Intelligent lock information monitoring method and system

Abstract

The invention relates to the technical field of intelligent lock information monitoring, in particular to an intelligent lock information monitoring method and system. The monitoring method comprises the following steps: acquiring state information of the intelligent lock by using various sensors; preprocessing the state information, wherein the preprocessing comprises data filtering, de-noising processing and time synchronization; key feature data of the state information are extracted and analyzed through an analysis algorithm; and analyzing and judging the working state and the fault condition of the intelligent lock by utilizing a multivariable analysis method. According to the intelligent lock information monitoring method provided by the invention, the state information of the intelligent lock is comprehensively collected through various sensors, and the monitoring of the working state of the intelligent lock and the early warning of faults are realized by combining data processing and analysis algorithms, so that the comprehensiveness and accuracy of monitoring are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of smart lock information monitoring, and in particular to a smart lock information monitoring method and system. Background Art

[0002] As part of the smart home ecosystem, the security and reliability of smart locks are directly related to the property safety of users. Traditional smart lock monitoring methods usually only focus on the basic switch status and battery power, which cannot fully reflect the working status and potential risks of smart locks. With the advancement of technology, smart locks have integrated more and more functions and sensors, such as fingerprint recognition, NFC communication, cameras, etc. Although these new functions have improved convenience, they have also increased the complexity of the system and potential failure points. Therefore, the problem faced by smart locks is that the existing monitoring methods often use simple threshold judgments or rule matching, which is difficult to cope with complex and changing usage scenarios and potential security threats. Summary of the invention

[0003] The present invention provides a smart lock information monitoring method and system to solve the problem of insufficient monitoring capability of existing smart locks.

[0004] The technical solution of the present invention to solve the above technical problems is as follows:

[0005] A method for monitoring smart lock information is provided, the steps of the monitoring method comprising:

[0006] Use multiple sensors to collect status information of smart locks, including battery power, lock switch status, door magnet status, abnormal vibration and positioning information;

[0007] Preprocess the status information, including data filtering, denoising and time synchronization;

[0008] Extract and analyze key feature data of status information through analysis algorithms;

[0009] Use multivariate analysis methods to analyze and determine the working status and fault conditions of smart locks;

[0010] Generate early warning information based on the analysis results and send it to the user terminal.

[0011] On the other hand, a smart lock information monitoring system is provided, the monitoring system comprising:

[0012] Data acquisition module, used to obtain various status information of smart locks;

[0013] The data preprocessing module is connected to the data acquisition module and is used to filter, remove noise and synchronize the state information;

[0014] A feature extraction module, connected to the data preprocessing module, is used to extract key features from the preprocessed data;

[0015] The state analysis module is connected to the feature extraction module to analyze the working state and fault conditions of the smart lock;

[0016] An alarm module, connected to a status analysis module, is used to generate and send warning information when a fault condition is detected;

[0017] The communication module is connected to the alarm module and is used to send the warning information to the user terminal;

[0018] The feature extraction and state analysis module processes the state information according to the following formula:

[0019] Among them, S a is the state analysis result, β0 is a constant term, β j is the regression coefficient of the jth state information, x j is the jth state information, ε is the error term, and k is the number of state information types.

[0020] The beneficial effects of the present invention are:

[0021] The smart lock information monitoring method provided by the present invention comprehensively collects the status information of the smart lock through multiple sensors, and combines data processing and analysis algorithms to realize the monitoring of the working status of the smart lock and early warning of faults, thereby improving the comprehensiveness and accuracy of monitoring. Compared with the traditional simple threshold judgment, the present invention has higher accuracy and prediction ability, can detect potential problems early, and reduce security risks. It provides technical support for the overall security management of smart home systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flow chart of a smart lock information monitoring method in one embodiment of the present invention;

[0023] Figure 2 This is a flow chart of state information processing in one embodiment of the present invention;

[0024] Figure 3 This is a flowchart of key feature data extraction in one embodiment of the present invention;

[0025] Figure 4 This is a schematic diagram of the key feature data extraction principle in one embodiment of the present invention;

[0026] Figure 5 A flowchart for determining a working fault state of a smart lock in one embodiment of the present invention;

[0027] Figure 6 This is a flowchart of automatic repair of a smart lock in one embodiment of the present invention;

[0028] Figure 7 A flowchart of security threat detection in one embodiment of the present invention;

[0029] Figure 8 A flowchart of smart lock performance optimization in one embodiment of the present invention;

[0030] Fig. 9 A schematic diagram of the performance optimization of a smart lock in one embodiment of the present invention;

[0031] Fig.10 This is a flow chart of multi-device intercommunication of smart locks in one embodiment of the present invention;

[0032] Fig.11 A schematic diagram of the intercommunication of multiple devices of a smart lock in one embodiment of the present invention;

[0033] Fig.12 This is an architecture diagram of a smart lock information monitoring system in one embodiment of the present invention;

[0034] In the accompanying drawings, the components represented by the reference numerals are listed as follows:

[0035] 100, data acquisition module; 200, data preprocessing module; 300, feature extraction module; 400, state analysis module; 500, alarm module; 600, communication module. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0037] The present invention provides the following preferred embodiments:

[0038] Embodiment 1

[0039] In order to solve the problems of existing smart lock monitoring methods such as single data collection, limited processing capacity, simple analysis method and lagging early warning mechanism, this embodiment proposes a specific implementation process of a smart lock information monitoring method.

[0040] like Figure 1 As shown, the steps of the monitoring method include:

[0041] S101. Use multiple sensors to collect status information of the smart lock, including battery power, lock switch status, door magnetic status, abnormal vibration and positioning information.

[0042] S102: pre-process the status information, including data filtering, denoising and time synchronization.

[0043] S103. Extract and analyze key feature data of the status information through an analysis algorithm.

[0044] S104: Analyze and determine the working status and fault conditions of the smart lock using a multivariate analysis method.

[0045] S105. Generate warning information according to the analysis results and send it to the user terminal.

[0046] The smart lock information monitoring method of this embodiment first uses a variety of sensors to collect the status information of the smart lock. These sensors include battery power sensors, lock body position sensors, door magnetic sensors, acceleration sensors and GPS modules. The battery power sensor uses a high-precision voltage detection chip, which can monitor the battery voltage changes in real time, and accurately reflect the remaining battery power in combination with the preset voltage threshold. The lock body position sensor uses a Hall sensor to determine the extension and retraction state of the lock tongue by detecting changes in the magnetic field, thereby determining the switch state of the lock. The door magnetic sensor uses a reed switch, which is installed between the door frame and the door leaf to detect the opening and closing state of the door. Abnormal vibration detection uses a three-axis acceleration sensor to capture vibrations in all directions to the lock body. The positioning information is obtained through the integrated GPS module, and when the indoor signal is weak, it can be switched to a Wi-Fi-based indoor positioning solution.

[0047] Furthermore, the collected state information is preprocessed to improve the accuracy of subsequent analysis. The Kalman filter is used for data filtering, which can effectively remove random noise in the sensor signal and improve the signal-to-noise ratio. The wavelet transform method is used for denoising, which can effectively suppress noise interference while retaining the important features of the signal through multi-scale analysis. The network time protocol (NTP) is used for time synchronization to ensure that the timestamps of all sensor data are accurately aligned, laying the foundation for subsequent multi-dimensional data fusion analysis.

[0048] It should be understood that the key feature data of the state information is extracted and analyzed through the analysis algorithm. This embodiment uses the principal component analysis (PCA) method for feature extraction, which can effectively reduce the data dimension and remove redundant information. For time series data, such as vibration signals, the fast Fourier transform (FFT) is used to extract frequency domain features, which can identify the frequency characteristics of abnormal vibrations. In addition, the convolutional neural network (CNN) model in deep learning is also introduced to extract deep features through automatic learning, thereby improving the expressiveness and generalization of features.

