An environmental parameter monitoring method and system based on a mobile internet of things

Abstract

The application discloses an environment parameter monitoring method and system based on a mobile Internet of Things, which comprises the following steps: collecting environment parameters, performing time synchronization and removing outliers to obtain primary monitoring data; generating a corresponding time-varying topology graph; embedding each node of the time-varying topology graph into a hyperbolic space, propagating the primary monitoring data, and obtaining a global estimation result of the environment parameters; monitoring the fluctuation of the environment parameters of each node in real time, and generating an abnormal trigger signal when the fluctuation exceeds a preset threshold; starting a geometric folding channel to obtain a folding result when the abnormal trigger signal is received; forming an abnormal cluster based on the folding result; and uploading the abnormal cluster information to the cloud for optimization. The application realizes stable estimation of environment parameters under a dynamic topology through a hyperbolic space embedding and an abnormal trigger geometric folding mechanism, and improves the accuracy and interpretability of monitoring.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring, and in particular to a method and system for monitoring environmental parameters based on the mobile Internet of Things. Background Technology

[0002] With the rapid development of the mobile Internet of Things (IoT), a large number of vehicle-mounted terminals, drone terminals, and wearable terminals are being used for environmental monitoring, continuously collecting multi-source environmental parameters such as temperature, humidity, gas concentration, particulate matter, noise intensity, light intensity, and wind speed. These terminals are highly dynamic and heterogeneous in space, and their data collection covers a wide range, providing important data support for air pollution monitoring, urban environmental governance, and emergency early warning.

[0003] Existing environmental monitoring methods largely rely on traditional convolutional neural networks, recurrent neural networks, or ordinary graph neural networks to fuse and analyze collected data. However, these methods mostly assume that sensor nodes are relatively fixed in space, making it difficult to adapt to the time-varying topological characteristics brought about by mobile terminals. Furthermore, differences in sampling frequency, delays, and drift between different sensors can easily lead to biases in the fusion results. In addition, during sudden environmental changes or localized outbreaks of pollution, traditional methods often rely on simple threshold detection or classifier output anomalies, lacking geometric interpretability.

[0004] Therefore, how to provide a method and system for monitoring environmental parameters based on mobile Internet of Things is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an environmental parameter monitoring method and system based on mobile Internet of Things (IoT). This invention effectively maintains the hierarchical structure and low-distortion representation of the dynamic topology by embedding and convolutionally propagating the time-varying topology formed by mobile IoT terminals in hyperbolic space, thereby improving the stability of large-scale environmental parameter monitoring. When an anomaly in environmental parameters is detected, the geometric folding channel triggered by the anomaly causes controlled compression of the local embedded region, automatically forming anomaly clusters, thus improving the sensitivity and interpretability of anomaly detection. Simultaneously, by continuously optimizing the trigger threshold and folding depth function in the cloud and transmitting the data back to the edge device, online self-calibration under edge-cloud collaboration is achieved, ensuring the real-time performance and long-term adaptability of the monitoring.

[0006] An environmental parameter monitoring method based on mobile Internet of Things according to an embodiment of the present invention includes the following steps:

[0007] Environmental parameters are collected by multiple mobile IoT terminals, and the environmental parameters are synchronized in time and outlier removal is performed by an edge gateway to obtain aligned primary monitoring data.

[0008] Based on the location, movement trajectory, and communication link of each mobile IoT terminal, a corresponding time-varying topology map is generated;

[0009] Each node of the time-varying topology graph is embedded into a hyperbolic space, and the primary monitoring data is propagated in the hyperbolic space using graph convolution kernels to obtain a global estimation result of environmental parameters.

[0010] Real-time monitoring of environmental parameter fluctuations at each node; when the fluctuation exceeds a preset threshold, an abnormal trigger signal is generated.

[0011] Upon receiving the abnormal trigger signal, the geometric folding channel is activated to locally fold the embedding space of the corresponding region, so that the embedding representations of neighboring nodes are compressed and clustered in the hyperbolic space, thus obtaining the folding result.

[0012] Based on the folding results, anomaly clusters are formed, and these anomaly clusters are marked as pollution sources or abnormal hotspot areas.

[0013] The abnormal cluster information is uploaded to the cloud, where the trigger threshold and folding depth function of the geometric folding channel are optimized, and the trigger threshold and folding depth function are sent back to the edge device.

[0014] Furthermore, the environmental parameters include:

[0015] At least one of temperature, humidity, gas concentration, particulate matter concentration, noise intensity, light intensity, and wind speed;

[0016] The time synchronization and outlier removal steps include:

[0017] Establish a unified time base by using network time protocols or high-precision time synchronization mechanisms to align data collected by different terminals to the same time scale;

[0018] Outlier removal is performed on the aligned data, including interval constraint judgment, historical mean deviation detection and multimodal cross-validation. When the abnormal conditions are met, the data is marked and removed. At the same time, node identifiers and collection location identifiers are added to the remaining data in the edge gateway to form primary monitoring data with spatiotemporal index.

[0019] Furthermore, the step of generating the corresponding time-varying topology graph includes:

[0020] Based on the location information of the terminal, the geographic coordinates and movement direction of the node are extracted, and the trajectory vector is constructed using the location information at continuous time.

