An intelligent automation monitoring method based on the Internet of Things

Abstract

The application relates to an intelligent automatic monitoring method based on an Internet of Things, which comprises the following steps: S1, collecting operation parameters and environment parameters of a monitoring target; S2, transmitting the collected data to a data processing platform through a network; S3, performing timestamp synchronization and data alignment, abnormal value and missing value processing, noise filtering and standardization or normalization processing on the collected data to obtain cleaned time series data; S4, extracting time domain features and frequency domain features from the cleaned time series data, and generating a fusion feature vector by adopting a feature level fusion or decision level fusion strategy; S5, intelligent analysis; and S6, generating graded early warning information according to the severity of the abnormal trend and pushing the information. The application can realize real-time collection, unified transmission, cleaning and fusion, intelligent analysis and graded early warning of multi-source heterogeneous data, so that the real-time performance, accuracy and intelligent level of monitoring are improved, and data support is provided for predictive maintenance.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) technology, and more specifically to an intelligent automated monitoring method based on IoT. Background Technology

[0002] Among existing monitoring technologies, the most common solutions mainly include regular manual inspections and simple threshold alarm systems.

[0003] Regular manual inspections typically involve technicians manually checking and recording equipment or environmental parameters on-site at fixed intervals. The data is usually recorded in paper forms or simple spreadsheets, and then manually analyzed later. This method has the following drawbacks: First, the fixed inspection cycle makes it impossible to detect sudden anomalies in real time, resulting in significant monitoring blind spots; second, it relies on manual operation, making it susceptible to factors such as personnel experience, fatigue, and negligence, leading to missed inspections or misjudgments; and third, it has high maintenance costs and low efficiency.

[0004] While simple threshold alarm systems can issue alerts when monitored data exceeds preset upper or lower limits, these systems typically rely solely on static thresholds for judgment. They lack the ability to continuously record historical data, analyze trends, and diagnose faults. They struggle to adapt to normal fluctuations in equipment operation under different conditions, such as startup, stable operation, and high load, leading to false alarms or missed alarms. Furthermore, the monitoring modules often operate independently, lacking a unified data fusion and processing platform, resulting in data silos and hindering the realization of the comprehensive value of multi-source data.

[0005] Therefore, there is an urgent need to provide an automated monitoring method that can realize real-time acquisition, unified transmission, cleaning and fusion, intelligent analysis and hierarchical early warning of multi-source heterogeneous data, so as to improve the real-time performance, accuracy and intelligence of monitoring, and provide data support for predictive maintenance. Summary of the Invention

[0006] To address the aforementioned shortcomings in existing technologies, this invention provides an intelligent automated monitoring method based on the Internet of Things, comprising the following steps: S1. Collect the operating parameters and environmental parameters of the monitored target by using multi-source heterogeneous sensors deployed at the monitored target location; S2. The collected data is transmitted to the data processing platform via the network through the IoT gateway; S3. Perform timestamp synchronization and data alignment, outlier and missing value processing, noise filtering, and standardization or normalization on the collected data to obtain cleaned time-series data. S4. Extract time-domain and frequency-domain features from the cleaned time-series data, and generate a fused feature vector using feature-level fusion or decision-level fusion strategies. S5. Intelligent analysis, including: constructing a dynamic baseline model based on historical and real-time data, and identifying abnormal trends that deviate from the normal operation mode based on the fused feature vector; S6. Generate graded early warning information based on the severity of the abnormal trend and push it out. At the same time, display the monitoring results, historical data and early warning information on the visualization interface.

[0007] Preferably, in step S3, outliers are identified using the 3σ principle, missing values ​​are filled using interpolation based on neighboring data points, the original signal is smoothed using a digital filter, and the processed data is standardized using the Z-score normalization method.

[0008] Preferably, interpolating and filling missing values ​​includes the following steps: Step S3-2-1: Traverse the time series data, identify data with empty values ​​and data marked as outliers by the 3σ principle, and record the timestamp position t of missing and outlier data. m ; Step S3-2-2, for each missing position t m Search for the k nearest valid normal values ​​on the time axis, and take the forward nearest point t. prev and corresponding value x prev Take the next nearest neighbor: t next and corresponding value x next ; Step S3-2-3: Perform the nearest neighbor mean interpolation calculation, as shown in the following formula: ; Step S3-2-4, boundary missing handling: If the beginning of the sequence is missing, the first valid value from the back is used to fill it; if the end of the sequence is missing, the last valid value from the front is used to fill it, so as to obtain time series data without missing or abnormal data.

