Sensor data transmission method, system and device based on internet of things

Abstract

The application discloses a sensor data transmission method, system and device based on Internet of Things, and relates to the technical field of industrial Internet of Things.The method comprises the following steps: dividing the sensor nodes based on the spatial position information and a preset physical field propagation model to obtain at least one physically associated cluster, wherein the sensor nodes in the same physically associated cluster meet a preset proximity condition in the spatial position, and the change of the time sequence signal meets the continuity constraint described by the physical field propagation model; for each physically associated cluster, constructing a corresponding space-time matrix based on the time sequence signals of the sensor nodes in the cluster; decomposing the space-time matrix to obtain at least one time base matrix and at least one space coefficient matrix; and encoding the time base matrix and the space coefficient matrix to generate an encoded data stream and transmitting the encoded data stream.The application has the effect of improving the data transmission efficiency.

Description

Technical Field

[0001] This application relates to the technical field of industrial Internet of Things (IoT), and in particular to IoT-based sensor data transmission methods, systems, and devices. Background Technology

[0002] The Industrial Internet of Things (IIoT) deploys a large number of sensor nodes within the workspace to monitor physical parameters such as temperature, pressure, and vibration in real time at high frequency. Sensors continuously generate massive amounts of time-series data at a predetermined sampling frequency. This data must be fully uploaded to a management platform via a sensor network to provide decision-making support for process optimization control, equipment health diagnostics, and safety risk warnings. However, with the continuous improvement of monitoring point density and sampling accuracy, existing data transmission mechanisms are revealing increasingly acute contradictions in terms of efficiency.

[0003] In the common raw data direct transmission mode, each sensor node independently pushes the full time-series signal it collects to the aggregation gateway, causing a sharp increase in the communication load of the sensor network layer, especially the wireless transmission link. Faced with high-frequency concurrent reporting from hundreds or thousands of nodes, the limited channel bandwidth resources are quickly filled, which can easily lead to packet collisions, a sharp increase in queuing delays, and large-scale packet loss and retransmission, severely weakening the real-time monitoring response capability of industrial sites. At the same time, edge nodes, due to being in a high duty cycle RF transceiver state for a long time, experience a sharp increase in energy consumption, which directly threatens the network's lifespan and the reliability of long-term maintenance-free operation in battery-powered scenarios.

[0004] To alleviate the aforementioned transmission bottlenecks, some systems have introduced lightweight strategies such as threshold-triggered reporting or fixed-interval downsampling. However, the threshold-triggered mechanism only activates the transmission link when the monitored quantity exceeds the limit. While this significantly reduces the traffic volume under normal conditions, it creates a blind spot in the observation of the continuous evolution of the physical field within the safe threshold range. This results in the backend losing the data foundation for tracing slow abnormal drifts and capturing early signs of degradation. Downsampling strategies, on the other hand, sacrifice temporal resolution to reduce the total amount of data, leading to irreversible loss of high-frequency fluctuation details and transient change characteristics. This makes it difficult to meet the stringent requirements for data fidelity in precision process control and micro-fault identification. More importantly, existing compression coding methods for single-node time-series signals only address intra- and inter-frame redundancy from a temporal perspective, failing to adequately utilize the inherent strong correlation of industrial physical fields in the spatial dimension. Based on the propagation mechanisms of physical fields such as heat conduction and mass diffusion, the signal changes sensed by spatially adjacent sensor nodes are not independent of each other, but rather follow specific continuity constraints and exhibit characteristics of collaborative evolution. Traditional compression methods fail to identify and remove this spatial structural redundancy, resulting in each node's compressed data stream still carrying a large number of strongly correlated information copies, fundamentally limiting the improvement in transmission efficiency.

[0005] In summary, effectively mining the spatial relationships between sensor nodes at the edge and achieving high-rate compression of spatiotemporal redundancy without compromising the accuracy of dynamic reconstruction of the physical field has become an urgent need to break through the bottleneck of data transmission efficiency in the Industrial Internet of Things and ensure the continuous availability of large-scale monitoring systems. Summary of the Invention

[0006] To improve data transmission efficiency, this application provides a sensor data transmission method, system, and device based on the Internet of Things.

[0007] Firstly, this application provides a sensor data transmission method based on the Internet of Things, employing the following technical solution: A sensor data transmission method based on the Internet of Things (IoT) is applied to an industrial IoT system, which includes a management platform, a sensor network platform, and an object platform connected in sequence. The method is executed by an edge computing node of the sensor network platform and includes: Acquire spatial location information and timing signals of each sensor node within the target space; Based on the spatial location information and the preset physical field propagation model, the sensor nodes are divided to obtain at least one physical association cluster. The sensor nodes in the same physical association cluster satisfy the preset proximity condition in spatial location, and the change of the time series signal satisfies the continuity constraint described by the physical field propagation model. For each of the physical association clusters, a corresponding spatiotemporal matrix is ​​constructed based on the time-series signals of each of the sensor nodes within the cluster; The spatiotemporal matrix is ​​decomposed to obtain at least one time base matrix and at least one spatial coefficient matrix. The time base matrix is ​​used to characterize the change pattern of the physical field within the cluster over time, and the spatial coefficient matrix is ​​used to characterize the response relationship of each sensor node within the cluster relative to the time base matrix. The total number of elements in the time base matrix and the spatial coefficient matrix is ​​less than the total number of elements in the spatiotemporal matrix. The time base matrix and the spatial coefficient matrix are encoded to generate an encoded data stream, which is then transmitted.

[0008] By adopting the above technical solution, the spatial location information and temporal signals of each sensor node in the target space are obtained. Then, based on the spatial location information and a preset physical field propagation model, the sensor nodes are divided to obtain at least one physical association cluster. The sensor nodes in the same physical association cluster satisfy the preset proximity condition in spatial location, and the changes in the temporal signals satisfy the continuity constraint described by the physical field propagation model. Then, for each physical association cluster, a corresponding spatiotemporal matrix is ​​constructed based on the temporal signals of each sensor node in the cluster. The spatiotemporal matrix is ​​then decomposed to obtain at least one time base matrix and at least one spatial coefficient matrix. The time base matrix is ​​used to characterize the change pattern of the physical field in the cluster over time, and the spatial coefficient matrix is ​​used to characterize the response relationship of each sensor node in the cluster relative to the time base matrix. The total number of elements in the time base matrix and the spatial coefficient matrix is ​​less than the total number of elements in the spatiotemporal matrix. Then, the time base matrix and the spatial coefficient matrix are encoded to generate an encoded data stream and transmitted. This invention divides physical correlation clusters through a physical field propagation model, effectively mining and utilizing the spatial structural redundancy between sensor nodes. This overcomes the shortcomings of traditional single-node compression in utilizing strong spatial correlations. Furthermore, by significantly reducing the total number of transmitted elements in the time base matrix and spatial coefficient matrix after spatiotemporal matrix decomposition, the data volume is greatly compressed, alleviating channel congestion, retransmission overhead, and edge node energy consumption problems in the original direct transmission mode. At the same time, this method fully preserves the continuous evolution characteristics of the physical field during high-rate compression, avoiding the loss of key details caused by threshold triggering and downsampling. Thus, while ensuring the accuracy of dynamic reconstruction, it improves data transmission efficiency.

