Atmospheric environment intelligent monitoring method and system based on Internet of Things

Abstract

The invention relates to the technical field of atmospheric environment monitoring, and discloses an intelligent atmospheric environment monitoring method and system based on the Internet of Things, and the method comprises the steps: collecting an original signal of a sensor through an edge monitoring node, extracting a time sequence fingerprint feature, comparing the time sequence fingerprint feature with a baseline fingerprint, and generating a confidence score; according to the method, through dynamic evaluation of the signal perturbation characteristics of the sensors, node state autonomous perception and data credibility quantification are realized, so that a monitoring network has the capability of automatically identifying and inhibiting abnormal data, and the reliability of the monitoring network is improved. Meanwhile, in combination with environment disturbance adaptive adjustment and a group consensus calibration mechanism, the robustness of a regional monitoring conclusion under a complex working condition is improved, and a basis is provided for predictive maintenance.

Description

TECHNICAL FIELD

[0001] The present application relates to an atmospheric environment intelligent monitoring method and system based on Internet of Things, and belongs to the technical field of atmospheric environment monitoring. BACKGROUND

[0002] The current mainstream monitoring scheme relies on distributed sensor nodes to collect environmental parameters, and guarantees data accuracy through static calibration and fixed threshold alarm mechanism. With the expansion of Internet of Things node scale and the complexity of application scenarios, the existing technology gradually reveals three essential limitations: 1. Low-cost sensors are limited by physical and chemical properties, and the drift of their measurement values caused by temperature and humidity changes, aging or pollution is inevitable. Although the existing scheme can alleviate this problem through periodic manual calibration, it cannot quantify the degree of node state degradation in real time, resulting in continuous pollution of the network by abnormal data; 2. Under complex working conditions such as strong wind and electromagnetic interference, the response characteristics of sensor signal fluctuations and internal equipment failures are highly similar, and the traditional threshold alarm mechanism cannot distinguish between external interference and real failure, which is prone to trigger false maintenance instructions; 3. The existing technology can only trigger a response after the node completely fails, and lacks early warning capability for gradual degradation. In a large-scale network, manual inspection and blind replacement of components result in waste of resources, and monitoring blind spots caused by sudden failures are more likely to cause environmental risk misjudgment.

[0003] Although the industry tries to improve robustness through redundant deployment or multi-source data fusion, it is still limited by the surge in hardware costs and the bottleneck of algorithm complexity, making it difficult to solve the systemic vulnerability problem caused by individual state invisibility. Therefore, how to give the monitoring network the ability to dynamically identify and suppress abnormal data, while realizing the quantitative evaluation and predictive maintenance of node health status, has become a technical problem to be solved by the present application. SUMMARY

[0004] The present application provides an atmospheric environment intelligent monitoring method based on Internet of Things, which mainly aims to solve the problems of data credibility out of control, operation and maintenance response lag and environmental disturbance misjudgment caused by node state drift in large-scale sensor networks.

[0005] To achieve the above purpose, the atmospheric environment intelligent monitoring method based on Internet of Things provided by the present application comprises the following steps: step a, the edge monitoring node collects the original output signal time series data sequence of the edge monitoring node itself sensor within a certain period of time.

[0006] Step b, the edge monitoring node pre-processes the time series data sequence, performs digital filtering operation to attenuate the environmental background noise component in the time series data sequence, and extracts the time series fingerprint feature capable of representing the current inherent physical and chemical properties of the sensor.

[0007] Step c, the edge monitoring node calculates the distance between the current time series fingerprint feature and the baseline fingerprint feature stored in the edge monitoring node in advance to obtain a fingerprint deviation value, which quantitatively reflects the deviation degree of the current inherent physical and chemical characteristics of the sensor compared with the baseline fingerprint feature.

[0008] Step d, the edge monitoring node generates a confidence score between 0 and 1 through a nonlinear mapping function according to the fingerprint deviation value, and the confidence score represents the reliability of the current measurement data of the sensor, wherein the confidence score decreases as the fingerprint deviation value increases.

[0009] Step e, the edge monitoring node reports a binary data set containing the measurement value of the edge monitoring node and the confidence score to the network.

[0010] Step f, the central platform receives the binary data set from multiple edge monitoring nodes, and weights and fuses the measurement values based on the confidence scores to dynamically reduce the weight of the measurement values with confidence scores lower than a predetermined threshold in the final regional atmospheric environment monitoring conclusion, thereby generating a robust regional atmospheric environment monitoring conclusion, and triggering a maintenance warning according to the continuous downward trend of the confidence scores of the edge monitoring nodes.

[0011] Preferably, in step b, the time series fingerprint feature includes the variance of the time series data sequence and the kurtosis of the time series data sequence, and the variance and the kurtosis together constitute a two-dimensional time series fingerprint vector for representing the fluctuation amplitude and outlier distribution characteristics of the signal. The time series fingerprint feature is obtained by performing fast Fourier transform on the time series data sequence to extract the energy ratio of a plurality of predetermined frequency bands, thereby forming a multi-dimensional time series fingerprint vector, and the energy ratio represents the frequency domain response mode of the sensor to the environmental perturbation.

[0012] Preferably, in step d, the nonlinear mapping function is a Sigmoid function, and the calculation formula is: wherein, is the confidence score, is the fingerprint deviation value, is the steepness adjustment parameter of the Sigmoid function, and is the center offset parameter of the Sigmoid function.

[0013] Preferably, the method further comprises the following steps: the edge monitoring node periodically broadcasts a heartbeat packet containing the identification information of the edge monitoring node and the fluctuation state bit obtained by the analog comparator through the Bluetooth Low Energy broadcast mode at a determined broadcast frequency, the fluctuation state bit indicating whether the measured value exceeds a certain fluctuation threshold; and the adjacent edge monitoring nodes receiving the heartbeat packet trigger the time sequence fingerprint feature extraction and fingerprint deviation value calculation of steps b and c after receiving the heartbeat packet indicating that the measured value fluctuation exceeds the preset threshold, and report the calculated edge monitoring node measured value and confidence score to the center platform through the main communication channel.

