Information technology data monitoring device and use method thereof

Abstract

The invention relates to the field of information technology data monitoring, and discloses an information technology data monitoring device and a use method thereof. The device comprises a plurality of heterogeneous sensor modules which are distributed in a monitoring area to collect multiple types of data such as light intensity and electromagnetic signals. The data acquisition unit synchronizes original data, the central processing module generates a comprehensive monitoring index through a multi-source data fusion algorithm after preprocessing of the edge computing node, and the dynamic weight adjustment module adaptively optimizes a fusion weight. The anomaly detection engine calculates the anomaly probability by using an auto-encoder model, and the data transmission unit uploads data through a LoRa and 5G dual-mode communication protocol according to the network condition. In addition, the adaptive learning module periodically updates model parameters. During use, the device is deployed according to the process, data is collected and processed, abnormity is detected, and the model is updated. The system can comprehensively collect data, accurately analyze and process the data, efficiently transmit the data, continuously optimize the model, and improve the monitoring accuracy and reliability.

Description

Technical Field

[0001] The present invention relates to the field of information technology data monitoring, and in particular to an information technology data monitoring device and a use method thereof. Background Art

[0002] In today's digital age, information technology data monitoring is crucial to many fields and is widely used in industrial production, intelligent security, environmental monitoring, building health monitoring, etc. However, existing data monitoring technologies have many problems and are difficult to meet the growing complex monitoring needs.

[0003] In terms of sensor deployment, the types of sensors in traditional monitoring systems are relatively single, and they can often only monitor a specific type of data, and cannot fully reflect the comprehensive conditions of the monitoring area. For example, in an industrial production workshop, only monitoring a single parameter such as temperature or pressure makes it difficult to detect potential failure risks caused by a combination of multiple factors. Even if some systems use multiple sensors, their distribution lacks scientific planning, sensors interfere with each other, and there are blind spots in data collection, which seriously affects the accuracy and completeness of the data. Taking a large-scale storage environment as an example, if the temperature and humidity sensors are not distributed reasonably, the temperature and humidity changes in certain key areas may be missed, and the deterioration of goods caused by abnormal temperature and humidity cannot be discovered in time.

[0004] At the data processing and analysis level, the data formats, sampling frequencies, and accuracies collected by different sensors vary greatly, making multi-source data fusion difficult. The early simple weighted average fusion method was unable to fully tap the intrinsic connections between the data, and the fusion results could not accurately reflect the actual situation. In complex traffic flow monitoring scenarios, when fusing multi-source data such as vehicle speed, flow, and density, traditional methods find it difficult to effectively integrate information, resulting in deviations in the judgment of traffic congestion conditions. At the same time, existing anomaly detection technologies rely on manual experience to set thresholds, which cannot adapt to dynamically changing monitoring environments and are prone to false alarms or missed reports. In the operation monitoring of power equipment, since the operating conditions of the equipment change over time, the fixed threshold detection method cannot detect early signs of equipment failure in a timely manner.

[0005] In the communication transmission link, the stability and efficiency of data transmission also face challenges. In areas with unstable network coverage, such as remote mountainous areas or underground mines, the traditional single communication protocol cannot guarantee reliable data transmission. If only relying on communication methods such as 4G or WiFi, the signal is easily interrupted, resulting in a large amount of monitoring data loss, affecting subsequent analysis and decision-making. In addition, the transmission of a large amount of raw data not only consumes a lot of network bandwidth, but also increases the burden of data storage and processing, reducing the operating efficiency of the entire monitoring system.

[0006] In terms of model optimization, model updates of existing monitoring systems often require centralized collection of all data for training, which not only consumes a lot of time and resources, but also faces the risk of data privacy leakage. In the field of medical health monitoring, patients' personal health data involves privacy. Centralized data collection and training may lead to data leakage and cause serious security issues. Moreover, due to differences in the environment and data characteristics of different monitoring areas, a single global model is difficult to adapt to all scenarios, resulting in reduced monitoring accuracy. Summary of the invention

[0007] The object of the present invention is to provide an information technology data monitoring device and a method of using the same to solve the problems raised in the above background technology.