[0009] Optionally, the step of dividing the sensor nodes based on the spatial location information and a preset physical field propagation model to obtain at least one physical association cluster, wherein sensor nodes within the same physical association cluster satisfy a preset proximity condition in spatial location, and the change in the time-series signal satisfies the continuity constraint described by the physical field propagation model, includes: Based on the spatial location information, determine the spatial geometric distance and relative orientation relationship between each sensor node; Based on a preset physical field propagation model, and according to the spatial geometric distance and the relative orientation relationship, the propagation attenuation characteristic value and propagation delay characteristic value of the physical field between any two sensor nodes are determined. Based on the propagation attenuation characteristic value and the propagation delay characteristic value, an association weight matrix is ​​constructed, wherein the association weight matrix is ​​used to characterize the signal coupling strength between the sensor nodes; Based on the spatial geometric distance and the association weight matrix, and with a preset spatial proximity threshold and a preset coupling strength threshold as constraints, a clustering operation is performed on the sensor nodes in the target space to output at least one physical association cluster.

[0010] By adopting the above technical solution, in order to achieve the division of physical association clusters, the spatial geometric distance and relative orientation relationship between each sensor node are determined based on spatial location information. Then, based on a preset physical field propagation model, and according to the spatial geometric distance and relative orientation relationship, the propagation attenuation characteristic value and propagation delay characteristic value of the physical field between any two sensor nodes are determined. Then, based on the propagation attenuation characteristic value and propagation delay characteristic value, an association weight matrix is ​​constructed, which is used to characterize the signal coupling strength between sensor nodes. Then, based on the spatial geometric distance and the association weight matrix, and with preset spatial proximity threshold and preset coupling strength threshold as constraints, a clustering operation is performed on the sensor nodes in the target space, and at least one physical association cluster is output.

[0011] Optionally, the step of constructing a corresponding spatiotemporal matrix based on the time-series signals of each sensor node within each of the physically associated clusters includes: For each of the physical association clusters, the timing signals of each of the sensor nodes within the cluster are obtained, and the timing signals are timestamped to obtain aligned data. Using the sensor node identifier as the row index and the sampling time as the column index, the aligned data is arranged by row to obtain the initial spatiotemporal matrix; The initial spatiotemporal matrix is ​​cleaned to obtain a spatiotemporal matrix, wherein the rows of the spatiotemporal matrix are used to represent the sensor nodes, and the columns of the spatiotemporal matrix are used to represent the sampling time.

[0012] By adopting the above technical solution, in order to construct the spatiotemporal matrix, for each physically associated cluster, the time-series signals of each sensor node in the cluster are obtained, and the time-series signals are timestamped to obtain aligned data. Then, the aligned data is arranged by row with the sensor node identifier as the row index and the sampling time as the column index to obtain the initial spatiotemporal matrix. Then, the initial spatiotemporal matrix is ​​cleaned to obtain the spatiotemporal matrix. In the spatiotemporal matrix, the rows are used to represent the sensor nodes, and the columns are used to represent the sampling times.

[0013] Optionally, the step of decomposing the spatiotemporal matrix to obtain at least one time basis matrix and at least one spatial coefficient matrix includes: The spatiotemporal matrix is ​​divided into time-frequency components to obtain a steady-state component matrix and a transient component matrix. The steady-state component matrix is ​​used to characterize the slow-changing trend of the physical field, and the transient component matrix is ​​used to characterize the transient fluctuations caused by physical events. The steady-state component matrix is ​​subjected to a first decomposition process to obtain at least one steady-state time basis matrix and at least one steady-state space coefficient matrix; The transient component matrix is ​​subjected to a second decomposition process to obtain at least one transient time basis matrix and at least one transient spatial coefficient matrix; The steady-state time base matrix and the transient time base matrix are combined to obtain the time base matrix, and the steady-state space coefficient matrix and the transient space coefficient matrix are combined to obtain the space coefficient matrix; In this matrix, the rows of the time base matrix represent different temporal variation patterns, and the columns represent sampling times. Similarly, the rows of the spatial coefficient matrix represent the sensor nodes, and the columns represent different spatial response patterns. M is the number of sensor nodes within the cluster, N is the number of columns in the spatiotemporal matrix, and K is the number of temporal variation patterns. The time base matrix has K rows and N columns, and the spatial coefficient matrix has M rows and K columns, satisfying K(M+N). <M×N。

[0014] By employing the above technical solution, in order to decompose the spatiotemporal matrix, the spatiotemporal matrix is ​​divided into time-frequency components to obtain a steady-state component matrix and a transient component matrix. The steady-state component matrix characterizes the gradual change trend of the physical field, while the transient component matrix characterizes the transient fluctuations caused by physical events. Then, the steady-state component matrix undergoes a first decomposition process to obtain at least one steady-state time basis matrix and at least one steady-state spatial coefficient matrix. Next, the transient component matrix undergoes a second decomposition process to obtain at least one transient time basis matrix and at least one transient spatial coefficient matrix. Finally, the steady-state time basis matrix... The time base matrix is ​​obtained by combining the transient time base matrix with the spatial coefficient matrix. The spatial coefficient matrix is ​​obtained by combining the steady-state spatial coefficient matrix with the transient spatial coefficient matrix. In the time base matrix, rows represent different time-varying patterns, and columns represent sampling times. In the spatial coefficient matrix, rows represent sensor nodes, and columns represent different spatial response patterns. M is the number of sensor nodes in the cluster, N is the number of columns in the spatiotemporal matrix, and K is the number of time-varying patterns. The time base matrix has K rows and N columns, and the spatial coefficient matrix has M rows and K columns, satisfying K(M+N). <M×N。

[0015] Optionally, the step of dividing the spatiotemporal matrix into time-frequency components to obtain a steady-state component matrix and a transient component matrix includes: Perform time-frequency transformation on the row vectors corresponding to each sensor node in the spatiotemporal matrix to obtain the frequency domain representation of each sensor node; Based on a preset frequency threshold, the frequency domain representation is divided into low-frequency components and high-frequency components; The low-frequency components are reconstructed by inverse transformation to obtain the steady-state component signals corresponding to each sensor node, and the steady-state component signals are combined to form a steady-state component matrix. The high-frequency components are reconstructed by inverse transformation to obtain the transient component signals corresponding to each sensor node, and the transient component signals are combined to form a transient component matrix.

[0016] By adopting the above technical solution, in order to achieve the time-frequency component division of the spatiotemporal matrix, the row vectors corresponding to each sensor node in the spatiotemporal matrix are subjected to time-frequency transformation to obtain the frequency domain representation of each sensor node. Then, based on a preset frequency threshold, the frequency domain representation is divided into low-frequency components and high-frequency components. The low-frequency components are then reconstructed by inverse transformation to obtain the steady-state component signals corresponding to each sensor node. The steady-state component signals are then combined to form a steady-state component matrix. The high-frequency components are then reconstructed by inverse transformation to obtain the transient component signals corresponding to each sensor node. The transient component signals are then combined to form a transient component matrix.

[0017] Optionally, the step of performing a first decomposition on the steady-state component matrix to obtain at least one steady-state time basis matrix and at least one steady-state spatial coefficient matrix includes: Principal component analysis is performed on the steady-state component matrix to extract at least one principal component as a reference steady-state vector, wherein the reference steady-state vector is used to characterize the common change trend of the row vectors of the steady-state component matrix; Using the benchmark steady-state vector as a reference, linear regression fitting is performed on each row vector in the steady-state component matrix to obtain the first scaling factor and the first deviation value corresponding to each sensor node; Based on the reference steady-state vector, the steady-state time base matrix is ​​generated, and based on the first scaling factor and the first deviation value, the steady-state space coefficient matrix is ​​generated.

[0018] By adopting the above technical solution, in order to decompose the steady-state component matrix, principal component analysis is performed on the steady-state component matrix to extract at least one principal component as a reference steady-state vector. The reference steady-state vector is used to characterize the common change trend of the row vectors of the steady-state component matrix. Then, with the reference steady-state vector as a reference, linear regression fitting is performed on each row vector in the steady-state component matrix to obtain the first scaling factor and the first deviation value corresponding to each sensor node. Then, based on the reference steady-state vector, a steady-state time base matrix is ​​generated, and based on the first scaling factor and the first deviation value, a steady-state space coefficient matrix is ​​generated.