[0049] It is understandable that the use of multivariate analysis methods to analyze and judge the working status and fault conditions of smart locks requires comprehensive consideration of multiple factors. This embodiment uses the support vector machine (SVM) algorithm to build a classification model, maps the extracted features to a high-dimensional space, and distinguishes between normal and abnormal states by finding the optimal separation hyperplane. For fault diagnosis, the random forest algorithm is used to improve the accuracy of classification by constructing multiple decision trees and taking majority votes. In order to achieve fault prediction, the long short-term memory network (LSTM) is introduced, which can effectively capture long-term dependencies and predict possible future fault trends of smart locks.

[0050] Furthermore, generating early warning information based on the analysis results and sending it to the user terminal is to ensure that the user is aware of the status of the smart lock in a timely manner. This embodiment designs a multi-level early warning mechanism, which divides the warning into three levels: low, medium, and high according to the severity and urgency of the fault. The generation of early warning information adopts a template design, which includes the type of fault, possible causes, scope of impact, and recommended operations. Information sending adopts a multi-channel strategy, including mobile phone APP push, text messages, and emails, to ensure that users can receive important notifications in a timely manner. In order to improve the security of information transmission, the AES encryption algorithm is used to encrypt the early warning information.

[0051] The benefit of this embodiment is that the comprehensiveness and accuracy of smart lock status monitoring are improved through multi-sensor data fusion and advanced signal processing technology. The use of machine learning and deep learning algorithms for feature extraction and status analysis can identify complex working modes and potential faults. The multi-level early warning mechanism and secure information transmission ensure that users can understand the working status of smart locks in a timely and accurate manner. Through the smart lock information monitoring method of this embodiment, the security and reliability of smart locks can be improved, providing assistance for the overall security management of smart home systems.

[0052] Embodiment 2

[0053] In order to solve the problem of single sensor and incomplete data collection in traditional smart lock monitoring methods, this embodiment further optimizes the implementation method of collecting smart lock status information using multiple sensors.

[0054] Specifically, the steps of collecting the status information of the smart lock using multiple sensors in this embodiment include:

[0055] Accelerometers and gyroscopes are used to collect the motion status and posture information of smart locks.

[0056] The battery management system collects information on battery power, charge and discharge status, and battery health.

[0057] Use door magnetic sensors and micro switches to collect the switch status of the lock and the open and closed status of the door.

[0058] Vibration sensors and sound sensors are used to detect abnormal vibration and sound information.

[0059] The ambient temperature and humidity data are collected through the temperature and humidity sensors.

[0060] Use GPS or indoor positioning module to obtain the location information of the smart lock.

[0061] This embodiment uses an acceleration sensor and a gyroscope to collect the motion state and posture information of the smart lock. The acceleration sensor uses a three-axis MEMS accelerometer, model LIS3DH, with a range of ±16g and a sampling rate of up to 5kHz, which can accurately capture the acceleration changes of the smart lock in three directions. The gyroscope uses the MPU-6050, which integrates a three-axis gyroscope and a three-axis accelerometer, and can provide 6 degrees of freedom (6DOF) motion tracking. By fusing acceleration and angular velocity data and using a complementary filtering algorithm, the attitude angle of the smart lock, including pitch angle, roll angle, and yaw angle, can be calculated in real time, thereby accurately reflecting the spatial attitude changes of the lock body.

[0062] Furthermore, the battery management system (BMS) is responsible for collecting battery-related information. This embodiment uses the BQ27Z561 battery management chip, which integrates a high-precision coulomb counter and voltage monitoring function. By communicating with the main controller through the I2C interface, parameters such as battery power, charge and discharge current, and battery temperature can be obtained in real time. The battery health (SOH) is evaluated by monitoring the change in battery internal resistance and capacity attenuation. The SOH estimation algorithm based on Kalman filtering is used to improve the accuracy of battery health assessment.

[0063] It should be understood that the combined use of door magnetic sensors and micro switches can improve the reliability of switch state detection. The door magnetic sensor uses the Hall effect sensor AH3362Q, which has low power consumption and high sensitivity. The micro switch uses the Omron D2F-L3-D type, which has high durability and stable triggering characteristics. The two are connected through a logic circuit, using the "AND" logic relationship. Only when the two sensors are triggered at the same time can they be judged as unlocked, which effectively reduces the misjudgment rate.

[0064] It is understandable that the coordinated work of vibration sensors and sound sensors can comprehensively monitor abnormal situations. The vibration sensor uses the piezoelectric accelerometer MiniSense 100, with a sensitivity of 1V / g and a frequency response range of 0-40kHz. The sound sensor uses the MEMS microphone SPH0645LM4H-B, which has low noise and wide-band response characteristics. The data of the two sensors are analyzed in real time by a digital signal processor (DSP), and the short-time Fourier transform (STFT) is used to extract the time-frequency features, which are matched with the preset abnormal pattern library to identify abnormal behaviors such as knocking and drilling.

[0065] Furthermore, the monitoring of ambient temperature and humidity affects the long-term stable operation of the smart lock. This embodiment uses Sensirion's SHT31 temperature and humidity sensor with a temperature measurement accuracy of ±0.2°C and a humidity measurement accuracy of ±2%RH. The sensor uses an I2C interface to communicate with the main controller and supports periodic sampling and alarm threshold setting. By monitoring changes in ambient temperature and humidity, lock failures caused by extreme environments can be prevented, such as battery performance degradation caused by low temperature or circuit board corrosion caused by high humidity.

[0066] This embodiment adopts a dual-mode positioning solution, combining GPS and indoor positioning technology. Outdoor positioning uses ublox's NEO-M8N GPS module, which supports three satellite navigation systems: GPS, GLONASS and Beidou, with a positioning accuracy of up to 2.5m CEP. Indoor positioning uses AOA (angle of arrival) technology based on Bluetooth 5.1, measures the signal arrival angle through multiple Bluetooth base stations, and combines the principle of triangulation to achieve centimeter-level positioning accuracy. Positioning data is uploaded to the cloud server in real time through the MQTT protocol, supports geographic fence function, and can trigger an alarm when the smart lock moves out of the preset area.

[0067] The benefit of this embodiment is that through the collaborative work of multiple high-precision sensors, all-round and multi-dimensional monitoring of the smart lock status is achieved. The combined use of acceleration sensors and gyroscopes improves the accuracy of motion status and posture detection. The battery management system not only monitors the power level, but also evaluates the battery health status, which helps prevent battery-related failures. The dual verification mechanism of the door magnetic sensor and the micro switch improves the reliability of switch status detection. The collaborative analysis of vibration and sound sensors enhances the ability to identify abnormal behavior. Ambient temperature and humidity monitoring provides a guarantee for the long-term reliable operation of smart locks. The dual-mode positioning solution ensures that the position of the smart lock can be accurately located in various scenarios. Through this multi-sensor fusion method, the comprehensiveness of smart lock status monitoring is improved, providing a rich and reliable data source for subsequent data analysis and decision-making.

[0068] Embodiment 3

[0069] In order to solve the problems of high-frequency noise, outliers, multi-source data asynchrony, and inconsistent data scales in the smart lock status information, this embodiment further optimizes the specific implementation method of status information preprocessing.

[0070] like Figure 2 As shown, the steps of preprocessing the status information include:

[0071] S301 , performing multi-scale filtering on the original data of the state information using a transformation algorithm to remove high-frequency noise and outliers.

[0072] S302: Time synchronization and alignment of multi-source data are performed to ensure consistency of data from different sensors.

[0073] S303: Perform data normalization processing to unify the scale ranges of different types of data.

[0074] S304: interpolate or estimate the missing data to ensure the integrity of the data.

[0075] Among them, the calculation formula of the transformation algorithm is:

[0076] Among them, W ψ (a,b) are the transformed coefficients, x(t) is the time series signal, ψ is the basis function, a is the scale parameter, and b is the time shift parameter.