[0014] Preferably, the method further comprises the following steps: the edge monitoring node identifies two time windows in which the rate of change of light intensity reaches a maximum value per day using the light-sensitive element to determine the dawn event window period and the dusk event window period; during the dawn event window period, the edge monitoring node performs the time sequence fingerprint feature extraction and confidence score evaluation of steps b and c, and reports a dawn report containing the identification information of the edge monitoring node, the current confidence score, and the current time sequence fingerprint to the center platform; the center platform collects the dawn reports, screens out a group of edge monitoring nodes with confidence scores higher than a previously determined high confidence score threshold, and calculates the average fingerprint of the current time sequence fingerprint of all edge monitoring nodes in the group of edge monitoring nodes with confidence scores higher than the previously determined high confidence score threshold; and the center platform broadcasts the average fingerprint as a correction parameter to all edge monitoring nodes in the network, and the edge monitoring nodes receive the correction parameter and perform a weighted moving update of their own baseline fingerprint feature based on a preset learning rate.

[0015] Preferably, the method further comprises the following steps: when the center platform determines that the edge monitoring node needs maintenance according to the confidence score of the edge monitoring node being continuously lower than a previously determined maintenance trigger threshold, the center platform suspends the dispatch of maintenance instructions and sends a echo request instruction to the edge monitoring node; the edge monitoring node drives its onboard piezoelectric electro-acoustic transducer to emit a physical ringing at a preset ringing frequency and for a preset duration after receiving the echo request instruction; the edge monitoring node then switches the input / output port mode of its piezoelectric electro-acoustic transducer from drive output to high-impedance analog input, and listens to the weak decay voltage signal generated by the residual vibration of the element structure after the ringing ends; and the edge monitoring node returns the integral value and / or peak value of the residual vibration signal to the center platform as a survival proof, and the center platform arbitrates whether to dispatch a hardware repair work order according to whether the valid survival proof is received within a preset time limit.

[0016] Preferably, the method further comprises the steps of: extracting the amplitude of the fundamental frequency and the amplitude of its second harmonic from the result of the fast Fourier transform; and generating, by the edge monitoring node, an index for characterizing the electrical health status of the edge monitoring node based on the ratio of the amplitude of the fundamental frequency to the amplitude of the second harmonic, which is reported in parallel with the sensor confidence score to assist the central platform in making a more comprehensive diagnosis of node failure.

[0017] Preferably, the method further comprises the steps of: analyzing, by the edge monitoring node, the short-term variance of the received signal strength indication of the signals received by the wireless communication module of the edge monitoring node from neighboring edge monitoring nodes to generate an environmental disturbance index; and when the environmental disturbance index exceeds a pre-determined environmental disturbance index threshold, adjusting, by the edge monitoring node, the fingerprint deviation threshold used to generate the confidence score within a pre-set range to distinguish between the time series fingerprint changes caused by internal state degradation of the edge monitoring node and external environmental disturbances.

[0018] Preferably, in step b, the preprocessing includes nonlinear filtering processing of the time series data sequence, which aims to enhance the micro-disturbance characteristics in the time series data sequence that reflect the inherent physical and chemical characteristics of the sensor, and suppress common mode interference caused by environmental changes, so that the extracted time series fingerprint features more purely characterize the state of the sensor itself.

[0019] An atmospheric environment intelligent monitoring system based on Internet of Things, comprising: an edge monitoring subsystem, specifically comprising: a sensor module for collecting raw output signal time series data; a preprocessing module coupled to the sensor module and configured to perform digital filtering operation on the time series data and extract time series fingerprint features; an evaluation module coupled to the preprocessing module and configured to calculate the distance between the current time series fingerprint features and the baseline fingerprint features to obtain a fingerprint deviation value, and generate a confidence score based on the fingerprint deviation value; and a communication module coupled to the evaluation module and configured to report the binary data set of the measurement value and the confidence score to the network.

[0020] A central platform subsystem, specifically comprising: a data receiving module for receiving binary data sets from a plurality of edge monitoring nodes; a fusion processing module coupled to the data receiving module and configured to perform weighted fusion on the measurement values based on the confidence scores to dynamically reduce the weight of the measurement values with confidence scores below a pre-determined threshold in the final regional atmospheric environment monitoring conclusion, thereby generating a robust regional atmospheric environment monitoring conclusion; and a maintenance warning module coupled to the fusion processing module and configured to trigger a maintenance warning according to the continuous downward trend of the confidence scores of the edge monitoring nodes.

[0021] Compared with the prior art, the beneficial effects of the present application are: 1. By extracting the time sequence micro-disturbance characteristics such as variance, kurtosis or frequency energy ratio of the sensor output signal as the internal health fingerprint, the edge node has the ability to continuously quantify the physical and chemical state of itself, and the node maps the distance between the current fingerprint and the baseline fingerprint to a dynamic confidence score, forming a binary data set containing the measurement value and its reliability. This mechanism enables each node to transform from a passive data acquisition unit to a state-aware intelligent monitoring unit, providing an endogenous reliability basis for subsequent data fusion, which reduces the risk of systematic data pollution caused by node state drift in large-scale Internet of Things.

[0022] 2. The edge node generates an environmental disturbance indicator by analyzing the short-term variance of the wireless communication signal strength, dynamically adjusts the fingerprint deviation threshold, and when the environment changes dramatically such as strong wind or electromagnetic interference causing signal fluctuations, the system automatically relaxes the tolerance of fingerprint changes to avoid misjudging external disturbances as node failures. At the same time, the center platform weights and fuses the multi-node measurement values based on the confidence score, so that low-reliability data is naturally suppressed in the regional conclusion. This cross-physical layer and application layer information coupling mechanism helps to improve the decision stability of the system under extreme working conditions, and improves the applicability in complex environments compared with traditional monitoring solutions.