[0008] To achieve the above object, the present invention provides the following technical solution: an information technology data monitoring device, the device comprising:

[0009] Multiple heterogeneous sensor modules are distributed at different locations in the monitoring area, and the sensor modules include light intensity sensors, electromagnetic signal sensors, temperature and humidity sensors, and motion acceleration sensors;

[0010] The data acquisition unit is connected to each sensor module to obtain the raw data of each sensor in real time and synchronize the data with the timestamp;

[0011] An edge computing node, deployed in the monitoring area, receives the synchronization data from the data acquisition unit and performs multimodal signal preprocessing;

[0012] The central processing module receives the pre-processed data and generates comprehensive monitoring indicators through multi-source data fusion algorithms;

[0013] Dynamic weight adjustment module, which adaptively adjusts the fusion weight of multi-source data based on real-time data distribution characteristics;

[0014] An anomaly detection engine, which uses an autoencoder model to calculate the anomaly probability of the comprehensive monitoring indicators;

[0015] The data transmission unit uses LoRa and 5G dual-mode communication protocols to upload anomaly detection results and raw data to the cloud server.

[0016] Preferably, the distribution of the heterogeneous sensor modules is as follows: a three-dimensional coordinate system is established with the geometric center of the monitoring area as the origin, light intensity sensors are deployed on the surface of the monitoring area, electromagnetic signal sensors are embedded in the interior of the monitoring area at intervals, temperature and humidity sensors are evenly distributed at the junction of the surface and the interior, and motion acceleration sensors are fixed at key structural points of the monitoring area.

[0017] Preferably, the multimodal signal preprocessing comprises the following steps:

[0018] Perform discrete wavelet transform on the light intensity sensor data to extract high-frequency components as light intensity fluctuation characteristics;

[0019] Fast Fourier transform is used on electromagnetic signal sensor data to extract the main frequency amplitude and harmonic energy ratio;

[0020] The temperature and humidity sensor data is filtered using a sliding window mean filter to eliminate environmental noise;

[0021] A dynamic time warping algorithm is used on the motion acceleration sensor data to align the time series data with different sampling frequencies.

[0022] Preferably, the multi-source data fusion algorithm is specifically:

[0023] Definition Credibility weight of sensor-like data ,in For the Variance of class sensor data;

[0024] Perform weighted summation on the preprocessed multimodal features based on the credibility weights to generate an initial fusion feature vector;

[0025] The initial fused feature vector is input into the graph convolutional network, the spatial correlation between sensors is modeled through the adjacency matrix, and the spatiotemporal fusion features are output.

[0026] Preferably, the workflow of the dynamic weight adjustment module includes:

[0027] Calculate the KL divergence of each sensor data in real time and detect the data distribution offset;

[0028] When the KL divergence of a certain type of sensor data exceeds the preset threshold, its credibility weight is dynamically adjusted down according to the exponential decay function. The calculation formula is:

[0029]

[0030] in is the credibility weight after the decrease, is the attenuation coefficient, is the KL divergence, is the current data distribution, The historical benchmark distribution.

[0031] Preferably, the specific implementation of the anomaly detection engine is:

[0032] Construct a deep autoencoder network. The encoder part contains a 3-layer fully connected network, and the hidden layer activation function is LeakyReLU.

[0033] The decoder part adopts a symmetrical structure, and the loss function is defined as the weighted sum of the reconstruction error and the sparse constraint of the latent space. The calculation formula is:

[0034]

[0035] in is the latent vector output by the encoder, is the original feature vector of the input data, is the reconstructed feature vector output by the autoencoder, and is a hyperparameter;

[0036] When the reconstruction error at a certain moment exceeds three times the standard deviation of the historical error distribution, it is considered an abnormal event.

[0037] Preferably, the working mode of the data transmission unit is: when the network coverage in the monitoring area is stable, the 5G protocol is preferentially used to transmit the original data and abnormal labels; when the 5G signal strength is lower than -90dBm, it switches to the LoRa protocol and only transmits the compressed feature vector and position code of the abnormal event.

[0038] Preferably, it also includes an adaptive learning module for:

[0039] Regularly download the latest monitoring data from the cloud server and update the parameters of the autoencoder model;

[0040] A federated learning framework is used to aggregate local model gradients of multiple edge computing nodes, generate global model parameters, and then synchronize them to each node.