[0077] This embodiment uses continuous wavelet transform (CWT) as the main transformation algorithm to perform multi-scale filtering on the original data of the state information. The basis function ψ in the calculation formula of the transformation algorithm uses Morlet wavelet, which has good time-frequency localization characteristics and is suitable for analyzing non-stationary signals. The selection range of scale parameter a is 2 j (j=1,2,…,J), J is the maximum number of decomposition levels, usually 5-8. The step size of the time shift parameter b is set to 1 / 4 of the sampling period to ensure sufficient time resolution. By adjusting the scale parameter a, the separation and extraction of different frequency components can be achieved. For the acceleration signal, a high-pass filter is used to remove low-frequency drift, and the filter cutoff frequency is set to 0.5Hz. For the voltage signal, a low-pass filter is used to suppress high-frequency noise, and the cutoff frequency is set to 100Hz.

[0078] Furthermore, time synchronization and alignment of multi-source data is performed to ensure data consistency. This embodiment uses the Network Time Protocol (NTP) to synchronize the clocks of each sensor and control the time accuracy to the millisecond level. For sensors with different sampling rates, interpolation and downsampling methods are used to unify the sampling frequency. Specifically, for low sampling rate data, such as temperature and humidity, cubic spline interpolation is used for upsampling; for high sampling rate data, such as acceleration, anti-aliasing filters are used for downsampling. Time alignment uses a sliding window method with a window size of 1 second and a step size of 0.1 second. The best alignment position is found by calculating the cross-correlation function.

[0079] It should be understood that data normalization plays an important role in unifying the scale range of different types of data. This embodiment adopts the Z-score normalization method, and the calculation formula is:

[0080] Among them, X is the original data, μ is the mean, and σ is the standard deviation. For non-normally distributed data, such as vibration amplitude, logarithmic transformation is used and then Z-score standardization is performed. In order to deal with outliers, the box plot method is used to identify outliers, and data points exceeding 1.5 times the interquartile range are regarded as outliers and replaced by the median.

[0081] It is understandable that in order to ensure data integrity, missing data needs to be processed. This embodiment adopts different processing strategies according to the type and proportion of missing data. For data that is randomly missing and the proportion is less than 5%, multiple imputation is used to generate multiple sets of possible filling values, and the average is taken as the final result. For continuously missing data segments, such as data missing caused by temporary sensor failure, a sequence prediction model based on a long short-term memory network (LSTM) is used for estimation. The input of the LSTM model is the data of 30 time steps before the missing data, and the output is the predicted missing value. The model structure includes two LSTM layers (128 units per layer) and a fully connected layer.

[0082] Furthermore, in order to improve the efficiency and real-time performance of preprocessing, this embodiment adopts a stream processing architecture. The data preprocessing module is designed as a series of pipelined processing units, each of which is responsible for a specific preprocessing task. Apache Flink is used as the stream processing engine, which supports millisecond-level data processing delays. In order to handle large amounts of sudden data, such as data surges caused by frequent lock switching, an adaptive batch processing mechanism is introduced to dynamically adjust the batch size to strike a balance between processing delay and throughput.

[0083] Furthermore, this embodiment also considers the adaptive adjustment of preprocessing parameters. An online learning algorithm based on a sliding window is used to regularly evaluate the preprocessing effect and automatically adjust the filter parameters, normalization range, etc. Evaluation indicators include signal-to-noise ratio, skewness and kurtosis of data distribution, etc. Through this adaptive mechanism, the system can adapt to different usage environments and modes and maintain the stability of the preprocessing effect.

[0084] The benefit of this embodiment is that by adopting signal processing technology and machine learning algorithms, high-quality preprocessing of smart lock status information is achieved. Continuous wavelet transform provides multi-scale signal analysis capabilities and effectively removes noise of various frequencies. Accurate time synchronization and alignment ensure the consistency of multi-source data, laying the foundation for subsequent data fusion analysis. Data normalization and missing value processing improve the quality and integrity of the data. The streaming processing architecture and adaptive parameter adjustment mechanism ensure the real-time and adaptability of the preprocessing. Through this comprehensive and sophisticated preprocessing method, the quality of the status information is significantly improved.

[0085] Embodiment 4

[0086] In order to solve the problems of incomplete features, excessive redundant information, insufficient expression ability, etc. in the process of extracting features of smart lock status information, this embodiment further optimizes the extraction and analysis method of key feature data.

[0087] like Figure 3 , Figure 4 As shown, the steps of extracting and analyzing key characteristic data of the state information through the analysis algorithm include:

[0088] S401, performing time domain analysis on the pre-processed state information to extract time domain features such as maximum value, minimum value, mean value, peak value, etc.

[0089] S402, using Fast Fourier Transform (FFT) to perform frequency domain conversion on the data, and extract the main frequency and frequency distribution.

[0090] S403. Calculate statistical characteristics of the state information, including mean, variance, skewness, kurtosis, and correlation coefficient.

[0091] S404. Use the principal component analysis (PCA) method to reduce feature dimensions and remove redundant information.

[0092] S405: Based on the feature importance evaluation algorithm, select the most representative feature set.

[0093] S406. Use convolutional neural network (CNN) to extract deep features and enhance feature expression capabilities.

[0094] The calculation formula of the principal component analysis (PCA) method is:

[0095] Z=XW, where Z is the data matrix after dimensionality reduction, X is the original data matrix, and W is the eigenvector matrix.

[0096] Specifically, this embodiment first performs time domain analysis on the preprocessed state information. For continuous data, such as acceleration signals, the sliding window method is used for segmented processing, the window size is set to 1 second, and the overlap rate is 50%. In each window, the time domain features such as maximum value, minimum value, mean value, peak value, root mean square value, waveform factor and peak factor are calculated. For discrete data, such as switch status, its change frequency and duration are counted. In addition, the time series decomposition technology is introduced, and the seasonal and trend decomposition (STL) method is used to decompose the time series into three components: trend, seasonality and residual, and extract features respectively.

[0097] Furthermore, the data is transformed into the frequency domain using the fast Fourier transform (FFT). The FFT is implemented using the FFTW library, which can adaptively select the optimal algorithm to improve computational efficiency. In order to improve the frequency resolution, the zero padding technique is used to extend the signal length to the power of 2. In the frequency domain, features such as the main frequency, frequency center, frequency variance, and energy concentration are extracted. For non-stationary signals, such as vibration data, the short-time Fourier transform (STFT) is used, and the window function uses the Hanning window, the window size is 256 points, and the overlap rate is 75% to obtain the time-frequency joint distribution information.

[0098] It should be understood that the statistical characteristics of the calculated state information can capture the overall distribution characteristics of the data. In addition to the basic mean and variance, this embodiment also calculates higher-order moments, including skewness (a measure of the asymmetry of the distribution) and kurtosis (a measure of the sharpness of the distribution). For multidimensional data, such as the three components of acceleration, the correlation coefficient matrix between them is calculated to capture the correlation between different dimensions. In order to deal with possible nonlinear relationships, mutual information is introduced as a nonlinear correlation measure.

[0099] It is understandable that the principal component analysis (PCA) method is used to reduce the feature dimension and remove redundant information. The column vector of the eigenvector matrix W in the calculation formula of PCA is the eigenvector of the covariance matrix of the original data, which is arranged in descending order according to the corresponding eigenvalue size. The present embodiment uses an iterative method to calculate eigenvalues ​​and eigenvectors, specifically using the QR algorithm, which performs well in numerical stability. The number of principal components is selected based on the cumulative explained variance ratio, and the principal components that explain 90% of the variance are usually retained. In order to process nonlinear data, the present embodiment also introduces kernel principal component analysis (KPCA), using the radial basis function (RBF) kernel, and the kernel parameters are optimized by grid search.

[0100] Further, the most representative feature set is selected based on the feature importance evaluation algorithm. In this embodiment, the random forest algorithm is used to calculate the feature importance score, and 100 decision trees are constructed, and the maximum depth of each tree is set to 10. The feature importance is calculated based on the reduction of the Gini impurity. In order to improve the stability of feature selection, the stability selection method is adopted to select the feature subset with the highest frequency through multiple random subsampling and feature selection.