[0019] Optionally, the step of performing a second decomposition on the transient component matrix to obtain at least one transient time basis matrix and at least one transient spatial coefficient matrix includes: Transient waveform detection is performed on each row vector in the transient component matrix to extract at least one transient waveform, wherein the transient waveform is a signal segment whose amplitude exceeds a preset event detection threshold; If the transient waveforms detected on different sensor nodes overlap in time, or the interval between timestamps is less than a preset time tolerance window, then the corresponding transient waveforms are determined to originate from the same physical event, and the corresponding transient waveforms are assigned to the same set of associated transient waveforms. For each of the associated transient waveform sets, after length normalization of the corresponding transient waveform, at least one principal component is extracted as a reference event waveform vector, wherein the reference event waveform vector is used to characterize the main common waveform shape of each transient waveform in the associated transient waveform set; Using the reference event waveform vector as a reference, calculate the second scaling factor, the second deviation value, and the time offset of each transient waveform in the associated transient waveform set relative to the reference event waveform vector; The reference event waveform vectors corresponding to each of the associated transient waveform sets are combined to generate a transient time base matrix, and the second scaling factor, the second deviation value, and the time offset corresponding to each of the sensor nodes are combined to generate a transient spatial coefficient matrix.

[0020] By adopting the above technical solution, in order to decompose the transient component matrix, transient waveform detection is performed on each row vector in the transient component matrix to extract at least one transient waveform. The transient waveform is a signal segment whose amplitude exceeds a preset event detection threshold. If transient waveforms detected on different sensor nodes overlap in time, or the interval between timestamps is less than a preset time tolerance window, then the corresponding transient waveforms are determined to originate from the same physical event, and are grouped into the same associated transient waveform set. Then, for each associated transient waveform set, the corresponding transient waveforms are subjected to length normalization processing. At least one principal component is extracted as a reference event waveform vector, which is used to characterize the main common waveform shape of each transient waveform in the associated transient waveform set. Then, with the reference event waveform vector as a reference, the second scaling factor, the second deviation value, and the time offset of each transient waveform in the associated transient waveform set relative to the reference event waveform vector are calculated. Then, the reference event waveform vectors corresponding to each associated transient waveform set are combined to generate a transient time basis matrix, and the second scaling factor, the second deviation value, and the time offset corresponding to each sensor node are combined to generate a transient spatial coefficient matrix.

[0021] Optionally, the step of encoding the time base matrix and the spatial coefficient matrix to generate an encoded data stream includes: Each row vector in the time base matrix is ​​treated as a data sequence. The sampling time sequence within each data sequence is kept in chronological order, and timestamp metadata is appended to obtain time base encoded data. The timestamp metadata includes the start time, sampling interval, and sequence length. The elements in the spatial coefficient matrix are associated and encapsulated with the corresponding sensor node identifiers to obtain spatial coefficient encoded data. The spatial extent information of the physically associated clusters is encoded into geometric description parameters to obtain geometric description parameter data, wherein the geometric description parameters include at least one of cluster boundary coordinates, cluster center coordinates, and cluster coverage radius; The time-based encoded data, the spatial coefficient encoded data, and the geometric description parameter data are combined and packaged to generate the encoded data stream.

[0022] By adopting the above technical solution, in order to encode the time base matrix and the spatial coefficient matrix, each row vector in the time base matrix is ​​treated as a data sequence, maintaining the chronological order of sampling times within each data sequence, and timestamp metadata is added to obtain time base encoded data. The timestamp metadata includes the start time, sampling interval, and sequence length. Then, the elements in the spatial coefficient matrix are associated and encapsulated with the corresponding sensor node identifiers to obtain spatial coefficient encoded data. Then, the spatial range information of the physical association clusters is encoded into geometric description parameters to obtain geometric description parameter data. The geometric description parameters include at least one of cluster boundary coordinates, cluster center coordinates, and cluster coverage radius. Finally, the time base encoded data, spatial coefficient encoded data, and geometric description parameter data are combined and packaged to generate an encoded data stream.

[0023] Secondly, this application also provides a sensor data transmission system based on the Internet of Things, which adopts the following technical solution: A sensor data transmission system based on the Internet of Things (IoT) is applied to an industrial IoT system. The industrial IoT system includes a management platform, a sensor network platform, and an object platform that are sequentially connected in communication. The edge computing nodes of the sensor network platform are configured with: The data acquisition module is used to acquire the spatial location information and timing signals of each sensor node within the target space; The clustering module is used to divide the sensor nodes based on the spatial location information and a preset physical field propagation model to obtain at least one physical association cluster. The sensor nodes in the same physical association cluster meet the preset proximity condition in spatial location, and the change of the time series signal meets the continuity constraint described by the physical field propagation model. The matrix construction module is used to construct a corresponding spatiotemporal matrix for each of the physical association clusters based on the time-series signals of each of the sensor nodes in the cluster. The matrix decomposition module is used to decompose the spatiotemporal matrix to obtain at least one time base matrix and at least one spatial coefficient matrix. The time base matrix is ​​used to characterize the change pattern of the physical field within the cluster over time, and the spatial coefficient matrix is ​​used to characterize the response relationship of each sensor node within the cluster relative to the time base matrix. The total number of elements in the time base matrix and the spatial coefficient matrix is ​​less than the total number of elements in the spatiotemporal matrix. The encoding and transmission module is used to encode the time base matrix and the spatial coefficient matrix to generate an encoded data stream and transmit it.

[0024] Thirdly, this application also provides a computer device, which adopts the following technical solution: A computer device includes a memory and a processor, the memory storing a computer program executable on the processor, the processor executing the computer program to implement the method described in the first aspect.

[0025] In summary, this application includes at least the following beneficial technical effects: acquiring the spatial location information and temporal signals of each sensor node in the target space; then, based on the spatial location information and a preset physical field propagation model, dividing the sensor nodes to obtain at least one physical association cluster, wherein the sensor nodes within the same physical association cluster satisfy a preset proximity condition in spatial location, and the changes in the temporal signals satisfy the continuity constraints described by the physical field propagation model; then, for each physical association cluster, constructing a corresponding spatiotemporal matrix based on the temporal signals of each sensor node within the cluster; then, decomposing the spatiotemporal matrix to obtain at least one time base matrix and at least one spatial coefficient matrix, wherein the time base matrix is ​​used to characterize the change pattern of the physical field within the cluster over time, and the spatial coefficient matrix is ​​used to characterize the response relationship of each sensor node within the cluster relative to the time base matrix; the total number of elements in the time base matrix and the spatial coefficient matrix is ​​less than the total number of elements in the spatiotemporal matrix; then, encoding the time base matrix and the spatial coefficient matrix to generate an encoded data stream and transmitting it. This invention divides physical correlation clusters through a physical field propagation model, effectively mining and utilizing the spatial structural redundancy between sensor nodes. This overcomes the shortcomings of traditional single-node compression in utilizing strong spatial correlations. Furthermore, by significantly reducing the total number of transmitted elements in the time base matrix and spatial coefficient matrix after spatiotemporal matrix decomposition, the data volume is greatly compressed, alleviating channel congestion, retransmission overhead, and edge node energy consumption problems in the original direct transmission mode. At the same time, this method fully preserves the continuous evolution characteristics of the physical field during high-rate compression, avoiding the loss of key details caused by threshold triggering and downsampling. Thus, while ensuring the accuracy of dynamic reconstruction, it improves data transmission efficiency. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the overall process of an embodiment of this application.