[0041] Preferably, the execution steps of the federated learning framework include:

[0042] Each edge node calculates the model gradient based on local data and adds differential privacy noise;

[0043] The central processing module aggregates gradients through a secure multi-party computing protocol and updates the global model;

[0044] The updated model parameters are encrypted and distributed to each edge node to replace the original model.

[0045] Preferably, the present invention also includes a method for using the information technology data monitoring device, comprising the following steps:

[0046] Deploy the above-mentioned information technology data monitoring device in the monitoring area and initialize the coordinate system and communication protocol of the sensor module;

[0047] The multimodal sensor data is synchronously collected at a preset frequency by a data acquisition unit and marked with time and space stamps;

[0048] Perform multimodal signal preprocessing at the edge computing node to generate standardized feature vectors; call the dynamic weight adjustment module to update the fusion weight according to the real-time data distribution;

[0049] The multi-source data fusion algorithm of the central processing module is used to generate comprehensive monitoring indicators, which are then input into the anomaly detection engine to calculate anomaly probabilities;

[0050] When the abnormal probability exceeds 0.95, the data transmission unit is triggered to upload the alarm information and related data to the cloud;

[0051] The global model parameters are downloaded through the adaptive learning module every month to complete the incremental update of the edge node model.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] The present invention realizes the comprehensive collection of multi-dimensional data of the monitoring area by deploying multiple heterogeneous sensor modules such as light intensity sensor, electromagnetic signal sensor, temperature and humidity sensor and motion acceleration sensor, and scientifically planning their distribution positions. A three-dimensional coordinate system is established with the geometric center of the monitoring area as the origin. The light intensity sensor is deployed on the surface layer, which can accurately capture the instantaneous changes in light intensity, which is of great significance for scenes such as agricultural greenhouse light monitoring and intelligent lighting system regulation; the electromagnetic signal sensor is embedded inside at intervals to effectively avoid interference and accurately monitor electromagnetic signals, playing a key role in electromagnetic radiation monitoring in areas with dense electronic equipment; the temperature and humidity sensors are evenly distributed at the surface layer and the internal junction, and the temperature and humidity changes are fully grasped to ensure the temperature and humidity stability of the storage environment and production workshop; the motion acceleration sensor is fixed at the key structural point to monitor the structural motion state in real time, which is indispensable in the safety monitoring of structures such as bridges and buildings. This multi-sensor collaborative and scientifically distributed method greatly improves the accuracy and comprehensiveness of data collection, and provides a rich and reliable data basis for subsequent analysis. In the data processing process, according to the characteristics of different sensor data, discrete wavelet transform, fast Fourier transform, sliding window mean filter and dynamic time warping algorithm are used for multimodal signal preprocessing to effectively extract features, eliminate noise and align time series data. The multi-source data fusion algorithm defines the credibility weight and combines the graph convolution network to fully consider the stability and spatial correlation of sensor data and generate accurate spatiotemporal fusion features. The dynamic weight adjustment module adaptively adjusts the fusion weight based on the real-time data distribution characteristics. When the data distribution changes, the weight of abnormal data is timely lowered to ensure the reliability of the fusion result. In industrial equipment fault monitoring, it can accurately integrate multi-source data such as vibration, temperature, and pressure, accurately judge the equipment operation status, and discover potential fault hazards in advance. Compared with traditional fusion methods, the accuracy of fault judgment is greatly improved, reducing equipment downtime and maintenance costs.

[0054] The anomaly detection engine builds a deep autoencoder network, and effectively learns the normal pattern of data by reasonably setting the encoder and decoder structure and loss function. When the reconstruction error exceeds 3 times the standard deviation of the historical error distribution, it is judged as an abnormal event. This model-based anomaly detection method is more accurate and adaptive than the traditional manual threshold setting. In intelligent security monitoring, it can timely detect abnormal behaviors, such as illegal intrusions, abnormal gatherings of people, etc., and quickly issue alarms to provide strong support for security prevention and reduce the risk of safety accidents. The data transmission unit adopts LoRa and 5G dual-mode communication protocols and switches intelligently according to network coverage. When the network coverage is stable, the high rate and low latency advantages of the 5G protocol are used to quickly upload the original data and abnormal labels; when the 5G signal strength is lower than -90dBm, it automatically switches to the LoRa protocol, only transmitting the compressed feature vector and position code of the abnormal event, reducing the amount of data transmission, and ensuring the stable transmission of data in harsh network environments. In scenarios such as environmental monitoring in remote areas and field operation data backhaul, data loss can be effectively avoided to ensure the continuity and integrity of monitoring data.