[0101] Furthermore, this embodiment uses a convolutional neural network (CNN) to extract deep features to enhance the expressive power of features. The CNN model adopts a 1D convolution structure, including 3 convolution layers and 2 pooling layers. The first convolution layer uses 32 3x1 convolution kernels, the second layer uses 64 3x1 convolution kernels, and the third layer uses 128 3x1 convolution kernels. Each convolution layer is followed by a ReLU activation function and batch normalization. The pooling layer uses maximum pooling with a pooling size of 2. The input of the network is the preprocessed original signal segment, and the output is a 256-dimensional feature vector. The model training uses the Adam optimizer, the learning rate is 0.001, the batch size is 64, and the number of training rounds is 100 rounds. In order to prevent overfitting, L2 regularization and Dropout technology are used, and the Dropout rate is set to 0.5.

[0102] Furthermore, in order to adapt to different types of data and tasks, this embodiment can also introduce an adaptive feature extraction mechanism. Through the meta-learning method, the system can automatically select the most appropriate feature extraction algorithm and parameters according to the characteristics of the input data. The specific implementation adopts the neural architecture search (NAS) technology based on reinforcement learning, and the search space includes different feature extraction algorithms, parameter settings and combinations. The reward function is designed as a performance indicator of downstream tasks (such as anomaly detection).

[0103] The benefit of this embodiment is that through a multi-dimensional and multi-scale feature extraction method, various characteristics of the smart lock state information are fully captured. Time domain and frequency domain analysis provide the basic characteristics of the signal, and the statistical characteristics reflect the overall distribution characteristics of the data. PCA and feature selection algorithms effectively reduce feature dimensions and remove redundant information. The introduction of CNN enhances the expressiveness of features and can capture complex nonlinear patterns. The adaptive feature extraction mechanism further improves the generalization ability of smart locks. Through this comprehensive feature extraction and analysis method, the characterization quality of the state information is improved, providing rich and effective feature inputs for state monitoring, anomaly detection and fault diagnosis of smart locks.

[0104] Embodiment 5

[0105] In order to solve the problems of insufficient accuracy, difficulty in identifying fault types and limited prediction ability in judging the working status and fault conditions of smart locks, this embodiment further optimizes the specific implementation method of the multivariate analysis method.

[0106] like Figure 5 As shown, the steps of analyzing and judging the working state and fault condition of the smart lock by using the multivariate analysis method include:

[0107] S501. Establish a normal working state model of the smart lock, including the normal range and change trend of various parameters.

[0108] S502: Calculate the deviation between the current state and the normal state model.

[0109] S503: Use a support vector machine (SVM) algorithm to classify the deviations and distinguish between normal and abnormal states.

[0110] S504: Identify specific fault types and causes based on a decision tree algorithm.

[0111] S505. Use the Bayesian network to calculate the probability and severity of the failure and assess the potential risk.

[0112] S506. Use the long short-term memory network (LSTM) to predict the fault development trend and realize fault early warning.

[0113] The decision function of the support vector machine algorithm is expressed as:

[0114] Among them, α i is the Lagrange multiplier, y i is the category label, K(xi,x) is the kernel function, and b is the bias term.

[0115] Specifically, this embodiment first establishes a normal working state model of the smart lock. The model uses a Gaussian mixture model (GMM) to represent the normal range and change trend of various parameters. The number of components of the GMM is determined by the Bayesian Information Criterion (BIC), and 3-5 components are usually selected. The model parameters are estimated by the expectation maximization (EM) algorithm, and the number of iterations is set to 100 or until convergence. In order to capture the correlation between parameters, the constraints of the covariance matrix are introduced, and a diagonal covariance matrix is ​​used to balance the model complexity and expressiveness. The model training data comes from the long-term normal state records, which are segmented by the sliding window method, with a window size of 24 hours and a step size of 1 hour.

[0116] Further, the deviation between the current state and the normal state model is calculated. This embodiment uses the Mahalanobis distance as the deviation metric, which takes into account the covariance structure between features. For each observation point, the Mahalanobis distance between it and the nearest GMM component center is calculated. In order to process multidimensional data, multivariate statistical process control (MSPC) technology is used, combined with Hotelling's T2 statistic and squared prediction error (SPE) statistic, to construct a composite index to comprehensively evaluate the degree of deviation.

[0117] It is important to understand that the support vector machine (SVM) algorithm is used to classify deviations and distinguish between normal and abnormal states. The kernel function K(xi,x) in the decision function of the support vector machine (SVM) uses the radial basis function (RBF), which is expressed as exp(-δ||xi-x|| 2 ). The kernel parameter δ and penalty parameter C are optimized by grid search and cross validation, and the search range is [2 -15 ,2 3 ] and [2 -5 ,2 15 ]. In order to deal with the class imbalance problem, weighted SVM is used, and the weight of minority class samples is set to the inverse of the majority class. The model training adopts the sequential minimal optimization (SMO) algorithm, and the convergence threshold is set to 0.001.

[0118] It is understandable that identifying specific fault types and causes based on decision tree algorithms can provide explainable diagnostic results. This embodiment uses a random forest algorithm to construct 100 decision trees, and the maximum depth of each tree is set to 15. The Gini index is used for feature selection, and the minimum number of leaf node samples is set to 5. In order to improve the generalization ability of the model, the self-service aggregation (Bagging) technology is introduced, and each tree is trained using 63.2% of randomly sampled data. In the process of generating the decision tree, the post-pruning technology is used, and the pruning standard is based on the error rate of the validation set.

[0119] Furthermore, the Bayesian network is used to calculate the probability and severity of the failure and evaluate the potential risk. The structure of the Bayesian network is determined by combining expert knowledge and data-driven methods. The network learning adopts the structural expectation maximization (SEM) algorithm, and the number of iterations is set to 100. The maximum likelihood estimation (MLE) method is used for parameter learning. In order to handle continuous variables, a Gaussian Bayesian network is used, assuming that each node obeys a Gaussian distribution. Risk assessment is based on the joint distribution of failure probability and severity, and Monte Carlo simulation is used to generate a risk distribution map.

[0120] Furthermore, this embodiment applies a long short-term memory network (LSTM) to predict the development trend of faults for fault warning. The LSTM model includes two LSTM layers, each with 128 units, followed by a fully connected layer. The input sequence length is 48 (corresponding to 48 hours of historical data), and the output is the predicted value for the next 24 hours. The model is trained using the Adam optimizer, with a learning rate of 0.001 and a batch size of 32. In order to improve the stability of the model, a Dropout layer is introduced, and the Dropout rate is set to 0.3. The loss function uses the mean square logarithmic error (MSLE), which can better handle prediction errors of different magnitudes. The warning threshold is set based on the confidence interval of the predicted value, and the uncertainty of the prediction is estimated by the Monte Carlo Dropout method.

[0121] The benefit of this embodiment is that through multi-model fusion and adaptive learning mechanism, comprehensive and accurate judgment of the working state of the smart lock and early warning of faults are achieved. The Gaussian mixture model provides a flexible normal state representation, and the support vector machine achieves high-precision anomaly detection. The random forest algorithm provides interpretable fault diagnosis results, and the Bayesian network quantifies the fault risk. The introduction of the LSTM network enables the system to have the ability to predict the development trend of faults. Through a comprehensive multivariate analysis method, the accuracy, interpretability and predictability of smart lock status monitoring and fault diagnosis are improved, providing decision support for the maintenance and management of smart locks.

[0122] Embodiment 6

[0123] In order to solve the problems of low repair efficiency, high operational risks and poor user experience in the process of smart lock fault repair, this embodiment further optimizes the specific implementation method of smart lock automatic repair.

[0124] like Figure 6 As shown, the monitoring method also includes automatic repair of the smart lock, and the steps of automatic repair of the smart lock include:

[0125] S601. Identify the fault type that can be repaired by software, including system abnormality, configuration error, or software bug.

