An internet of things multi-sensor fusion intelligent monitoring system
Patent Information
- Application Number
- CN202610971520.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-01
- Publication Date
- 2026-09-29
AI Technical Summary
在监测过程中，当单个传感器发生漂移时，其异常读数会被误判为设备状态变化，引发虚假报警；当多个传感器同时漂移时，真实故障信号可能被掩盖，导致漏报；
该一种物联网多传感器融合的智能监测系统，通过对各传感器的原始数据获取，依据监测物理过程的守恒特征建立约束方程，识别发生数值漂移的传感器并实施在线修正，输出修正后传感器数据流的时序数据，实现传感器漂移的在线精准识别与自校准，有效区分传感器故障与物理过程异常，通过对传感器数据流的时序数据进行时间轴反转处理，计算原始时序与反转时序之间的波形相似度作为时间非对称指数，以全局非对称指数分析整个监测系统的不可逆退化程度，以局部平均非对称指数定位异常热点区域，以非对称指数场的梯度方向指示异常传播的主流方向，通过全局非对称指数与局部偏异方向的双尺度输出，实现对传感器监测系统整体健康状态以及可逆性退化评估目的，以各传感器节点为顶点构建动态拓扑网络，并将全局非对称指数注入各顶点的状态演化方程，当全局非对称指数突破临界阈值时，触发异常预警信号，提高告警时的准确性，计算各传感器数据流之间的传递熵，构建因果影响有向图，识别异常传播的源发节点与响应节点，将识别出的源发节点信息回传至数据校准模块，对源发节点的传感器实施优先校验与加速校准，实现传递熵因果分析实现异常源点精准定位，提高源头定位的准确性。
Smart Images
Figure CN122839256A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of sensor data fusion technology, and in particular to an intelligent monitoring system for multi-sensor fusion in the Internet of Things. Background Technology
[0002] The IoT multi-sensor fusion monitoring system achieves status monitoring and anomaly early warning for targets such as industrial equipment, pipeline networks, and large structures by collaboratively sensing and integrating information from various types of sensors deployed in the monitoring area. It is a key technology in the construction of industrial internet and smart facilities. Multi-sensor fusion monitoring technology is the core means to realize equipment status perception and early fault warning. By deploying various types of sensors such as temperature, pressure, vibration, and flow on the monitoring section, the system can capture physical characteristic parameters of equipment operation from different dimensions, providing a data foundation for status assessment and anomaly diagnosis. In recent years, with the advancement of edge computing and intelligent algorithms, multi-sensor fusion monitoring technology has made certain progress. Existing technologies mainly focus on consistency verification methods based on redundant sensors, parameter inversion methods based on physical models, and anomaly detection methods based on data-driven approaches. Existing IoT multi-sensor fusion monitoring systems mainly suffer from the following technical shortcomings: During monitoring, when a single sensor drifts, its abnormal readings may be misinterpreted as changes in equipment status, triggering false alarms; when multiple sensors drift simultaneously, the true fault signal may be masked, leading to missed alarms. Existing monitoring systems only focus on whether the signal amplitude exceeds a threshold, while ignoring the temporal directional characteristics of the signal waveform. The amplitude threshold method can only respond when the fault has developed to a certain extent and the signal strength is sufficient to trigger an alarm. It cannot provide early warning when the signal amplitude change is not yet significant, but the temporal symmetry of the waveform has already begun to be broken.
[0003] To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention
[0004] The present invention adopts the following technical solution: an intelligent monitoring system for multi-sensor fusion of Internet of Things, including a data calibration module, which is used to acquire the raw data of each sensor, establish constraint equations based on the conservation characteristics of the monitored physical process, identify sensors that have experienced numerical drift and implement online correction, and output the time-series data of the corrected sensor data stream; The irreversible situational awareness module is used to reverse the timeline of the time data of each pair of sensor data streams, calculate the waveform similarity between the original time series and the reversed time series as the time asymmetry index, extract the global asymmetry index and local deviation direction, and output the degree of reversibility degradation. The group state evolution module is used to construct a dynamic topology network with each sensor node as a vertex and inject the global asymmetric index into the state evolution equation of each vertex. When the global asymmetric index exceeds the critical threshold, an abnormal warning signal is triggered. The cross-layer collaborative control module is used to receive local deviation direction, abnormal early warning signals and time-series data of each sensor data stream, calculate the transmission entropy between each sensor data stream, construct a causal influence directed graph, and identify the source node and response node of abnormal propagation. The backtracking management module transmits the identified source node information back to the data calibration module, which performs priority verification and accelerated calibration on the sensors of the source node.