[0055] The adaptive learning module updates the autoencoder model parameters by regularly downloading the latest monitoring data from the cloud server, thereby improving the model's adaptability and accuracy to new data. Using the federated learning framework, each edge node calculates the model gradient based on local data and adds differential privacy noise. The central processing module aggregates the gradients through a secure multi-party computing protocol to update the global model, and then distributes the encrypted model parameters to each node. This approach not only achieves collaborative optimization of the model and improves monitoring accuracy, but also protects local data privacy. In areas with extremely high requirements for data privacy, such as financial risk monitoring and medical health data monitoring, it can not only make full use of data from all parties to improve monitoring results, but also ensure data security and meet strict privacy regulations.

[0056] The information technology data monitoring device of the present invention is simple and convenient to use, covering a series of processes including device deployment, data collection, preprocessing, fusion analysis, anomaly detection, data transmission, and model update. Multimodal sensor data is synchronously collected at a preset frequency and time and space stamps are marked to ensure the time and space consistency of the data; incremental updates are completed by downloading global model parameters through the adaptive learning module every month, so that the device always maintains good performance. In practical applications, it reduces the user's usage threshold and maintenance costs, improves the operating efficiency and reliability of the monitoring system, can be quickly deployed and applied to various monitoring scenarios, and provides efficient data monitoring solutions for different industries. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a working principle diagram of the information technology data monitoring device of the present invention;

[0058] Figure 2 This is the workflow diagram of the dynamic weight adjustment module;

[0059] Figure 3 It is the workflow diagram of the anomaly detection engine;

[0060] Figure 4 This is a step diagram of a method for using the information technology data monitoring device of the present invention. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] See also Figure 1-4 The present invention provides a technical solution: an information technology data monitoring device, which is mainly used to perform multi-dimensional data monitoring of the monitoring area and realize data processing, analysis and transmission. The device includes multiple heterogeneous sensor modules, a data acquisition unit, an edge computing node, a central processing module, a dynamic weight adjustment module, an anomaly detection engine, a data transmission unit and an adaptive learning module.

[0063] Multiple heterogeneous sensor modules are distributed in different locations in the monitoring area. These sensor modules include light intensity sensors, electromagnetic signal sensors, temperature and humidity sensors, and motion acceleration sensors. They are responsible for collecting different types of environmental data in the monitoring area and obtaining raw data from multiple dimensions such as light intensity, electromagnetic signals, temperature and humidity, and motion acceleration.

[0064] The data acquisition unit is connected to each sensor module. Its main function is to obtain the raw data of each sensor in real time and synchronize the timestamps of these data. Through timestamp synchronization, the data collected by different sensors are ensured to be consistent in the time dimension, which is convenient for subsequent data analysis and processing.

[0065] The edge computing nodes are deployed in the monitoring area and receive the data synchronized by the data acquisition unit after time stamp. The edge computing nodes perform multimodal signal preprocessing on these synchronized data and convert the raw data into standardized feature vectors that are easier to analyze and process.

[0066] The central processing module receives the data pre-processed by the edge computing node, and fuses the multi-modal data collected by different sensors through a multi-source data fusion algorithm to generate comprehensive monitoring indicators. Comprehensive monitoring indicators can more comprehensively and accurately reflect the overall status of the monitoring area.

[0067] The dynamic weight adjustment module adaptively adjusts the fusion weight of multi-source data based on the real-time data distribution characteristics. This module can reasonably adjust the importance of different sensor data in the fusion process according to the real-time changes of the data, making the fusion result more accurate.

[0068] The anomaly detection engine uses the autoencoder model to calculate the anomaly probability of the comprehensive monitoring indicators generated by the central processing module. By comparing the difference between current data and historical data, it determines whether there is an abnormal situation in the monitoring area.