[0023] 3. The high-confidence node group reports fingerprint data during this period, and the center platform calculates the average fingerprint and broadcasts it to the entire network. The node aligns the baseline fingerprint to the collective consensus with a gradual learning rate, allowing the network to update the reference baseline autonomously when the environmental background changes slowly, such as seasonal changes. This mechanism, which uses astronomical events as synchronization signals and group consensus as correction basis, gives the system the ability to adapt to long-term environmental drift, helping to reduce the risk of collective misjudgment caused by fixed baseline.

[0024] 4. Extract the fundamental frequency and harmonic amplitude ratio from the Fast Fourier Transform result to generate an electrical health indicator independent of the sensor state, such as capacitor aging and poor grounding. At the same time, verify the survival state of the core hardware by driving the piezoelectric element to ring and detecting its structural residual vibration signal. These two mechanisms reuse existing computing resources and hardware components to form orthogonal diagnostic dimensions for sensor probes, power supply systems, and main control hardware, allowing operation and maintenance instructions to accurately distinguish between software anomalies and physical damage, reducing the cost of ineffective maintenance caused by misjudgment. The node broadcasts a minimalist heartbeat packet through Bluetooth on a daily basis, and only when it receives an abnormal broadcast from a neighboring node does it trigger fingerprint calculation and main channel reporting. This design converts continuous calculation into an event-driven mode, allowing the system to sleep at a micro-amp level in normal state, combined with the directional activation strategy calibrated during the dawn window period, ensuring the integrity of the function while reducing the energy consumption of battery-powered nodes, supporting the deployment of large-scale networks without human operation and maintenance. BRIEF DESCRIPTION OF DRAWINGS

[0025] Fig. 1 The structure and information flow diagram of the atmospheric environment intelligent monitoring system based on the Internet of Things of the present application.

[0026] Fig. 2 The trend curve diagram of the confidence score S of different nodes of the present application changes with time.

[0027] Fig. 3 The timing diagram of the abnormal event cooperative response triggered by the Bluetooth broadcast of the present application.

[0028] The object implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0029] It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0030] The present application discloses an atmospheric environment intelligent monitoring method and system based on the Internet of Things, which is composed of distributed edge monitoring nodes and a central platform, wherein each edge monitoring node independently performs localized analysis of sensor signals to evaluate its own state and generate data containing credibility quantization, and the central platform is responsible for gathering data from multiple nodes, performing data fusion and system-level decision-making through a group consensus mechanism, forming a closed-loop monitoring system from individual state self-awareness to group intelligent decision-making.

[0031] In view of the fact that the output signal of the sensor in the monitoring network is easily affected by the unpredictable drift caused by its own aging, pollution or environmental changes due to the limitation of physical and chemical characteristics, the core mechanism of the present application starts from the autonomous quantitative evaluation of the health state of the edge monitoring node; specifically, each edge monitoring node is configured to continuously collect its sensor original output signal within a certain time period to form a time series data sequence, the node first performs preprocessing on the sequence, which is preferably a nonlinear filtering process, aiming to enhance the micro-disturbance characteristics in the sequence that can reflect the inherent physical and chemical characteristics of the sensor, and suppress the common-mode interference caused by the severe change of the environment, so that the subsequent extracted time series fingerprint feature more purely represents the state of the sensor itself, the time series fingerprint feature can be specifically implemented as a two-dimensional time series fingerprint vector composed of the variance and kurtosis of the time series data sequence, reflecting the signal fluctuation amplitude and outlier distribution characteristics, or by performing fast Fourier transform on the sequence, extracting the energy ratio of a plurality of predetermined frequency bands to form a multi-dimensional time series fingerprint vector representing the frequency domain response mode of the sensor to the environmental micro-disturbance; the node calculates the distance between the current time series fingerprint feature and the baseline fingerprint feature stored in the device to obtain a scalar fingerprint deviation value , the deviation value which is then plugged into a non-linear mapping function to generate a confidence score between 0 and 1 , which is preferably a sigmoid function of the form , where the parameters and are calibrated following a deterministic procedure, i.e. during the stable operation period of the initial deployment of the device, all computed fingerprint deviation values are recorded and their average is set as the center shift parameter , while the steepness adjustment parameter is solved by setting a constraint that the value of reaches when equals a certain value, e.g. 0.2, thus ensuring that the confidence score can impose a non-linear penalty on deviations beyond the normal range; finally, the edge monitoring node reports a binary data set containing the current measurement value and the corresponding confidence score through the main communication channel, by which each node is transformed from a passive data source to an intelligent agent capable of real-time quantitative endorsement of its own data quality.

[0032] To balance the real-time performance of the monitoring and the power consumption limit under large-scale deployment, the invention also introduces an event-driven computing and communication architecture; in the normal monitoring state, the edge monitoring node periodically broadcasts a minimalist heartbeat packet through the Bluetooth Low Energy broadcast mode at a certain broadcast frequency, which only contains the identification information of the node and a fluctuation state bit obtained by the analog comparator, which is used to indicate whether the measurement value of the node exceeds a certain fluctuation threshold, after receiving the heartbeat packet, the adjacent edge monitoring node is only triggered to perform the series of operations of the aforementioned time sequence fingerprint feature extraction, fingerprint deviation value calculation and reporting of the complete binary data set containing the measurement value and the confidence score through the main communication channel when it finds that the fluctuation state bit in the heartbeat packet indicates that the measurement value fluctuates abnormally; this event-driven mode converts continuous computation into event-driven mode, so that the system is in a micro-ampere sleep state for most of the time, prolonging the life cycle of the battery-powered node and providing feasibility for the unattended operation and deployment of the super-large-scale network; after receiving the binary data set from multiple edge monitoring nodes in the network, the primary task of the center platform is to generate a robust regional atmospheric environment monitoring conclusion that can resist individual node anomalies, given that some nodes may report data with very low confidence due to failure or severe drift, the center platform adopts a weighted fusion algorithm based on the confidence score, which is specifically implemented as the final regional monitoring value is the weighted average of all node measurement values, and the weight corresponding to each node measurement value is the confidence score reported by the node , while setting a previously determined confidence threshold, the confidence score If the threshold is lower, the weight will be dynamically reduced or even set to zero. In other words, the center platform also maintains a time series database of confidence scores for each node and runs a trend analysis algorithm. Once it detects that a node's confidence score is showing a continuous downward trend and is lower than a preset maintenance trigger threshold, the system automatically triggers a maintenance warning. This mechanism not only ensures the robustness of regional monitoring conclusions, but also changes the traditional post-remedy operation and maintenance mode to a pre-warning predictive maintenance mode.