[0021] The edge connection relationships between nodes are determined based on the communication link quality, geographical proximity, and topological history between nodes;

[0022] In the generated topology, a weight is assigned to each edge, which is calculated by combining the stability of the communication link, spatial distance, and similar motion trends;

[0023] The nodes and edges are then sliced ​​and updated over time to form a time-varying topology graph.

[0024] Furthermore, the step of adding weights to each edge in the generated topology includes:

[0025] Based on the historical transmission records of the communication link from the edge gateway, the average packet loss rate, round-trip delay variance, and connection hold duration of the link are calculated and normalized to obtain the first weighted component representing communication stability.

[0026] Based on the geographic coordinates of two nodes in the same time slice, their spatial Euclidean distance is calculated, and the distance is transformed by a Gaussian kernel function to obtain a second weighted component representing spatial proximity.

[0027] Based on the position sequence of nodes in continuous time slices, the corresponding motion trajectory vectors are generated. By calculating the difference between the cosine similarity of the motion direction and the velocity magnitude, a third weight component representing the similar motion trend is obtained.

[0028] The first weight component, the second weight component, and the third weight component are fused according to a preset weighting formula to obtain the comprehensive weight value of the edge, wherein the weighting formula is a linear weighting function, and the fusion result is limited to a preset numerical range.

[0029] Furthermore, the generation of the global estimation results of the environmental parameters includes:

[0030] Based on the node set of the time-varying topology graph, an initial feature vector is established according to the spatiotemporal index of each node. The feature vector includes environmental parameter observations, node position coordinates, and edge weight information.

[0031] The initial feature vector is mapped from Euclidean space to the tangent space of hyperbolic space using an exponential mapping function, and then the node embedding representation in hyperbolic space is obtained through a logarithmic mapping function.

[0032] On the hyperbolic space embedding representation, an adjacency matrix with geodesic distance as the metric is constructed, and convolution propagation is performed using the adjacency matrix and convolution kernel parameters to update the embedding features of the nodes;

[0033] The node embedding features after convolutional propagation are fused with the primary monitoring data to generate environmental parameter estimation results covering the entire map.

[0034] Furthermore, the step of generating the abnormal trigger signal includes:

[0035] Based on the node embedding features after convolutional propagation and the primary monitoring data, the changes in environmental parameters of each node in continuous time slices are calculated. The changes include the difference value, the rate of change, and the direction of change.

[0036] The change is compared with the historical statistical baseline of the node to generate a first detection index representing the degree of short-term deviation.

[0037] The change is compared with the average change trend within the same topological neighborhood to generate a second detection index representing spatial inconsistency.

[0038] The first detection index and the second detection index are input into the threshold judgment function. If the output result exceeds the preset threshold, an abnormal trigger signal is generated, and the identifier of the trigger node, the trigger time and the trigger intensity are recorded.

[0039] The abnormal trigger signal is passed as the activation condition for the geometric folding channel.

[0040] Furthermore, the step of generating the folding result includes:

[0041] Based on the trigger node, trigger time, and trigger intensity in the abnormal trigger signal, the local embedding region to be folded is determined;

[0042] Within the local embedding region, a folding depth parameter is calculated based on the trigger intensity, and the folding depth parameter defines the shrinkage ratio of the geodesic distance between embedding points;

[0043] A folding transformation is performed on the node embedding representation of the local embedding region. The folding transformation includes applying a nonlinear compression function in hyperbolic space to reduce the original geodesic distance by the folding depth parameter.

[0044] After the folding transformation, the embedding representation of the affected nodes is uniformly reorganized, and the folding result is output.

[0045] Furthermore, the step of forming anomaly clusters based on the folding results and marking the anomaly clusters as pollution sources or anomalous hotspot areas includes:

[0046] In the folded embedding representation, a clustering algorithm based on geodesic distance is used to divide the shrunken clustered node set into several clusters, with each cluster corresponding to a candidate anomaly region;

[0047] Calculate the center location, coverage area, and number of member nodes of each candidate anomaly region to obtain the spatial attributes of the anomaly cluster;

[0048] We performed weighted average and variance analysis on the environmental parameters collected from the member nodes of each anomaly cluster to obtain the parameter characteristics of the anomaly cluster.

[0049] The spatial attributes and parameter features are input into the labeling rule base for judgment. If the pollution source conditions are met, it is labeled as a pollution source. If the local disturbance conditions are met, it is labeled as an abnormal hotspot area.

[0050] The output includes anomaly cluster labeling results containing geographic range, parameter type, and confidence level.

[0051] Furthermore, the step of uploading the abnormal cluster information to the cloud and having the cloud optimize the trigger threshold and folding depth function of the geometric folding channel and then transmit it back to the edge device includes:

[0052] The results of anomaly cluster annotation formed on the edge side, along with trigger node identifiers, folding depth parameters, and historical monitoring data, will be uploaded to the cloud processing module.

[0053] Statistical analysis is performed on the labeling results of multiple batches of abnormal clusters in the cloud to calculate the fitness distribution of the trigger threshold, and the trigger threshold is adaptively updated based on the fluctuation characteristics of environmental parameters.