[0005] Furthermore, for acquiring the raw data from each sensor, constraint equations are established based on the conservation characteristics of the monitored physical process. The specific process is as follows: In the industrial monitoring area, the raw measurement values of all sensors are read uniformly through the edge gateway, and the data of each channel are interpolated and aligned according to a unified time axis so that the sensor data of different sampling frequencies have corresponding measurement values at the same time point. The aligned data form the raw data matrix. Based on the attributes of the monitoring area, select the corresponding physical conservation relationship, establish constraint equations based on the conservation characteristics of the monitored physical process, substitute the readings of each sensor into the conservation equations to calculate the real-time residuals, and for each channel, use the conservation equations and sensor readings to back-calculate the theoretical estimate to calculate the cumulative deviation and smooth it to obtain the drift trend. When the drift trend of the sensor data stream continues to exceed the set drift discrimination threshold, it is determined that there is numerical drift in the channel, and the confidence evaluation factor of the channel is calculated. For channels identified as drifting, the drift coefficients are fitted using the least squares method with theoretical estimates and measured values within the window, a linear correction is applied, and a confidence factor is calculated.
[0006] In the above formula, n is the sensor data stream index. Here, e is the natural constant and serves as the drift threshold. When the residual is greater than the standard value, the residual is recalculated after replacing the readings of each sensor data stream with the theoretical estimate. If the residual returns to normal after replacing the sensor data stream, the sensor data stream is determined to be the source of the anomaly. If replacing any single channel is ineffective, the physical process itself is determined to have undergone a substantial change, triggering the upper-level module to analyze and output the corrected timing number and drift status flag.
[0007] Furthermore, the time-series data of each pair of sensor data streams is time-reversed, and the waveform similarity between the original and reversed time series is calculated as a time asymmetry index. The specific processing and calculation process is as follows: The system receives corrected multi-channel timing data from the data calibration module, pairs all sensor data streams together, and generates sensor data pairs. Two sequences within the synchronization time window are extracted and preprocessed to obtain a normalized inverted sequence with zero mean. and the original time series ; For the first channel in the data pair, the time-series data is reversed along the time axis to obtain... This moves the value originally at time t to position T-t+1, the second channel. Remain unchanged; For a given time offset Range of values Calculate the normalized cross-correlation function between the reversed sequence and the original time series. :
[0008] In the above formula, T represents the discrete time point, and the maximum value of the cross-correlation function is taken within the range of the offset values. Define the time asymmetry exponent for data pairs. , .
[0009] Furthermore, the global asymmetric index and local bias directions are extracted, and the reversibility degradation result is output. The specific process is as follows: Integrate the asymmetric exponents of all sensor data stream pairs to calculate the global asymmetric exponent. The global asymmetric index reflects the overall irreversibility of the entire monitoring system at the current moment:
[0010] For each sensor data stream, calculate the temporal asymmetry index of the current channel compared to other channels, and average it with the temporal asymmetry indices obtained from other channels to obtain the local average asymmetry index of the current channel.