[0033] To address the challenge of complex working conditions, such as strong electromagnetic interference or severe airflow disturbance, which may cause sensor signals to fluctuate similarly to their own degradation characteristics, the present invention introduces an environmental disturbance adaptive adjustment mechanism. Each edge monitoring node uses its wireless communication module to analyze the short-term variance of the received signal strength indication of the signals received from neighboring nodes to generate a real-time quantitative environmental disturbance indicator. When the indicator exceeds an environmental disturbance indicator threshold set based on historical data statistics, it means that the node is currently in a rapidly changing external environment. At this time, the node temporarily and proportionally adjusts the fingerprint deviation threshold used to generate the confidence score to distinguish between internal state degradation and external environmental disturbance-induced temporal fingerprint changes, thereby avoiding mistakenly attributing external environmental transient effects to node failure and enhancing the decision stability of the monitoring system in real complex scenarios. Considering that the environmental background itself has slow long-term changes, such as temperature and humidity baseline drift caused by seasonal changes, if all nodes' baseline fingerprint characteristics remain unchanged for a long time, it may cause collective judgment bias in the entire monitoring network. Therefore, the present invention designs a natural rhythm-based group consensus calibration mechanism. Each edge monitoring node uses its onboard photosensitive element to identify the two time windows when the daily light intensity change rate reaches its maximum, thereby determining the dawn and dusk event window period. During the daily dawn event window period, all nodes in the network are awakened to perform temporal fingerprint feature extraction and confidence assessment, and report the dawn report containing their own identification information, current confidence score and current temporal fingerprint feature to the center platform. After collecting the reports, the center platform selects nodes with confidence scores higher than the high confidence score threshold as the consensus group and calculates the average fingerprint of the group's temporal fingerprint features. Subsequently, the center platform broadcasts this average fingerprint as a correction parameter to the entire network. After receiving it, each node aligns its own baseline fingerprint characteristics to the average fingerprint based on a weighted moving update algorithm based on a preset learning rate. This mechanism, which uses astronomical events as synchronization signals and high-confidence group consensus as calibration criteria, enables the monitoring network to adaptively update the baseline fingerprint based on group consensus, allowing it to adapt to long-term environmental background changes.

[0034] Moreover, in order to achieve more accurate fault diagnosis, the application further provides a multi-dimensional fault diagnosis system; firstly, in the scene where the edge monitoring node performs fast Fourier transform on the signal, the node is configured to additionally extract the amplitude of the power frequency fundamental wave and the amplitude of the second harmonic of the power frequency fundamental wave, generate an index capable of representing the electrical health status of the node power supply system by calculating the ratio of the two, and report the index in parallel with the sensor confidence score, thereby providing orthogonal diagnostic basis for distinguishing between sensor probe failure and node circuit failure for the center platform; secondly, when the center platform determines that a node needs maintenance, a back echo request instruction can be sent to the target node before dispatching the maintenance instruction, the node drives the onboard piezoelectric electroacoustic transducer element to emit a physical ringing after receiving the instruction, and immediately switches the element to a high impedance analog input mode to listen to the weak decay voltage signal generated by the structural residual vibration of the element, finally, the node returns the integral value or peak value of the residual vibration signal as survival proof, and the center platform arbitrates whether to dispatch a hardware maintenance work order according to whether the valid survival proof is received within the preset time limit, this survival test mechanism using the onboard element for self-tapping and self-echoing provides a basis for remotely judging whether the node master hardware is physically intact at a very low cost, thereby improving the accuracy of the operation and maintenance instruction.

[0035] Embodiment 1: In a large-scale petrochemical industrial park in continuous operation, hundreds of edge monitoring nodes are deployed to monitor the micro leakage of a key chemical raw material, the challenge of this scene is that the intermittent start and stop of high-power pump groups in the production process will cause local electromagnetic interference, at the same time, the sudden temperature drop at night is easy to form condensate water on the surface of the sensor probe, both of these two external factors can cause the sensor output signal to produce fluctuations similar to the characteristics of the real leakage signal, causing the monitoring system to frequently issue false alarms, interfering with the normal production order; on a certain night, the 3rd edge monitoring node in the southeast corner of the park started to continuously rise and exceed the first alarm threshold, at the same time, a large pump group within 50 meters of it entered the predetermined operation period, in this case, the traditional system cannot distinguish whether this is a false event induced by external interference or a real leakage event masked by noise; however, under the technical solution of the application, the 3rd edge monitoring node starts its internal processing flow at the same time as it collects abnormal signals, it first calculates the fingerprint deviation value of the current time series data sequence , the value presents a high amplitude deviation, accordingly, it calculates the confidence score by the Sigmoid function The drop to 0.2 or below constitutes an internal evaluation conclusion of the state of the node itself; at the same time, the node and several nodes adjacent thereto generate a high environmental disturbance index by analyzing the short-term variance of the signal strength indication received by the respective wireless communication modules, which is a consensus judgment of the external environmental state; here, a synergistic mechanism is embodied, that is, the generation of the environmental disturbance index provides the necessary context for the interpretation of the confidence score, and the quantification of the confidence score provides the final basis for decision-making under environmental disturbance. Specifically, when node 3 detects a high environmental disturbance index, its internal algorithm increases the fingerprint deviation threshold for generating the confidence score according to the regulations, trying to accommodate a part of the signal fluctuations caused by external disturbance, but since the fingerprint deviation value caused by the deterioration of its internal state is too large, its final confidence score is still maintained at a low level, and a healthy node adjacent thereto, although it is also subjected to strong environmental disturbance, its fingerprint deviation is still within an acceptable range under the relaxed threshold, so its confidence score only decreases slightly and is maintained at about 0.8; in this way, the information received by the center platform is no longer a single or indistinguishable alarm signal, but a data matrix containing multiple dimensions and cross-verification, among which the measurement value of node 3 is automatically suppressed in the weighted fusion calculation of the regional atmospheric environment monitoring conclusion because its confidence score is lower than the previously determined threshold, thereby avoiding a false alarm.