[0054] The relationship between the folding depth parameter and the anomaly detection accuracy is fitted in the cloud to obtain the optimized folding depth function;

[0055] The updated trigger threshold and fold depth function are fed back to the edge device as optimization parameters.

[0056] An environmental parameter monitoring system based on mobile Internet of Things according to an embodiment of the present invention includes the following modules:

[0057] The data acquisition and preprocessing module is used to collect environmental parameters through multiple mobile IoT terminals, and to use an edge gateway to perform time synchronization and outlier removal on the environmental parameters to obtain aligned primary monitoring data.

[0058] The topology building module is used to generate a corresponding time-varying topology map based on the location, movement trajectory and communication link of each mobile IoT terminal;

[0059] The hyperbolic embedding and convolutional propagation module is used to embed each node of the time-varying topology graph into the hyperbolic space, and propagate the primary monitoring data in the hyperbolic space using graph convolution kernels to obtain the global estimation results of environmental parameters.

[0060] An anomaly detection module is used to monitor the fluctuations of environmental parameters of each node in real time, and generate an anomaly trigger signal when the fluctuations exceed a preset threshold.

[0061] The geometric folding module is used to activate the geometric folding channel when the abnormal trigger signal is received, and to locally fold the embedding space of the corresponding region, so that the embedding representation of the neighboring nodes is compressed and clustered in the hyperbolic space to obtain the folding result.

[0062] An anomaly cluster generation and labeling module is used to generate anomaly clusters based on the folding results and label the anomaly clusters as pollution sources or abnormal hotspot areas.

[0063] The cloud optimization module is used to upload the abnormal cluster information to the cloud, whereby the cloud optimizes the trigger threshold and folding depth function of the geometric folding channel, and then sends the updated trigger threshold and folding depth function back to the edge device.

[0064] The beneficial effects of this invention are:

[0065] This invention effectively maintains the hierarchical relationship and low-distortion representation of the dynamic topology by embedding and convolutional propagation of the time-varying topology formed by mobile IoT terminals in hyperbolic space. It can still obtain stable global environmental parameter estimation results under large-scale mobile node conditions, thereby improving the comprehensiveness and robustness of monitoring.

[0066] This invention employs anomaly-triggered geometric folding channels when abnormal environmental parameters are detected. This causes controlled compression of the locally embedded region in hyperbolic space, automatically aggregating abnormal nodes into anomaly clusters. This mechanism not only improves the sensitivity of anomaly detection but also enhances the ability to locate and interpret anomaly sources through the visual aggregation of geometric folding.

[0067] This invention achieves online self-calibration under edge-cloud collaboration by continuously optimizing the trigger threshold and folding depth function in the cloud and sending the optimization results back to the edge device. This ensures both the real-time monitoring of the edge side and the adaptability and accuracy of the system in long-term operation, thus possessing strong engineering application value. Attached Figure Description

[0068] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0069] Figure 1 This is a flowchart of an environmental parameter monitoring method based on mobile Internet of Things proposed in this invention;

[0070] Figure 2 This is a flowchart of the time-varying topology generation process for an environmental parameter monitoring method based on mobile Internet of Things proposed in this invention.

[0071] Figure 3 This is a schematic diagram of the anomaly triggering and geometric folding of an environmental parameter monitoring method based on mobile Internet of Things proposed in this invention.

[0072] Figure 4 This is a data processing flowchart of an environmental parameter monitoring method based on mobile Internet of Things proposed in this invention. Detailed Implementation

[0073] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0074] refer to Figure 1 - Figure 4 An environmental parameter monitoring method based on mobile Internet of Things includes the following steps:

[0075] Environmental parameters are collected by multiple mobile IoT terminals, and the environmental parameters are synchronized in time and outlier removal is performed by an edge gateway to obtain aligned primary monitoring data.

[0076] Based on the location, movement trajectory, and communication link of each mobile IoT terminal, a corresponding time-varying topology map is generated;

[0077] Each node of the time-varying topology graph is embedded into a hyperbolic space, and the primary monitoring data is propagated in the hyperbolic space using graph convolution kernels to obtain a global estimation result of environmental parameters.

[0078] Real-time monitoring of environmental parameter fluctuations at each node; when the fluctuation exceeds a preset threshold, an abnormal trigger signal is generated.

[0079] Upon receiving the abnormal trigger signal, the geometric folding channel is activated to locally fold the embedding space of the corresponding region, so that the embedding representations of neighboring nodes are compressed and clustered in the hyperbolic space, thus obtaining the folding result.

[0080] Based on the folding results, anomaly clusters are formed, and these anomaly clusters are marked as pollution sources or abnormal hotspot areas.

[0081] The abnormal cluster information is uploaded to the cloud, where the trigger threshold and folding depth function of the geometric folding channel are optimized, and the trigger threshold and folding depth function are sent back to the edge device.