[0009] Preferably, in steps S4 and S5, the standardized time-series data is recursively identified and analyzed in batches and time windows according to this week, this month, this year, or a custom time range, so as to obtain abnormal change trends at different time scales.

[0010] Preferably, step S5 includes: step S5-2-1, historical normal baseline initialization: extract historical normal operating condition fusion feature data within multiple windows of this week, this month, and this year, calculate the mean, standard deviation, and normal fluctuation range of each dimension feature, fit the initial baseline parameters of the equipment's standard operating state, and establish a benchmark model for normal equipment operation. Step S5-2-2, Real-time data dynamic correction: Input the fused feature vector collected in real time into the model, and combine it with the recursive analysis results of the custom time window to iteratively update the mean, fluctuation threshold and trend range of the baseline in real time, so that the baseline can be adaptively adjusted according to changes in equipment operating conditions and environmental conditions. Step S5-2-3, Dynamic Baseline Output: Form a dynamic baseline that takes into account both short-term stability and long-term adaptability, simulating the normal operation mode of the equipment under different time periods and operating conditions.

[0011] Preferably, in step S5, sequential least squares quadratic programming (SLSQP) is used to optimize the parameters of the anomaly analysis model, and a REM matrix is ​​established to organize, fuse, and infer the analysis feature points in order to improve the accuracy of anomaly identification.

[0012] Preferably, SLSQP is used to optimize the model parameters, and REM matrix is ​​used to organize, fuse, and infer the analysis feature points, including: Step S5-4-1: Construct optimization objectives and constraints: The objective function is to maximize the anomaly identification accuracy and minimize the false alarm rate and false negative rate; Equation and inequality constraints are set in combination with the sensor physical range, equipment operating condition limits, and feature value range.

[0013] Step S5-4-2, Parameter initialization: Initialize the core parameters of the model, such as the anomaly threshold, deviation weight, and trend discrimination coefficient, and construct the optimal solution space for the parameters.

[0014] Step S5-4-3, SLSQP Iterative Optimization: Based on the samples of multi-time window recursive analysis, the model parameters are corrected and converged to the optimal parameter combination by least squares fitting and quadratic programming iterative solution. Step S5-4-4, Optimal parameter update: After the iteration converges, replace the initial parameters of the model to complete the adaptive optimization of the anomaly analysis model.

[0015] Compared with the prior art, the present invention has at least the following beneficial effects: From reactive response to proactive prevention: Through dynamic baseline modeling and anomaly trend identification, this invention can provide early warnings when early signs of failure appear, thus shifting from traditional "reactive maintenance" to "proactive prevention." This significantly reduces unplanned downtime.

[0016] Improve monitoring accuracy and reliability: By integrating multi-source heterogeneous data, time synchronization, outlier processing, and dynamic baseline analysis, false alarms caused by normal fluctuations in operating conditions can be significantly reduced, while the ability to detect potential anomalies can be improved.

[0017] Improve operational efficiency: The system replaces manual inspections with automated data collection, analysis, and push notifications, and outputs historical curves and diagnostic reports through a visual interface, which helps reduce the burden on technical personnel and shorten the fault location time.

[0018] Maximizing the value of data: Through continuous collection, long-term storage, and multi-dimensional analysis, this invention can provide data support for equipment performance optimization, condition assessment, and lifecycle management.

[0019] High system integration: This invention integrates data collection, transmission, cleaning, fusion, analysis, early warning and display into a unified platform, avoiding the isolation of each monitoring module and improving overall collaborative capabilities. Attached Figure Description

[0020] Figure 1 This invention relates to a flowchart illustrating an intelligent automated monitoring method based on the Internet of Things. Detailed Implementation

[0021] To better understand this invention, the following description, in conjunction with the accompanying drawings and examples, will further illustrate the invention.

