similarity calculation model is as follows:

[0072] ;

[0073] Among them: DPS t represents the Doppler profile similarity; Dref(v) represents the Doppler velocity distribution reference spectrum generated by historical normal data training; Dobs(v) represents the real-time observed Doppler velocity distribution function, which can be extracted by the radar signal processing module; v represents the moving speed of the target object.

[0074] It should be noted that both Dref(v) and Dobs(v) are functions that represent the Doppler velocity distribution. Dref(v) is generated by collecting and analyzing a large amount of historical normal data (for example, radar echo signals of normally moving targets) using machine learning or statistical analysis methods to generate a reference spectrum. For example, historical data is acquired through the millimeter-wave radar signal processing module and trained using data analysis tools such as Python and MATLAB to generate the reference spectrum. Dobs(v) can be extracted by collecting radar echo signals in real time through the millimeter-wave radar signal processing module and performing signal processing (such as Fourier transform) to extract the Doppler velocity distribution. Doppler profile similarity is used to detect abnormal behavior of moving targets, such as sudden acceleration or direction changes. If the DPS value is less than 0.6, it indicates that the currently observed Doppler velocity distribution differs significantly from the reference spectrum, indicating a possible anomaly.

[0075] This formula calculates Doppler profile similarity by comparing the difference between the Doppler velocity distribution reference spectrum Dref(v) generated by training with historical normal data and the Doppler velocity distribution function Dobs(v) observed in real time. Specifically, the numerator of the formula calculates the sum of the squares of the differences between the reference spectrum and the observed spectrum, reflecting the degree of difference between the two. The denominator calculates the sum of the squares of the sum of the reference spectrum and the observed spectrum, which is used to normalize the numerator so that the similarity result falls between 0 and 1. The closer the similarity is to 1, the more similar the real-time observed Doppler profile is to the historical normal data; conversely, the greater the difference.

[0076] In IoT communication systems, base station echo signals are affected by a variety of factors, such as environmental changes, equipment aging, and signal interference, leading to variations in the Doppler velocity distribution. Traditional monitoring methods struggle to effectively quantify the difference between the real-time Doppler profile and the normal state, making it difficult to detect signal anomalies in a timely manner, which in turn affects the accurate assessment of base station signal status. This formula, which calculates Doppler profile similarity, accurately quantifies the degree of similarity between the real-time observed Doppler velocity distribution and historical normal data, providing an intuitive and reliable indicator for monitoring base station signal stability.

[0077] Step 204: The perception parameter data also includes the point cloud data of the target scene collected by the three-dimensional sensor, and the collected point cloud data is pre-processed; the three-dimensional space is divided into multiple spatial bins of equal size, and the number of three-dimensional point clouds in each spatial bin is counted by calculating the index of the point cloud coordinates in the bin coordinate system, and then the probability density of the three-dimensional point cloud in each spatial bin is calculated; the probability density in the spatial bin is input into the point cloud structure entropy calculation model to obtain the point cloud structure entropy PSE t , the calculation formula of the point cloud structure entropy calculation model is as follows:

[0078] ;

[0079] Among them, PSE t represents the point cloud structure entropy; P k It represents the probability density of the three-dimensional point cloud in the spatial bin k, where k represents the sequence number of the spatial bin and is a positive integer. K represents the total number of spatial bins, which is determined by the maximum detection distance of the lidar.

[0080] This formula calculates the point cloud structural entropy by calculating the probability density of a 3D point cloud in different spatial bins. Specifically, the maximum detection space of the lidar is first divided into multiple bins. The frequency of point cloud occurrence in each bin is then counted to obtain the probability density pk. Finally, the information entropy formula is used to calculate the weighted sum of the probability densities of all bins to obtain the point cloud structural entropy. The point cloud structural entropy reflects the complexity and uncertainty of the spatial distribution of the point cloud data. Higher entropy values ​​indicate a more complex and disordered spatial distribution of the point cloud data.