[0082] In this embodiment, the environmental parameters include:

[0083] At least one of temperature, humidity, gas concentration, particulate matter concentration, noise intensity, light intensity, and wind speed; specifically, mobile IoT terminals can flexibly select parameter combinations based on the type of sensors they carry and the deployment scenario. For example, vehicle-mounted terminals focus on collecting gas concentration and particulate matter concentration, drone terminals focus on collecting wind speed and temperature and humidity, and wearable terminals focus on collecting noise intensity and light intensity. Through multi-terminal, multi-modal parameter collection, comprehensive coverage of the environmental conditions of the target area can be achieved.

[0084] The time synchronization and outlier removal steps include:

[0085] A unified time reference is established by using network time protocols or high-precision time synchronization mechanisms to align data collected by different terminals to the same time scale. After receiving data uploaded by multiple mobile IoT terminals, the edge gateway aligns the timestamps of each data packet with the unified reference, so that the sampling of different terminals at the same time can be mapped to a consistent time scale, thereby ensuring the temporal consistency of subsequent fusion analysis.

[0086] In terms of outlier removal, the edge gateway performs multiple checks on the aligned primary data.

[0087] Perform interval constraint judgment: When the collected value exceeds the preset physical range (e.g., the temperature is below -50℃ or above 80℃, or the gas concentration is above a reasonable threshold), it is directly judged as abnormal and removed.

[0088] Historical mean deviation detection: By calculating the mean and standard deviation of a node over a past time window, when the current sampled value deviates from the mean by more than a set multiple of the standard deviation, it is determined to be an outlier.

[0089] Perform multimodal cross-validation: When there are significant contradictions in the data collected by different sensors in the same spatial neighborhood (such as abnormally large differences in gas concentration between adjacent nodes), the outlier is marked and removed.

[0090] After removing outliers, the edge gateway adds node and collection location identifiers to each retained data record to form primary monitoring data with spatiotemporal index. This primary monitoring data not only contains numerical information but also includes a clear collection source and geographical location, which is helpful for subsequent topology construction and hyperbolic space embedding.

[0091] In this embodiment, the step of generating the corresponding time-varying topology graph includes:

[0092] Mobile IoT terminals acquire their latitude and longitude coordinates via a Global Positioning System (GPS) receiver and record their velocity magnitude and direction via an inertial measurement unit. The edge gateway synchronously collects the coordinates and velocity vectors of each node under a unified time reference and calculates the node's trajectory vector based on the position information from two consecutive time slices. The trajectory vector includes displacement increment, velocity change, and direction angle.

[0093] The edge gateway establishes edge connections based on communication link data and spatial location relationships between nodes. Specifically: when the signal strength between two nodes is greater than -70dBm and the round-trip delay is less than 50 milliseconds, an edge connection is established; when the Euclidean distance between nodes is less than 100 meters, an edge connection is also established; and when at least 3 out of 5 historical time slices remain connected, the edge connection relationship remains unchanged.

[0094] In the generated topology graph, each edge is assigned a weight. The weight is calculated as follows:

[0095] Based on communication link statistics, the average packet loss rate, round-trip delay variance and connection hold duration of the link in the past 10 time slices are calculated. The communication stability component is obtained by summing the three after linear normalization.

[0096] Based on Euclidean distance between nodes Through function Obtain the spatial proximity component;

[0097] Based on the node trajectory vector, the cosine similarity of its velocity direction and the difference between its velocity magnitude are calculated, and the weighted sum of the two is taken as the motion trend similarity component.

[0098] The edge gateway linearly weights the three types of components according to a weight ratio of 0.4:0.3:0.3 to obtain the comprehensive weight value of the edge.

[0099] The edge gateway updates the state of all nodes and edges in 1-second time slices to form a time-varying topology graph. This time-varying topology graph accurately describes the topology and connectivity of the mobile IoT terminal during the time evolution process and serves as the input for subsequent hyperbolic space embedding and convolutional propagation.

[0100] In this embodiment, the generation of the global estimation results of the environmental parameters includes:

[0101] The edge gateway constructs an initial feature vector for each node based on the spatiotemporal index of each node in the time-varying topology graph. This feature vector consists of three parts: first, the environmental parameter values ​​collected by the node, including real-time observation data such as temperature, humidity, and gas concentration; second, the latitude and longitude coordinates of the node in the current time slice; and third, the comprehensive weight value of the edges connected to the node.

[0102] Using the exponential mapping function in hyperbolic geometry, the initial eigenvectors are mapped from the Euclidean tangent space to the hyperbolic space. Specifically, at the origin of the tangent space, for each node's eigenvector... Calculate the exponential mapping This yields the embedding coordinates of the nodes in hyperbolic space; then, through a logarithmic mapping function... Transform the embedded coordinates back into tangent space for distance metrics and convolution operations with the embeddings of other nodes.

[0103] Based on the hyperbolic space embedding results, an adjacency matrix is ​​constructed. Each element of the adjacency matrix is ​​determined by the geodesic distance between node pairs, and the geodesic distance is calculated using the following formula:

[0104] ;

[0105] in Embed coordinates for nodes in hyperbolic space.