[0027] Figure 2 This is a structural diagram of one application scenario of the system in this application embodiment.

[0028] Figure 3 This is a structural diagram of another application scenario of the system according to an embodiment of this application.

[0029] Figure 4 This is a structural block diagram of the computer device described in this application. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0031] This application discloses a sensor data transmission method based on the Internet of Things.

[0032] Reference Figure 1 A sensor data transmission method based on the Internet of Things (IoT) is applied to an industrial IoT system. The industrial IoT system includes a management platform, a sensor network platform, and an object platform that are sequentially connected in communication. The method is executed by the edge computing nodes of the sensor network platform and includes: Step S11: Obtain the spatial location information and timing signals of each sensor node in the target space.

[0033] It should be noted that in step S11, the edge computing node collects two types of key data from all sensor nodes deployed in the target space through communication links with each sensor node in the target platform: The first type of data is spatial location information, including but not limited to the three-dimensional coordinates, relative installation elevation, and process area identifier of each sensor node in the target spatial coordinate system. This information can be preset and stored locally on the edge node during the sensor deployment phase, or it can be dynamically acquired through the positioning module. The second type of data is time-series signals, which are time series of physical field parameters continuously collected by each sensor node at preset sampling intervals, such as temperature change curves, pressure change curves, or vibration amplitude change curves. The edge computing node can set a time window length and periodically acquire time-series signal segments of all sensor nodes within that time window.

[0034] Step S12: Based on spatial location information and a preset physical field propagation model, the sensor nodes are divided to obtain at least one physical association cluster. The sensor nodes in the same physical association cluster satisfy a preset proximity condition in spatial location, and the changes in the time series signal satisfy the continuity constraint described by the physical field propagation model.

[0035] It should be noted that the evolution of the physical field within the industrial space is not an independent process for each sensor node, but rather the result of continuous propagation in space following specific physical laws. Step S12, based on this prior physical knowledge, structurally groups the sensor nodes: the edge computing node first determines whether the Euclidean distance or geodesic distance along the process path between nodes meets a preset proximity condition based on spatial location information. For example, whether the distance between two sensor nodes is less than the effective influence radius of the physical field within a given time scale. Nodes that initially meet the proximity condition are grouped into the same candidate cluster. Then, the edge computing node introduces a preset physical field propagation model. This model can be the difference form corresponding to Fourier's law describing heat conduction, the discretized expression of Fick's law describing material diffusion, or an empirical formula describing elastic wave attenuation, etc. For nodes within a candidate cluster, the edge computing node checks whether the trend of their temporal signals meets the continuity constraint described by the physical field propagation model, i.e., whether the signal differences between nodes match the spatial distance, propagation direction, and medium characteristics. Only when both spatial proximity and physical propagation consistency are satisfied is the corresponding set of nodes determined as a physically associated cluster. This partitioning method ensures that sensor nodes within the same physical cluster are not only spatially adjacent, but also have real causal or strong correlations in physical processes. This provides clear physical boundaries for subsequent low-rank decomposition and avoids the loss of compression accuracy caused by forcibly classifying physically unrelated nodes into the same decomposition unit.

[0036] Step S13: For each physically associated cluster, construct the corresponding spatiotemporal matrix based on the time-series signals of each sensor node within the cluster.

[0037] It should be noted that for each physically associated cluster obtained in step S12, the edge computing node aligns and combines the time-series signals of all sensor nodes within the cluster within the same time window to construct a two-dimensional data matrix that can simultaneously express information in both the time and spatial dimensions, namely, a spatiotemporal matrix. Specifically, the rows of this spatiotemporal matrix can correspond to different sensor nodes within the cluster, and the columns can correspond to different sampling times within the time window. The elements in the matrix are the physical field parameter values ​​collected by a specific sensor node at a specific sampling time. For example, for a physically associated cluster containing M sensor nodes and N sampling times within the time window, the constructed spatiotemporal matrix has a size of M rows and N columns. Since the sensor nodes within the same physically associated cluster are governed by the same physical field propagation process, their spatiotemporal matrix theoretically has an inherent low-rank characteristic, that is, there is a strong correlation between the row vectors and column vectors of the matrix determined by physical continuity. The purpose of constructing the spatiotemporal matrix is ​​to centrally express the spatial correlation and temporal evolution information scattered in the time-series signals of various nodes within a single data container for subsequent decomposition.

[0038] Step S14: Decompose the spatiotemporal matrix to obtain at least one time base matrix and at least one spatial coefficient matrix. The time base matrix is ​​used to characterize the change pattern of the physical field within the cluster over time, and the spatial coefficient matrix is ​​used to characterize the response relationship of each sensor node within the cluster relative to the time base matrix. The total number of elements in the time base matrix and the spatial coefficient matrix is ​​less than the total number of elements in the spatiotemporal matrix.

[0039] It should be noted that step S14 uses a matrix factorization algorithm to perform a low-rank approximation of the spatiotemporal matrix constructed in step S13, thereby achieving efficient compression of spatiotemporal joint redundancy. Specifically, edge computing nodes can use numerical methods such as singular value decomposition, nonnegative matrix factorization, or principal component analysis to approximate the original spatiotemporal matrix of size M×N as a product of a time base matrix and a spatial coefficient matrix. The column vectors of the time base matrix constitute a set of basic patterns describing the changes of the physical field over time, such as several time curves reflecting the overall upward trend of temperature, periodic fluctuations, or sudden pulse shapes. The row vectors of the spatial coefficient matrix correspond to the combined weights of each sensor node on this set of time base patterns, reflecting the response strength of each node's spatial location to the time-varying patterns. By controlling the decomposition rank, different degrees of compression ratio adjustment can be achieved. Due to the high correlation of node signals within the physical cluster, only a small number of base patterns much smaller than min(M,N) are needed to accurately reconstruct the original spatiotemporal matrix, making the sum of the total number of elements in the decomposed time base matrix and spatial coefficient matrix significantly smaller than the total number of elements in the original spatiotemporal matrix, M×N. This compression method differs from traditional node-by-node time series compression. While eliminating temporal redundancy, it also removes spatial structural redundancy caused by the propagation of physical fields, thereby achieving a higher data compression ratio while ensuring reconstruction accuracy.

[0040] Step S15: Encode the time base matrix and the spatial coefficient matrix to generate an encoded data stream and transmit it.

[0041] It should be noted that after matrix decomposition, both the time base matrix and the spatial coefficient matrix obtained by the edge computing nodes are in numerical matrix form. Directly transmitting the raw floating-point representation would still consume a significant amount of bandwidth. This step further encodes the decomposition results to reduce transmission volume. Specific encoding methods may include: quantizing matrix elements to map high-precision floating-point numbers to integer values ​​with a finite number of bits; using entropy coding such as Huffman coding or arithmetic coding to perform lossless compression on the quantized numerical sequence; and attaching descriptive metadata to the time base matrix and spatial coefficient matrix, including cluster identifiers, sensor node member lists, start and end times of the time window, decomposition rank, and quantization step size. The encoded data stream generated after encoding exists in binary bitstream form. The edge computing nodes transmit this encoded data stream to the management platform through the sensor network platform. Compared to the traditional method of uploading the raw time-series signals of each sensor node separately, the data volume of the encoded data stream transmitted in this step is significantly reduced, effectively reducing transmission bandwidth consumption and node power consumption. After receiving the encoded data stream, the management platform can approximately restore the original spatiotemporal matrix through the corresponding decoding and matrix multiplication reconstruction process, thereby obtaining the time-series signals of each sensor node in the target space within the corresponding time window, supporting subsequent process monitoring and diagnostic analysis tasks.

