Intelligent monitoring and prediction analysis platform based on Internet of Things and big data
By employing a multi-module collaborative sensing layer, a redundant transmission layer, and an improved prediction model, the problems of compatibility, transmission efficiency, and prediction accuracy of the monitoring system are solved, achieving high-precision, stable, and adaptive intelligent monitoring and predictive analysis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-15
- Publication Date
- 2026-04-07
AI Technical Summary
Existing monitoring systems suffer from poor compatibility at the perception layer, low efficiency at the transmission layer, high data processing latency, and weak prediction accuracy, making it impossible to achieve refined management.
A multi-module collaborative perception layer is used for high-precision data acquisition and preprocessing, a dual-mode redundant transmission layer ensures stable data transmission, a data layer provides efficient storage and quality control, and an improved Attention-LSTM model is introduced into the application layer for accurate prediction.
It achieves high-precision data acquisition, stable transmission, rapid processing, and highly reliable prediction, thereby improving the overall accuracy and adaptability of the monitoring system.
Smart Images
Figure CN121814788A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of interdisciplinary technologies of the Internet of Things (IoT), big data processing, and intelligent prediction, specifically to an intelligent monitoring and predictive analysis platform based on IoT and big data. Background Technology
[0002] With the deep integration of industrialization and informatization, various industries have placed higher demands on the real-time performance, data accuracy, and predictive foresight of monitoring systems. Existing traditional monitoring systems face several technical bottlenecks: (i) The compatibility of sensing layer devices is poor, and they mostly use a single sensor or a combination of similar sensors, resulting in limited data acquisition dimensions. Affected by factors such as temperature drift and environmental interference, the measurement accuracy is low. For example, in industrial vibration monitoring, the error of a single sensor often exceeds ±2Hz, which cannot meet the needs of early fault warning for equipment.
[0003] (ii) The transmission layer relies on a single communication method. 4G communication has problems such as high energy consumption and poor signal coverage in remote areas. Wired communication limits deployment flexibility and the packet loss rate is generally as high as 10% or more.
[0004] (iii) The data processing layer is mostly in stand-alone processing mode, lacking efficient fusion and cleaning mechanisms. Redundant and abnormal data lead to increased processing delays.
[0005] (iv) The prediction layer uses traditional algorithms such as ARIMA and SVM, which have weak fitting ability for nonlinear time series data such as equipment vibration and pollutant concentration, resulting in large short-term prediction errors and making it difficult to achieve "early warning".
[0006] The aforementioned problems mean that traditional monitoring systems can only achieve "post-event monitoring" and cannot provide effective support for refined management. There is an urgent need for an intelligent monitoring and prediction solution that is optimized across the entire chain. Summary of the Invention
[0007] To address the problems mentioned in the background section, the present invention aims to provide an intelligent monitoring and predictive analysis platform based on the Internet of Things and big data, which has the advantages of high data acquisition accuracy, high transmission efficiency, high prediction reliability, and strong compatibility.
[0008] To achieve the above objectives, the present invention provides the following technical solution: an intelligent monitoring and predictive analysis platform based on the Internet of Things and big data, comprising: The perception layer includes a sensor access module, a data preprocessing submodule, an edge computing submodule, a sensor health monitoring module, and a sampling frequency adaptive module. The perception layer achieves high-precision and adaptive data acquisition preprocessing through the collaboration of multiple modules. The edge computing submodule completes accurate data processing based on error calibration and weighted fusion algorithms, while performing local anomaly detection and uploading only key data to reduce transmission pressure. The transmission layer includes a LoRa communication submodule, a 4G / 5G backup submodule, a communication optimization submodule, a traffic control module, and a breakpoint resume module. The transmission layer constructs a dual-mode redundant and highly efficient and controllable communication system. The data layer includes a distributed storage submodule, a real-time computing submodule, a data quality control submodule, a data traceability module, and a cold / hot data stratification module. The data layer is responsible for efficient data storage, accurate processing, and full lifecycle management. The data quality control submodule ensures that data quality meets analysis standards through outlier detection and data integrity verification. The application layer includes a prediction model submodule, a visualization interaction submodule, an intelligent alarm submodule, a model operation and maintenance submodule, an operation and maintenance decision recommendation module, an energy consumption optimization module, and a multi-scenario adaptation module. The application layer focuses on data value transformation and personalized service output. The prediction model submodule is based on an improved Attention-LSTM model to accurately realize the prediction and analysis of time series data such as equipment status and environmental changes.