[0081] In IoT perception environments, LiDAR point cloud data is susceptible to environmental factors such as weather changes and obstacle movement, as well as equipment status changes such as variations in scanning accuracy and wear of mechanical components, leading to changes in the spatial distribution of point cloud data. Traditional point cloud data processing methods struggle to effectively quantify these changes in spatial distribution, making it difficult to promptly detect anomalies in point cloud data, which in turn affects the accuracy of environmental perception. By calculating point cloud structural entropy using this formula, the complexity and variability of point cloud data's spatial distribution can be accurately quantified, providing an intuitive and reliable metric for monitoring the stability and consistency of point cloud data. This ensures the quality and reliability of point cloud data, thereby improving the environmental perception capabilities and decision-making accuracy of IoT perception systems, reducing perception errors or system misjudgments caused by point cloud data anomalies, and ensuring the efficient and stable operation of the entire IoT system.

[0082] Step 205: Analyze the device status data to obtain a hardware anomaly index based on the following formula:

[0083] ;

[0084] Among them, HAL t Indicates the hardware abnormality index; T J Indicates the junction temperature of the device chip, collected by the temperature sensor; T th Indicates the temperature threshold of the chip, which is generally 85 degrees Celsius according to the chip specification. mac Indicates the maximum temperature allowed by the device chip under normal working conditions. You can refer to the parameters defined in the device specification to ensure that the chip operates within a safe temperature range. u Indicates memory usage, which can be read from the device driver; M toIndicates the total memory capacity. This is a parameter in the device hardware configuration used to evaluate memory usage. ɑ represents the weighting factor for temperature status; β represents the weighting factor for memory status, where ɑ + β = 1. Here, ɑ can be 0.7, and β can be 0.3.

[0085] It should be noted that the device chip here can be a core processing chip or key function chip in an IoT device, such as a central processing unit (CPU) chip, a graphics processing unit (GPU) chip, a baseband chip, or an application-specific integrated circuit (ASIC) chip, or the chip with the highest junction temperature among them can be selected as the device chip junction temperature.

[0086] This formula calculates the hardware anomaly index by comprehensively considering two key factors: the device's chip junction temperature and memory usage. Specifically, the difference between the device's chip junction temperature and the chip's temperature threshold is calculated, then divided by the difference between the chip's maximum allowable temperature and the temperature threshold. This yields a normalized temperature-related index, reflecting the extent to which the temperature exceeds the normal range. Simultaneously, the ratio of memory usage to total memory capacity is calculated to yield a memory usage-related index, reflecting the severity of memory usage. These two indicators are then assigned corresponding weighting coefficients, and finally, the two are added together to produce the hardware anomaly index. The weighting coefficients can be adjusted based on actual conditions.

[0087] During the operation of IoT devices, hardware anomalies can be caused by a variety of factors, such as chip overheating and excessive memory usage. Traditional hardware monitoring methods often focus on a single metric, making it difficult to fully and accurately reflect the overall operating status of the hardware and prone to omissions or false alarms. Furthermore, the hardware parameters and operating conditions of different devices vary significantly, making a universal and effective method for quantifying hardware anomalies lacking. This formula calculates the hardware anomaly index, comprehensively considering two key factors: chip temperature and memory usage. This provides a straightforward and reliable indicator for monitoring the hardware status of IoT devices.

[0088] Step 206: Analyze the environmental interference data to obtain the multipath interference factor, based on the following formula:

[0089] ;

[0090] Among them, MIF t Represents the multipath interference factor; SY l Indicates the l The time delay of the multipath component can be obtained by analyzing the channel impulse response CIR; l Indicates the lThe channel coefficients of the multipaths can be extracted from the CIR data; L represents the number of multipaths and can be determined by the CIR peak detection algorithm.

[0091] This formula calculates the multipath interference factor by analyzing channel impulse response data. Specifically, it first obtains the delay and channel coefficient of each multipath component. The delay reflects the time difference in signal transmission along different paths, while the channel coefficient reflects the attenuation and phase change of the signal along different paths. Next, the product of the delay of each multipath component and the square of its channel coefficient modulus is calculated. This product is summed for all multipath components and finally divided by the number of multipath paths to obtain the multipath interference factor. This factor quantifies the degree of signal interference caused by multipath effects; larger values ​​indicate more severe multipath interference.