[0126] S602: Query the fault repair database and select the best repair strategy.

[0127] S603: Generate a repair instruction set including specific operation steps.

[0128] S604: Perform compatibility check on the repair instruction set to prevent the introduction of new faults.

[0129] S605. Remotely execute repair instructions, including software restart, configuration reset, or firmware update.

[0130] S606: Verify the repair result to confirm whether the fault is resolved.

[0131] S607: Feedback the repair status and results to the user and provide operation suggestions.

[0132] Specifically, this embodiment first identifies the fault types that can be repaired by software through a deep learning model. The model adopts a hybrid structure of a convolutional neural network (CNN) and a long short-term memory network (LSTM), which can process time series data and static features at the same time.

[0133] Furthermore, the fault repair database is queried to select the best repair strategy. The database is implemented using the graph database Neo4j, which can efficiently represent and query complex fault-repair relationships. Each node in the database represents a fault type or repair strategy, the edges between nodes represent their associations, and the weight of the edge reflects the success rate of the repair strategy.

[0134] It should be understood that generating a repair instruction set containing specific operating steps is a key link in automatic repair. This embodiment uses a reasoning system based on a knowledge graph to generate repair instructions. The knowledge graph contains entities and relationships such as device components, operating instructions, and dependencies. The reasoning process uses a forward chaining algorithm to gradually derive the repair steps based on the fault symptoms. It can be understood that compatibility checking of the repair instruction set is to prevent the introduction of new faults. This embodiment constructs a rule-based expert system, and the rule base contains compatibility rules for information such as device model, software version, and hardware configuration. The inspection process uses the Rete algorithm, which can efficiently match multiple rules. For complex compatibility problems, a solution method based on constraint satisfaction problems is introduced, and a backtracking search algorithm is used to find a solution that satisfies all compatibility constraints.

[0135] Furthermore, the process of remotely executing repair instructions uses a secure remote procedure call (RPC) mechanism. The communication protocol uses gRPC, which supports two-way streaming and automatic reconnection. In order to ensure the reliability of execution, a transaction management mechanism is introduced, and each repair instruction is treated as an atomic operation, which supports rollback operations. During the execution process, real-time logging and state synchronization are used to ensure the traceability of operations. For repair operations that require a long time to execute, such as firmware updates, breakpoint resume technology is used to support resumed transmission after interruption.

[0136] This embodiment verifies the repair results through multi-dimensional indicators. First, the statistical process control (SPC) method is used to monitor the changing trend of key performance indicators (KPIs). Secondly, anomaly detection algorithms, such as Isolation Forest, are applied to detect whether there are abnormalities in the system behavior after the repair. Thirdly, the transfer learning technology is used to migrate the pre-trained normal state model to the repaired system to evaluate the similarity of the system state. Finally, the system performance before and after the repair is compared through A / B testing.

[0137] The benefit of this embodiment is that it realizes efficient and automatic repair of smart lock faults through intelligent fault identification, accurate repair strategy selection, secure remote execution mechanism and comprehensive result verification. Compatibility check and secure execution mechanism reduce the risk of the repair process. Multi-dimensional result verification ensures the effectiveness of the repair.

[0138] Embodiment 7

[0139] In order to solve the problems of high false alarm rate, delayed detection and untimely response in security threat detection of smart locks, this embodiment further optimizes the specific implementation method of security threat detection.

[0140] like Figure 7 As shown, the monitoring method also includes security threat detection, and the steps of security threat detection include:

[0141] S701. Real-time monitoring of abnormal unlocking attempts, including frequent failed password inputs or unauthorized NFC card use or unauthorized biometric features.

[0142] S702: Analyze unlocking pattern changes using machine learning algorithms to identify potential fraud or counterfeiting behaviors.

[0143] S703. Detect suspicious physical interference, such as lock picking or violent destruction, through sensor data fusion technology.

[0144] S704. Continuously evaluate the security of network connections and detect potential risks of network attacks or data hijacking.

[0145] S705. Trigger an emergency security response when a high-risk threat is detected, including locking the device, sending an alarm, and notifying security personnel.

[0146] S706. Record detailed security event logs to support subsequent forensic analysis and security policy optimization.

[0147] The anomaly score evaluation formula for real-time monitoring of abnormal unlocking attempts is:

[0148] Among them, θ i is the observed value, μ i is the mean of normal behavior, σ i is the standard deviation, and m is the number of features.

[0149] This embodiment first monitors abnormal unlocking attempts in real time through multimodal fusion technology. The system uses a heterogeneous sensor network, including high-precision accelerometers, acoustic sensors, photoelectric sensors, and electromagnetic field sensors. The sensor data is preliminarily processed by the edge computing device, and the sliding window method is used to calculate the features in real time, with a window size of 5 seconds and a step size of 1 second. Feature extraction uses a combination of time domain and frequency domain methods, including statistical features, wavelet transform coefficients, and Mel-frequency cepstral coefficients (MFCC). The anomaly score evaluation uses the above formula, where the mean and standard deviation of normal behavior are dynamically updated through an online learning algorithm to adapt to environmental changes. In order to process multidimensional features, the Mahalanobis distance is introduced as an anomaly metric, taking into account the covariance structure between features.

[0150] Furthermore, a deep learning algorithm is used to analyze changes in unlocking patterns and identify potential deception or counterfeiting behaviors. This embodiment adopts a model structure that combines a recurrent neural network (RNN) and an attention mechanism. The RNN part uses a bidirectional long short-term memory network (Bi-LSTM), which contains 3 layers and 128 units in each layer to capture the temporal dependencies of the unlocking sequence. The attention mechanism uses multi-head self-attention (Multi-head Self-attention) with 8 heads, which can focus on key information in the sequence. The model input is a sequence of unlocking events in the past 24 hours, and each event contains features such as timestamp, unlocking method, and duration. The output is the abnormal score of the unlocking behavior. The model training adopts a contrastive learning method and uses a triplet loss function, which can effectively distinguish between normal and abnormal behaviors.

[0151] It should be understood that the physical security of smart locks is ensured by detecting suspicious physical interference through sensor data fusion technology. This embodiment adopts a multi-sensor data fusion framework based on the Dempster-Shafer evidence theory. First, the data of each sensor is preprocessed and feature extracted, and the time-frequency features are extracted using the wavelet packet decomposition method. Then, the support vector data description (SVDD) algorithm is used to build a single-class classifier for each sensor for anomaly detection. The kernel function of SVDD selects the radial basis function (RBF), and the kernel parameters are optimized by cross-validation. Next, a basic probability assignment (BPA) is constructed based on the output of each classifier, and the evidence of multiple sensors is fused through the Dempster combination rule. Finally, the Pignistic Probability Transformation is used to convert the fusion result into a decision probability.

[0152] It is understandable that in order to prevent remote attacks, it is necessary to continuously evaluate the security of network connections. To this end, this embodiment designs a multi-level network security monitoring system. At the network layer, deep packet inspection (DPI) technology is used to identify abnormal traffic patterns in combination with rule matching and machine learning methods. At the transport layer, a secure communication protocol based on TLS1.3 is implemented, and a certificate transparency (Certificate Transparency) mechanism is introduced to prevent man-in-the-middle attacks. At the application layer, OAuth 2.0 authorization framework and JWT (JSON Web Token) are used for identity authentication and authorization. In addition, the system also integrates an intrusion detection system (IDS), adopts an anomaly-based detection method, and uses the isolation forest (Isolation Forest) algorithm to detect network anomalies in real time.

[0153] Furthermore, the process of triggering an emergency security response when a high-risk threat is detected uses a decision-making system based on a combination of rules and machine learning. The rule engine is implemented based on Drools and contains predefined security policies and response rules. The machine learning part adopts a reinforcement learning method, specifically using a deep Q network (DQN), which can dynamically optimize the decision-making strategy based on historical response effects. Response measures include a multi-level locking mechanism, a distributed alarm system, and security personnel scheduling. The locking mechanism adopts a progressive strategy, from soft locking to hard locking, to minimize the risk of misoperation. The alarm system uses message queues (such as Apache Kafka) to achieve high-throughput real-time message distribution. Security personnel scheduling uses a heuristic algorithm that takes into account factors such as geographic location, professional skills, and current workload.