[0022] This application provides an intelligent automated monitoring method based on the Internet of Things, such as... Figure 1 As shown, it includes the following steps: Step S1: Data Acquisition. Operating parameters and environmental parameters are collected using multi-source heterogeneous sensors deployed at the monitoring target. These parameters may include temperature, humidity, vibration, displacement, pressure, etc.

[0023] Step S2: Data transmission. The collected data is uploaded to the cloud platform or edge server via wired or wireless networks through the IoT gateway using communication protocols such as HTTP, TCP, and MQTT.

[0024] Step S3: Data Cleaning and Preprocessing: The collected data undergoes timestamp synchronization and data alignment, outlier and missing value handling, noise filtering, and standardization or normalization to obtain cleaned time-series data. This specifically includes the following steps: Step S3-1: Timestamp synchronization and data alignment are performed on data from different sources and with different sampling frequencies; Step S3-2: Identify and remove outliers using the 3σ principle, and fill missing values ​​by interpolation: use interpolation based on neighboring data points to fill missing values.

[0025] Specifically, the effective acquired values ​​X=[x1,x2,x3,…x] in the time series data i ,…x nCalculate the mean μ (the average of all data in the valid collection) and the standard deviation σ (a measure of the dispersion of the data); Thresholds are defined based on the 3σ principle, with the normal data range being [μ-3σ, μ+3σ]. Data values ​​below the lower limit μ-3σ or above the upper limit μ+3σ are considered outliers. For outliers, anomaly labels are added to the data (for subsequent replacement or padding), while preserving the timestamp correspondence.

[0026] The steps for filling missing values ​​using interpolation based on neighboring data points are as follows: Step S3-2-1: Traverse the time series data, identify data with empty values ​​(NaN / Null) and data marked as outliers by the 3σ principle, and record the timestamp position t of the missing / outlier data. m .

[0027] Step S3-2-2, for each missing position t m Search for the k nearest valid normal values ​​on the time axis (k=2 / 4 is commonly used in engineering), and take the forward nearest point t. prev and corresponding value x prev Take the next nearest neighbor: t next and corresponding value x next ; Step S3-2-3 performs nearest neighbor mean interpolation calculation, which is highly real-time and suitable for rapidly collected IoT data, as shown in the following formula: .

[0028] Step S3-2-4, boundary missing handling: If the beginning of the sequence is missing, the first valid value from the back is used to fill it; if the end of the sequence is missing, the last valid value from the front is used to fill it, so as to obtain time series data without missing or abnormal data.

[0029] Step S3-3 involves using a digital filter to suppress noise and smooth the original signal, including: Step S3-3-1: Determine the filter window size. Set the window length W (commonly used: 3 / 5 / 7, odd number) according to the acquisition frequency, which represents the number of neighboring data points used for each smoothing.

[0030] Step S3-3-2: Perform filtering calculations using moving average filtering (to suppress random noise): ; i: Output the index of the i-th data point after smoothing; x j : The j-th sampled value of the original acquired time-series data; y i The i-th filtered value after filtering and smoothing. k: Number of points on one side .

[0031] Step S3-4 involves using Z-score standardization to unify the dimensions of the processed data. Z-score standardization transforms the data into a standard normal distribution with a mean of 0 and a standard deviation of 1, eliminating the influence of dimensions and numerical magnitude. This includes the following steps: Calculate data statistics: For the filtered data Y, calculate the mean μY and the standard deviation σY.

[0032] Point-by-point standardization calculations are performed to obtain the standard value after Z-score standardization. : ; If the standardized data satisfy the following conditions: new mean μY≈0; new standard deviation σY≈1, then we obtain mean-centered, dimensionless standard time series data.

[0033] Step S4: Feature extraction and fusion.

[0034] Temporal and frequency domain features are extracted from the cleaned time-series data, and feature-level or decision-level fusion is used to form a high-dimensional fusion feature vector.

[0035] Step S4-1: Arrange all time-domain features in a fixed order to construct a time-domain feature vector: F t =[μ,σ,R,x max ,x min [,Z,C]; Where μ is the mean, reflecting the overall steady-state level of the monitoring data and characterizing the normal operating status of the equipment. As shown in the following formula: ; Where σ is the standard deviation, which characterizes the degree of discrete fluctuation in time series data and can identify states such as abnormal equipment vibration and unstable operating conditions, as shown in the following formula: ; Where, x max For the maximum value, x min The minimum value reflects the extreme amplitude characteristics of the monitoring data, capturing out-of-limit and extreme abnormal operating conditions.