[0092] In IoT communications, multipath occurs when signals propagate through complex environments, such as urban buildings and terrain, where signals are reflected, scattered, and refracted. This causes the received signal to contain multiple copies of the signal along different paths. These copies interfere with each other, causing signal distortion and increased bit error rates. Existing methods struggle to accurately quantify the extent of multipath interference, making it difficult to effectively assess and improve communication quality. Calculating the multipath interference factor using this formula precisely quantifies the degree of multipath interference on the signal, providing a straightforward and specific metric for evaluating channel quality and communication reliability.

[0093] Step 3: By comprehensively analyzing the Doppler profile similarity, point cloud structure entropy and multipath interference factor, the comprehensive feature fusion coefficient is obtained; and the comprehensive anomaly probability is calculated based on the comprehensive feature fusion coefficient.

[0094] Step 301: Use a convolutional neural network to extract features from the dynamic entropy of the signal-to-noise ratio, beamforming distortion, Doppler profile similarity, and point cloud structure entropy to obtain feature vector 1. Then, use a recurrent neural network to process the hardware anomaly index and multipath interference factor to obtain feature vector 2. The two feature vectors are then fused to obtain a comprehensive feature fusion coefficient based on the following formula:

[0095] ;

[0096] Among them, F t Represents the comprehensive feature fusion coefficient; Conv1d represents one-dimensional convolution, which represents the convolution operation used to process one-dimensional sequence data; it extracts local features by sliding a convolution kernel on the input data; It represents the joint distribution of the dynamic entropy of the signal-to-noise ratio, beamforming distortion, Doppler profile similarity, and point cloud structure entropy at time t. GRU is a type of recurrent neural network RNN, which is used to process time series data. It represents the continuous state of the hardware anomaly index and multipath interference factor from time tg to time t within a time series segment. Here, g is the size of the time window, which is used to capture the changing trend of HAI and MIF over a period of time. ⊕ represents the feature fusion operation, which is used to concatenate or add two feature vectors element by element to generate a joint feature vector.

[0097] This formula processes different types of data using convolutional neural networks and recurrent neural networks, respectively, and then fuses the two resulting feature vectors to produce a comprehensive feature fusion coefficient. Specifically, a one-dimensional convolution is first used to extract the temporal joint distribution of four features: the dynamic entropy of the signal-to-noise ratio, beamforming distortion, Doppler profile similarity, and point cloud structural entropy. This produces feature vector 1, which captures the local characteristics of these features. Simultaneously, a recurrent neural network is used to process the continuous states of the hardware anomaly index and multipath interference factor from time t−g to t, producing feature vector 2, which captures the temporal trends of these two features. Finally, through a feature fusion operation, the two feature vectors are concatenated or element-by-element added to generate a comprehensive feature fusion coefficient, achieving deep fusion of multi-source heterogeneous data and fully utilizing the information contained in each feature.

[0098] In anomaly perception and analysis of multi-source IoT data, different types of feature data (such as dynamic entropy of the signal-to-noise ratio, beamforming distortion, Doppler profile similarity, point cloud structural entropy, hardware anomaly index, and multipath interference factor) have different characteristics and temporal dimensions. Traditional feature processing methods struggle to simultaneously process both static and time series features, failing to fully utilize the information contained in each feature. This results in incomplete and inaccurate feature representation, impacting anomaly perception performance. This formula, which calculates the comprehensive feature fusion coefficient, effectively integrates feature information from different data sources. This includes local features of dynamic entropy of the signal-to-noise ratio, beamforming distortion, Doppler profile similarity, and point cloud structural entropy extracted via a convolutional neural network, as well as the time series features of the hardware anomaly index and multipath interference factor processed via a recurrent neural network. This fusion method fully leverages the advantages of convolutional neural networks in processing local features and recurrent neural networks in processing time series data, so that the obtained comprehensive feature fusion coefficient more comprehensively and accurately reflects the operating status of the Internet of Things system, provides more powerful feature expression for subsequent anomaly perception analysis, improves the accuracy and timeliness of anomaly perception, reduces the omission and false alarm of anomalies, and enhances the performance and reliability of the anomaly perception analysis method of the entire Internet of Things system, enabling it to more effectively deal with anomalies in complex Internet of Things environments.