[0036] Furthermore, the scheme alleviates the contradiction between high sensitivity and high reliability in the monitoring field. Traditional schemes necessarily amplify noise interference in order to capture weak signals and improve sensitivity. However, the present scheme changes the problem from how to distinguish real information from signals to how to evaluate the reliability of signal sources by introducing the dimension of confidence score. When the confidence score of a node decreases due to its own state deterioration, regardless of its output value, the system also limits its speaking power, achieving the reliability of the entire system conclusion without sacrificing the sensitivity of individual nodes; the next day after the above event, the center platform automatically triggers a maintenance warning according to the continuous downward trend of the confidence score of the 3rd edge monitoring node. Before dispatching the work order, the system sends a echo request instruction to the node, which drives the on-board piezoelectric electro-acoustic transducer element to emit a sound after receiving the instruction. However, it fails to return an effective survival proof to the center platform, and the system arbitrates that the node has physical hardware damage and generates a maintenance work order containing explicit fault location, guiding the operation and maintenance personnel to directly replace the sensor core component of the node.

[0037] Example 2: To verify the effectiveness of the method of the present application in distinguishing between sensor self-state degradation and external environmental disturbance, as well as its robustness in group consensus fusion decision-making, a test platform consisting of an environmental simulation cabin and four edge monitoring nodes deployed therein was built. The platform can control the target gas concentration, temperature and humidity, and electromagnetic field intensity in the cabin. The hardware configuration and software algorithm of the four edge monitoring nodes are consistent with the above, and before the start of the test, the baseline fingerprint characteristics of each node were obtained by calibrating in standard clean air for twenty-four hours. The calculation period of the time series fingerprint characteristics was set to three hundred seconds. The setting of this parameter aims to balance the real-time of data acquisition and the data processing load and energy consumption of the system. The decision rule set is that the calculation period should be significantly longer than the nominal T90 response time of the used sensor to obtain a statistically stable fingerprint, while it should be shorter than the time scale of typical fault mode change. Three hundred seconds is an engineering example obtained for the characteristics of the sensors used in this test under this decision rule.

[0038] The test process is divided into three consecutive stages. In the first stage, all nodes are stably running in standard clean air for twelve hours as a baseline reference period. In the second stage, a small amount of weak organic acid solvent is continuously titrated to the surface of the sensor probe of node B to simulate the gradual performance degradation of the sensor caused by chemical corrosion. After observing and recording the fingerprint deviation value of node B showing a consistent monotonic increasing trend, and its confidence score has correspondingly continuously decreased for six hours. After that, the third stage test is started, that is, on the basis of the previous two stages, an alternating electromagnetic field with a frequency of and an intensity of is applied in the whole test space by the built-in Helmholtz coil in the cabin to simulate the external environmental disturbance in the industrial field. In the third stage, the wireless communication module of all nodes detects a jump in the short-term variance of the received signal strength indication, and generates a high environmental disturbance index. At this time, the fingerprint deviation values of the healthily running nodes A, C, and D all show a short and small amplitude jump, but since their environmental disturbance self-adaptive adjustment mechanism is triggered, their confidence scores only show a small amplitude fluctuation and then recover to stability. However, the fingerprint deviation value of node B further superimposes a disturbance pulse on its already high baseline, resulting in a further decrease in its confidence score and maintaining at a very low level. Table 1 is a data snapshot at a certain time in the third stage.

[0039] Table 1: State data table of each node at a certain time in the third stage of the test.

[0040]

[0041] Referring to Table 1, the data shows that the internal mechanism of the application can distinguish events of different properties. The low confidence score of node B is the result of the combined action of its internal state and external environment, and the root cause is the continuous deterioration of the internal state. The system recognizes this fact through the trend of the confidence score which has been continuously decreasing in the second stage. Nodes A, C and D, although exposed to the same external disturbance, their environmental disturbance adaptive adjustment mechanism can distinguish the influence of external disturbance through dynamic adjustment of the confidence score evaluation model, so as to maintain the high reliability of their own data. When the center platform performs weighted fusion, it assigns weights based on the confidence scores of each node , so that the data contribution of the severely deteriorated node B is set to zero, and the weights of other healthy nodes are correspondingly increased, so as to ensure that the regional monitoring conclusion of the final output is not affected by a single faulty node.

[0042] Embodiment 3: This embodiment combines Figs. 1 to 3 the atmospheric environment intelligent monitoring method and system based on the Internet of Things, as shown in Fig. 1 , the edge monitoring node part first collects the original output signal through the sensor module, and then transmits the signal into the preprocessing module to perform digital filtering and feature extraction operation, to obtain the time sequence fingerprint features including variance, kurtosis, frequency energy ratio, etc., to describe the sensor micro-disturbance response mode. The extracted fingerprint features are sent to the evaluation module to calculate the fingerprint deviation value D through baseline comparison, and the confidence score S formula generates a value representing the current data reliability. At this time, the system executes the environmental disturbance adaptive adjustment mechanism, judges whether it is in a severe disturbance situation through RSSI variance analysis, temporarily relaxes the deviation threshold if the condition is met, and finally the communication module reports the binary data set containing the measurement value and the confidence to the center platform. The center platform collects multi-node data through the data receiving module, and submits it to the fusion processing module to perform confidence weighted fusion operation to generate a robust regional conclusion. In order to maintain the consistency of the monitoring network fingerprint baseline, the system introduces a group consensus calibration mechanism to update the node baseline based on the average fingerprint in the dawn window period. The system also has a maintenance and early warning module to analyze and predict the trend of node confidence, and timely identify potential failures. If the prediction result is abnormal, start the hardware survival test process, and use the piezoelectric element residual vibration detection to evaluate whether the node hardware is intact. Finally, the center platform outputs the regional atmospheric environment monitoring conclusion.