[0154] Furthermore, this embodiment designs a high-performance security event logging system to support subsequent forensic analysis and security policy optimization. The system uses the distributed time series database InfluxDB to store detailed event logs, supporting high write throughput and fast time range queries. The log data uses a structured format, including fields such as timestamp, event type, device ID, and detailed description.

[0155] The benefit of this embodiment is that the security and reliability of the smart lock are improved through a multi-dimensional and multi-level security threat detection mechanism. Multimodal sensor fusion and deep learning algorithms improve the accuracy and real-time performance of abnormal behavior detection. The data fusion framework based on evidence theory enhances the stability of physical interference detection. The multi-level network security monitoring system provides comprehensive network protection.

[0156] Embodiment 8

[0157] In order to solve the problems of performance degradation, increased energy consumption and decreased user experience in the long-term operation of smart locks, this embodiment further optimizes the implementation method of smart lock performance optimization.

[0158] like Figure 8 , Fig. 9 As shown, the monitoring method also includes smart lock performance optimization, and the steps of smart lock performance optimization include:

[0159] S801. Continuously collect long-term operation data of the smart lock, including usage frequency, battery consumption, and fault records.

[0160] S802: Analyze user usage patterns, including unlocking time distribution, common unlocking methods, and usage habits.

[0161] S803. Use machine learning algorithms to identify performance bottlenecks, including response delays, battery life, or network connections.

[0162] S804: Generate an optimization strategy based on the analysis result, including adjusting the sleep strategy, optimizing the network connection, or updating the encryption algorithm.

[0163] S805. Optimize performance by remotely updating the smart lock software, including firmware upgrades and parameter adjustments.

[0164] S806. Evaluate and verify the optimization effect and collect user feedback.

[0165] Specifically, this embodiment first continuously collects the long-term operation data of the smart lock through a distributed data acquisition system. The system adopts an edge computing architecture and deploys a lightweight data acquisition agent locally on the smart lock, which is responsible for real-time collection of data such as usage frequency, battery consumption, and fault records. The agent program is written in C language to ensure minimal system resource usage. The data collection frequency adopts an adaptive strategy and is dynamically adjusted according to the battery power and usage frequency, usually between 1 minute and 1 hour. After local preprocessing and compression, the collected data is transmitted to the cloud data center through low-power wide area network (LPWAN) technology, such as LoRaWAN. The cloud uses the time series database InfluxDB to store raw data, which supports high write throughput and fast time range queries. In order to ensure the integrity and reliability of the data, the system implements a data integrity verification mechanism based on block chain.

[0166] Furthermore, a multi-dimensional time series analysis method is used to analyze user usage patterns. First, the seasonal and trend decomposition (STL) algorithm is used to decompose the unlocking time series to identify periodic patterns of different scales such as days, weeks, and months. Then, the dynamic time warping (DTW) algorithm is applied to calculate the similarity of usage patterns of different users or different periods to achieve user clustering and pattern evolution analysis. For the analysis of commonly used unlocking methods, a Markov chain model is used to construct an unlocking method transfer matrix to reveal the dynamic changes of user preferences. The mining of usage habits is achieved through a sequence pattern mining algorithm (such as PrefixSpan), which can discover frequently occurring behavior sequences. In order to process large-scale data, the analysis process uses the Apache Spark distributed computing framework to ensure computing efficiency.

[0167] It should be understood that machine learning algorithms are used to identify performance bottlenecks. This embodiment designs a performance diagnosis system that integrates multiple models. For the response delay problem, a long short-term memory network (LSTM) is used to establish a time series prediction model that can capture long-term dependencies and predict possible delay peaks in the future. The battery life analysis uses a random forest algorithm to consider various factors such as frequency of use, ambient temperature, network connection status, etc. to build a battery life prediction model. The network connection quality assessment uses a support vector regression (SVR) model to predict future connection stability based on historical connection data. The outputs of these models are fused through ensemble learning methods (such as Stacking) to obtain a comprehensive performance score. The model training adopts a federated learning framework to protect user privacy while utilizing data scattered across devices.

[0168] It is understandable that the optimization strategy is generated based on the analysis results. This embodiment adopts an adaptive strategy generation system based on reinforcement learning. The system uses a deep Q network (DQN) as the core algorithm. The state space contains current performance indicators, user usage patterns, environmental factors, etc., and the action space includes adjusting sleep strategies, optimizing network connection parameters, updating encryption algorithms, etc. The reward function design comprehensively considers factors such as performance improvement, energy consumption reduction, and user satisfaction. In order to accelerate the learning process, a Prioritized Experience Replay mechanism is introduced to give priority to learning more valuable experiences. In addition, the system also integrates a rule-based expert system to process known optimization scenarios, which complements the reinforcement learning model.

[0169] Furthermore, the process of remotely updating the smart lock software to achieve performance optimization adopts a safe and reliable remote update mechanism. The update package adopts differential update (Delta Update) technology to only transmit the changed parts and reduce the amount of data transmission. The update process uses a dual partition (A / B) solution to ensure that it can roll back to the previous stable version when the update fails. In order to prevent security threats during the update process, a signature verification mechanism based on a hardware security module (HSM) is adopted to ensure the authenticity and integrity of the update package. Breakpoint resume technology is used during the update process to support continued transmission after interruption and improve the update success rate. For the adjustment of key parameters, the system implements a progressive update strategy, which is fully promoted after verifying the effect through small batch A / B testing.

[0170] This embodiment designs a comprehensive optimization effect evaluation and user feedback collection system. The evaluation process uses before-after comparative analysis, combined with statistical hypothesis testing methods (such as paired t-test) to quantify the degree of performance improvement. The monitoring of system performance indicators uses the open source monitoring tool Prometheus, which is combined with Grafana for visual display. User feedback is collected using a multi-channel strategy, including in-application feedback, regular questionnaires, and social media monitoring. Feedback data is subjected to sentiment analysis and topic extraction through natural language processing technology to identify key issues of concern to users.

[0171] The benefit of this embodiment is that the long-term operating performance of the smart lock is improved through a comprehensive and intelligent performance optimization mechanism. The distributed data acquisition system and the multi-dimensional time series analysis method provide rich data support. The multi-model fusion performance diagnosis system realizes accurate bottleneck identification. The strategy generation system based on reinforcement learning can adaptively propose optimization solutions. The safe and reliable remote update mechanism ensures the smooth implementation of the optimization measures. The comprehensive effect evaluation and feedback collection system provides strong support for continuous optimization. Through this comprehensive performance optimization method, the service life of the smart lock is extended and the energy efficiency is improved.

[0172] Embodiment 9

[0173] In order to solve the problems of low collaboration efficiency among smart home devices, inconsistent security policies, and fragmented user experience, this embodiment further optimizes the collaborative working mechanism between smart locks and other smart home devices.

[0174] like Fig.10 , Fig.11 As shown, the monitoring method also includes the collaborative work of multiple devices, and the steps of the collaborative work of multiple devices include:

[0175] S901. Establish a communication interface between the smart lock and other smart home devices, such as smart cameras, access control systems, etc.

[0176] S902. Realize data sharing and status synchronization between devices to improve the overall security and convenience of the system.

[0177] S903. Develop security strategies and alarm mechanisms based on comprehensive information from multiple devices.

[0178] S904, supports scene linkage, automatically starts camera recording if an abnormality is detected.

[0179] S905. Provide a unified user interface to facilitate users to centrally manage and control multiple smart devices.

[0180] S906. Realize mutual backup and failover between devices to improve the reliability and fault tolerance of the system.