[0099] It should be noted that the number of convolution kernels of Conv1D can be set to 2, corresponding to the perception dimension and communication dimension; the convolution kernel size selects an appropriate convolution kernel size, such as kernel_size=3, to extract local features; output dimension: the output of Conv1D is a two-dimensional matrix with a shape of (T,2), where T is the length of the input sequence and 2 is the number of convolution kernels. The two dimensions correspond to the perception dimension such as DPS and PSE, and the communication dimension such as SNDE and BFD; the number of hidden units of GRU is set to 2, corresponding to the device dimension and the environment dimension; the input of GRU is Output dimension: The output of GRU is a two-dimensional matrix with shape (T,2), where T is the length of the time series and 2 is the number of hidden units. The two dimensions correspond to the device dimension such as HAI and the environment dimension such as MIF. The input of feature fusion (⊕) is the output of Conv1D and GRU, and its output dimension is the concatenated F t The dimensions are 2+2=4, and each dimension corresponds to the communication dimension, perception dimension, device dimension, and environment dimension.

[0100] Step 302: Calculate the comprehensive abnormality probability based on the comprehensive feature fusion coefficient, according to the following formula:

[0101] ;

[0102] Among them, Pan t represents the comprehensive abnormal probability; σ represents the Sigmoid function, which can be expressed as , used to map the output value to the range from 0 to 1, λ r The attention weights of the feature vectors r of different dimensions representing the comprehensive feature fusion coefficient are used to measure the importance of each dimension feature vector in anomaly detection. They can be calculated using the softmax function, referring to the formula: , where wa is the attention weight matrix, which is used to learn the importance of each feature dimension. It is usually initialized using a random method, such as Gaussian distribution or Xavier initialization, and is learned through a known backpropagation method and optimizer such as Adam; r represents the sequence number of the feature vectors of different dimensions of the comprehensive feature fusion coefficient, and takes values ​​of 1, 2, 3, or 4, corresponding to the communication dimension, perception dimension, device dimension, and environment dimension, respectively; MLP r It is a frontal neural network with four groups of layer perceptrons with corresponding feature vectors r of different dimensions, which is used to process the information of the corresponding feature dimensions and extract the contribution of each dimension feature to the abnormality probability.

[0103] This formula calculates the comprehensive anomaly probability through a multi-layer perceptron (MLP) and an attention mechanism. Specifically, the multi-layer perceptron is first used to process the four feature dimensions of the comprehensive feature fusion coefficient separately, and the contribution of each dimension feature to the anomaly probability is extracted. Then, the importance of each feature dimension in anomaly detection is measured by the attention weight, and the features of different dimensions are weighted. Finally, the weighted features are mapped to the range of 0 to 1 through the Sigmoid function to obtain the comprehensive anomaly probability. The attention weight is calculated by the softmax function, which reflects the relative importance of each feature dimension to anomaly detection under the current feature fusion coefficient, so that the model can automatically learn which features are more critical in anomaly detection. The multi-layer perceptron is used to perform nonlinear transformations on each feature dimension to extract deeper feature representations.

[0104] In anomaly perception and analysis of multi-source data in the Internet of Things (IoT), different features contribute to anomalies to varying degrees, and complex nonlinear relationships exist between features. Traditional methods struggle to effectively measure the importance of individual features or fully exploit the deep connections between them, resulting in inaccurate anomaly probability calculations and failing to meet the requirements for precise anomaly detection. This formula for calculating the combined anomaly probability fully accounts for the varying contributions of different features to anomaly detection. It leverages an attention mechanism to dynamically adjust the weights of each feature, allowing the model to focus more on features related to anomalies. Furthermore, the nonlinear transformation capabilities of the multi-layer perceptron (MLP) enable deeper exploration of the complex relationships between features and extraction of deeper feature representations. This approach, based on the attention mechanism and MLP, effectively improves the accuracy of the combined anomaly probability calculation, providing a more reliable basis for subsequent anomaly assessment and enhancing the performance of anomaly perception analysis. This enables the system to more accurately identify anomalies, reduce false positives and false negatives, and enhance the reliability and stability of IoT systems, ensuring timely detection and resolution of potential anomalies.