[0069] The data transmission unit uses LoRa and 5G dual-mode communication protocols to upload abnormal detection results and raw data to the cloud server. In different network environments, choose the appropriate communication protocol to ensure that data can be uploaded stably and efficiently.

[0070] The adaptive learning module regularly downloads the latest monitoring data from the cloud server, updates the parameters of the autoencoder model, and improves the accuracy and adaptability of the model. At the same time, the federated learning framework is used to aggregate the local model gradients of multiple edge computing nodes, generate global model parameters, and synchronize them to each node to achieve collaborative optimization of the model.

[0071] The present invention will be further described below in conjunction with Examples 1 to 5:

[0072] Embodiment 1:

[0073] This embodiment clarifies the specific distribution mode of heterogeneous sensor modules in the monitoring area to improve the accuracy and comprehensiveness of data collection.

[0074] In practical applications, in order to ensure that the heterogeneous sensor modules can effectively collect data of different locations and types in the monitoring area, a specific distribution method is adopted. A three-dimensional coordinate system is established with the geometric center of the monitoring area as the origin. The establishment of this coordinate system provides a spatial reference basis for the precise deployment of sensors.

[0075] The light intensity sensor is deployed on the surface of the monitoring area. Because the light intensity changes significantly on the surface of the monitoring area, this deployment method can obtain real-time change data of light intensity to the greatest extent. For example, in outdoor monitoring scenes, the change of light intensity can be perceived in time, which is of great significance for some light-sensitive equipment or environmental monitoring.

[0076] Electromagnetic signal sensors are embedded in the monitoring area at intervals. Electromagnetic signals have certain penetration and complexity when propagating in space. The interval embedding method can avoid mutual interference between sensors, and at the same time, it can comprehensively monitor the electromagnetic signal conditions at different locations in the monitoring area. For example, in areas with dense electronic equipment, it can accurately monitor electromagnetic radiation intensity and other data.

[0077] Temperature and humidity sensors are evenly distributed at the junction of the surface and the interior. This position can monitor the changes in the temperature and humidity of the surface environment, and take into account the temperature and humidity of the internal environment. For some storage environments or production workshops with strict requirements on temperature and humidity, it can fully and accurately obtain temperature and humidity data to ensure the stability of the environment.

[0078] The motion accelerometer is fixed at the key structural point of the monitoring area. The key structural point is usually an important support point for the structural stability of the entire monitoring area. By fixing the motion accelerometer here, the motion state changes of the structural point can be monitored in time. In scenarios such as building monitoring and bridge monitoring, it plays an important role in the assessment of structural safety.

[0079] Embodiment 2:

[0080] This embodiment elaborates on the specific steps of multimodal signal preprocessing, performs targeted processing on data collected by different sensors, and provides more effective features for subsequent data fusion and analysis.

[0081] For the light intensity sensor data, discrete wavelet transform is performed to extract high-frequency components as light intensity fluctuation characteristics. Light intensity in the actual environment will be affected by many factors and cause fluctuations. Discrete wavelet transform can effectively decompose the light intensity signal into components of different frequencies. By extracting high-frequency components, the rapid changes in light intensity can be highlighted. For example, when monitoring areas with frequent lighting changes, such as stage lighting areas, high-frequency components can accurately reflect the instantaneous changes in light intensity, which is of great significance for lighting control and monitoring of related scenes.

[0082] For electromagnetic signal sensor data, fast Fourier transform is used to extract the main frequency amplitude and harmonic energy ratio. Electromagnetic signals contain multiple frequency components. Fast Fourier transform can convert electromagnetic signals in the time domain to the frequency domain, so as to clearly analyze its main frequency amplitude and harmonic energy ratio. In power system monitoring, by analyzing these characteristics, it can be determined whether the power equipment is operating normally. If the harmonic energy ratio is abnormal, it may mean that there is a hidden danger of equipment failure.

[0083] The temperature and humidity sensor data is easily affected by environmental noise, and the environmental noise can be eliminated by sliding window mean filtering. Sliding window mean filtering is a commonly used signal filtering method. It averages the temperature and humidity data within a certain time window to remove the interference of noise, making the temperature and humidity data smoother and more accurate. For example, in indoor environmental monitoring, it can more accurately reflect the real temperature and humidity conditions in the room.