[0042] In the above implementation, the spatial location information and temporal signals of each sensor node in the target space are acquired. Then, based on the spatial location information and a preset physical field propagation model, the sensor nodes are divided to obtain at least one physical association cluster. The sensor nodes in the same physical association cluster satisfy a preset proximity condition in spatial location, and the changes in the temporal signals satisfy the continuity constraint described by the physical field propagation model. Then, for each physical association cluster, a corresponding spatiotemporal matrix is ​​constructed based on the temporal signals of each sensor node in the cluster. The spatiotemporal matrix is ​​then decomposed to obtain at least one time base matrix and at least one spatial coefficient matrix. The time base matrix is ​​used to characterize the change pattern of the physical field in the cluster over time, and the spatial coefficient matrix is ​​used to characterize the response relationship of each sensor node in the cluster relative to the time base matrix. The total number of elements in the time base matrix and the spatial coefficient matrix is ​​less than the total number of elements in the spatiotemporal matrix. Then, the time base matrix and the spatial coefficient matrix are encoded to generate an encoded data stream, which is then transmitted. This invention divides physical correlation clusters through a physical field propagation model, effectively mining and utilizing the spatial structural redundancy between sensor nodes. This overcomes the shortcomings of traditional single-node compression in utilizing strong spatial correlations. Furthermore, by significantly reducing the total number of transmitted elements in the time base matrix and spatial coefficient matrix after spatiotemporal matrix decomposition, the data volume is greatly compressed, alleviating channel congestion, retransmission overhead, and edge node energy consumption problems in the original direct transmission mode. At the same time, this method fully preserves the continuous evolution characteristics of the physical field during high-rate compression, avoiding the loss of key details caused by threshold triggering and downsampling. Thus, while ensuring the accuracy of dynamic reconstruction, it improves data transmission efficiency.

[0043] As a further implementation of the method, based on spatial location information and a preset physical field propagation model, sensor nodes are divided to obtain at least one physical association cluster, wherein sensor nodes within the same physical association cluster satisfy a preset proximity condition in spatial location, and the changes in the temporal signal satisfy the continuity constraint described by the physical field propagation model. This includes the following steps: Step S21: Determine the spatial geometric distance and relative orientation relationship between each sensor node based on the spatial location information.

[0044] Step S22: Based on the preset physical field propagation model and according to the spatial geometric distance and relative orientation relationship, determine the propagation attenuation characteristic value and propagation delay characteristic value of the physical field between any two sensor nodes.

[0045] Step S23: Based on the propagation attenuation characteristic value and the propagation delay characteristic value, construct the correlation weight matrix, whereby the correlation weight matrix is ​​used to characterize the signal coupling strength between sensor nodes.

[0046] Step S24: Based on spatial geometric distance and association weight matrix, and with preset spatial proximity threshold and preset coupling strength threshold as constraints, perform clustering operation on sensor nodes in the target space and output at least one physical association cluster.

[0047] It should be noted that from steps S21 to S23, by combining the spatial distribution of sensor nodes with the continuous attenuation law of physical field propagation, the clustering based on geometric distance is elevated to a more physically causal association partitioning method. Specifically, firstly, the geometric distance and relative orientation between each node are calculated using spatial location information. Then, a preset physical field propagation model is introduced to map the distance and orientation to specific propagation attenuation characteristic values ​​and propagation delay characteristic values. This ensures that the signal influence between any two nodes is no longer determined solely by Euclidean distance, but also considers factors such as directionality and medium attenuation. Based on these propagation characteristic values, an association weight matrix is ​​constructed. The elements in the matrix represent the actual signal coupling strength between nodes. Nodes with greater strength are more likely to produce a coordinated response to the same field source disturbance. Then, during the clustering stage, a dual constraint of spatial proximity threshold and coupling strength threshold is applied simultaneously. That is, only nodes that are both geometrically close enough and signal propagation-related enough will be assigned to the same physical association cluster. This ensures that each cluster is not only geographically adjacent, but its temporal signal changes also meet the continuity required by the physical field propagation model.

[0048] In the above implementation, in order to achieve the division of physical association clusters, the spatial geometric distance and relative orientation relationship between each sensor node are determined based on the spatial location information. Then, based on the preset physical field propagation model, and according to the spatial geometric distance and relative orientation relationship, the propagation attenuation characteristic value and propagation delay characteristic value of the physical field between any two sensor nodes are determined. Then, based on the propagation attenuation characteristic value and propagation delay characteristic value, an association weight matrix is ​​constructed, wherein the association weight matrix is ​​used to characterize the signal coupling strength between sensor nodes. Then, based on the spatial geometric distance and the association weight matrix, with preset spatial proximity threshold and preset coupling strength threshold as constraints, a clustering operation is performed on the sensor nodes in the target space, and at least one physical association cluster is output.

[0049] As a further implementation of the method, the step of constructing a corresponding spatiotemporal matrix based on the time-series signals of each sensor node within each physically associated cluster includes: Step S31: For each physical association cluster, obtain the timing signals of each sensor node within the cluster, and perform timestamp alignment processing on the timing signals to obtain aligned data.

[0050] Step S32: Using the sensor node identifier as the row index and the sampling time as the column index, the aligned data is arranged by row to obtain the initial spatiotemporal matrix.

[0051] Step S33: Perform data cleaning operation on the initial spatiotemporal matrix to obtain a spatiotemporal matrix, wherein the rows of the spatiotemporal matrix are used to represent sensor nodes, and the columns of the spatiotemporal matrix are used to represent sampling times.

[0052] In the above implementation, in order to construct the spatiotemporal matrix, for each physical association cluster, the timing signals of each sensor node in the cluster are obtained, and the timing signals are timestamped to obtain aligned data. Then, the aligned data is arranged by row with the sensor node identifier as the row index and the sampling time as the column index to obtain the initial spatiotemporal matrix. Then, the initial spatiotemporal matrix is ​​cleaned to obtain the spatiotemporal matrix. The rows of the spatiotemporal matrix are used to represent the sensor nodes, and the columns of the spatiotemporal matrix are used to represent the sampling times.

[0053] As a further implementation of the method, the step of decomposing the spatiotemporal matrix to obtain at least one time basis matrix and at least one spatial coefficient matrix includes: Step S41: Divide the spatiotemporal matrix into time-frequency components to obtain a steady-state component matrix and a transient component matrix. The steady-state component matrix is ​​used to characterize the gradual change trend of the physical field, and the transient component matrix is ​​used to characterize the transient fluctuations caused by physical events.

[0054] Step S42: Perform a first decomposition process on the steady-state component matrix to obtain at least one steady-state time basis matrix and at least one steady-state space coefficient matrix.

[0055] Step S43: Perform a second decomposition on the transient component matrix to obtain at least one transient time basis matrix and at least one transient spatial coefficient matrix.

[0056] Step S44: Combine the steady-state time base matrix and the transient time base matrix to obtain the time base matrix, and combine the steady-state space coefficient matrix and the transient space coefficient matrix to obtain the space coefficient matrix.