[0011]
[0012] Local average asymmetric index This reflects the degree of local irreversibility at the location of the sensor. The global average asymmetric index is obtained by taking a quadratic average of the local average asymmetric indices of all channels. As a global baseline,
[0013] The global baseline reflects the overall irreversibility of the entire monitoring system at the current moment. When the local average asymmetric index of the channel is greater than the set global standard deviation, it is marked as an abnormal hot spot. Spatial interpolation is performed on the abnormal area to generate an asymmetric index field, and the gradient direction is calculated as the mainstream direction of abnormal propagation. The global asymmetric exponent is recorded in chronological order to form a time-series evolution curve, and the rate of change over time is calculated. :
[0014] Output the global asymmetric index, a list of anomalous hotspot channels and their confidence levels, the biased direction vector, and the time-series evolution curve of the asymmetric index as the reversibility degradation result; In the above formula, N is the total number of sensor data streams. Let p be the confidence factor for channel p. Let be the confidence evaluation factor for channel q, i be the current sensor data stream index, and j be the indices of all sensor data streams except i. For time intervals, For the previous moment The global asymmetric exponent.
[0015] Furthermore, a dynamic topology network is constructed with each sensor node as a vertex, and a global asymmetric exponent is injected into the state evolution equation of each vertex. The specific process is as follows: A dynamic topology network is constructed using each sensor node within the monitoring area as a vertex and the physical adjacency relationships and signal propagation characteristics between sensors as edges. The edge connection rules are as follows: Physical adjacency edge: If the spatial distance between two sensors is less than a set threshold, an undirected edge is established; Signal propagation edge: If an abnormal signal from a sensor has a measurable response delay in another sensor, then a directed edge is established with the edge weight being the propagation delay time. The network topology is dynamically updated according to the device's operating status: when a sensor is marked as faulty by the data calibration module, it will be temporarily removed from the network, and will be rejoined to the network when the faulty sensor is restored. The state evolution of each vertex follows a nonlinear dynamic equation, modulated in real time by a global asymmetric exponent.
[0016] Furthermore, when the global asymmetric exponent exceeds the critical threshold, an anomaly warning signal is triggered, and the specific process is as follows: During monitoring, the statistical distribution of the global asymmetric index is recorded. Based on the combined order parameter of the spatial mean and spatial variance of all vertex state values, the order parameter sequence under normal operating conditions is collected { Calculate the mean and standard deviation and define the critical threshold. ,
[0017] Real-time monitoring of the evolution trajectory of the order parameter, calculation of its proximity to the critical threshold, and extraction of the order parameter sensitivity index at the critical transition moment. ,
[0018] In the above formula, the sensitivity index reflects the severity of the response to an anomaly and is used to assess the severity of the anomaly. When the sensitivity index is greater than the warning threshold, an anomaly warning signal is triggered.
[0019] Furthermore, the transfer entropy matrix between the data streams of each sensor is calculated, and a causal influence directed graph is constructed. The specific process is as follows: For two sensor data streams, the transfer entropy Given the known history of channel i, the reduction in uncertainty of the future value of channel i after incorporating the history of channel j:
[0020] In the above, Let be the measurement value of channel i at time t+1. Let i be the measurement value of channel i at time t. Let H be the measurement value of channel j at the current moment, and let H be the information entropy. The quantile value is extracted based on the propagation entropy value as the influence threshold. When the propagation entropy is greater than the influence threshold, a mail transfer is established from node j to node i, generating a weighted directed graph and outputting a causal influence directed graph.
[0021] Furthermore, the identified source node information is sent back to the data calibration module to perform priority verification and accelerated calibration on the sensors of the source node. The specific process is as follows: Based on the out-degree and in-degree of each node, where the out-degree represents how many other nodes a node influences and the in-degree represents how many other nodes it influences, anomaly propagation paths are extracted from the originating node along the out-degree direction. These paths are then combined with local deviation directions as spatial constraints. Nodes with an out-degree greater than their in-degree are source nodes, and nodes with an in-degree greater than their out-degree are response nodes. The source node flag is received and stored in the priority processing queue in descending order of confidence level. This triggers the calibration process, which dynamically adjusts the data acquisition frequency and verification frequency of the source node.