[0106] During convolution propagation, a graph convolution kernel is used to operate on the adjacency matrix and node features. During each convolution update, the node feature vectors aggregate information from their neighboring nodes according to a weighted distribution. The aggregation function is defined as:

[0107] ;

[0108] in For the first Layer node embedding features The normalized adjacency weight is calculated based on geodesic distance. The convolution kernel parameter matrix, This is the activation function.

[0109] The node embedding features after convolutional propagation are fused with the primary monitoring data to obtain a global estimation result of environmental parameters covering the entire map. The fusion method is as follows: the convolutional output of the same node is weighted and summed with the original environmental parameter value in a ratio of 0.7:0.3 to generate an environmental parameter estimation result covering the entire map.

[0110] In this embodiment, the step of generating the abnormal trigger signal includes:

[0111] After acquiring the node embedding features and primary monitoring data following convolutional propagation, the edge gateway calculates the parameter changes of each node across consecutive time slices. These changes include:

[0112] The parameter difference is defined as the current observation value minus the previous observation value.

[0113] The rate of change of a parameter is defined as the ratio of the difference value to the time interval.

[0114] The direction of parameter change is defined as the sign of the rate of change.

[0115] The edge gateway generates two types of detection metrics for each node:

[0116] The first detection metric is to compare the difference value of a node with its historical mean and standard deviation over the past 30 seconds. When the difference value exceeds the historical mean ± 3 times the standard deviation, an abnormal deviation score is output.

[0117] The second detection metric is to compare the difference value of a node with the average change of the same adjacent subgraph. When the difference between the two exceeds twice the standard deviation of the neighborhood, the spatial inconsistency score is output.

[0118] The first detection index and the second detection index are input into the threshold determination function. The function is defined as follows:

[0119] ;

[0120] in, Represents a node The primary detection indicator This indicates the second detection indicator. It is fixed at 0.6.

[0121] when When the node is determined to have triggered an anomaly at the current moment, the threshold is determined. Set to 0.8.

[0122] The edge gateway generates an anomaly trigger signal. The signal includes the unique identifier of the triggering node, the timestamp of the trigger time, and the trigger strength (i.e., the magnitude by which it exceeds the threshold). This anomaly trigger signal serves as the input condition for the geometric folding channel and is passed to subsequent steps.

[0123] In this embodiment, the step of generating the folding result includes:

[0124] The edge gateway parses the abnormal trigger signal, reading the trigger node identifier, trigger timestamp, and trigger strength. Centered on the trigger node, it extracts its first-order and second-order neighbor nodes from the time-varying topology graph, forming a local embedding region to be folded.

[0125] The folding depth parameter is calculated based on the anomaly trigger strength. Specifically: when the trigger strength value is... At that time, folding depth parameter According to the formula Determined, and limited to an interval This folding depth parameter is used to control the shrinkage ratio of the geodesic distance.

[0126] A folding transformation is performed on the node coordinates within the locally embedded region. This folding transformation is achieved by applying a nonlinear compression function in hyperbolic space, defined as:

[0127] ;

[0128] in For nodes The original geodetic distance, Let be the geodesic distance after folding. Through this function, the geodesic distance within the folded region is proportionally shortened, thereby causing the embedding representation of relevant nodes in hyperbolic space to shrink and cluster.

[0129] After the folding transformation is completed, the edge gateway performs a consistent reshaping of the embedded representations of the affected nodes. Specifically, it performs a smooth correction on the distances between the boundary nodes of the folded region and their neighboring nodes in the non-folded region to ensure the embedding continuity between the folded and non-folded regions.

[0130] The output folding result is used as input to the anomaly cluster generation and labeling step for subsequent anomaly clustering and classification.

[0131] In this embodiment, the step of forming anomaly clusters based on the folding results and marking the anomaly clusters as pollution sources or anomalous hotspot areas includes:

[0132] After receiving the folded embedding results, the edge gateway performs clustering analysis using the geodesic distance matrix. A hierarchical clustering algorithm based on geodesic distance is employed, merging nodes in the folded region according to their geodesic distance from smallest to largest. Merging stops when the maximum geodesic distance within a cluster is less than 0.3. This ultimately generates several candidate clusters, each corresponding to a potential anomaly region.

[0133] Calculate the spatial properties of each candidate cluster. The cluster center is the weighted average of the embedded coordinates of all nodes within the cluster. The coverage of the cluster is determined by the geodesic distance from the center to the farthest node, and the size of the cluster is characterized by the number of nodes.

[0134] Environmental parameter features are extracted for each cluster. A weighted average of the environmental parameter values ​​for all nodes within the cluster is calculated, and the variance index is then determined. The weighting coefficients are related to the fold depth parameter; the greater the fold depth, the higher the weight of the corresponding node's parameter. The final environmental parameter features of the cluster are expressed as follows: ,in For weighted average, This is the weighted variance.

[0135] Spatial attributes and parametric features are input into the annotation rule base for judgment. The annotation rule base contains two types of rules:

[0136] When the coverage area of ​​a cluster is greater than 50 meters and the average gas concentration or particulate matter concentration exceeds the national standard limit, the cluster is marked as a pollution source.

[0137] When a cluster's coverage area is less than 50 meters and the mean of its parameters is normal but the variance exceeds the threshold, the cluster is marked as an abnormal hotspot area.