[0105] Step 4: Set a comprehensive evaluation threshold, analyze and evaluate the comprehensive abnormality probability based on the comprehensive evaluation threshold, and then determine whether to start the diagnostic instruction.

[0106] Step 401: Calculate the comprehensive evaluation threshold based on the comprehensive abnormal probability of the historical time period, based on the following formula:

[0107] ;

[0108] Among them, TH t represents the comprehensive evaluation threshold; μ t-1 Represents the moving average of the comprehensive abnormal probability in the previous time period; τ t-1 It represents the standard deviation of the comprehensive abnormal probability in the previous time period, ρ is a constant, and τt-1 Multiplication indicates the multiple of the standard deviation and can be adjusted according to actual needs. Here, the value can be 1-3.

[0109] This formula calculates a comprehensive assessment threshold based on the combined anomaly probability over a historical time period. Specifically, the moving average of the combined anomaly probability over the previous time period is calculated to reflect the average level of historical anomaly probability. The standard deviation of the combined anomaly probability over the previous time period is also calculated to reflect the fluctuations in the historical anomaly probability. The standard deviation is then multiplied by a constant and added to the moving average to obtain the comprehensive assessment threshold. The constant ρ can be adjusted based on actual needs and is typically set between 1 and 3. It controls the degree of flexibility of the threshold. A larger ρ value results in a more lenient threshold, while a smaller ρ value results in a more stringent threshold.

[0110] In anomaly perception and analysis of multi-source data in the IoT, there is a lack of adaptive methods for setting assessment thresholds. Traditional methods often use fixed thresholds, which are difficult to adapt to the dynamic nature of anomaly probabilities over time and are prone to false positives and missed negatives. Anomaly probability distributions vary significantly across scenarios, making fixed thresholds inadequate for diverse needs. By calculating a comprehensive assessment threshold using this formula, the threshold can be dynamically adjusted based on the average level and fluctuations of historical anomaly probabilities, adaptively reflecting the current anomaly probability distribution. During the anomaly assessment process, if the comprehensive anomaly probability exceeds this dynamic threshold, an alert is issued promptly, effectively improving the accuracy and timeliness of anomaly assessments and reducing false positives and missed negatives. This dynamic threshold adjustment method enhances the flexibility and adaptability of anomaly perception analysis, enabling the system to better adapt to diverse scenarios and dynamically changing IoT environments. It improves the performance and reliability of the entire anomaly perception system and provides strong support for the timely detection and resolution of anomalies.

[0111] Step 402: Analyze and evaluate the comprehensive abnormality probability based on the comprehensive evaluation threshold, and determine whether to start the diagnosis instruction: When Pan t >TH t When the diagnostic process is started, t ≤TH t The diagnostic process is not started.

[0112] This step determines whether to initiate the diagnostic process by comparing the comprehensive abnormality probability with the comprehensive assessment threshold. Specifically, when the comprehensive abnormality probability is greater than the comprehensive assessment threshold, the system is considered to be abnormal and the diagnostic process is initiated. Conversely, when the comprehensive abnormality probability is less than or equal to the comprehensive assessment threshold, the system is considered to be operating normally and the diagnostic process is not initiated. This threshold-based judgment method effectively enables rapid response and handling of abnormal situations.