[0022] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: This IoT multi-sensor fusion intelligent monitoring system acquires raw data from various sensors, establishes constraint equations based on the conservation characteristics of the monitored physical process, identifies sensors experiencing numerical drift, and implements online correction. It outputs the time-series data of the corrected sensor data stream, achieving accurate online identification and self-calibration of sensor drift. This effectively distinguishes between sensor faults and physical process anomalies. By reversing the time-series data of the sensor data stream, the system calculates the waveform similarity between the original and reversed time series as a time asymmetry index. The global asymmetry index is used to analyze the irreversible degradation degree of the entire monitoring system, the local average asymmetry index is used to locate abnormal hotspots, and the gradient direction of the asymmetry index field indicates the mainstream of anomaly propagation. The system employs a dual-scale output of global asymmetric index and local deviation direction to assess the overall health status and reversible degradation of the sensor monitoring system. A dynamic topology network is constructed with each sensor node as a vertex, and the global asymmetric index is injected into the state evolution equation of each vertex. When the global asymmetric index exceeds a critical threshold, an anomaly warning signal is triggered, improving the accuracy of alarms. The system calculates the transfer entropy between the data streams of each sensor, constructs a causal influence directed graph, identifies the source and response nodes of anomaly propagation, and sends the identified source node information back to the data calibration module. Priority verification and accelerated calibration are then performed on the sensors of the source nodes. This transfer entropy causal analysis enables precise location of the anomaly source, improving the accuracy of source location. Attached Figure Description
[0023] Figure 1 A schematic diagram of the overall system flow structure of the present invention is shown. Detailed Implementation
[0024] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.
[0025] Example 1: like Figure 1 As shown, an intelligent monitoring system for IoT multi-sensor fusion includes a data calibration module, an irreversible situational awareness module, a group state evolution module, a cross-layer collaborative control module, and a backtracking management module. The data calibration module is used to acquire raw data from each sensor, establish constraint equations based on the conservation characteristics of the monitored physical process, identify sensors that have experienced numerical drift and implement online correction, and output the time-series data of the corrected sensor data stream. The irreversible situational awareness module is used to reverse the timeline of the time series data of each pair of sensor data streams, calculate the waveform similarity between the original time series and the reversed time series as the time asymmetry index, extract the global asymmetry index and local deviation direction, and output the reversibility degradation result. The group state evolution module is used to construct a dynamic topology network with each sensor node as a vertex and inject the global asymmetric index into the state evolution equation of each vertex. When the global asymmetric index exceeds the critical threshold, an abnormal warning signal is triggered. The cross-layer collaborative control module is used to receive local deviation direction, abnormal early warning signals and time-series data of each sensor data stream, calculate the transmission entropy between each sensor data stream, construct a causal influence directed graph, and identify the source node and response node of abnormal propagation. The backtracking management module transmits the identified source node information back to the data calibration module, which performs priority verification and accelerated calibration on the sensors of the source node.
[0026] The raw data from each sensor is used to acquire data, and constraint equations are established based on the conservation characteristics of the monitored physical process. The specific process is as follows: In the industrial monitoring area, the raw measurement values of all sensors are read uniformly through the edge gateway, and the data of each channel are interpolated and aligned according to a unified time axis so that the sensor data of different sampling frequencies have corresponding measurement values at the same time point. The aligned data form the raw data matrix. Based on the attributes of the monitoring area, select the corresponding physical conservation relationship, establish constraint equations based on the conservation characteristics of the monitored physical process, substitute the readings of each sensor into the conservation equations to calculate the real-time residuals, and for each channel, use the conservation equations and sensor readings to back-calculate the theoretical estimate to calculate the cumulative deviation and smooth it to obtain the drift trend. When the drift trend of the sensor data stream continues to exceed the set drift discrimination threshold, it is determined that there is numerical drift in the channel, and the confidence evaluation factor of the channel is calculated. For channels identified as drifting, the drift coefficients are fitted using the least squares method with theoretical estimates and measured values within the window, a linear correction is applied, and a confidence factor is calculated.
[0027] In the above formula, n is the sensor data stream index. Here, e is the natural constant and serves as the drift threshold. When the residual is greater than the standard value, the residual is recalculated after replacing the readings of each sensor data stream with the theoretical estimate. If the residual returns to normal after replacing the sensor data stream, the sensor data stream is determined to be the source of the anomaly. If replacing any single channel is ineffective, the physical process itself is determined to have undergone a substantial change, triggering the upper-level module to analyze and output the corrected timing number and drift status flag.