[0138] Generate abnormal cluster annotation results. The results include the cluster's center coordinates, coverage area, main parameter types, annotation category, and confidence score, and are pushed to the visualization interface for alerts and remediation suggestions.

[0139] In this embodiment, the step of uploading the abnormal cluster information to the cloud, optimizing the trigger threshold and folding depth function of the geometric folding channel by the cloud, and then sending it back to the edge device includes:

[0140] After generating and labeling abnormal clusters, the edge gateway uploads the results along with relevant contextual information to the cloud processing module. This information includes: the center coordinates of the abnormal cluster, its coverage area, labeling category, key parameter characteristics (mean and variance) within the cluster, trigger node identifier, trigger timestamp, and the folding depth parameter at the time of triggering. The upload frequency is set to once every 10 seconds, or immediately upon the appearance of a new abnormal cluster.

[0141] The cloud processing module performs statistical analysis on the abnormal cluster results from multiple batches. Specifically, this includes: counting the number of abnormal clusters and the false positive rate under the trigger threshold, calculating the correlation curve between the trigger threshold and the detection accuracy; and simultaneously, statistically analyzing the relationship between the folding depth parameter and the aggregation degree of abnormal clusters, and establishing a fitting function model between the two.

[0142] The cloud platform updates the trigger threshold and fold depth function based on the above statistical results. The trigger threshold is determined by maximizing the F1-score, a comprehensive indicator of precision and recall. The fold depth function is optimized by minimizing the variance within anomaly clusters and maximizing the difference between anomaly clusters. The optimized fold depth function adopts a piecewise linear function form. For example, when the trigger intensity is in the range [0,5], the fold depth is linearly related to the intensity; when the trigger intensity exceeds 5, the fold depth is fixed at 0.5.

[0143] An optimized parameter package is generated in the cloud, which includes a new trigger threshold, optimized folding depth function parameters, and an updated timestamp, and is then sent to the edge gateway via an encrypted channel.

[0144] Upon receiving the parameter packet, the edge gateway immediately updates its local anomaly detection module and geometric folding channel module, enabling subsequent anomaly detection and folding operations to be based on the latest thresholds and functions, thereby achieving online self-calibration and continuous optimization of the monitoring system.

[0145] An environmental parameter monitoring system based on mobile Internet of Things includes the following modules:

[0146] The data acquisition and preprocessing module is used to collect environmental parameters through multiple mobile IoT terminals, and to use an edge gateway to perform time synchronization and outlier removal on the environmental parameters to obtain aligned primary monitoring data.

[0147] The topology building module is used to generate a corresponding time-varying topology map based on the location, movement trajectory and communication link of each mobile IoT terminal;

[0148] The hyperbolic embedding and convolutional propagation module is used to embed each node of the time-varying topology graph into the hyperbolic space, and propagate the primary monitoring data in the hyperbolic space using graph convolution kernels to obtain the global estimation results of environmental parameters.

[0149] An anomaly detection module is used to monitor the fluctuations of environmental parameters of each node in real time, and generate an anomaly trigger signal when the fluctuations exceed a preset threshold.

[0150] The geometric folding module is used to activate the geometric folding channel when the abnormal trigger signal is received, and to locally fold the embedding space of the corresponding region, so that the embedding representation of the neighboring nodes is compressed and clustered in the hyperbolic space to obtain the folding result.

[0151] An anomaly cluster generation and labeling module is used to generate anomaly clusters based on the folding results and label the anomaly clusters as pollution sources or abnormal hotspot areas.

[0152] The cloud optimization module is used to upload the abnormal cluster information to the cloud, whereby the cloud optimizes the trigger threshold and folding depth function of the geometric folding channel, and then sends the updated trigger threshold and folding depth function back to the edge device.

[0153] Example 1:

[0154] To verify the feasibility of this invention in practice, it was applied to an air quality monitoring scenario on a main road in a city. Two vehicle-mounted terminals, one drone terminal, one cleaning robot terminal, and one wearable terminal were deployed in the test area. The sampling period was one second. The edge gateway handled time synchronization and outlier removal, while the cloud was responsible for parameter optimization and data transmission.

[0155] Table 1. Primary monitoring data after collection and alignment.

[0156] node Terminal type PM2.5 (µg / m³) <![CDATA[NO2(ppm)]]> Temperature (°C) humidity(%) Wind speed (m / s) C1 In-vehicle 58 0.035 29.4 48 1.2 C2 In-vehicle 172 0.096 29.7 46 1.5 U1 drones 154 0.082 29.1 44 3.2 R1 robot 138 0.071 29.5 49 1.0 W1 Wearable 64 0.039 29.2 47 1.1

[0157] After the edge gateway removes outliers from the data in Table 1, the system constructs a time-varying topology graph based on the latitude and longitude coordinates, movement trajectories, and communication links of each node, and calculates edge weights. Unlike traditional methods that directly use Euclidean distance or simple adjacency, this invention employs hyperbolic space embedding to maintain topological hierarchical relationships and uses geodesic distance propagation to more realistically represent the connections between nodes. When the monitoring data shows significant fluctuations, a geometric folding channel is triggered, compressing and clustering outlier nodes in hyperbolic space.