[0009] As a preferred embodiment of the present invention, the edge computing submodule of the perception layer adopts an STM32H743 microcontroller.
[0010] As a preferred embodiment of the present invention, the error calibration of the edge computing submodule of the sensing layer adopts a method combining linear calibration and nonlinear compensation, and the calculation formula for the calibrated sensor measurement value is as follows: Where \(k\) and \(b\) are linear calibration parameters, and the nonlinear compensation function is: + The multi-sensor data fusion uses a weighted average algorithm, and the fused monitoring values are: The formula for calculating the weighting coefficient is: In the formula, \(\sigma_i\) represents the sensor measurement accuracy index.
[0011] Preferably, the LoRa communication submodule of the transmission layer supports SF5-SF12 spreading factors and adjustable bandwidth of 125kHz / 250kHz / 500kHz. The 4G / 5G backup submodule serves as a backup link and automatically switches over when the LoRa communication packet loss rate exceeds 5%. The LoRa communication efficiency calculation formula is as follows: The genetic algorithm is used to optimize the SF, BW, and CR parameters to maximize \(\eta\), where the transmission rate is calculated using the following formula: .
[0012] As a preferred embodiment of the present invention, the real-time computing submodule quickly completes data cleaning, multi-source fusion and feature extraction, providing high-quality data support for predictive analysis. The extracted time series features include mean, variance and peak value. The data processing adopts an incremental update mechanism to reduce the occupation of storage and computing resources by redundant data.
[0013] As a preferred embodiment of the present invention, the Attention-LSTM model of the application layer introduces an attention layer on top of LSTM, and the relevant formula for calculating the attention weights is as follows: The model is optimized using the Huber loss function, which is calculated using the following formula: Where \(\delta=1.0\), and the training effect is improved by using the Adam optimizer and cosine annealing learning rate strategy.
[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention employs a multi-module collaborative sensing layer to achieve high-precision adaptive data acquisition and preprocessing. It supports multiple sensor types for rapid expansion of acquisition dimensions. Data preprocessing and edge computing improve data quality and reduce transmission pressure. Sensor health monitoring avoids interference from erroneous data. Adaptive sampling frequency balances monitoring accuracy and energy consumption. The transmission layer features dual-mode redundant communication to ensure stable and efficient data transmission. Communication optimization and flow control improve transmission efficiency, and breakpoint resume ensures data integrity. The data layer provides efficient storage and full lifecycle management to safeguard data value. The application layer offers accurate prediction and personalized service output, comprehensively enhancing monitoring and prediction capabilities and scenario adaptability. This device boasts advantages such as high data acquisition accuracy, high transmission efficiency, high prediction reliability, and strong compatibility.
[0015] 2. This invention utilizes a specific microcontroller in the edge computing submodule of the perception layer to provide stable and reliable hardware support for accurate data processing. This facilitates the efficient operation of error calibration and weighted fusion algorithms, ensuring accurate and efficient local anomaly detection. It can quickly complete data processing tasks and filter key data for uploading, effectively reducing data transmission pressure. At the same time, it ensures the timeliness and accuracy of data processing, providing a high-quality data foundation for subsequent data storage, analysis, and prediction, and further improving the overall platform's operating efficiency and monitoring and prediction reliability. Attached Figure Description
[0016] Figure 1 This is a system diagram of the present invention; Figure 2 This is a system diagram of the sensing layer of the present invention; Figure 3 This is a diagram of the transport layer system of the present invention; Figure 4 This is a data layer system diagram of the present invention; Figure 5 This is a system diagram of the application layer of the present invention; Figure 6 This is a graph of the piecewise loss function of the present invention. Detailed Implementation
[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0018] like Figures 1 to 6 As shown, an intelligent monitoring and predictive analysis platform based on the Internet of Things and big data includes: The perception layer comprises a sensor access module, a data preprocessing submodule, an edge computing submodule, a sensor health monitoring module, and a sampling frequency adaptive module. Through multi-module collaboration, the perception layer achieves high-precision, adaptive data acquisition and preprocessing. The sensor access module supports hot-swapping and compatible access of multiple types and manufacturers of sensors, allowing for rapid expansion of acquisition dimensions without additional adaptation. The data preprocessing submodule filters, denoises, and standardizes the format of the raw data acquired by the sensors, improving the data signal-to-noise ratio and consistency. The edge computing submodule performs accurate data processing based on error calibration and weighted fusion algorithms, while also performing local anomaly detection, uploading only critical data to