[0028] This process is used to reverse the timeline of each pair of sensor data streams, and calculate the waveform similarity between the original and reversed timelines as a time asymmetry index. The specific processing and calculation steps are as follows: The system receives corrected multi-channel timing data from the data calibration module, pairs all sensor data streams together, and generates sensor data pairs. A total of N(N-1) / 2N(N-1) / 2 channel pairs are formed. Two sequences within the synchronization time window are extracted, and after preprocessing, a normalized inverted sequence with zero mean is obtained. and the original time series ; For the first channel in the data pair, the time-series data is reversed along the time axis to obtain... This moves the value originally at time t to position T-t+1, the second channel. Remain unchanged; For a given time offset Range of values Calculate the normalized cross-correlation function between the reversed sequence and the original time series. :
[0029] In the above formula, T represents the discrete time point, and the maximum value of the cross-correlation function is taken within the range of the offset values. The maximum value reflects the best positive and negative matching degree of the channel pair, and defines the time asymmetry exponent of the data pair. , ; when When the forward and reverse time sequences are completely reversible, the physical process is in an ideal reversible state. when When the value is 0, it indicates the presence of irreversible components; the larger the value, the greater the degree of irreversibility. when When ≈1, it means that the forward and reverse directions are almost unrelated, and the system is in a highly irreversible state; The cross-correlation function measures the degree of matching between the forward and reverse sequences of channel q and channel p when the time sequence of channel q is reversed and shifted by τ units. The closer the cross-correlation value is to 1, the higher the matching between the forward and reverse sequences of the two channels, which means that the physical process has good time reversibility. The further the cross-correlation value deviates from 1, the more irreversible the energy dissipation or directional process exists.
[0030] The global asymmetric index and local anisotropic directions are extracted, and the reversible degradation result is output. The specific process is as follows: Integrate the asymmetric exponents of all sensor data stream pairs to calculate the global asymmetric exponent. The global asymmetric index reflects the overall irreversibility of the entire monitoring system at the current moment:
[0031] For each sensor data stream, calculate the temporal asymmetry index of the current channel compared to other channels, and average it with the temporal asymmetry indices obtained from other channels to obtain the local average asymmetry index of the current channel.
[0032]
[0033] Local average asymmetric index This reflects the degree of local irreversibility at the location of the sensor. The global average asymmetric index is obtained by taking a quadratic average of the local average asymmetric indices of all channels. As a global baseline,
[0034] The global baseline reflects the overall irreversibility of the entire monitoring system at the current moment. When the local average asymmetry index of a channel is greater than the set global standard deviation, it is marked as an abnormal hotspot region. For sensor pairs near the abnormal hotspot region, if the temporal asymmetry index is higher than that of the adjacent region, the direction of anomaly propagation is determined by the direction of the maximum gradient of the asymmetry index. Spatial interpolation is performed on the abnormal region to generate an asymmetry index field, and the gradient direction is calculated. This represents the mainstream direction of abnormal propagation:
[0035] The global asymmetric exponent is recorded in chronological order to form a time-series evolution curve, and the rate of change over time is calculated. :
[0036] Output the global asymmetric index, a list of anomalous hotspot channels and their confidence levels, the biased direction vector, and the time-series evolution curve of the asymmetric index as the reversibility degradation result; In the above formula, N is the total number of sensor data streams. Let p be the confidence factor for channel p. Let be the confidence evaluation factor for channel q, i be the current sensor data stream index, and j be the indices of all sensor data streams except i. For time intervals, For the previous moment The global asymmetric exponent, To generalize the asymmetric exponential of discrete sensor locations into a continuous field function using the Kriging interpolation method, the direction of the maximum gradient is the mainstream direction of anomaly propagation.