[0158] Based on this, the system uses clustering algorithms to analyze the folding results and combines parameter thresholds to label abnormal clusters.

[0159] Table 2. Anomaly Cluster Labeling Results

[0160] Abnormal clusters Includes nodes Average PM2.5 (µg / m³) <![CDATA[Average NO2 (ppm)]]> Coverage area (m) Label Category A C2, U1 164.8 0.091 72 pollution source B R1 138 0.071 30 Abnormal hotspots

[0161] As shown in Table 2, C2 and U1 showed synchronous increases in PM2.5 and NO2 in adjacent areas. After geometric folding, they were clustered into the same anomalous cluster and labeled as pollution sources. Although R1 did not exceed the absolute threshold, its parameters fluctuated significantly and it was labeled as an anomalous hotspot by the system. The values ​​of the remaining nodes were stable, forming a comparative baseline.

[0162] Experimental results show that, under dynamic topological conditions, this invention can more sensitively and accurately identify pollution sources and anomalous hotspots through hyperbolic space embedding and geometric folding mechanisms.

[0163] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for monitoring environmental parameters based on mobile Internet of Things, characterized in that, Includes the following steps: Environmental parameters are collected by multiple mobile IoT terminals, and the environmental parameters are synchronized in time and outlier removal is performed by an edge gateway to obtain aligned primary monitoring data. Based on the location, movement trajectory, and communication link of each mobile IoT terminal, a corresponding time-varying topology map is generated; Each node of the time-varying topology graph is embedded into a hyperbolic space, and the primary monitoring data is propagated in the hyperbolic space using graph convolution kernels to obtain a global estimation result of environmental parameters. Real-time monitoring of environmental parameter fluctuations at each node; when the fluctuation exceeds a preset threshold, an abnormal trigger signal is generated. Upon receiving the abnormal trigger signal, the geometric folding channel is activated to locally fold the embedding space of the corresponding region, so that the embedding representations of neighboring nodes are compressed and clustered in the hyperbolic space, thus obtaining the folding result. Based on the folding results, anomaly clusters are formed, and these anomaly clusters are marked as pollution sources or abnormal hotspot areas. The abnormal cluster information is uploaded to the cloud, where the trigger threshold and folding depth function of the geometric folding channel are optimized, and the trigger threshold and folding depth function are sent back to the edge device.

2. The environmental parameter monitoring method based on mobile Internet of Things according to claim 1, characterized in that, The environmental parameters include: At least one of temperature, humidity, gas concentration, particulate matter concentration, noise intensity, light intensity, and wind speed; The time synchronization and outlier removal steps include: Establish a unified time base by using network time protocols or high-precision time synchronization mechanisms to align data collected by different terminals to the same time scale; Outlier removal is performed on the aligned data, including interval constraint judgment, historical mean deviation detection and multimodal cross-validation. When the abnormal conditions are met, the data is marked and removed. At the same time, node identifiers and collection location identifiers are added to the remaining data in the edge gateway to form primary monitoring data with spatiotemporal index.

3. The environmental parameter monitoring method based on mobile Internet of Things according to claim 1, characterized in that, The steps for generating the corresponding time-varying topology graph include: Based on the location information of the terminal, the geographic coordinates and movement direction of the node are extracted, and the trajectory vector is constructed using the location information at continuous time. The edge connection relationships between nodes are determined based on the communication link quality, geographical proximity, and topological history between nodes; In the generated topology, a weight is assigned to each edge, which is calculated by combining the stability of the communication link, spatial distance, and similar motion trends; The nodes and edges are then sliced ​​and updated over time to form a time-varying topology graph.

4. The environmental parameter monitoring method based on mobile Internet of Things according to claim 1, characterized in that, The step of adding weights to each edge in the generated topology includes: Based on the historical transmission records of the communication link from the edge gateway, the average packet loss rate, round-trip delay variance, and connection hold duration of the link are calculated and normalized to obtain the first weighted component representing communication stability. Based on the geographic coordinates of two nodes in the same time slice, their spatial Euclidean distance is calculated, and the distance is transformed by a Gaussian kernel function to obtain a second weighted component representing spatial proximity. Based on the position sequence of nodes in continuous time slices, the corresponding motion trajectory vectors are generated. By calculating the difference between the cosine similarity of the motion direction and the velocity magnitude, a third weight component representing the similar motion trend is obtained. The first weight component, the second weight component, and the third weight component are fused according to a preset weighting formula to obtain the comprehensive weight value of the edge, wherein the weighting formula is a linear weighting function, and the fusion result is limited to a preset numerical range.

5. The environmental parameter monitoring method based on mobile Internet of Things according to claim 1, characterized in that, The generation of the global estimation results for the environmental parameters includes: Based on the node set of the time-varying topology graph, an initial feature vector is established according to the spatiotemporal index of each node. The feature vector includes environmental parameter observations, node position coordinates, and edge weight information. The initial feature vector is mapped from Euclidean space to the tangent space of hyperbolic space using an exponential mapping function, and then the node embedding representation in hyperbolic space is obtained through a logarithmic mapping function. On the hyperbolic space embedding representation, an adjacency matrix with geodesic distance as the metric is constructed, and convolution propagation is performed using the adjacency matrix and convolution kernel parameters to update the embedding features of the nodes; The node embedding features after convolutional propagation are fused with the primary monitoring data to generate environmental parameter estimation results covering the entire map.