[0036] R is the peak-to-peak value, the difference between the maximum and minimum values, representing the overall fluctuation range of the data, as shown in the following formula: R = x max -x min ; Where Z represents kurtosis, which is sensitive to pulse abrupt changes and transient anomalies in time-series data, adapting to the impact characteristics of equipment failures. As shown in the following formula: ; Where C is the waveform factor, reflecting the regularity of the data waveform and characterizing the stability of equipment operation, as shown in the following formula: .

[0037] Step S4-2: Convert the time-domain time series data into a frequency-domain signal using Fast Fourier Transform (FFT) to extract core frequency-domain features and overcome the limitations of time-domain analysis.

[0038] Frequency domain feature extraction is as follows: S4-2-1, Time-domain to frequency-domain transformation: Perform a fast Fourier transform on the cleaned time-series data to convert the discrete time-domain signal into a discrete frequency-domain signal, obtain the corresponding frequency amplitude sequence, and complete the time-frequency domain transformation.

[0039] S4-2-2, Filtering effective frequency domain components: Eliminate low-frequency noise and invalid zero-frequency components, and retain amplitude and frequency information within the effective operating frequency range of the monitoring equipment.

[0040] S4-2-3, Extracting core frequency domain features: Selecting common frequency domain features for IoT monitoring, including the spectral mean E mean Spectral standard deviation E std , main frequency amplitude A main , main frequency f main Spectral energy E total Mean square frequency f rms These represent the overall energy level, frequency distribution dispersion, core operating frequency, and vibration energy characteristics of the frequency domain signal, respectively, as shown in the following formula: F f =[E mean E std ,f main A main E total ,f rms ].

[0041] Step S4-3: Feature fusion to construct a high-dimensional fused feature vector. This application adopts a feature-level fusion method (the mainstream preferred method, with higher accuracy than decision-level fusion). This method integrates time-domain and frequency-domain features at the feature level, retaining all effective feature information and achieving multi-dimensional information complementarity. Compared with decision-level fusion, it has the advantages of high information integrity, strong feature representation ability, and high recognition accuracy. The specific fusion steps are as follows: Step S4-3-1, Feature Normalization and Unification: Normalize the extracted time-domain feature vector F t and frequency domain eigenvector F f A second-order unification process is performed to eliminate differences in the numerical magnitude of different features, prevent high-magnitude features from covering low-magnitude effective features, and ensure a balanced weight for each feature.

[0042] Step S4-3-2, Feature splicing and fusion: Following the fixed order of "time domain features first, frequency domain features second", the two sets of one-dimensional feature vectors are spliced ​​and fused horizontally to achieve complementary time and frequency domain information.

[0043] Step S4-3-3: After splicing, a high-dimensional fused feature vector containing temporal fluctuation features, steady-state features, frequency energy features, and vibration features is generated. The final fused vector expression is: F=[F t ,F f The final constructed high-dimensional fusion feature vector simultaneously encompasses the macroscopic variation patterns of data in the time dimension and the microscopic implicit features in the frequency dimension, completely solving the problems of one-sided representation and low monitoring accuracy of single features, and providing complete and accurate feature data support for subsequent equipment status identification, anomaly warning, and intelligent monitoring and judgment.

[0044] Step S5: Intelligent analysis, recursively analyze data in batches and time windows according to this week, this month, this year or a custom time range; Step S5-1: Multi-scale time window hierarchical segmentation and batch recursive analysis. To take into account short-term transient anomalies, medium- and long-term operating condition drift, and long-term equipment aging trends, the full-volume time-series fusion feature data is hierarchically segmented into this week, this month, this year, and custom time ranges to achieve batch and window recursive analysis.

[0045] Step S5-1-1. Multi-scale time window construction: According to the monitoring business needs, four levels of analysis windows are divided, namely weekly short-cycle window, monthly medium-cycle window, annual long-cycle window and user-defined variable time window, covering different anomaly types such as short-term fluctuations, medium-term steady-state deviations and long-term trend drifts.