[0181] This embodiment first establishes a communication interface between the smart lock and other smart home devices by designing a general device interconnection protocol. The protocol is customized based on the IoT standard protocol MQTT (Message Queuing Telemetry Transport) and uses a publish / subscribe model to achieve communication between devices. The protocol stack is divided into three layers: the physical layer uses low-power Bluetooth (BLE 5.0) or Wi-Fi (IEEE 802.11ac) to achieve short-range high-speed communication; the network layer uses 6LoWPAN technology and supports IPv6 address allocation; the application layer is based on MQTT-SN (MQTT for Sensor Networks) to meet the needs of resource-constrained devices.

[0182] Furthermore, a distributed state management system is used to achieve data sharing and state synchronization between devices. The system is built on the open source framework Apache ZooKeeper and implements distributed coordination services with high availability and consistency. Each smart home device acts as a ZooKeeper node and shares its state information through a hierarchical namespace. The state update adopts an optimistic concurrency control mechanism and uses version numbers to resolve concurrency conflicts.

[0183] It should be understood that the purpose of formulating security policies and alarm mechanisms based on the comprehensive information of multiple devices is to improve overall security. This embodiment designs an intelligent decision-making system based on a combination of rule engine and machine learning. The rule engine is implemented using Drools, supports complex event processing (CEP), and can identify timing patterns and combined events of device states. The machine learning part uses an integrated learning method, combining algorithms such as decision trees, random forests, and gradient boosting trees (XGBoost) to build an anomaly detection model. Model training uses a federated learning framework to utilize data scattered across devices while protecting user privacy.

[0184] It is understandable that supporting scene linkage is to improve the intelligence and response speed of the system. This embodiment implements an event-driven scene linkage engine. The engine uses a directed acyclic graph (DAG) model to describe the trigger relationship between devices, and supports complex conditional logic and action sequences. Event processing adopts the Actor model. Each device acts as an independent Actor, and asynchronous communication is achieved through message passing to improve the concurrent processing capability of the system. In order to handle highly concurrent event streams, the system integrates the stream processing framework Apache Flink to support real-time complex event processing. The configuration of linkage rules adopts a visual drag-and-drop interface, and the underlying layer automatically generates DSL (Domain-Specific Language) scripts, which is convenient for users to customize complex scenes.

[0185] Furthermore, a micro-frontend architecture is used to provide a unified user interface. This architecture is implemented based on the Web Components standard, and the control interface of each smart device is encapsulated as an independent custom element. The main framework uses Single-SPA to implement routing and lifecycle management between applications.

[0186] This embodiment realizes mutual backup between devices and system-level fault tolerance by designing redundant backup and failover mechanisms. The master-slave mode is adopted to designate backup devices for key devices (such as smart locks). The state synchronization adopts the Raft consistency algorithm to ensure the state consistency between the master and backup devices. Fault detection adopts the gossip protocol to realize distributed fault detection through periodic heartbeat messages. When the failure of the master device is detected, a new master device is selected to take over the function through the leader election algorithm. In order to improve the availability of the system, an automated fault recovery mechanism is introduced, including strategies such as software restart, configuration rollback, and firmware downgrade. The system also implements predictive maintenance based on machine learning, predicts potential faults by analyzing device operation data, and realizes proactive fault handling.

[0187] The benefit of this embodiment is that by building a highly integrated and intelligent multi-device collaborative working mechanism, the overall performance of the smart home system is significantly improved. The universal device interconnection protocol realizes seamless communication between devices. The distributed state management system ensures real-time sharing and consistency of data. The intelligent decision-making system provides flexible and powerful security policy formulation capabilities. The scene linkage engine enhances the automation and response speed of the system.

[0188] Embodiment 10

[0189] In order to solve the problems of incomplete data collection, low processing efficiency, insufficient feature extraction and low state analysis accuracy in the process of smart lock information monitoring, this embodiment proposes an overall architecture of a smart lock information monitoring system and specific implementation methods of each module.

[0190] like Fig.12 As shown, the monitoring system includes:

[0191] The data acquisition module 100 is used to obtain various status information of the smart lock;

[0192] The data preprocessing module 200 is connected to the data acquisition module 100 and is used to filter, remove noise and synchronize the state information;

[0193] A feature extraction module 300, connected to the data preprocessing module 200, is used to extract key features from the preprocessed data;

[0194] The state analysis module 400 is connected to the feature extraction module 300 and is used to analyze the working state and fault conditions of the smart lock;

[0195] The alarm module 500, connected to the status analysis module 400, is used to generate and send warning information when a fault condition is detected;

[0196] The communication module 600 is connected to the alarm module 500 and is used to send the warning information to the user terminal;

[0197] The feature extraction and state analysis module 400 processes the state information according to the following formula:

[0198] Among them, S a is the state analysis result, β0 is a constant term, β j is the regression coefficient of the jth state information, x j is the jth state information, ε is the error term, and k is the number of state information types.

[0199] The data acquisition module 100 of this embodiment adopts multi-sensor fusion technology and integrates high-precision MEMS accelerometers, magnetometers, photoelectric sensors, temperature and humidity sensors, and current and voltage sensors. The sensor sampling frequency can be dynamically adjusted from 10Hz to 1kHz to meet different monitoring needs. In order to improve the reliability of data acquisition, the system adopts a redundant design, and key sensors are equipped with spare units. A real-time operating system (RTOS) such as FreeRTOS is used during data acquisition to ensure accurate control of the sampling timing. In addition, the module also integrates an edge computing unit, using a low-power ARMCortex-M4F processor, supporting local preliminary data processing and compression, and reducing transmission bandwidth requirements.

[0200] Furthermore, the data preprocessing module 200 implements a set of complex signal processing procedures. First, the Kalman filter is used to denoise the original data, and the filter parameters are dynamically adjusted through an adaptive algorithm to adapt to different noise environments. Then, a multi-scale analysis is performed using a wavelet transform to effectively separate the trend, cycle, and noise components in the signal. For the time synchronization problem of multi-source data, the system implements a high-precision timestamp mechanism based on the Network Time Protocol (NTP), combined with data interpolation technology, to align data from different sources into a unified time series. In addition, the module also integrates an outlier detection algorithm, using the local outlier factor (LOF) method to identify and process outliers.

[0201] It should be understood that the feature extraction module 300 is used for data analysis. The feature extraction module 300 adopts a multi-level feature extraction strategy. In the time domain, statistical features such as mean, variance, skewness, kurtosis, etc. are calculated; in the frequency domain, spectral features, including main frequency, harmonic ratio, etc., are extracted by fast Fourier transform (FFT); in the time-frequency domain, wavelet packet decomposition (WPD) is applied to extract time-frequency features. For complex nonlinear features, a deep feature learning method based on autoencoders is introduced. The feature selection process adopts a hybrid feature selection algorithm based on mutual information and L1 regularization to effectively reduce the feature dimension and improve the efficiency of subsequent analysis.

[0202] It is understandable that the state analysis module 400 constructs a generalized linear model framework based on the state information calculation formula. In order to deal with nonlinear relationships, the model introduces basis function extension and adopts radial basis function (RBF) as the kernel function. The model parameter estimation adopts the least squares method with L2 regularization, and the optimal regularization parameter is determined by cross-validation. In order to improve the generalization ability of the model, an integrated learning method is introduced, and a strong learner composed of multiple basic models is constructed by combining Bagging and Boosting techniques. The model training adopts an online learning strategy, supports incremental updates, and can adapt to state changes during the use of smart locks.

[0203] Furthermore, the alarm module 500 adopts a multi-level threshold strategy to generate different levels of early warning information according to the severity of the status analysis results. The early warning generation process takes into account the statistical characteristics of historical data and adopts dynamic threshold technology to adaptively adjust the early warning trigger conditions. In order to reduce false alarms, the system introduces a rule-based expert system and combines the output of the machine learning model for secondary judgment. The format of the early warning information follows the universal alarm protocol (SNMP) and supports seamless integration with mainstream monitoring systems. In addition, the module also implements the priority management and automatic merging functions of the early warning information to effectively avoid alarm storms.