[0043] As shown in Fig. 2As shown in the figure, the horizontal axis represents time (hours), and the vertical axis represents the confidence score S, with a scale range from 0 to 1. In the legend, healthy node A is marked with a solid line, deteriorated node B with a dashed line, and healthy node C with a dotted line. The maintenance trigger threshold is indicated by long and short dashes. It can be observed from the figure that the confidence scores S of healthy nodes A and C remain consistently above approximately 0.9 with minimal fluctuations, indicating that their internal fingerprint deviation values ​​are stable and the sensor is in good condition. However, the confidence score S of deteriorated node B shows a continuous downward trend, gradually dropping from an initial value close to 0.9 to below 0.3, significantly lower than the 0.5 level set for the maintenance trigger threshold, indicating that its sensor fingerprint deviation value is continuously increasing and its condition is continuously deteriorating.

[0044] like Fig. 3 As shown, the process begins with an anomaly at node 1. When this node detects a fluctuation in the measured value and meets a set threshold condition, its analog comparator is triggered. Subsequently, it broadcasts a heartbeat containing the fluctuation status bit to surrounding nodes via the Bluetooth broadcast channel, achieving low-power diffusion of event awareness. After receiving the heartbeat packet, neighboring nodes 2 and 3 respectively process the event. Specifically, both receive the heartbeat packet and detect the abnormal fluctuation status, then perform local computation operations, sequentially completing the two steps of fingerprint feature extraction and confidence score calculation, forming a basis for judging their own status and environmental disturbances. Afterward, the neighboring nodes report complete data including sensor measurements and confidence scores. Each node reports a complete data set to the central platform. After receiving data from multiple neighboring nodes, the central platform performs multi-source fusion judgment to complete a comprehensive assessment of the sudden anomaly, i.e., it performs comprehensive analysis of the anomaly event. The process shown in the figure demonstrates the advantages of the event-driven reporting mechanism in this invention. Through Bluetooth broadcasting and the transmission of fluctuation status bits between nodes, it can quickly link with surrounding nodes to conduct collaborative confirmation and confidence assessment when the measurement value fluctuation first appears. This achieves a fast, accurate, and distributed response to anomalies without significantly increasing power consumption, and provides the central platform with a basis for judgment, avoiding single-point false alarms and improving the overall stability and decision-making accuracy of the system.

[0045] Embodiment 4: When the technical solution of the present application is deployed in an urban traffic hub environment with coexistence of slowly-varying seasonal characteristics and sudden strong disturbances, in order for the internal algorithm parameters to obtain a setting matched with the dynamic characteristics of this specific scenario, a set of targeted offline calibration and parameterization procedures need to be performed; the challenge of this scenario lies in the fact that the system not only needs to adapt to the slow drift of temperature and humidity background in units of quarters, but also must be immune to the instantaneous electromagnetic and vibration disturbances generated when the subway trains enter and exit the station, which lasts for tens of seconds, thus precise configuration requirements are put forward for the convergence rate of the group consensus calibration mechanism and the response sensitivity of the environmental disturbance adaptive adjustment mechanism; before the system is formally put into operation, first, the time series fingerprint feature extraction algorithm based on fast Fourier transform in the edge monitoring node is configured, the operation process of this algorithm is that the node collects a raw output signal sequence with a length of sample points, after applying the Hanning window function, fast Fourier transform is performed, from the obtained frequency spectrum, a number of non-overlapping frequency bands are pre-defined, the setting of the frequency bands aims to separate different physical processes, one frequency band is to represent the slow baseline drift of the sensor, another frequency band is to represent its effective response to the target gas, and another frequency band is to represent high-frequency noise, then, the algorithm calculates the signal energy in each frequency band , and the total signal energy , finally, the generated dimensional time series fingerprint vector each component of which is defined by the ratio of the frequency band energy to the total energy, that is , the normalization process aims to eliminate the influence of the overall intensity change of the signal, so that the fingerprint vector reflects the frequency domain structural characteristics of the signal.

[0046] Further, the core adjustment logic of the environmental disturbance adaptive adjustment mechanism is calibrated, in this mechanism, the relationship between the adjusted fingerprint deviation threshold , the original threshold , and the environmental disturbance index is determined as a linear model: , where is the environmental disturbance index threshold that triggers adjustment, and is the coefficient that determines the adjustment sensitivity; to determine the value of the parameter , a healthy edge monitoring node is subjected to a disturbance signal generated by a signal generator, which is consistent with the received signal strength indication fluctuation characteristics of the subway train passing through the station in the field, and the value of is gradually adjusted until the confidence score It can be maintained above 0.85 at this time. The value is adopted as the configuration parameter for this scenario. This procedure ensures that the system will not overreact and misclassify healthy nodes as abnormal when faced with predictable strong disturbances. Finally, the learning rate in the group consensus calibration mechanism is... This parameter, when configured, determines the speed at which a node's baseline fingerprint features align to the network's average fingerprint; Parameter The system is set based on the long-term environmental change rate of the deployment location. The setup process involves first obtaining the daily historical temperature and humidity data released by the meteorological department for the past five years, calculating the maximum rate of change of the monthly average temperature and humidity during the seasonal transitions each year, and using the average of this rate of change as an indicator of the long-term stability of the region's environmental background. Then, according to a pre-defined mapping table, the interval to which this rate of change indicator belongs is mapped to a specific... The value, this mapping table corresponds a higher rate of change range to a larger value. The value corresponds to a smaller range of change rates. This process ensures that the baseline update rate matches the rhythm of natural environmental evolution, preventing baseline instability due to overly rapid updates or the system's inability to keep up with seasonal environmental drift due to overly slow updates. Through the aforementioned series of parameter calibrations and algorithm configurations based on objective data and deterministic processes, all the adaptive and self-learning mechanisms within the entire monitoring network are given clear, reproducible, and highly coupled engineering implementation details with the specific deployment environment before being put into actual operation.