reduce transmission pressure. The sensor health monitoring module tracks the working status of each sensor in real time, promptly identifying and marking anomalies such as drift and malfunctions to prevent erroneous data from flowing into subsequent processes. The sampling frequency adaptive module dynamically adjusts the acquisition frequency according to data fluctuations, increasing the frequency to ensure monitoring accuracy when data fluctuations are severe and decreasing the frequency to save energy when data is stable. The transport layer includes a LoRa communication submodule, a 4G / 5G backup submodule, a communication optimization submodule, a flow control module, and a breakpoint resume module. The transport layer constructs a dual-mode redundant and highly efficient controllable communication system. The LoRa communication submodule serves as the main link to achieve low-power, long-distance data transmission, adapting to the wide coverage requirements of industrial scenarios. The 4G / 5G backup submodule automatically switches when the main link fails, serving as a redundant link to ensure uninterrupted data transmission. The communication optimization submodule dynamically optimizes parameters such as the spread factor and bandwidth of LoRa communication through a genetic algorithm to maximize communication efficiency. The flow control module performs hierarchical management of data transmission, limiting the amount of non-critical data transmission to avoid bandwidth congestion affecting core data transmission. The breakpoint resume module temporarily stores the data to be transmitted locally when communication is interrupted and automatically retransmits it after the connection is restored, ensuring the integrity of data transmission. The data layer comprises a distributed storage submodule, a real-time computing submodule, a data quality control submodule, a data traceability module, and a cold / hot data tiering module. The data layer is responsible for efficient data storage, accurate processing, and full lifecycle management. The distributed storage submodule employs a tiered storage strategy, storing raw data and structured data via HDFS and HBase respectively, supporting petabyte-level capacity and millisecond-level read / write speeds. The data quality control submodule ensures data quality meets analytical standards through outlier detection and data integrity verification. The data traceability module assigns a unique identifier to each data entry, recording the entire process from data collection source, processing steps, and storage location, achieving end-to-end data traceability. The cold / hot data tiering module stores frequently accessed hot data on high-speed devices to ensure access speed, while archiving infrequently accessed cold data to low-cost storage media, balancing storage performance and cost. The application layer comprises a prediction model submodule, a visualization and interaction submodule, an intelligent alarm submodule, a model operation and maintenance submodule, an operation and maintenance decision recommendation module, an energy consumption optimization module, and a multi-scenario adaptation module. The application layer focuses on data value transformation and personalized service output. The prediction model submodule, based on an improved Attention-LSTM model, accurately predicts and analyzes time-series data such as equipment status and environmental changes. The visualization and interaction submodule displays monitoring data, prediction results, and equipment status in intuitive charts and dashboards, while also supporting user-defined parameter configurations. The intelligent alarm submodule alerts users when predicted values exceed limits or anomalies are detected. The system provides real-time alerts through multiple channels such as SMS and app push notifications to remind relevant personnel to handle issues promptly. The model operation and maintenance submodule updates model parameters regularly through an incremental training mechanism, allowing it to quickly adapt to scenario changes with only a small amount of new data. The operation and maintenance decision recommendation module combines prediction results with the operation and maintenance knowledge base to generate specific and actionable equipment maintenance suggestions to guide actual operation and maintenance work. The energy consumption optimization module analyzes the correlation between equipment operating parameters and energy consumption, recommending low-energy-consumption operating solutions that meet production needs. The multi-scenario adaptation module has built-in preset configurations for industries such as industry, environment, and transportation, supporting one-click scenario switching and quickly adapting to different application requirements without complex debugging.
[0019] refer to Figure 1 The edge computing submodule of the perception layer uses an STM32H743 microcontroller.
[0020] As a technical optimization of this invention, the edge computing submodule of the perception layer adopts a specific microcontroller to provide stable and reliable hardware support for accurate data processing. This facilitates the efficient operation of error calibration and weighted fusion algorithms, ensures accurate and efficient local anomaly detection, and enables the rapid completion of data processing tasks and the uploading of key data. This effectively reduces data transmission pressure while ensuring the timeliness and accuracy of data processing. It provides a high-quality data foundation for subsequent data storage, analysis, and prediction, further improving the overall platform's operating efficiency and monitoring and prediction reliability.