[0113] The lack of an effective anomaly assessment and decision-making mechanism in IoT multi-source data anomaly perception and analysis makes it difficult to accurately determine when to initiate the diagnostic process. Traditional methods often rely on fixed thresholds or manual experience, which cannot adapt to the dynamic changes in anomaly probability and are prone to false positives or missed negatives. This anomaly assessment method automatically and accurately determines whether to initiate the diagnostic process based on the comparison of the comprehensive anomaly probability and the comprehensive assessment threshold. This effectively improves the accuracy and timeliness of anomaly assessment, reduces false positives and missed negatives, ensures the system can take timely action when anomalies occur, enhances the reliability and practicality of the entire anomaly perception and analysis system, and provides a strong guarantee for the stable operation of the IoT system.

[0114] Step 5: Determine the fault type by analyzing historical fault data and comprehensive abnormality probability.

[0115] Step 501: By analyzing the comprehensive abnormality probability, the posterior probability of different fault causes is obtained, based on the following formula:

[0116] ;

[0117] Among them, P (X i |Pan t ) represents the observed comprehensive abnormal probability Pan t In the case of failure reason X i The posterior probability of occurrence; P (Pan t-z |X i ) indicates that the fault reason is X i In the event of a fault, the historical comprehensive abnormal probability Pan observed in the historical fault log t-z The probability of P(X i ) indicates the cause of the fault X i Prior probability of occurrence; i The serial number representing the fault cause, which is a positive integer.

[0118] It should be noted that P (Pan t-z |X i ) A large number of fault records can be collected from historical fault logs or historical fault case libraries, including the fault cause X i and the corresponding historical comprehensive abnormal probability Pan t-z ;Fault cause X i It can include AAU hardware failure, CU software error, core network congestion, etc. The historical comprehensive abnormality probability Pan t-z The comprehensive abnormal probability calculated by the calculation method of step 1 to step 3 under historical conditions is P (Pan t-z |X i ) = Fault cause Xi Pan t-z Number of times / failure cause X i The total number of occurrences. P(X i ) can collect a large number of fault records from historical fault logs or historical fault case libraries, and the calculation method is: P (X i ) = Fault Cause X i Number of occurrences / total number of failures.

[0119] In the anomaly perception and analysis of multi-source data in the Internet of Things, traditional methods have difficulty accurately identifying the specific fault cause that causes the anomaly, and cannot effectively use historical fault data to assist in judgment, which easily leads to inaccurate fault diagnosis and low efficiency. When multiple fault causes exist simultaneously, traditional methods are even more difficult to determine the most likely root cause.

[0120] This step accurately calculates the posterior probabilities of different fault causes under the currently observed comprehensive anomaly probability, thereby quickly and accurately locating the most likely fault cause. This not only improves the accuracy of fault diagnosis, but also greatly enhances diagnostic efficiency, reducing the time and workload of manual troubleshooting. For example, in a complex IoT system, if multiple abnormalities occur simultaneously, such as decreased communication quality, device hardware anomalies, and increased environmental interference, this method can comprehensively consider these abnormal manifestations and, combined with historical fault data, accurately determine which fault cause or causes are responsible for the current abnormal condition. This allows operations and maintenance personnel to quickly implement targeted maintenance measures, promptly restore the system to normal operation, reduce the impact of the fault on system performance and business operations, and improve the availability and reliability of the entire IoT system.

[0121] Step 502: The method for determining the fault type is: traverse P(X i |Pan t ) under different fault causes Xi, P (X i |Pan t ) are prioritized from large to small. At this time, P(X i |Pan t ) The fault cause X corresponding to the maximum value i The first priority fault cause is used as the fault type for this judgment. The priorities of the other fault causes are based on P (X i |Pan t ) values, decreasing from large to small.

[0122] If the first priority fault cause is eliminated, the remaining fault causes will be checked in order of priority.