[0046] Step S5-1-2, Time Series Data Slicing: Based on the boundaries of each time window, the continuous time series fusion feature dataset is sliced ​​and grouped to decompose the massive streaming monitoring data into independent window batch data, avoiding the loss of time series details caused by global batch analysis.

[0047] Step S5-1-3, Window-by-Window Recursive Iterative Analysis: A recursive traversal mechanism is adopted to conduct feature statistics and state analysis window by window and batch by batch. Short-cycle windows are used to correct instantaneous fluctuation baselines, and medium- and long-cycle windows are used to correct slowly drifting baselines. Through multi-level recursive iteration, comprehensive and thorough mining of data features is achieved, solving the problems of one-sided, missed, and misjudged analysis on a single time scale.

[0048] Step S5-2: Construct a dynamic baseline model based on historical and real-time data, and update the dynamic baseline according to the operating conditions of the monitored object. By combining the results of multi-time window recursive analysis and integrating historical normal data with real-time monitoring data, an adaptively updatable dynamic baseline model for normal equipment operation is constructed to replace the fixed static threshold and adapt to dynamic changes in operating conditions.

[0049] Step S5-2-1, Historical Normal Baseline Initialization: Extract historical normal operating condition fusion feature data from multiple windows this week, this month, and this year, calculate the mean, standard deviation, and normal fluctuation range of each dimension of features, fit the initial baseline parameters of the equipment's standard operating state, and establish a benchmark model for normal equipment operation.

[0050] Step S5-2-2, Real-time Data Dynamic Correction: Input the fused feature vector collected in real time into the model, and combine it with the recursive analysis results of the custom time window to iteratively update the mean, fluctuation threshold and trend range of the baseline in real time, so that the baseline can be adaptively adjusted with the slow changes in equipment operating conditions and environmental conditions.

[0051] Step S5-2-3, Dynamic Baseline Output: Finally, a dynamic baseline is formed that takes into account both short-term stability and long-term adaptability, accurately simulating the normal operation mode of the equipment under different time periods and working conditions.

[0052] Step S5-3, Abnormal Trend Identification Based on Baseline Deviation. The current fused feature vector is compared with the dynamic baseline model to identify abnormal trends that deviate from the normal operating mode.

[0053] Step S5-3-1, Multi-window feature deviation calculation: Compare the real-time fused feature vectors of each batch and each time window with the dynamic baseline of the corresponding time period one by one, and calculate the real-time deviation, deviation rate and trend slope of each dimension feature.

[0054] Step S5-3-2, Abnormal Trend Judgment: If the single-window feature deviation exceeds the normal threshold, or if continuous same-direction offset or continuous expansion of deviation occurs in multi-window recursive analysis, then the current data is determined to deviate from the normal operating mode of the equipment, and different types of abnormal states such as instantaneous abnormality, slow drift, and trend failure are identified.

[0055] In step S5-4, in the preferred method, SLSQP is used to optimize the model parameters, and REM matrix is ​​used to organize, fuse and infer the analysis feature points.

[0056] To address the issues of fixed parameters and poor adaptability in traditional anomaly models, Sequential Least Squares Quadratic Programming (SLSQP) is employed to iteratively optimize the core parameters of the anomaly analysis model, thereby improving the model's fitting accuracy. Specifically: Step S5-4-1: Construct optimization objectives and constraints: The objective function is to maximize the anomaly identification accuracy and minimize the false alarm rate and false negative rate; Equation and inequality constraints are set in combination with the sensor physical range, equipment operating condition limits, and feature value range.

[0057] Step S5-4-2, Parameter initialization: Initialize the core parameters of the model, such as the anomaly threshold, deviation weight, and trend discrimination coefficient, and construct the optimal solution space for the parameters.

[0058] Step S5-4-3, SLSQP Iterative Optimization: Based on the samples from the multi-time-window recursive analysis, the model parameters are continuously corrected through least squares fitting and quadratic programming iterative solution, gradually converging to the optimal parameter combination.

[0059] Step S5-4-4, Optimal parameter update: After the iteration converges, replace the initial parameters of the model to complete the adaptive optimization of the anomaly analysis model and improve the model's adaptability to anomalies of multiple scales and multiple working conditions.