[0037] A dynamic topology network is constructed with each sensor node as a vertex, and a global asymmetric exponent is injected into the state evolution equation of each vertex. The specific process is as follows: A dynamic topology network is constructed using each sensor node within the monitoring area as a vertex and the physical adjacency relationships and signal propagation characteristics between sensors as edges. The edge connection rules are as follows: Physical adjacency edge: If the spatial distance between two sensors is less than a set threshold, an undirected edge is established; Signal propagation edge: If an abnormal signal from a sensor has a measurable response delay in another sensor, then a directed edge is established with the edge weight being the propagation delay time. The network topology is dynamically updated according to the device's operating status: when a sensor is marked as faulty by the data calibration module, it will be temporarily removed from the network, and will be rejoined to the network when the faulty sensor is restored. The state evolution of each vertex follows a nonlinear dynamic equation, which is modulated in real time according to a global asymmetric exponent.
[0038] When the global asymmetric exponent exceeds the critical threshold, an anomaly warning signal is triggered. The specific process is as follows: During monitoring, the statistical distribution of the global asymmetric index is recorded. Based on the combined order parameter of the spatial mean and spatial variance of all vertex state values, the order parameter sequence under normal operating conditions is collected { Calculate the mean and standard deviation and define the critical threshold. ,
[0039] Real-time monitoring of the evolution trajectory of the order parameter, calculation of its proximity to the critical threshold, and extraction of the order parameter sensitivity index at the critical transition moment. ,
[0040] In the above formula, the sensitivity index reflects the severity of the response to an anomaly and is used to assess the severity of the anomaly. When the sensitivity index is greater than the warning threshold, an anomaly warning signal is triggered.
[0041] The transfer entropy matrix between the data streams of each sensor is calculated, and a causal influence directed graph is constructed. The specific process is as follows: For two sensor data streams, the transfer entropy Given the known history of channel i, the reduction in uncertainty of the future value of channel i after incorporating the history of channel j:
[0042] In the above, Let be the measurement value of channel i at time t+1. Let i be the measurement value of channel i at time t. Let H be the measurement value of channel j at the current moment, and let H be the information entropy. The quantile value is extracted based on the propagation entropy value as the influence threshold. When the propagation entropy is greater than the influence threshold, a mail transfer is established from node j to node i, generating a weighted directed graph and outputting a causal influence directed graph.
[0043] The identified source node information is sent back to the data calibration module to perform priority verification and accelerated calibration on the sensors of the source node. The specific process is as follows: Based on the out-degree and in-degree of each node, where the out-degree represents how many other nodes a node influences and the in-degree represents how many other nodes it influences, anomaly propagation paths are extracted from the originating node along the out-degree direction. These paths are then combined with local deviation directions as spatial constraints. Nodes with an out-degree greater than their in-degree are source nodes, and nodes with an in-degree greater than their out-degree are response nodes. The source node flag is received and stored in the priority processing queue in descending order of confidence level. This triggers the calibration process, which dynamically adjusts the data acquisition frequency and verification frequency of the source node.
[0044] The degree of irreversibility is a quantitative indicator used to measure the extent to which a physical process deviates from an ideal reversible state. The numerical range is [0,1] - the closer the value is to 0, the closer the process is to reversible (ideal state); the closer the value is to 1, the more irreversible the process is.
[0045] The size of the interval and threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value.
[0046] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation. In the two embodiments provided in this application, it should be understood that the disclosed apparatus and system can be implemented in other ways; for example, the apparatus embodiments described above are merely illustrative, and the division of modules is merely a logical functional division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed; furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of the apparatus or module may be electrical, mechanical or other forms. The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.
Claims
1. An intelligent monitoring system for IoT multi-sensor fusion, characterized in that, include The data calibration module is used to acquire raw data from each sensor, establish constraint equations based on the conservation characteristics of the monitored physical process, identify sensors that have experienced numerical drift and implement online correction, and output the time-series data of the corrected sensor data stream. The irreversible situational awareness module is used to reverse the timeline of the time data of each pair of sensor data streams, calculate the waveform similarity between the original time series and the reversed time series as the time asymmetry index, extract the global asymmetry index and local deviation direction, and output the degree of reversibility degradation. The group state evolution module is used to construct a dynamic topology network with each sensor node as a vertex and inject the global asymmetric index into the state evolution equation of each vertex. When the global asymmetric index exceeds the critical threshold, an abnormal warning signal is triggered. The cross-layer collaborative control module is used to receive local deviation direction, abnormal early warning signals and time-series data of each sensor data stream, calculate the transmission entropy between each sensor data stream, construct a causal influence directed graph, and identify the source node and response node of abnormal propagation. The backtracking management module transmits the identified source node information back to the data calibration module, which performs priority verification and accelerated calibration on the sensors of the source node.