[0123] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0124] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0125] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. An AI-based method for analyzing anomaly perception of multi-source data in the Internet of Things, characterized by: include: Step 1: Obtain IoT communication quality data, perception parameter data, device status data, and environmental interference data, and perform preprocessing; Step 2: By analyzing and calculating the communication quality data, the dynamic entropy value of the signal-to-noise ratio of the physical layer channel of the communication system is obtained. SND t and beamforming distortion (BFD) t ; By analyzing and calculating the perception parameter data, the Doppler profile similarity DPS of the base station signal data is obtained. t and point cloud structural entropy PSE t ; Calculate the hardware anomaly index HAL of the base station equipment by using the equipment status data t ; By analyzing and calculating the environmental interference data, the multipath interference factor MIF of the channel is obtained t ; Step 3: Dynamic entropy value of signal-to-noise ratio SND t , beamforming distortion BFD t , Doppler profile similarity DPS t , point cloud structure entropy PSE t and multipath interference factor MIF t Conduct comprehensive analysis to obtain the comprehensive feature fusion coefficient; Specifically: The dynamic entropy value of the signal-to-noise ratio is calculated through the convolutional neural network SND t , beamforming distortion BFD t , Doppler profile similarity DPS t and point cloud structural entropy PSE t Perform feature extraction to obtain feature vector 1; Then the hardware anomaly index HAL is calculated through a recurrent neural network. t and multipath interference factor MIF t After processing, the obtained feature vector is 2, and then the two feature vectors are fused to obtain the comprehensive feature fusion coefficient. The formula is as follows: ； Among them, F t Represents the comprehensive feature fusion coefficient; Conv1d represents one-dimensional convolution, which represents the convolution operation used to process one-dimensional sequence data; GRU is a type of recurrent neural network RNN ​​used to process time series data; [HAI t-g:t， MIF t-g:t ] represents the continuous state of the hardware anomaly index and multipath interference factor in a time series segment, tg:t represents a time series segment, represents the time period from time tg to time t, and g represents the size of the time window; ⊕ represents the feature fusion operation, which is used to splice or add two feature vectors element by element to generate a joint feature vector; The comprehensive abnormality probability is calculated based on the comprehensive feature fusion coefficient, and the formula is as follows: ； Among them, Pan t represents the comprehensive abnormal probability; σ represents the Sigmoid function, which is used to map the output value to the range of 0 to 1; λ r The attention weights of the feature vectors r of different dimensions of the comprehensive feature fusion coefficient are calculated by the softmax function, referring to the formula: , where wa is the attention weight matrix; r represents the sequence number of the feature vectors of different dimensions of the comprehensive feature fusion coefficient, which takes values ​​of 1, 2, 3 or 4, corresponding to the communication dimension, perception dimension, device dimension and environment dimension respectively; MLP r It is a frontal neural network with layer perceptrons corresponding to feature vectors r of different dimensions, used to process information of corresponding feature dimensions; Step 4: Set a comprehensive evaluation threshold, analyze and evaluate the comprehensive abnormality probability based on the comprehensive evaluation threshold, and then determine whether to initiate the diagnostic instruction; Step 5: Determine the fault type by analyzing historical fault data and comprehensive abnormality probability.

2. The AI-based IoT multi-source data anomaly perception and analysis method according to claim 1 is characterized in that: The communication quality data includes obtaining the instantaneous signal-to-noise ratio of each subcarrier from the network base station or access point, aggregating the instantaneous signal-to-noise ratio data of all subcarriers collected into a data processing unit, counting the number of subcarriers, summing the aggregated instantaneous signal-to-noise ratio data, and then dividing it by the number of subcarriers to obtain the average value of the instantaneous signal-to-noise ratio of all subcarriers; then, based on the instantaneous signal-to-noise ratio of the subcarrier and the average value of the instantaneous signal-to-noise ratio of all subcarriers, the dynamic entropy value SND of the signal-to-noise ratio of the subcarrier is calculated through the entropy calculation model t； The communication quality data also includes the theoretical beam weight matrix extracted from the design document or configuration file of the beamforming algorithm; and the actual channel response matrix calculated by the channel estimation algorithm by performing channel measurement between the IoT device and the base station, sending pilot signals and receiving feedback channel state information, as the actual beam weight matrix; the beamforming distortion degree BFD is obtained by inputting the theoretical beam weight matrix and the actual beam weight matrix into the calculation model of the beamforming distortion degree. t .