[0060] Step S5-5: Feature organization, fusion, and intelligent reasoning based on the REM matrix. A REM feature analysis matrix is ​​established to perform regularization, fusion, and deep reasoning on discrete feature points from multiple windows, batches, and dimensions, further improving anomaly detection accuracy.

[0061] Step S5-5-1, REM matrix initialization: Construct a two-dimensional REM analysis matrix with the time window as the vertical dimension and the time-frequency fusion feature dimension as the horizontal dimension. This matrix is ​​used to store the feature data and bias results of multiple batches of recursive analysis.

[0062] Step S5-5-2, Discrete Feature Point Consolidation and Input: Enter all optimized feature points, baseline deviation values, and trend parameters from this week, this month, this year, and the custom window into the corresponding positions in the matrix to achieve structured and ordered organization of discrete data.

[0063] Step S5-5-3, Multidimensional feature fusion of the matrix: horizontally fuse multidimensional time-frequency feature information under a single window, and vertically fuse time-series trend information at different time scales to achieve deep fusion of local features and global trends.

[0064] Step S5-5-4, Matrix Intelligent Reasoning and Judgment: Based on the feature correlation, time series change pattern and multi-scale deviation features within the REM matrix, comprehensive reasoning is carried out to accurately distinguish between normal fluctuations, operating condition drift and real fault anomalies, effectively reducing the probability of misjudgment and missed judgment, and significantly improving the anomaly identification accuracy of intelligent monitoring.

[0065] Step S6: Tiered early warning and visualization display.

[0066] Based on the severity of abnormal trends, different levels of early warning information are automatically generated and pushed to the responsible persons through App, SMS, email and other means; Meanwhile, the monitoring status, historical curves, early warning information, and diagnostic reports are displayed in real time on the web or mobile devices.

[0067] Specifically, step S6-1 involves a quantitative assessment of the severity of the anomaly.

[0068] The system extracts core anomaly evaluation indicators based on REM matrix fusion inference results and dynamic baseline deviation data, including parameters such as feature deviation rate, anomaly duration frames, trend fluctuation slope, and multi-window recursive offset degree, to comprehensively quantify and score the identified anomaly trends. It abandons the single threshold judgment method and combines different anomaly characteristics such as short-term instantaneous fluctuations, medium- and long-term operating condition drift, and persistent anomaly offsets to comprehensively assess the severity, risk level, and failure evolution speed of current equipment anomalies, providing accurate data support for early warning classification.

[0069] Step S6-2: Multi-level early warning classification and rule matching Combining industrial IoT monitoring standards and equipment operation and maintenance specifications, abnormal states are categorized into four warning levels—Indicative, Moderate, Severe, and Critical—according to their severity from low to high. Preset thresholds and trigger rules for each level are also provided. Specifically, minor data fluctuations and short-term recoverable deviations are classified as Indicative warnings; minor deviations from the baseline without a continuing deterioration trend are classified as Moderate warnings; continuously widening deviations and significantly abnormal operating conditions are classified as Severe warnings; and states significantly exceeding the normal baseline range, posing equipment failure, safety hazards, and downtime risks are classified as Critical warnings. The system automatically matches the corresponding warning level based on the quantitative assessment results.

[0070] Step S6-3: Intelligent generation of standardized early warning information.

[0071] The system automatically assembles and generates structured and standardized early warning messages based on the matched warning level, anomaly type, monitoring location, anomaly occurrence time, multi-window analysis results, deviation values, and risk descriptions. The warning information comprehensively includes the abnormal device number, monitoring dimension, anomaly start time, real-time deviation data, anomaly evolution trend, risk level, and preliminary maintenance suggestions, avoiding issues such as vague warning content and missing information, and ensuring that maintenance personnel accurately grasp the details of equipment anomalies.

[0072] Step S6-4: Differentiated early warning push through multiple channels.

[0073] The system establishes a multi-terminal integrated push mechanism, implementing differentiated push strategies for different levels of early warning information. It accurately pushes information to relevant personnel through three main channels: mobile app, SMS, and email. Specifically, alerts and general warnings are primarily pushed via the IoT monitoring app, with warning records maintained. More severe warnings are simultaneously pushed via app pop-ups and email notifications, reminding maintenance personnel to verify information promptly. Critical warnings activate an emergency push mechanism, linking strong app pop-ups, rapid SMS pushes, and expedited email delivery to ensure that high-risk anomaly information reaches management personnel without delay or omission.