[0021] refer to Figure 2 The edge computing submodule error calibration of the perception layer adopts a combination of linear calibration and nonlinear compensation. The formula for calculating the calibrated sensor measurement value is as follows: Where \(k\) and \(b\) are linear calibration parameters, and the nonlinear compensation function is: + The multi-sensor data fusion uses a weighted average algorithm, and the fused monitoring values are: The formula for calculating the weighting coefficient is: In the formula, \(\sigma_i\) is the sensor measurement accuracy index, where k is the linear calibration coefficient, which is obtained by fitting the sensor calibration experiment. It has no fixed value and needs to be determined by actual measurement according to the sensor type (such as temperature and vibration sensors). The range is usually 0.95-1.05. b is the linear calibration offset, which is the fitting parameter of the calibration experiment. The unit is consistent with the sensor measurement value (such as ℃ for temperature sensors and Pa for pressure sensors). It is used to correct the fixed error of the system. Temperature interference error, in °C, refers to the measurement deviation caused when the ambient temperature deviates from the standard calibration temperature. Humidity interference error, in %RH (relative humidity), refers to the sensor measurement error caused by changes in ambient humidity. : Air pressure interference error, in kPa, is the effect of atmospheric pressure fluctuations on the sensor measurement results.
[0022] As a technical optimization scheme of the present invention, the edge computing submodule employs an error calibration method that combines linear calibration and nonlinear compensation, which significantly improves the accuracy of sensor measurement data. The weighted fusion algorithm ensures the reliability of multi-sensor data fusion and reduces the impact of data deviation on subsequent analysis. Through precise data processing and local anomaly detection, invalid data is effectively filtered out, reducing transmission costs and preventing erroneous data from flowing into subsequent processes. This provides the platform with accurate and reliable data support, improves the overall monitoring accuracy and predictive analysis accuracy, and enhances the platform's practicality and credibility in real-world applications.
[0023] refer to Figure 1 The LoRa communication submodule in the transport layer supports SF5-SF12 spreading factors and adjustable bandwidth of 125kHz / 250kHz / 500kHz. The 4G / 5G backup submodule serves as a backup link, automatically switching when the LoRa communication packet loss rate exceeds 5%. The LoRa communication efficiency calculation formula is as follows: The genetic algorithm is used to optimize the SF, BW, and CR parameters to maximize \(\eta\), where the transmission rate is calculated using the following formula: .
[0024] As a technical optimization scheme of this invention, the LoRa communication submodule of the transmission layer supports multi-parameter adjustment, and the genetic algorithm of the communication optimization submodule dynamically optimizes the parameters to maximize communication efficiency and meet the data transmission needs in different scenarios. The 4G / 5G backup submodule ensures link redundancy and ensures that data transmission is not interrupted when the main link is abnormal. The flow control module reasonably allocates bandwidth resources to avoid congestion affecting core data transmission. The breakpoint resume module ensures the integrity of data transmission. Overall, the stability, efficiency and integrity of data transmission are improved, providing a strong guarantee for the real-time transmission and reliable storage of platform data. Among them, η is the LoRa communication efficiency, which is dimensionless and ranges from 0 to 1. The larger the value, the more efficient the communication resource utilization. β is the spreading factor influence coefficient, which is dimensionless and has a fixed value of 0.15. It is used to quantify the influence weight of the spreading factor on link attenuation. SF is the spreading factor, which is dimensionless and ranges from 5 to 12 (the document supports SF5-SF12). It is used to improve the signal anti-interference capability. γ is the basic attenuation constant, which is dimensionless and has a fixed value of 2.3. It is determined by the characteristics of the LoRa communication protocol itself.
[0025] refer to Figure 1 The real-time computing submodule quickly completes data cleaning, multi-source fusion, and feature extraction, providing high-quality data support for predictive analysis. The extracted time series features include mean, variance, and peak value. The data processing adopts an incremental update mechanism to reduce the occupation of storage and computing resources by redundant data.
[0026] As a technical optimization of this invention, the real-time computing submodule of the data layer rapidly completes data cleaning, multi-source fusion, and feature extraction through an efficient data processing mechanism, providing high-quality data support for predictive analysis and effectively improving the analytical efficiency and accuracy of the predictive model. The incremental update mechanism reduces the occupation of storage and computing resources by redundant data, lowers the platform's operating costs, and ensures the timeliness and efficiency of data processing. This helps the data layer better fulfill its functions of data storage, processing, and lifecycle management, providing a solid data foundation for the stable operation and efficient decision-making of the entire platform.