3. The AI-based IoT multi-source data anomaly perception and analysis method according to claim 1 is characterized in that: The perception parameter data includes using the millimeter wave radar signal processing module to collect a large amount of historical normal data, and through Fourier transform, converting the time domain signal into the frequency domain signal to extract the Doppler velocity distribution characteristics, generating the Doppler velocity distribution reference spectrum, and real-time acquisition of the radar echo signal to obtain the real-time Doppler velocity distribution function; by inputting the Doppler velocity distribution reference spectrum generated by historical normal data training and the real-time observed Doppler velocity distribution function into the Doppler profile similarity calculation model, the Doppler profile similarity DPS is calculated. t ; The perception parameter data also includes the point cloud data of the target scene collected by the three-dimensional sensor, and the collected point cloud data is pre-processed; the three-dimensional space is divided into multiple spatial bins of equal size, and the number of three-dimensional point clouds in each spatial bin is counted by calculating the index of the point cloud coordinates in the bin coordinate system, and then the probability density of the three-dimensional point cloud in each spatial bin is calculated; the probability density in the spatial bin is input into the point cloud structure entropy calculation model to obtain the point cloud structure entropy PSE t .

4. The AI-based IoT multi-source data anomaly perception and analysis method according to claim 1 is characterized in that: By analyzing the device status data, the hardware anomaly index is obtained based on the following formula: ； Among them, HAL t Indicates the hardware abnormality index; T J Indicates the junction temperature of the device chip; T th Indicates the temperature threshold of the chip; T mac Indicates the maximum temperature allowed by the device chip under normal working conditions; M u Indicates memory usage; M to represents the total memory capacity; ɑ represents the weight coefficient of the temperature state; β represents the weight coefficient of the memory state, and ɑ+β=1.

5. The AI-based IoT multi-source data anomaly perception and analysis method according to claim 1 is characterized in that: By analyzing the environmental interference data, the multipath interference factor is obtained based on the following formula: ； Among them, MIF t Represents the multipath interference factor; SY l Indicates the l The time delay of the multipath component; h l Indicates the l The channel coefficient of the multipath; L represents the number of multipaths.

6. The AI-based IoT multi-source data anomaly perception and analysis method according to claim 5 is characterized in that: The comprehensive evaluation threshold is calculated based on the comprehensive abnormal probability of the historical time period, and the formula is as follows: ； Among them, TH t represents the comprehensive evaluation threshold; μ t-1 Represents the moving average of the comprehensive abnormal probability in the previous time period; τ t-1 It represents the standard deviation of the comprehensive abnormal probability in the previous time period, ρ is a constant, and τ t-1 Multiplication indicates multiples of the standard deviation; Based on the comprehensive evaluation threshold, the comprehensive abnormal probability is analyzed and evaluated, and the method for determining whether to start the diagnostic instruction is as follows: t >TH t When the diagnostic process is started, t ≤TH t The diagnostic process is not started.

7. The AI-based IoT multi-source data anomaly perception and analysis method according to claim 6 is characterized in that: By analyzing the comprehensive abnormal probability, the posterior probability of different fault causes is obtained, based on the following formula: ； Among them, P (X i |Pan t ) represents the observed comprehensive abnormal probability Pan t In the case of failure reason X i The posterior probability of occurrence; P (Pan t-z |X i ) indicates that the fault reason is X i In the event of a fault, the historical comprehensive abnormal probability Pan observed in the historical fault log t-z The probability of P(X i ) indicates the cause of the fault X i Prior probability of occurrence; i The serial number representing the fault cause, which is a positive integer.

8. The AI-based IoT multi-source data anomaly perception and analysis method according to claim 7 is characterized in that: The method to determine the fault type is: traverse P (X i |Pan t ) under different fault causes Xi, P (X i |Pan t ) are prioritized from large to small. At this time, P(X i |Pan t ) The fault cause X corresponding to the maximum value i The first priority fault cause is used as the fault type for this judgment. The priorities of the other fault causes are based on P (X i |Pan t ) values, decreasing from large to small; If the first priority fault cause is eliminated, the remaining fault causes will be checked in order of priority.

Images