[0074] Step S6-5: Retention and closed-loop recording of early warning data.

[0075] All generated early warning information, push time, recipients, early warning level, and original abnormal data are automatically stored in the system database, forming a complete early warning log ledger. This provides data support for subsequent equipment status tracing, fault analysis, operation and maintenance review, and model iteration and optimization, realizing closed-loop management of the entire process of equipment anomaly monitoring, early warning, operation and maintenance, and review.

[0076] Based on the above method, this application also provides an Internet of Things-based intelligent automated monitoring system for continuous status monitoring of industrial equipment, production line equipment, or environmental areas.

[0077] The system comprises a sensing layer, a network layer, and a platform layer.

[0078] The sensing layer is installed at the monitoring site and includes at least two of the following: temperature sensors, humidity sensors, vibration sensors, displacement sensors, and pressure sensors. These sensors are used to continuously acquire operational and environmental status data of the monitored object. Each sensor can be selected and combined according to the actual needs of the monitored object.

[0079] The network layer includes IoT gateways and communication modules, used to receive raw data from the sensing layer and upload the raw data to the platform layer via wired networks or wireless networks such as 5G, Wi-Fi, and LoRa through communication protocols such as HTTP, TCP, and MQTT. The platform layer can be deployed in a cloud platform, edge server, or cloud-edge collaborative architecture.

[0080] The platform layer includes the following functional modules: Data preprocessing module: used to timestamp and align data from different sensors and sampling frequencies according to a unified time base; For data points that significantly deviate from the normal range, outliers are identified and removed using the 3σ principle. For missing data caused by link interruption or instantaneous packet loss, interpolation based on neighboring data points is used to fill in the missing data. For random noise in the original signal, a digital filter is used for smoothing. For data with different dimensions and numerical ranges, the maximum-minimum normalization method or the Z-score standardization method are used for unified processing to eliminate the influence of dimensions.

[0081] Feature extraction and fusion module: This is used to extract multidimensional features from cleaned time-series data. The extracted features include at least time-domain features such as mean, variance, and peak value, and may also include frequency-domain features and time-frequency-domain features.

[0082] After obtaining multiple features, data from different sensors are integrated through feature-level fusion or decision-level fusion to form a high-dimensional fusion feature vector representing the state of the monitored object, thereby improving the completeness and accuracy of the state representation.

[0083] Intelligent Analysis Module: Used to build dynamic baseline models based on historical normal operation data and real-time collected data.

[0084] When the monitored object is in different operating stages such as startup, stable operation and high load, the dynamic baseline model can be updated adaptively to adapt to the normal fluctuation range under different operating conditions.

[0085] In addition, the intelligent analysis module can perform recursive analysis of standardized data in batches and time windows according to this week, this month, this year or a custom time range, thereby identifying subtle abnormal trends at different time scales.

[0086] In a preferred embodiment, SLSQP is used to optimize the parameters of the anomaly analysis model, and a REM matrix is ​​established to organize, fuse, and infer anomaly feature points, thereby improving the ability to identify anomaly trends under complex working conditions.

[0087] Early warning push module: It is used to determine the severity of anomalies based on the degree of deviation, duration, and direction of evolution of abnormal trends, and to output at least three levels of early warning: notification, warning, and alarm.

[0088] Early warning information can be automatically pushed to the responsible person via App, SMS, email, etc., so that timely response measures can be taken.

[0089] Visualization module: Used to display the real-time status, historical data curves, early warning information, and diagnostic reports of monitored objects on web or mobile devices.

[0090] Managers can use this module to view the current status of equipment, abnormal changes, and analysis results, thereby improving decision-making efficiency.