6. The environmental parameter monitoring method based on mobile Internet of Things according to claim 1, characterized in that, The steps for generating the abnormal trigger signal include: Based on the node embedding features after convolutional propagation and the primary monitoring data, the changes in environmental parameters of each node in continuous time slices are calculated. The changes include the difference value, the rate of change, and the direction of change. The change is compared with the historical statistical baseline of the node to generate a first detection index representing the degree of short-term deviation. The change is compared with the average change trend within the same topological neighborhood to generate a second detection index representing spatial inconsistency. The first detection index and the second detection index are input into the threshold judgment function. If the output result exceeds the preset threshold, an abnormal trigger signal is generated, and the identifier of the trigger node, the trigger time and the trigger intensity are recorded. The abnormal trigger signal is passed as the activation condition for the geometric folding channel.

7. The environmental parameter monitoring method based on mobile Internet of Things according to claim 1, characterized in that, The steps for generating the folding result include: Based on the trigger node, trigger time, and trigger intensity in the abnormal trigger signal, the local embedding region to be folded is determined; Within the local embedding region, the folding depth parameter is calculated based on the trigger intensity; A folding transformation is performed on the node embedding representation of the local embedding region. The folding transformation includes applying a nonlinear compression function in hyperbolic space to reduce the original geodesic distance by the folding depth parameter. After the folding transformation, the embedding representation of the affected nodes is uniformly reorganized, and the folding result is output.

8. The environmental parameter monitoring method based on mobile Internet of Things according to claim 1, characterized in that, The step of forming anomaly clusters based on the folding results and marking the anomaly clusters as pollution sources or anomalous hotspot areas includes: In the folded embedding representation, a clustering algorithm based on geodesic distance is used to divide the shrunken clustered node set into several clusters, with each cluster corresponding to a candidate anomaly region; Calculate the center location, coverage area, and number of member nodes of each candidate anomaly region to obtain the spatial attributes of the anomaly cluster; We performed weighted average and variance analysis on the environmental parameters collected from the member nodes of each anomaly cluster to obtain the parameter characteristics of the anomaly cluster. The spatial attributes and parameter features are input into the labeling rule base for judgment. If the pollution source conditions are met, it is labeled as a pollution source. If the local disturbance conditions are met, it is labeled as an abnormal hotspot area. The output includes anomaly cluster labeling results containing geographic range, parameter type, and confidence level.

9. The environmental parameter monitoring method based on mobile Internet of Things according to claim 1, characterized in that, The steps of uploading abnormal cluster information to the cloud, optimizing the trigger threshold and folding depth function of the geometric folding channel by the cloud, and then sending the optimized information back to the edge device include: The results of anomaly cluster annotation formed on the edge side, along with trigger node identifiers, folding depth parameters, and historical monitoring data, will be uploaded to the cloud processing module. Statistical analysis is performed on the labeling results of multiple batches of abnormal clusters in the cloud to calculate the fitness distribution of the trigger threshold, and the trigger threshold is adaptively updated based on the fluctuation characteristics of environmental parameters. The relationship between the folding depth parameter and the anomaly detection accuracy is fitted in the cloud to obtain the optimized folding depth function; The updated trigger threshold and fold depth function are fed back to the edge device as optimization parameters.

10. An environmental parameter monitoring system based on mobile Internet of Things, executing the environmental parameter monitoring method based on mobile Internet of Things as described in any one of claims 1 to 9, characterized in that, Includes the following modules: The data acquisition and preprocessing module is used to collect environmental parameters through multiple mobile IoT terminals, and to use an edge gateway to perform time synchronization and outlier removal on the environmental parameters to obtain aligned primary monitoring data. The topology building module is used to generate a corresponding time-varying topology map based on the location, movement trajectory and communication link of each mobile IoT terminal; The hyperbolic embedding and convolutional propagation module is used to embed each node of the time-varying topology graph into the hyperbolic space, and propagate the primary monitoring data in the hyperbolic space using graph convolution kernels to obtain the global estimation results of environmental parameters. An anomaly detection module is used to monitor the fluctuations of environmental parameters of each node in real time, and generate an anomaly trigger signal when the fluctuations exceed a preset threshold. The geometric folding module is used to activate the geometric folding channel when the abnormal trigger signal is received, and to locally fold the embedding space of the corresponding region, so that the embedding representation of the neighboring nodes is compressed and clustered in the hyperbolic space to obtain the folding result. An anomaly cluster generation and labeling module is used to generate anomaly clusters based on the folding results and label the anomaly clusters as pollution sources or abnormal hotspot areas. The cloud optimization module is used to upload the abnormal cluster information to the cloud, whereby the cloud optimizes the trigger threshold and folding depth function of the geometric folding channel, and then sends the updated trigger threshold and folding depth function back to the edge device.

Images