[0204] Furthermore, the communication module 600 of this embodiment adopts a multi-protocol support design. Near-field communication adopts Bluetooth low energy (BLE 5.0) technology to support short-distance data transmission with low power consumption and high bandwidth. For long-distance communication, appropriate technology is selected according to the deployment environment, such as NB-IoT, LoRa or 4G / 5G network. Data transmission adopts MQTT protocol, supports publish / subscribe mode, and realizes flexible message routing. In order to ensure communication security, all data transmission is encrypted with TLS1.3, and a device identity authentication mechanism based on PKI is implemented.

[0205] The benefit of this embodiment is that by building a highly integrated and intelligent information monitoring system, the comprehensiveness, accuracy and real-time performance of smart lock status monitoring are significantly improved. Multi-sensor fusion and edge computing technology improve the quality and efficiency of data acquisition. Complex signal processing processes ensure the reliability of data. Multi-level feature extraction strategies and deep learning methods provide rich feature representations. The state analysis framework based on generalized linear models and integrated learning achieves high-precision state evaluation and prediction. Multi-level early warning strategies and expert systems improve the accuracy of alarms.

Claims

1. A smart lock information monitoring method, characterized in that: The steps of the monitoring method include: Use multiple sensors to collect status information of smart locks, including battery power, lock switch status, door magnet status, abnormal vibration and positioning information; Preprocessing the state information, including data filtering, denoising and time synchronization; Extracting and analyzing key feature data of the state information through an analysis algorithm; Use multivariate analysis methods to analyze and determine the working status and fault conditions of smart locks; Generate early warning information based on the analysis results and send it to the user terminal.

2. The smart lock information monitoring method according to claim 1, characterized in that: The step of collecting the status information of the smart lock using multiple sensors includes: Use the accelerometer and gyroscope to collect the motion status and posture information of the smart lock; Collect battery power, charge and discharge status, and battery health information through the battery management system; Use door magnetic sensors and micro switches to collect the lock's switch status and the door's open and closed status; Utilize vibration sensors and sound sensors to detect abnormal vibration and sound information; Collect environmental temperature and humidity data through temperature and humidity sensors; Use GPS or indoor positioning module to obtain the location information of the smart lock.

3. The smart lock information monitoring method according to claim 1, characterized in that: The step of preprocessing the status information comprises: Using a transformation algorithm to perform multi-scale filtering on the raw data of the state information to remove high-frequency noise and outliers; Time synchronization and alignment of multi-source data to ensure data consistency from different sensors; Perform data normalization to unify the scale range of different types of data; Interpolation or estimation of missing data is used to ensure data completeness; The calculation formula of the transformation algorithm is: Among them, W ψ (a,b) are the transformed coefficients, x(t) is the time series signal, ψ is the basis function, a is the scale parameter, and b is the time shift parameter.

4. The smart lock information monitoring method according to claim 1, characterized in that: The step of extracting and analyzing the key feature data of the state information by an analysis algorithm comprises: Perform time domain analysis on the preprocessed state information to extract time domain features such as maximum value, minimum value, mean value, and peak value; Use fast Fourier transform (FFT) to convert the data into frequency domain and extract the main frequency and frequency distribution; Calculate the statistical characteristics of the state information, including mean, variance, skewness, kurtosis, and correlation coefficient; The principal component analysis (PCA) method is used to reduce feature dimensions and remove redundant information; Based on the feature importance evaluation algorithm, the most representative feature set is selected; Use convolutional neural network (CNN) to extract deep features and enhance feature expression capabilities; The calculation formula of the principal component analysis (PCA) method is: Z=XW, where Z is the data matrix after dimensionality reduction, X is the original data matrix, and W is the eigenvector matrix.

5. The smart lock information monitoring method according to claim 1, characterized in that: The step of analyzing and judging the working state and fault condition of the smart lock by using the multivariate analysis method comprises: Establish a normal working state model of the smart lock, including the normal range and change trend of various parameters; Calculate the deviation between the current state and the normal state model; The support vector machine (SVM) algorithm is used to classify deviations and distinguish between normal and abnormal states; Identify specific fault types and causes based on decision tree algorithms; Use Bayesian networks to calculate the probability and severity of failures and assess potential risks; Apply long short-term memory network (LSTM) to predict fault development trend and realize fault early warning; The decision function of the support vector machine algorithm is expressed as: Among them, α i is the Lagrange multiplier, y i is the category label, K(xi,x) is the kernel function, and b is the bias term.

6. The smart lock information monitoring method according to claim 1, characterized in that: The monitoring method also includes automatic repair of the smart lock, and the steps of automatic repair of the smart lock include: Identify the types of faults that can be fixed by software, including system anomalies, configuration errors, or software bugs; Query the fault repair database and select the best repair strategy; Generate a repair instruction set containing specific operation steps; Performing compatibility check on the repair instruction set to prevent the introduction of new faults; Remotely execute repair instructions, including software reboot, configuration reset, or firmware update; Verify the repair results and confirm whether the fault has been resolved; Provide users with feedback on repair status and results and provide action suggestions.

7. The smart lock information monitoring method according to claim 1, characterized in that: The monitoring method further includes security threat detection, and the steps of security threat detection include: Real-time monitoring of abnormal unlocking attempts; Use machine learning algorithms to analyze unlocking pattern changes and identify potential fraud or counterfeiting; Detect suspected physical interference through sensor data fusion technology; Continuously evaluate the security of network connections to detect potential network attacks or data hijacking risks; Trigger an emergency security response when a high-risk threat is detected, including locking the device, sending an alarm, and notifying security personnel; Record detailed security event logs to support subsequent forensic analysis and security policy optimization; The abnormal score evaluation formula for real-time monitoring of abnormal unlocking attempts is: Among them, θ i is the observed value, μ i is the mean of normal behavior, σ i is the standard deviation, and m is the number of features.

8. The smart lock information monitoring method according to claim 1, characterized in that: The monitoring method further includes smart lock performance optimization, and the steps of smart lock performance optimization include: Continuously collect long-term operation data of smart locks, including usage frequency, battery consumption, and fault records; Analyze user usage patterns, including unlocking time distribution, common unlocking methods and usage habits; Use machine learning algorithms to identify performance bottlenecks, including response latency, battery life, or network connectivity; Generate optimization strategies based on the analysis results, including adjusting sleep strategies, optimizing network connections, or updating encryption algorithms; Optimize performance by remotely updating smart lock software, including firmware upgrades and parameter adjustments; Evaluate and verify optimization results and collect user feedback.

9. The smart lock information monitoring method according to claim 8, characterized in that: The monitoring method further includes the collaborative work of multiple devices, and the steps of the collaborative work of multiple devices include: Establish communication interface between smart locks and other smart home devices; Realize data sharing and status synchronization between devices; Formulate security strategies and alarm mechanisms based on comprehensive information from multiple devices; Support scene linkage, automatically start camera recording if abnormality is detected; Provide a unified user interface to facilitate users to centrally manage and control multiple smart devices; Realize mutual backup and failover between devices.

10. A smart lock information monitoring system, characterized in that: The monitoring system comprises: Data acquisition module, used to obtain various status information of smart locks; A data preprocessing module, connected to the data acquisition module, for filtering, denoising and time synchronization of the state information; A feature extraction module, connected to the data preprocessing module, for extracting key features from the preprocessed data; A state analysis module, connected to the feature extraction module, for analyzing the working state and fault conditions of the smart lock; An alarm module, connected to the status analysis module, for generating and sending an early warning message when a fault condition is detected; A communication module, connected to the alarm module, for sending the warning information to the user terminal; The feature extraction and state analysis module processes the state information according to the following formula: Among them, S a is the state analysis result, β0 is a constant term, β j is the regression coefficient of the jth state information, x j is the jth state information, ε is the error term, and k is the number of state information types.

Images