[0057] In this matrix, the rows of the time base matrix represent different temporal variation patterns, and the columns represent sampling times. The rows of the spatial coefficient matrix represent sensor nodes, and the columns represent different spatial response patterns. M is the number of sensor nodes in the cluster, N is the number of columns in the spatiotemporal matrix, and K is the number of temporal variation patterns. The time base matrix has K rows and N columns, and the spatial coefficient matrix has M rows and K columns, satisfying K(M+N). <M×N。

[0058] It should be noted that, from step S41 to step S44, the time-frequency component division is first performed on the spatio-temporal matrix, which is split into a steady-state component matrix representing the slow variation trend of the physical field and a transient component matrix representing the transient fluctuation caused by physical events. This pre-separation processing aims to prevent the slow trend from submerging weak anomalies or transient spikes from interfering with trend estimation; then a first decomposition process is performed on the steady-state component matrix to obtain a steady-state time basis matrix and a steady-state space coefficient matrix, and a second decomposition process is performed on the transient component matrix to obtain a transient time basis matrix and a transient space coefficient matrix. The two paths of decomposition can each apply different constraint strategies to adapt to different signal characteristics; then the steady-state and transient time basis matrices are combined into a final time basis matrix, and the steady-state and transient space coefficient matrices are combined into a final space coefficient matrix, thereby forming a time basis matrix with dimensions of K×N (rows corresponding to K time variation patterns, columns corresponding to N sampling instants) and a space coefficient matrix with dimensions of M×K (rows corresponding to M sensor nodes, columns corresponding to K spatial response patterns); the inequality K(M+N)<M×N herein indicates that the number K of independent variation patterns retained by decomposition is far smaller than the number of nodes M and the number of instants N, so that the inherent spatio-temporal redundancy and continuity of the physical field are utilized to compress the original large-scale observation data into a small number of latent feature patterns with clear physical interpretations, and the key dynamic information of the monitored object is extracted while reducing the dimensionality.

[0059] In the above embodiments, in order to realize the decomposition of the spatio-temporal matrix, time-frequency component division is performed on the spatio-temporal matrix to obtain a steady-state component matrix and a transient component matrix, wherein the steady-state component matrix is used to represent the slow variation trend of the physical field, and the transient component matrix is used to represent the transient fluctuation caused by physical events. Then a first decomposition process is performed on the steady-state component matrix to obtain at least one steady-state time basis matrix and at least one steady-state space coefficient matrix, then a second decomposition process is performed on the transient component matrix to obtain at least one transient time basis matrix and at least one transient space coefficient matrix, then the steady-state time basis matrix and the transient time basis matrix are combined to obtain a time basis matrix, and the steady-state space coefficient matrix and the transient space coefficient matrix are combined to obtain a space coefficient matrix, wherein the rows of the time basis matrix are used to represent different time variation patterns, the columns are used to represent sampling instants, the rows of the space coefficient matrix are used to represent sensor nodes, the columns are used to represent different spatial response patterns, M is the number of sensor nodes in the cluster, N is the number of columns of the spatio-temporal matrix, K is the number of time variation patterns, the number of rows of the time basis matrix is K and the number of columns is N, the number of rows of the space coefficient matrix is M and the number of columns is K, and K(M+N)<M×N is satisfied.

[0060] As a further embodiment of the method, the step of performing time-frequency component division on the spatio-temporal matrix to obtain a steady-state component matrix and a transient component matrix comprises: Step S51: Perform time-frequency transformation on the row vectors corresponding to each sensor node in the spatiotemporal matrix to obtain the frequency domain representation of each sensor node.

[0061] Step S52: Based on a preset frequency threshold, the frequency domain representation is divided into low-frequency components and high-frequency components.

[0062] Step S53: Perform inverse transformation and reconstruction on the low-frequency components to obtain the steady-state component signals corresponding to each sensor node, and combine the steady-state component signals to form a steady-state component matrix.

[0063] Step S54: Perform inverse transform reconstruction on the high-frequency components to obtain the transient component signals corresponding to each sensor node, and combine the transient component signals to form a transient component matrix.

[0064] It should be noted that from steps S51 to S54, wavelet transform is first performed on the row vectors corresponding to each sensor node in the spatiotemporal matrix to obtain a multi-scale frequency domain representation. The advantages of wavelet transform in time-frequency localization analysis are used to achieve fine decomposition of non-stationary signals. Then, based on a preset frequency threshold, the wavelet coefficients are divided into low-frequency components and high-frequency components. The low-frequency part corresponds to the approximate coefficients dominated by slow processes such as environmental drift and seasonal changes, while the high-frequency part corresponds to the detail coefficients caused by events such as sudden leakage and vibration. Next, the low-frequency components are reconstructed by inverse wavelet transform to recover the smooth steady-state component signals of each node after removing transient fluctuations, and these are stacked horizontally to form a steady-state component matrix. At the same time, the high-frequency components are reconstructed by inverse wavelet transform to obtain transient component signals containing only short-term abrupt changes, and these are combined to form a transient component matrix. Through this wavelet threshold-based separation method, the original mixed observation data can be decoupled into two independent matrices with clear physical connotations while preserving the temporal continuity of the physical field.

[0065] In the above implementation, in order to achieve the time-frequency component division of the spatiotemporal matrix, the row vectors corresponding to each sensor node in the spatiotemporal matrix are subjected to time-frequency transformation to obtain the frequency domain representation corresponding to each sensor node. Then, based on a preset frequency threshold, the frequency domain representation is divided into low-frequency components and high-frequency components. The low-frequency components are then reconstructed by inverse transformation to obtain the steady-state component signals corresponding to each sensor node. The steady-state component signals are then combined to form a steady-state component matrix. The high-frequency components are then reconstructed by inverse transformation to obtain the transient component signals corresponding to each sensor node. The transient component signals are then combined to form a transient component matrix.

[0066] As a further implementation of the method, the step of performing a first decomposition process on the steady-state component matrix to obtain at least one steady-state time basis matrix and at least one steady-state spatial coefficient matrix includes: Step S61: Perform principal component analysis on the steady-state component matrix and extract at least one principal component as a reference steady-state vector, wherein the reference steady-state vector is used to characterize the common change trend of the row vectors of the steady-state component matrix.

[0067] Step S62: Using the benchmark steady-state vector as a reference, perform linear regression fitting on each row vector in the steady-state component matrix to obtain the first scaling factor and the first deviation value corresponding to each sensor node.

[0068] Step S63: Generate a steady-state time base matrix based on the reference steady-state vector, and generate a steady-state space coefficient matrix based on the first scaling factor and the first deviation value.

[0069] It should be noted that from steps S61 to S63, principal component analysis is first performed on the steady-state component matrix to extract the most dominant principal components as reference steady-state vectors. These reference steady-state vectors capture the slow-changing trend common to all sensor nodes, representing the main evolution mode of the background physical field in the time dimension. Subsequently, using the extracted reference steady-state vectors as a reference sequence, linear regression fitting is performed on the row vectors corresponding to each sensor node in the steady-state component matrix to solve for the scaling factor and deviation value of each node relative to the common trend. The scaling factor reflects the difference in the response amplitude of the node to the global steady-state mode, and the deviation value is... The baseline offset caused by differences in local environment or sensor characteristics of each node was compensated. Finally, the reference steady-state vector was arranged in rows to form a steady-state time basis matrix (rows correspond to different principal component modes, columns correspond to sampling times). At the same time, the first scaling factor and the first deviation value of each node were organized in an orderly manner to form a steady-state spatial coefficient matrix (rows correspond to sensor nodes, columns correspond to coefficients and deviation terms in each mode). This achieved a low-rank expression of steady-state components, that is, the original steady-state data matrix was approximated by a combination of a small number of global time modes and node-specific parameters. This decomposition strategy fully exploited the high spatial correlation and temporal smoothness of the slowly changing physical field process.

[0070] In the above implementation, in order to decompose the steady-state component matrix, principal component analysis is performed on the steady-state component matrix to extract at least one principal component as a reference steady-state vector. The reference steady-state vector is used to characterize the common change trend of the row vectors of the steady-state component matrix. Then, with the reference steady-state vector as a reference, linear regression fitting is performed on each row vector in the steady-state component matrix to obtain the first scaling factor and the first deviation value corresponding to each sensor node. Then, based on the reference steady-state vector, a steady-state time base matrix is ​​generated, and based on the first scaling factor and the first deviation value, a steady-state space coefficient matrix is ​​generated.