[0047] Example 5: To ensure that each edge monitoring node of the present invention has a stable and traceable initial state when put into use, all nodes undergo a standardized baseline fingerprint solidification procedure before leaving the factory. This procedure places each node under test in a constant temperature and humidity shielded chamber with a standard atmospheric environment. After one hour of thermal stabilization, the node starts to run continuously for eight hours. During this period, it continuously calculates and records its normalized time-series fingerprint vector generated based on fast Fourier transform at five-minute intervals. After the eight-hour test cycle, the system performs statistical processing on all collected fingerprint vectors, removes outliers, calculates its arithmetic mean, and uses this final average vector as the unique baseline fingerprint feature of the node that cannot be changed after leaving the factory, and burns it into its onboard non-volatile memory.

[0048] Immediately after the solidification of the baseline fingerprint feature, the system proceeds to feature calibration of the node's hardware liveness proof mechanism, the test system sends a ping request instruction to the node, upon receiving the instruction, the node drives its on-board piezoelectric electro-acoustic transducer element to ring at a pre-set frequency, and immediately after, it collects the decay voltage signal generated by the element structure's residual vibration, the node performs a fast Fourier transform on the decay signal and extracts the piezoelectric element structure's main resonance frequency from it and quality factor These two parameters together constitute the hardware resonance signature corresponding to the individual physical characteristics of the node, and are stored together with the baseline fingerprint feature; when the node receives a liveness challenge from the central platform in future actual operation, it will repeat the above self-test process, and return the newly measured resonance frequency and quality factor as the liveness proof return, the central platform arbitrates the physical integrity and authenticity of the node by comparing the returned parameters with the pre-stored hardware resonance signature within the pre-set error tolerance.

[0049] Embodiment 6: In the system's whole life cycle management, in order to solidify the determinacy and reproducibility of its operation, a set of standardized engineering procedures covering from factory calibration to in-service node management is implemented; in the hardware resonance signature calibration process, in order to determine the driving parameters of the piezoelectric electro-acoustic transducer element, the test system, after initially identifying the main resonance frequency , will perform a sweep excitation within a pre-set frequency range centered on this frequency, and take the frequency point that produces the maximum amplitude response as the node's final pre-set ringing frequency, while the pre-set duration is set to be five times the signal decay time constant calculated according to the quality factor , this setting aims to capture the complete residual vibration signal and limit energy consumption.

[0050] The system's internal logic judgment conditions also follow quantitative setting procedures after deployment, among which the condition triggering the maintenance warning is algorithmized as the confidence score of an edge monitoring node , which is a moving average containing the past data points, continuously falling below a maintenance trigger threshold for more than twenty-four hours, here, the size of the observation window is determined according to the node's reporting frequency to cover a complete day-night cycle, while the maintenance trigger threshold The confidence score of each healthy node in the past week is dynamically calculated by the central platform according to the average of all confidence scores higher than 0.9 in the whole network, minus twice the standard deviation; similarly, the environmental disturbance index threshold for triggering the adaptive adjustment mechanism of environmental disturbance is not a fixed value, but is periodically self-corrected by each node according to the average of the received signal strength indicator variance in the past seventy-two hours, identified and marked by the central platform during the smooth period.

[0051] When a newly manufactured edge monitoring node carrying the initial baseline fingerprint feature of the factory is added to an already running mature network, in order to smoothly integrate it, a set of new node network synchronization procedures need to be performed; the new node is set to listen mode within the initial seventy-two hours of deployment, in this mode it normally performs all calculations, but its measured values and confidence scores do not participate in the regional weighted fusion of the central platform; the node normally receives the average fingerprint representing the group consensus broadcast by the central platform during the dawn event window period, and with a larger initial learning rate aligns its own baseline fingerprint feature to the average fingerprint; when the distance between the node's own fingerprint and the network average fingerprint is less than a preset synchronization completion threshold for three consecutive calibration periods, the central platform automatically switches it to active mode, making its data officially integrated into the fusion calculation of the whole network.

[0052] It is obvious to those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application.

[0053] Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application and are not limiting, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application.