[0084] Motion acceleration sensors may have inconsistent sampling frequencies due to different deployment locations and acquisition devices. A dynamic time warping algorithm is used to align time series data with different sampling frequencies. This algorithm can align motion acceleration data with different sampling frequencies on the time axis without changing the data characteristics, facilitating subsequent unified analysis of the motion state. In scenarios where multiple sensors collaborate to monitor moving objects, data consistency and comparability are ensured.

[0085] Embodiment 3:

[0086] In the multi-source data fusion algorithm, we first define Credibility weight of sensor-like data Here For the The variance of sensor data can reflect the degree of data dispersion. The smaller the variance, the higher the credibility of the data, and the greater the corresponding credibility weight. In this way, the importance of each sensor data in the fusion process is determined according to its stability.

[0087] The preprocessed multimodal features are weighted and summed based on the credibility weights to generate an initial fused feature vector. The preprocessed features of different sensors are weighted and added according to their respective credibility weights to obtain a preliminary fused feature vector that integrates the information of multiple sensor data.

[0088] The initial fusion feature vector is input into the graph convolutional network, and the spatial correlation between sensors is modeled through the adjacency matrix to output the spatiotemporal fusion features. The graph convolutional network can take into account the spatial position relationship between sensors and describe the connection relationship between sensors through the adjacency matrix. Therefore, in the fusion process, not only the characteristics of the data itself are considered, but also the spatial information can be integrated, so that the fusion result can more accurately reflect the actual situation of the monitoring area.

[0089] In the workflow of the dynamic weight adjustment module, the KL divergence of each sensor data is calculated in real time to detect the data distribution offset. KL divergence can measure the difference between two probability distributions. By calculating the KL divergence of the current data distribution and the historical benchmark distribution, it is possible to promptly detect whether the data distribution has shifted.

[0090] When the KL divergence of a certain type of sensor data exceeds the preset threshold, its credibility weight is dynamically adjusted down according to the exponential decay function. The calculation formula is: .

[0091] in is the credibility weight after the decrease, is the attenuation coefficient, which can be adjusted according to actual conditions. is the KL divergence, is the current data distribution, is the historical benchmark distribution. When the distribution of a certain type of sensor data changes significantly, its weight in the fusion process is reduced to avoid abnormal data from having too much impact on the fusion result and ensure the reliability of the fusion result.

[0092] Embodiment 4:

[0093] This embodiment describes in detail the specific implementation of the anomaly detection engine and the working mode of the data transmission unit to achieve accurate anomaly detection and efficient data transmission.

[0094] The anomaly detection engine builds a deep autoencoder network. The encoder part contains a 3-layer fully connected network, and the hidden layer activation function is LeakyReLU. The LeakyReLU function can effectively solve the gradient vanishing problem, allowing the network to better learn the characteristics of the data during training. The encoder extracts and compresses the input comprehensive monitoring indicator data and converts it into a low-dimensional potential vector.

[0095] The decoder part adopts a symmetrical structure, corresponding to the encoder, to reconstruct the latent vector into an approximate representation of the original data. The loss function is defined as the weighted sum of the reconstruction error and the sparse constraint of the latent space, and the calculation formula is: .in is the latent vector output by the encoder, is the original feature vector of the input data, is the reconstructed feature vector output by the autoencoder, and are hyperparameters. By adjusting these two hyperparameters, the importance of reconstruction error and latent space sparsity constraints can be balanced.

[0096] When the reconstruction error at a certain moment exceeds 3 times the standard deviation of the historical error distribution, it is judged as an abnormal event. This is based on statistical principles. Under normal circumstances, the reconstruction error of data should fluctuate within a certain range. When the reconstruction error exceeds 3 times the standard deviation, it means that there is a large difference between the current data and the historical data, and it is likely that an abnormal situation has occurred.

[0097] The working mode of the data transmission unit is to use the 5G protocol to transmit raw data and abnormal labels when the network coverage in the monitoring area is stable. The 5G protocol has the characteristics of high speed and low latency, and can quickly and accurately upload a large amount of raw data and abnormal labels to the cloud server, which is convenient for subsequent data analysis and processing.