[0071] As a further implementation of the method, the step of performing a second decomposition process on the transient component matrix to obtain at least one transient time basis matrix and at least one transient spatial coefficient matrix includes: Step S71: Perform transient waveform detection on each row vector in the transient component matrix and extract at least one transient waveform, wherein the transient waveform is a signal segment whose amplitude exceeds the detection threshold of a preset event.

[0072] Step S72: If the transient waveforms detected on different sensor nodes overlap in time, or the interval between timestamps is less than a preset time tolerance window, then the corresponding transient waveforms are determined to originate from the same physical event, and the corresponding transient waveforms are assigned to the same set of associated transient waveforms.

[0073] Step S73: For each set of associated transient waveforms, after length normalization of the corresponding transient waveforms, at least one principal component is extracted as a reference event waveform vector, wherein the reference event waveform vector is used to characterize the main common waveform shape of each transient waveform in the set of associated transient waveforms.

[0074] Step S74: Using the reference event waveform vector as a reference, calculate the second scaling factor, second deviation value, and time offset of each transient waveform in the associated transient waveform set relative to the reference event waveform vector.

[0075] Step S75: Combine the reference event waveform vectors corresponding to each associated transient waveform set to generate a transient time base matrix, and combine the second scaling factor, second deviation value and time offset corresponding to each sensor node to generate a transient spatial coefficient matrix.

[0076] It should be noted that from steps S71 to S75, transient waveform detection based on amplitude thresholds is first performed on the row vectors of each sensor node in the transient component matrix. Signal segments exceeding the background noise level are selected as candidate transient waveforms by using a preset event detection threshold. Subsequently, the detection results are correlated and judged according to the principle of spatiotemporal continuity of physical event propagation. If the transient waveforms extracted from different nodes overlap on the time axis or the difference between their starting timestamps is less than a preset time tolerance window, these waveforms are determined to originate from the same physical event and are classified into the same associated transient waveform set, thereby establishing the correspondence between the event and the multi-node response. For each associated transient waveform set, length normalization is first performed on each waveform to eliminate the duration inconsistency caused by differences in propagation distance or sampling start offset. Then, principal component analysis is used to extract the reference event waveform vector from the normalized waveform set. This vector represents the dominant waveform shape shared by all waveforms in the set, i.e., the dominant waveform shape shared by all waveforms in the set. The typical evolution profile of physical events on the time axis is then established. Next, using the reference event waveform vector as a template, parameter fitting is performed on each original transient waveform in the set to solve for the second scaling factor (reflecting the relative intensity of the event signal received by each node), the second deviation value (compensating for local baseline disturbances at each node), and the time offset (recording the time delay difference of the event signal arriving at different nodes). This allows for a unified parameterized expression of the multi-node response to the same event. Finally, the reference event waveform vectors of all associated transient waveform sets are concatenated row by row to form a transient time basis matrix (rows correspond to different event types or different event instances, and columns correspond to sampling times). The second scaling factor, second deviation value, and time offset obtained by each sensor node for each event are then organized into a transient spatial coefficient matrix (rows correspond to sensor nodes, and columns correspond to relevant parameters of each event). This compresses sparse and sudden transient fluctuation data into a low-dimensional representation composed of a small number of event waveform prototypes and their spatial propagation parameters.

[0077] In the above implementation, in order to decompose the transient component matrix, transient waveform detection is performed on each row vector in the transient component matrix to extract at least one transient waveform. The transient waveform is a signal segment whose amplitude exceeds the detection threshold of a preset event. If the transient waveforms detected on different sensor nodes overlap in time, or the interval between timestamps is less than a preset time tolerance window, it is determined that the corresponding transient waveforms originate from the same physical event, and the corresponding transient waveforms are assigned to the same associated transient waveform set. Then, for each associated transient waveform set, the corresponding transient waveforms are length normalized, and at least one principal component is extracted as a reference event waveform vector. The reference event waveform vector is used to characterize the main common waveform shape of each transient waveform in the associated transient waveform set. Then, with the reference event waveform vector as a reference, the second scaling factor, the second deviation value, and the time offset of each transient waveform in the associated transient waveform set relative to the reference event waveform vector are calculated. Then, the reference event waveform vectors corresponding to each associated transient waveform set are combined to generate a transient time basis matrix, and the second scaling factor, the second deviation value, and the time offset corresponding to each sensor node are combined to generate a transient spatial coefficient matrix.

[0078] As a further implementation of the method, the step of encoding the time base matrix and the spatial coefficient matrix to generate an encoded data stream includes: Step S81: Each row vector in the time base matrix is ​​treated as a data sequence, the sampling time order within each data sequence is maintained, and timestamp metadata is appended to obtain time base encoded data. The timestamp metadata includes the start time, sampling interval, and sequence length.

[0079] Step S82: Associate and encapsulate the elements in the spatial coefficient matrix with the corresponding sensor node identifiers to obtain spatial coefficient encoded data.

[0080] Step S83: Encode the spatial extent information of the physically associated clusters into geometric description parameters to obtain geometric description parameter data, wherein the geometric description parameters include at least one of cluster boundary coordinates, cluster center coordinates, and cluster coverage radius.

[0081] Step S84: Combine and package the time base encoded data, spatial coefficient encoded data, and geometric description parameter data to generate an encoded data stream.

[0082] In the above implementation, in order to encode the time base matrix and the spatial coefficient matrix, each row vector in the time base matrix is ​​treated as a data sequence, maintaining the chronological order of sampling times within each data sequence, and timestamp metadata is appended to obtain time base encoded data. The timestamp metadata includes the start time, sampling interval, and sequence length. Then, the elements in the spatial coefficient matrix are associated and encapsulated with the corresponding sensor node identifiers to obtain spatial coefficient encoded data. Then, the spatial range information of the physical association cluster is encoded into geometric description parameters to obtain geometric description parameter data. The geometric description parameters include at least one of cluster boundary coordinates, cluster center coordinates, and cluster coverage radius. Finally, the time base encoded data, spatial coefficient encoded data, and geometric description parameter data are combined and packaged to generate an encoded data stream.

[0083] This application also discloses a sensor data transmission system based on the Internet of Things.

[0084] refer to Figure 2 A sensor data transmission system based on the Internet of Things (IoT) is applied to an industrial IoT system. The industrial IoT system includes a management platform, a sensor network platform, and an object platform that are sequentially connected in communication. The edge computing nodes of the sensor network platform are configured with: The data acquisition module is used to acquire the spatial location information and timing signals of each sensor node within the target space; The clustering module is used to divide sensor nodes based on spatial location information and a preset physical field propagation model to obtain at least one physical association cluster. Sensor nodes within the same physical association cluster satisfy a preset proximity condition in spatial location, and the changes in the temporal signal satisfy the continuity constraint described by the physical field propagation model. The matrix construction module is used to construct a corresponding spatiotemporal matrix for each physically associated cluster based on the time-series signals of each sensor node within that cluster. The matrix decomposition module is used to decompose the spatiotemporal matrix to obtain at least one time base matrix and at least one spatial coefficient matrix. The time base matrix is ​​used to characterize the change pattern of the physical field within the cluster over time, and the spatial coefficient matrix is ​​used to characterize the response relationship of each sensor node within the cluster relative to the time base matrix. The total number of elements in the time base matrix and the spatial coefficient matrix is ​​less than the total number of elements in the spatiotemporal matrix. The encoding and transmission module is used to encode the time base matrix and the spatial coefficient matrix, generate an encoded data stream, and transmit it.