[0027] refer to Figure 6 The application layer Attention-LSTM model introduces an attention layer on top of LSTM. The formula for calculating attention weights is as follows: The model is optimized using the Huber loss function, which is calculated using the following formula: Where \(\delta=1.0\), and the training effect is improved by using the Adam optimizer and cosine annealing learning rate strategy. Let be the attention score at time t, which is dimensionless and represents the importance of the hidden state at time t. The attention layer weight matrix has dimensions of (hidden layer dimension × hidden layer dimension) and is obtained through optimization during the model training process. Attention layer bias vector, with dimensions (1 × hidden layer dimension), used for model training optimization parameters; : Attention weight at time t, dimensionless, ranging from 0 to 1, with a total of 1, used to weight and highlight key time series data; T: Time series length, dimensionless, the total number of time steps of the model input data; y: True value, with units consistent with the prediction target (e.g., equipment vibration amplitude m / s², pollutant concentration mg / m³).
[0028] As a technical optimization scheme of this invention, an attention layer is introduced by applying the Attention-LSTM model, which can accurately capture key information in time series data and improve the accuracy and reliability of data prediction and analysis such as device status and environmental changes. The Huber loss function optimizes model training, enhances the robustness of the model to abnormal data, and combined with the Adam optimizer and cosine annealing learning rate strategy, improves the model training effect and convergence speed. The incremental training mechanism of the model operation and maintenance submodule allows the model to quickly adapt to scene changes, ensuring the long-term stable and efficient operation of the prediction model, providing strong support for the platform's accurate prediction and intelligent decision-making, and enhancing the platform's practical value and market competitiveness.
[0029] The working principle and usage process of this invention are as follows: During use, the perception layer starts running, and the sensor access module automatically identifies and is compatible with connected sensors of various types and manufacturers, expanding the data acquisition dimensions without additional adaptation. Subsequently, the data preprocessing submodule filters, denoises, and standardizes the format of the raw data collected by the sensors, improving the signal-to-noise ratio and consistency. Simultaneously, the sensor health monitoring module tracks the working status of each sensor in real time, promptly identifying and marking anomalies such as drift and malfunctions to prevent erroneous data from flowing into subsequent processes. The sampling frequency adaptive module dynamically adjusts the acquisition frequency based on data fluctuations, ensuring a balance between monitoring accuracy and energy consumption. The edge computing submodule performs accurate data processing based on error calibration and weighted fusion algorithms, while also performing local anomaly detection, uploading only critical data to the transmission layer.
[0030] In the transport layer, the LoRa communication submodule serves as the main link for low-power, long-distance data transmission. The communication optimization submodule dynamically optimizes LoRa communication parameters using a genetic algorithm to maximize communication efficiency. The flow control module implements hierarchical management of data transmission to prevent bandwidth congestion from affecting core data transmission. If the LoRa communication link malfunctions, the 4G / 5G backup submodule automatically switches to the redundant link to ensure uninterrupted data transmission. If an interruption occurs during communication, the breakpoint resumption module temporarily stores the data to be transmitted locally and automatically retransmits it after the connection is restored to ensure data integrity. After receiving data from the transport layer, the data layer uses a hierarchical storage strategy to store raw data and structured data separately. The data quality control submodule ensures data quality through outlier detection and data integrity verification. The data traceability module assigns a unique identifier to each piece of data and records relevant information throughout the process to achieve end-to-end traceability. The cold / hot data stratification module stores frequently accessed hot data on high-speed devices and archives infrequently accessed cold data on low-cost storage media. The real-time computing submodule uses an incremental update mechanism to quickly complete data cleaning, multi-source fusion, and feature extraction, providing high-quality data support for the application layer.