[0098] When the 5G signal strength is lower than -90dBm, it means that the 5G network coverage is unstable, and the LoRa protocol is switched to at this time. The LoRa protocol has the advantages of low power consumption and long-distance transmission. In this case, only the compressed feature vector and position code of the abnormal event are transmitted, which reduces the amount of data transmission and ensures that the data can be successfully uploaded in a poor network environment.

[0099] Embodiment 5:

[0100] This embodiment describes in detail the working process of the adaptive learning module and the method of using the information technology data monitoring device to achieve continuous optimization of the model and effective operation of the device.

[0101] The adaptive learning module regularly downloads the latest monitoring data from the cloud server and updates the parameters of the autoencoder model. As time goes by and the monitoring environment changes, new data can provide more information for the model. By updating the parameters, the autoencoder model can better adapt to the new data distribution and improve the accuracy of anomaly detection.

[0102] The adaptive learning module uses a federated learning framework to aggregate the local model gradients of multiple edge computing nodes, generate global model parameters, and then synchronize them to each node. In the execution step of the federated learning framework, each edge node calculates the model gradient based on local data and adds differential privacy noise. Adding differential privacy noise can participate in model training and optimization while protecting local data privacy and prevent data leakage.

[0103] The central processing module aggregates the gradients through a secure multi-party computing protocol and updates the global model. The secure multi-party computing protocol ensures that the privacy of each party’s data is protected during the process of aggregating the gradients, while accurately calculating the update direction of the global model.

[0104] The updated model parameters are encrypted and distributed to each edge node to replace the original model. Encrypted distribution ensures the security of model parameters during transmission. Each edge node uses the updated model parameters to achieve collaborative optimization of the model and improve the performance of the entire monitoring system.

[0105] The method for using the information technology data monitoring device comprises the following steps: first, deploying the device of the present invention in the monitoring area, initializing the coordinate system and communication protocol of the sensor module, and ensuring that the sensor module can accurately collect data and communicate with other modules.

[0106] The data acquisition unit collects multimodal sensor data synchronously at a preset frequency and marks the time and space stamps. The preset frequency is set according to the actual monitoring needs, and the marking of time and space stamps ensures the temporal and spatial consistency of the data.

[0107] Perform multimodal signal preprocessing at the edge computing node to generate standardized feature vectors. The preprocessed data is more convenient for subsequent analysis and processing.

[0108] Call the dynamic weight adjustment module to update the fusion weight according to the real-time data distribution to improve the accuracy of multi-source data fusion.

[0109] The multi-source data fusion algorithm of the central processing module is used to generate comprehensive monitoring indicators, which are then input into the anomaly detection engine to calculate the anomaly probability. Through the calculation of comprehensive monitoring indicators and anomaly probabilities, it is determined whether there are abnormal conditions in the monitoring area.

[0110] When the abnormal probability exceeds 0.95, the data transmission unit is triggered to upload the alarm information and related data to the cloud. The abnormal situation is promptly notified to relevant personnel for processing.

[0111] The global model parameters are downloaded monthly through the adaptive learning module to complete the incremental update of the edge node model. The model is updated regularly to ensure that the performance of the monitoring device is always in good condition.

[0112] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0113] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An information technology data monitoring device, characterized in that: include: Multiple heterogeneous sensor modules are distributed at different locations in the monitoring area, and the sensor modules include light intensity sensors, electromagnetic signal sensors, temperature and humidity sensors, and motion acceleration sensors; The data acquisition unit is connected to each sensor module to obtain the raw data of each sensor in real time and synchronize the data with the timestamp; An edge computing node, deployed in the monitoring area, receives the synchronization data from the data acquisition unit and performs multimodal signal preprocessing; The central processing module receives the pre-processed data and generates comprehensive monitoring indicators through multi-source data fusion algorithms; Dynamic weight adjustment module, which adaptively adjusts the fusion weight of multi-source data based on real-time data distribution characteristics; An anomaly detection engine, which uses an autoencoder model to calculate the anomaly probability of the comprehensive monitoring indicators; The data transmission unit uses LoRa and 5G dual-mode communication protocols to upload abnormal detection results and raw data to the cloud server; The workflow of the dynamic weight adjustment module includes: Calculate the KL divergence of each sensor data in real time and detect the data distribution offset; When the KL divergence of a certain type of sensor data exceeds the preset threshold, its credibility weight is dynamically adjusted down according to the exponential decay function. The calculation formula is: ； in is the credibility weight after the decrease, is the attenuation coefficient, is the KL divergence, is the current data distribution, The historical benchmark distribution.