2. The IoT multi-sensor fusion intelligent monitoring system according to claim 1, characterized in that, The raw data from each sensor is used to acquire data, and constraint equations are established based on the conservation characteristics of the monitored physical process. The specific process is as follows: In the industrial monitoring area, the raw measurement values of all sensors are read uniformly through the edge gateway, and the data of each channel are interpolated and aligned according to a unified time axis so that the sensor data of different sampling frequencies have corresponding measurement values at the same time point. The aligned data form the raw data matrix. Based on the attributes of the monitoring area, select the corresponding physical conservation relationship, establish constraint equations based on the conservation characteristics of the monitored physical process, substitute the readings of each sensor into the conservation equations to calculate the real-time residuals, and for each channel, use the conservation equations and sensor readings to back-calculate the theoretical estimate to calculate the cumulative deviation and smooth it to obtain the drift trend. When the drift trend of the sensor data stream continues to exceed the set drift discrimination threshold, it is determined that there is numerical drift in the channel, and the confidence evaluation factor of the channel is calculated. For channels identified as drifting, the drift coefficients are fitted using the least squares method with theoretical estimates and measured values within the window, a linear correction is applied, and a confidence factor is calculated. ； In the above formula, n is the sensor data stream index. Here, e is the natural constant and serves as the drift threshold. When the residual is greater than the standard value, the residual is recalculated after replacing the readings of each sensor data stream with the theoretical estimate. If the residual returns to normal after replacing the sensor data stream, the sensor data stream is determined to be the source of the anomaly. If replacing any single channel is ineffective, the physical process itself is determined to have undergone a substantial change, triggering the upper-level module to analyze and output the corrected timing number and drift status flag.
3. The intelligent monitoring system for IoT multi-sensor fusion according to claim 1, characterized in that, This process is used to reverse the timeline of each pair of sensor data streams, and calculate the waveform similarity between the original and reversed timelines as a time asymmetry index. The specific processing and calculation steps are as follows: The system receives corrected multi-channel timing data from the data calibration module, pairs all sensor data streams together, and generates sensor data pairs. Two sequences within the synchronization time window are extracted and preprocessed to obtain a normalized inverted sequence with zero mean. and the original time series ; For the first channel in the data pair, the time-series data is reversed along the time axis to obtain... This moves the value originally at time t to position T-t+1, the second channel. Remain unchanged; For a given time offset Range of values Calculate the normalized cross-correlation function between the reversed sequence and the original time series. : In the above formula, T represents the discrete time point, and the maximum value of the cross-correlation function is taken within the range of the offset values. Define the time asymmetry exponent for data pairs. , .
4. The IoT multi-sensor fusion intelligent monitoring system according to claim 1, characterized in that, The global asymmetric index and local anisotropic directions are extracted, and the reversible degradation result is output. The specific process is as follows: Integrate the asymmetric exponents of all sensor data stream pairs to calculate the global asymmetric exponent. The global asymmetric index reflects the overall irreversibility of the entire monitoring system at the current moment: ； For each sensor data stream, calculate the temporal asymmetry exponent of the current channel compared to other channels, and average it with the temporal asymmetry exponents obtained from other channels to obtain the local average asymmetry exponent of the current channel. ； Local average asymmetric index This reflects the degree of local irreversibility at the location of the sensor. The global average asymmetric index is obtained by taking a quadratic average of the local average asymmetric indices of all channels. As a global baseline, ； The global baseline reflects the overall irreversibility of the entire monitoring system at the current moment. When the local average asymmetric index of the channel is greater than the set global standard deviation, it is marked as an abnormal hot spot. Spatial interpolation is performed on the abnormal area to generate an asymmetric index field, and the gradient direction is calculated as the mainstream direction of abnormal propagation. The global asymmetric exponent is recorded in chronological order to form a time-series evolution curve, and the rate of change over time is calculated. : ； Output the global asymmetric index, a list of anomalous hotspot channels and their confidence levels, the biased direction vector, and the time-series evolution curve of the asymmetric index as the reversibility degradation result; In the above formula, N is the total number of sensor data streams. Let p be the confidence factor for channel p. Let be the confidence evaluation factor for channel q, i be the current sensor data stream index, and j be the indices of all sensor data streams except i. For time intervals, For the previous moment The global asymmetric exponent.