[0091] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0092] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0093] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0094] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0095] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. An intelligent automated monitoring method based on the Internet of Things, characterized in that, Includes the following steps: S1. Collect the operating parameters and environmental parameters of the monitored target by using multi-source heterogeneous sensors deployed at the monitored target location; S2. The collected data is transmitted to the data processing platform via the network through the IoT gateway; S3. Perform timestamp synchronization and data alignment, outlier and missing value processing, noise filtering, and standardization or normalization on the collected data to obtain cleaned time-series data. S4. Extract time-domain and frequency-domain features from the cleaned time-series data, and generate a fused feature vector using feature-level fusion or decision-level fusion strategies. S5. Intelligent analysis, including: constructing a dynamic baseline model based on historical and real-time data, and identifying abnormal trends that deviate from the normal operation mode based on the fused feature vector; S6. Generate graded early warning information based on the severity of the abnormal trend and push it out. At the same time, display the monitoring results, historical data and early warning information on the visualization interface.

2. The intelligent automated monitoring method based on the Internet of Things according to claim 1, characterized in that, In step S3, outliers are identified using the 3σ principle, missing values ​​are filled using interpolation based on neighboring data points, the original signal is smoothed using a digital filter, and the processed data is standardized using the Z-score normalization method.

3. The intelligent automated monitoring method based on the Internet of Things according to claim 1, characterized in that, Filling in missing values ​​by interpolation includes the following steps: Step S3-2-1: Traverse the time series data, identify data with empty values ​​and data marked as outliers by the 3σ principle, and record the timestamp position t of missing and outlier data. m ; Step S3-2-2, for each missing position t m Search for the k nearest valid normal values ​​on the time axis, and take the forward nearest point t. prev and corresponding value x prev Take the next nearest neighbor: t next and corresponding value x next ; Step S3-2-3: Perform the nearest neighbor mean interpolation calculation, as shown in the following formula: ； Step S3-2-4, Boundary missing handling: If the beginning of the sequence is missing, the first valid value from the back is used to fill it; if the end of the sequence is missing, the last valid value from the front is used to fill it, so as to obtain time series data without missing or abnormal data.

4. The intelligent automated monitoring method based on the Internet of Things according to claim 1, characterized in that, In steps S4 and S5, the standardized time-series data are recursively identified and analyzed in batches and time windows according to this week, this month, this year, or a custom time range, in order to obtain abnormal change trends at different time scales.

5. The intelligent automated monitoring method based on the Internet of Things according to claim 4, characterized in that, Step S5 includes: Step S5-2-1, Historical Normal Baseline Initialization: Extract historical normal operating condition fusion feature data from multiple windows this week, this month, and this year, calculate the mean, standard deviation, and normal fluctuation range of each dimension of features, fit the initial baseline parameters of the equipment's standard operating state, and establish a benchmark model for normal equipment operation. Step S5-2-2, Real-time data dynamic correction: Input the fused feature vector collected in real time into the model, and combine it with the recursive analysis results of the custom time window to iteratively update the mean, fluctuation threshold and trend range of the baseline in real time, so that the baseline can be adaptively adjusted according to changes in equipment operating conditions and environmental conditions. Step S5-2-3, Dynamic Baseline Output: Form a dynamic baseline that takes into account both short-term stability and long-term adaptability, simulating the normal operation mode of the equipment under different time periods and operating conditions.

6. The intelligent automated monitoring method based on the Internet of Things according to claim 1, characterized in that, In step S5, sequential least squares quadratic programming (SLSQP) is used to optimize the parameters of the anomaly analysis model, and a REM matrix is ​​established to organize, fuse, and infer the analysis feature points in order to improve the accuracy of anomaly identification.

7. The intelligent automated monitoring method based on the Internet of Things according to claim 6, characterized in that, SLSQP was used to optimize the model parameters, and REM matrices were used to organize, fuse, and infer the feature points, including: Step S5-4-1, construct optimization objectives and constraints: take maximizing the anomaly identification accuracy and minimizing the false alarm rate and false negative rate as the objective function; set equality and inequality constraints in combination with sensor physical range, equipment operating condition limits, and feature value range; Step S5-4-2, Parameter initialization: Initialize the core parameters of the model, such as the anomaly threshold, deviation weight, and trend discrimination coefficient, and construct the optimal solution space for the parameters; Step S5-4-3, SLSQP Iterative Optimization: Based on the samples of multi-time window recursive analysis, the model parameters are corrected and converged to the optimal parameter combination by least squares fitting and quadratic programming iterative solution. Step S5-4-4, Optimal parameter update: After the iteration converges, replace the initial parameters of the model to complete the adaptive optimization of the anomaly analysis model.

Images