2. The device according to claim 1, characterized in that The distribution method of the heterogeneous sensor modules is specifically as follows: a three-dimensional coordinate system is established with the geometric center of the monitoring area as the origin, the light intensity sensor is deployed on the surface of the monitoring area, the electromagnetic signal sensor is embedded in the monitoring area at intervals, the temperature and humidity sensors are evenly distributed at the junction of the surface and the interior, and the motion acceleration sensor is fixed at the key structural point of the monitoring area.

3. The device according to claim 2, characterized in that The multimodal signal preprocessing comprises the following steps: Perform discrete wavelet transform on the light intensity sensor data to extract high-frequency components as light intensity fluctuation characteristics; Fast Fourier transform is used on electromagnetic signal sensor data to extract the main frequency amplitude and harmonic energy ratio; The temperature and humidity sensor data is filtered using a sliding window mean filter to eliminate environmental noise; A dynamic time warping algorithm is used on the motion acceleration sensor data to align the time series data with different sampling frequencies.

4. The device according to claim 3, characterized in that The multi-source data fusion algorithm is specifically as follows: Definition Credibility weight of sensor-like data ,in For the Variance of class sensor data; Perform weighted summation on the preprocessed multimodal features based on the credibility weights to generate an initial fusion feature vector; The initial fused feature vector is input into the graph convolutional network, the spatial correlation between sensors is modeled through the adjacency matrix, and the spatiotemporal fusion features are output.

5. The device according to claim 1, characterized in that The specific implementation of the anomaly detection engine is as follows: Construct a deep autoencoder network. The encoder part contains a 3-layer fully connected network, and the hidden layer activation function is LeakyReLU. The decoder part adopts a symmetrical structure, and the loss function is defined as the weighted sum of the reconstruction error and the sparse constraint of the latent space. The calculation formula is: ； in is the latent vector output by the encoder, is the original feature vector of the input data, is the reconstructed feature vector output by the autoencoder, and is a hyperparameter; When the reconstruction error at a certain moment exceeds three times the standard deviation of the historical error distribution, it is considered an abnormal event.

6. The device according to claim 1, characterized in that The working mode of the data transmission unit is: When the network coverage in the monitoring area is stable, the 5G protocol is used to transmit raw data and abnormal labels; When the 5G signal strength is lower than -90dBm, it switches to the LoRa protocol and only transmits the compressed feature vector and position code of the abnormal event.

7. The device according to claim 1, characterized in that Also includes adaptive learning modules for: Regularly download the latest monitoring data from the cloud server and update the parameters of the autoencoder model; A federated learning framework is used to aggregate local model gradients of multiple edge computing nodes, generate global model parameters, and then synchronize them to each node.

8. The device according to claim 7, characterized in that The execution steps of the federated learning framework include: Each edge node calculates the model gradient based on local data and adds differential privacy noise; The central processing module aggregates gradients through a secure multi-party computing protocol and updates the global model; The updated model parameters are encrypted and distributed to each edge node to replace the original model.

9. A method for using an information technology data monitoring device, characterized in that: The following steps are involved: Deploy the device as described in any one of claims 1 to 8 in the monitoring area, and initialize the coordinate system and communication protocol of the sensor module; The multimodal sensor data is synchronously collected at a preset frequency by a data acquisition unit and marked with time and space stamps; Perform multimodal signal preprocessing at the edge computing node to generate standardized feature vectors; Call the dynamic weight adjustment module to update the fusion weight according to the real-time data distribution; The multi-source data fusion algorithm of the central processing module is used to generate comprehensive monitoring indicators, which are then input into the anomaly detection engine to calculate anomaly probabilities; When the abnormal probability exceeds 0.95, the data transmission unit is triggered to upload the alarm information and related data to the cloud; The global model parameters are downloaded through the adaptive learning module every month to complete the incremental update of the edge node model.

Images