5. The IoT multi-sensor fusion intelligent monitoring system according to claim 1, characterized in that, A dynamic topology network is constructed with each sensor node as a vertex, and a global asymmetric exponent is injected into the state evolution equation of each vertex. The specific process is as follows: A dynamic topology network is constructed using each sensor node within the monitoring area as a vertex and the physical adjacency relationships and signal propagation characteristics between sensors as edges. The edge connection rules are as follows: Physical adjacency edge: If the spatial distance between two sensors is less than a set threshold, an undirected edge is established; Signal propagation edge: If an abnormal signal from a sensor has a measurable response delay in another sensor, then a directed edge is established with the edge weight being the propagation delay time. The network topology is dynamically updated according to the device's operating status: when a sensor is marked as faulty by the data calibration module, it will be temporarily removed from the network, and will be rejoined to the network when the faulty sensor is restored. The state evolution of each vertex follows a nonlinear dynamic equation, modulated in real time by a global asymmetric exponent.
6. The IoT multi-sensor fusion intelligent monitoring system according to claim 1, characterized in that, When the global asymmetric exponent exceeds the critical threshold, an anomaly warning signal is triggered. The specific process is as follows: During the monitoring process, the statistical distribution of the global asymmetric index is recorded. Based on the combined order parameter of the spatial mean and spatial variance of all vertex state values, the order parameter sequence under normal operating conditions is collected, and the mean is calculated. and standard deviation and define the critical threshold. , ； Real-time monitoring of the evolution trajectory of the order parameter, calculation of its proximity to the critical threshold, and extraction of the order parameter sensitivity index at the critical transition moment. , ； In the above formula, the sensitivity index reflects the severity of the response to an anomaly and is used to assess the severity of the anomaly. When the sensitivity index is greater than the warning threshold, an anomaly warning signal is triggered.
7. The IoT multi-sensor fusion intelligent monitoring system according to claim 1, characterized in that, The transfer entropy matrix between the data streams of each sensor is calculated, and a causal influence directed graph is constructed. The specific process is as follows: For two sensor data streams, the transfer entropy Given that the history of channel i itself is known, the reduction in uncertainty of the future value of channel i after incorporating the history of channel j is as follows: ； In the above, Let be the measurement value of channel i at time t+1. Let be the measurement value of channel i at time t. Let H be the measurement value of channel j at the current moment, and let H be the information entropy. The quantile value is extracted based on the propagation entropy value as the influence threshold. When the propagation entropy is greater than the influence threshold, a mail transfer is established from node j to node i to generate a weighted directed graph and output the causal influence directed graph.
8. The intelligent monitoring system for IoT multi-sensor fusion according to claim 1, characterized in that, The identified source node information is sent back to the data calibration module to perform priority verification and accelerated calibration on the sensors of the source node. The specific process is as follows: Based on the out-degree and in-degree of each node, where the out-degree represents the number of other nodes it influences and the in-degree represents the number of other nodes it is influenced by, anomaly propagation paths are extracted from the originating node along the out-degree direction. These paths are then combined with local deviation directions as spatial constraints. Nodes with an out-degree greater than their in-degree are source nodes, and nodes with an in-degree greater than their out-degree are response nodes. The source node flag is received and stored in the priority processing queue in descending order of confidence level. This triggers the calibration process, which dynamically adjusts the data acquisition frequency and verification frequency of the source node.