[0031] The application layer prediction model submodule uses an improved Attention-LSTM model to accurately predict and analyze time series data such as equipment status and environmental changes. The visualization and interaction submodule displays monitoring data, prediction results and equipment status in an intuitive way and supports user-defined parameter configuration. When predicted values exceed limits or anomalies are detected, the intelligent alarm submodule provides real-time alerts through multiple channels to relevant personnel for handling. The operation and maintenance decision recommendation module combines prediction results with the operation and maintenance knowledge base to generate equipment maintenance suggestions to guide actual operation and maintenance work. The energy consumption optimization module analyzes the correlation between equipment operating parameters and energy consumption and recommends low-energy operation solutions. The multi-scenario adaptation module allows users to switch preset industry configurations with one click to quickly adapt to different application needs. At the same time, the model operation and maintenance submodule updates model parameters regularly through an incremental training mechanism to ensure that the model can quickly adapt to changes in scenarios. The 4G / 5G backup submodule uses EC20 industrial-grade modules, supports multiple frequency bands, has a maximum downlink rate of 1Gbps, and automatically switches when the LoRa packet loss rate exceeds 5%, with a switching time of ≤500ms. The application layer is based on the Attention-LSTM prediction model, with supporting data visualization, alarm, and model update modules. The prediction model is deployed on a server equipped with an NVIDIA Tesla V100 GPU, uses the TensorFlow framework, introduces a dual attention layer to highlight key data, and improves robustness through the Huber loss function.
[0032] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0033] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A smart monitoring and predictive analysis platform based on the Internet of Things and big data, characterized in that, include: The perception layer includes a sensor access module, a data preprocessing submodule, an edge computing submodule, a sensor health monitoring module, and a sampling frequency adaptive module. The perception layer achieves high-precision and adaptive data acquisition preprocessing through the collaboration of multiple modules. The edge computing submodule completes accurate data processing based on error calibration and weighted fusion algorithms, while performing local anomaly detection and uploading only key data to reduce transmission pressure. The transmission layer includes a LoRa communication submodule, a 4G / 5G backup submodule, a communication optimization submodule, a traffic control module, and a breakpoint resume module. The transmission layer constructs a dual-mode redundant and highly efficient and controllable communication system. The data layer includes a distributed storage submodule, a real-time computing submodule, a data quality control submodule, a data traceability module, and a cold / hot data stratification module. The data layer is responsible for efficient data storage, accurate processing, and full lifecycle management. The data quality control submodule ensures that data quality meets analysis standards through outlier detection and data integrity verification. The application layer includes a prediction model submodule, a visualization interaction submodule, an intelligent alarm submodule, a model operation and maintenance submodule, an operation and maintenance decision recommendation module, an energy consumption optimization module, and a multi-scenario adaptation module. The application layer focuses on data value transformation and personalized service output. The prediction model submodule is based on an improved Attention-LSTM model to accurately realize the prediction and analysis of time series data such as equipment status and environmental changes.
2. The intelligent monitoring and predictive analysis platform based on the Internet of Things and big data according to claim 1, characterized in that: The edge computing submodule of the perception layer uses an STM32H743 microcontroller.
3. The intelligent monitoring and predictive analysis platform based on the Internet of Things and big data according to claim 2, characterized in that: The edge computing submodule error calibration of the sensing layer adopts a combination of linear calibration and nonlinear compensation. The formula for calculating the calibrated sensor measurement value is as follows: Where \(k\) and \(b\) are linear calibration parameters, and the nonlinear compensation function is: + The multi-sensor data fusion uses a weighted average algorithm, and the fused monitoring values are: The formula for calculating the weighting coefficient is: In the formula, \(\sigma_i\) represents the sensor measurement accuracy index.
4. The intelligent monitoring and predictive analysis platform based on the Internet of Things and big data according to claim 3, characterized in that: The LoRa communication submodule of the transmission layer supports SF5-SF12 spreading factors and adjustable bandwidth of 125kHz / 250kHz / 500kHz. The 4G / 5G backup submodule serves as a backup link and automatically switches over when the LoRa communication packet loss rate exceeds 5%. The LoRa communication efficiency calculation formula is as follows: The genetic algorithm is used to optimize the SF, BW, and CR parameters to maximize \(\eta\), where the transmission rate is calculated using the following formula: 。 5. The intelligent monitoring and predictive analysis platform based on the Internet of Things and big data according to claim 4, characterized in that: The real-time computing submodule quickly completes data cleaning, multi-source fusion, and feature extraction, providing high-quality data support for predictive analysis. The extracted time series features include mean, variance, and peak value. The data processing adopts an incremental update mechanism to reduce the occupation of storage and computing resources by redundant data.
6. The intelligent monitoring and predictive analysis platform based on the Internet of Things and big data according to claim 5, characterized in that: The Attention-LSTM model in the application layer introduces an attention layer on top of LSTM. The formula for calculating the attention weights is as follows: The model is optimized using the Huber loss function, which is calculated using the following formula: Where \(\delta=1.0\), and the training effect is improved by using the Adam optimizer and cosine annealing learning rate strategy.