Claims

1. An atmospheric environment intelligent monitoring method based on Internet of Things, characterized in that, The method comprises the following steps: Step a, the edge monitoring node collects the original output signal time series data sequence of the sensor of the edge monitoring node itself within a certain time period; step b, the edge monitoring node pre-processes the time series data sequence, performs digital filtering operation to attenuate the environmental background noise component in the time series data sequence, and extracts the time series fingerprint feature capable of representing the current inherent physical and chemical characteristics of the sensor; Step c, the edge monitoring node calculates the distance between the current time series fingerprint feature and the baseline fingerprint feature stored in the edge monitoring node in advance to obtain a fingerprint deviation value, which quantitatively reflects the deviation degree of the current inherent physical and chemical characteristics of the sensor compared with the baseline fingerprint feature; step d, the edge monitoring node generates a confidence score between 0 and 1 through a nonlinear mapping function according to the fingerprint deviation value, which represents the reliability of the current measurement data of the sensor, wherein the confidence score decreases as the fingerprint deviation value increases; step e, the edge monitoring node reports the binary data set containing the measurement value and the confidence score of the edge monitoring node to the network; step f, the center platform receives the binary data set from multiple edge monitoring nodes, and performs weighted fusion on the measurement values based on the confidence scores to dynamically reduce the weight of the measurement values with confidence scores lower than a predetermined threshold in the final regional atmospheric environment monitoring conclusion, thereby generating a robust regional atmospheric environment monitoring conclusion, and triggering a maintenance warning according to the continuous downward trend of the confidence scores of the edge monitoring nodes. 2.The atmospheric environment intelligent monitoring method based on the Internet of Things according to claim 1, characterized in that, In step b, the time series fingerprint feature includes the variance of the time series data sequence and the kurtosis of the time series data sequence, and the variance and the kurtosis together constitute a two-dimensional time series fingerprint vector for representing the fluctuation amplitude and outlier distribution characteristics of the signal. The time series fingerprint feature is obtained by performing fast Fourier transform on the time series data sequence to extract the energy ratio of a plurality of predetermined frequency bands, thereby forming a multi-dimensional time series fingerprint vector. The energy ratio represents the frequency domain response mode of the sensor to the environmental perturbation. 3.The atmospheric environment intelligent monitoring method based on the Internet of Things according to claim 1, characterized in that, In step d, the nonlinear mapping function is a Sigmoid function, and its calculation formula is: wherein, is a confidence score, is a fingerprint deviation value, is a steepness adjustment parameter of the Sigmoid function, and is a center offset parameter of the Sigmoid function. 4.The atmospheric environment intelligent monitoring method based on the Internet of Things according to claim 1, characterized in that, The method further comprises the following steps: the edge monitoring node periodically broadcasts a heartbeat packet through the Bluetooth low power broadcast mode at a certain broadcast frequency, the heartbeat packet containing the identification information of the edge monitoring node and the fluctuation state bit obtained through the analog comparator, the fluctuation state bit indicating whether the measurement value exceeds a certain fluctuation threshold; and the adjacent edge monitoring node receiving the heartbeat packet, after receiving the heartbeat packet indicating that the measurement value fluctuation exceeds the preset threshold, triggers the extraction of the time series fingerprint feature and the calculation of the fingerprint deviation value in steps b and c, and reports the calculated measurement value and confidence score of the edge monitoring node through the main communication channel. 5.The atmospheric environment intelligent monitoring method based on the Internet of Things according to claim 1, characterized in that, The method further comprises the following steps: the edge monitoring node identifies two time windows in which the rate of change of the daily light intensity reaches a maximum value using the light-sensitive element, and determines the dawn event window period and the dusk event window period; in the dawn event window period, the edge monitoring node performs the time sequence fingerprint feature extraction and confidence score evaluation of steps b and c, and reports a dawn report containing the edge monitoring node identification information, the current confidence score, and the current time sequence fingerprint to the central platform; the central platform collects the dawn reports, screens out a group of edge monitoring nodes with confidence scores higher than a pre-determined high confidence score threshold, and calculates the average fingerprint of the current time sequence fingerprint of all edge monitoring nodes in the group; and the central platform broadcasts the average fingerprint as a correction parameter to all edge monitoring nodes in the network, and after the edge monitoring nodes receive the correction parameter, the baseline fingerprint feature of the edge monitoring nodes is updated based on the preset learning rate. 6.The atmospheric environment intelligent monitoring method based on the Internet of Things according to claim 1, characterized in that, The method further comprises the following steps: when the central platform determines that the edge monitoring node needs maintenance according to the continuous confidence score of the edge monitoring node being lower than a pre-determined maintenance trigger threshold, the central platform suspends the dispatch of the maintenance instruction and sends a echo request instruction to the edge monitoring node; after the edge monitoring node receives the echo request instruction, the edge monitoring node drives the piezoelectric electro-acoustic transducer element on board to emit a physical ringing at a preset ringing frequency and for a preset duration; the edge monitoring node then switches the input / output port mode of the piezoelectric electro-acoustic transducer element from driving output to high-impedance analog input, and listens to the weak decay voltage signal generated by the element structure after the ringing; and the edge monitoring node returns the integral value and / or peak value of the residual vibration signal to the central platform as a survival proof, and the central platform arbitrates whether to dispatch a hardware repair work order according to whether the valid survival proof is received within a preset time limit. 7.The atmospheric environment intelligent monitoring method based on the Internet of Things according to claim 3, characterized in that, The method further comprises the following steps: from the results of the fast Fourier transform, the amplitude of the fundamental wave and the amplitude of the second harmonic of the fundamental wave are extracted; and based on the ratio of the amplitude of the fundamental wave to the amplitude of the second harmonic, the edge monitoring node generates an index for characterizing the electrical health status of the edge monitoring node, which is reported in parallel with the sensor confidence score to assist the central platform in more comprehensive node fault diagnosis. 8.The atmospheric environment intelligent monitoring method based on the Internet of Things according to claim 1, characterized in that, The method further comprises the following steps: the edge monitoring node analyzes the short-term variance of the received signal strength indication of the signals received by the wireless communication module from the adjacent edge monitoring nodes to generate an environmental disturbance index; and when the environmental disturbance index exceeds a pre-determined environmental disturbance index threshold, the edge monitoring node adjusts the fingerprint deviation threshold used to generate the confidence score within a preset range to distinguish between the time sequence fingerprint changes caused by internal state degradation and external environmental disturbance. 9.The atmospheric environment intelligent monitoring method based on the Internet of Things according to claim 1, characterized in that, In step b, the preprocessing includes nonlinear filtering processing on the time series data sequence, and the nonlinear filtering processing aims to enhance the micro-disturbance characteristics reflecting the inherent physical and chemical characteristics of the sensor in the time series data sequence, and suppress the common mode interference caused by the environmental drastic change, so that the extracted time series fingerprint feature is more pure to represent the sensor state itself.

10. An atmospheric environment intelligent monitoring system based on Internet of Things, characterized in that, Comprise: The edge monitoring subsystem specifically comprises: a sensor module for collecting raw output signal time series data; a preprocessing module coupled to the sensor module and configured to perform digital filtering operation on the time series data and extract time series fingerprint features; an evaluation module coupled to the preprocessing module and configured to calculate the distance between the current time series fingerprint features and the baseline fingerprint features to obtain a fingerprint deviation value, and generate a confidence score according to the fingerprint deviation value; a communication module coupled to the evaluation module and configured to report the measurement value and the binary data set of the confidence score to the network; a central platform subsystem specifically comprising: a data receiving module for receiving the binary data set from multiple edge monitoring nodes; a fusion processing module coupled to the data receiving module and configured to weight and fuse the measurement values based on the confidence score, so as to dynamically reduce the weight of the measurement value with a confidence score lower than a predetermined threshold in the final regional atmospheric environment monitoring conclusion, thereby generating a robust regional atmospheric environment monitoring conclusion; a maintenance warning module coupled to the fusion processing module and configured to trigger a maintenance warning according to the continuous downward trend of the confidence score of each edge monitoring node.

Images