[0085] The overall framework of another application scenario of the IoT-based sensor data transmission system of the present invention is as follows: Figure 3As shown, it can include a user platform, service platform, management platform, sensor network platform, and object platform that interact sequentially, forming a five-platform architecture based on the Internet of Things. Among them, the edge computing nodes of the sensor network platform are configured with data acquisition modules, clustering modules, matrix construction modules, matrix decomposition modules, and encoding and transmission modules.

[0086] Specifically, the IoT-based sensor data transmission system of this embodiment includes a sensor network platform, which includes edge computing nodes. The edge computing nodes are configured to: acquire spatial location information and temporal signals of each sensor node within a target space; divide the sensor nodes based on the spatial location information and a preset physical field propagation model to obtain at least one physical association cluster, wherein sensor nodes within the same physical association cluster satisfy a preset proximity condition in spatial location, and the changes in temporal signals satisfy the continuity constraint described by the physical field propagation model; for each physical association cluster, construct a corresponding spatiotemporal matrix based on the temporal signals of each sensor node within the cluster; decompose the spatiotemporal matrix to obtain at least one time base matrix and at least one spatial coefficient matrix, wherein the time base matrix is ​​used to characterize the change pattern of the physical field within the cluster over time, and the spatial coefficient matrix is ​​used to characterize the response relationship of each sensor node within the cluster relative to the time base matrix, and the total number of elements in the time base matrix and the spatial coefficient matrix is ​​less than the total number of elements in the spatiotemporal matrix; encode the time base matrix and the spatial coefficient matrix to generate an encoded data stream, and transmit it.

[0087] The IoT-based sensor data transmission system of the present invention can implement any of the IoT-based sensor data transmission methods, and the specific working process of the IoT-based sensor data transmission system of the present invention can refer to the corresponding process in the above-mentioned IoT-based sensor data transmission methods.

[0088] This application also discloses a computer device.

[0089] refer to Figure 4 A computer device includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement any of the above-described methods for sensor data transmission based on the Internet of Things.

[0090] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A sensor data transmission method based on the Internet of Things, characterized in that, Applied to an industrial Internet of Things (IIoT) system, the IIoT system includes a management platform, a sensor network platform, and an object platform that are sequentially and communicatively connected. The method is executed by an edge computing node of the sensor network platform and includes: Acquire spatial location information and timing signals of each sensor node within the target space; Based on the spatial location information and the preset physical field propagation model, the sensor nodes are divided to obtain at least one physical association cluster. The sensor nodes in the same physical association cluster satisfy the preset proximity condition in spatial location, and the change of the time series signal satisfies the continuity constraint described by the physical field propagation model. For each of the physical association clusters, a corresponding spatiotemporal matrix is ​​constructed based on the time-series signals of each of the sensor nodes within the cluster; The spatiotemporal matrix is ​​decomposed to obtain at least one time base matrix and at least one spatial coefficient matrix. The time base matrix is ​​used to characterize the change pattern of the physical field within the cluster over time, and the spatial coefficient matrix is ​​used to characterize the response relationship of each sensor node within the cluster relative to the time base matrix. The total number of elements in the time base matrix and the spatial coefficient matrix is ​​less than the total number of elements in the spatiotemporal matrix. The time base matrix and the spatial coefficient matrix are encoded to generate an encoded data stream, which is then transmitted.

2. The sensor data transmission method based on the Internet of Things according to claim 1, characterized in that, The step of dividing the sensor nodes based on the spatial location information and a preset physical field propagation model to obtain at least one physical association cluster, wherein sensor nodes within the same physical association cluster satisfy a preset proximity condition in spatial location, and the changes in the time-series signal satisfy the continuity constraint described by the physical field propagation model, includes: Based on the spatial location information, determine the spatial geometric distance and relative orientation relationship between each sensor node; Based on a preset physical field propagation model, and according to the spatial geometric distance and the relative orientation relationship, the propagation attenuation characteristic value and propagation delay characteristic value of the physical field between any two sensor nodes are determined. Based on the propagation attenuation characteristic value and the propagation delay characteristic value, an association weight matrix is ​​constructed, wherein the association weight matrix is ​​used to characterize the signal coupling strength between the sensor nodes; Based on the spatial geometric distance and the association weight matrix, and with a preset spatial proximity threshold and a preset coupling strength threshold as constraints, a clustering operation is performed on the sensor nodes in the target space to output at least one physical association cluster.

3. The sensor data transmission method based on the Internet of Things according to claim 1, characterized in that, The step of constructing a corresponding spatiotemporal matrix for each of the physically associated clusters based on the time-series signals of each of the sensor nodes within that cluster includes: For each of the physical association clusters, the timing signals of each of the sensor nodes within the cluster are obtained, and the timing signals are timestamped to obtain aligned data. Using the sensor node identifier as the row index and the sampling time as the column index, the aligned data is arranged by row to obtain the initial spatiotemporal matrix; The initial spatiotemporal matrix is ​​cleaned to obtain a spatiotemporal matrix, wherein the rows of the spatiotemporal matrix are used to represent the sensor nodes, and the columns of the spatiotemporal matrix are used to represent the sampling time.

4. The sensor data transmission method based on the Internet of Things according to claim 1, characterized in that, The step of decomposing the spatiotemporal matrix to obtain at least one time basis matrix and at least one spatial coefficient matrix includes: The spatiotemporal matrix is ​​divided into time-frequency components to obtain a steady-state component matrix and a transient component matrix. The steady-state component matrix is ​​used to characterize the slow-changing trend of the physical field, and the transient component matrix is ​​used to characterize the transient fluctuations caused by physical events. The steady-state component matrix is ​​subjected to a first decomposition process to obtain at least one steady-state time basis matrix and at least one steady-state space coefficient matrix; The transient component matrix is ​​subjected to a second decomposition process to obtain at least one transient time basis matrix and at least one transient spatial coefficient matrix; The steady-state time base matrix and the transient time base matrix are combined to obtain the time base matrix, and the steady-state space coefficient matrix and the transient space coefficient matrix are combined to obtain the space coefficient matrix; In this matrix, the rows of the time base matrix represent different temporal variation patterns, and the columns represent sampling times. Similarly, the rows of the spatial coefficient matrix represent the sensor nodes, and the columns represent different spatial response patterns. M is the number of sensor nodes within the cluster, N is the number of columns in the spatiotemporal matrix, and K is the number of temporal variation patterns. The time base matrix has K rows and N columns, and the spatial coefficient matrix has M rows and K columns, satisfying K(M+N). <M×N。 5. The sensor data transmission method based on the Internet of Things according to claim 4, characterized in that, The step of dividing the spatiotemporal matrix into time-frequency components to obtain a steady-state component matrix and a transient component matrix includes: Perform time-frequency transformation on the row vectors corresponding to each sensor node in the spatiotemporal matrix to obtain the frequency domain representation of each sensor node; Based on a preset frequency threshold, the frequency domain representation is divided into low-frequency components and high-frequency components; The low-frequency components are reconstructed by inverse transformation to obtain the steady-state component signals corresponding to each sensor node, and the steady-state component signals are combined to form a steady-state component matrix. The high-frequency components are reconstructed by inverse transformation to obtain the transient component signals corresponding to each sensor node, and the transient component signals are combined to form a transient component matrix.

6. The sensor data transmission method based on the Internet of Things according to claim 4, characterized in that, The step of performing a first decomposition on the steady-state component matrix to obtain at least one steady-state time basis matrix and at least one steady-state spatial coefficient matrix includes: Principal component analysis is performed on the steady-state component matrix to extract at least one principal component as a reference steady-state vector, wherein the reference steady-state vector is used to characterize the common change trend of the row vectors of the steady-state component matrix; Using the benchmark steady-state vector as a reference, linear regression fitting is performed on each row vector in the steady-state component matrix to obtain the first scaling factor and the first deviation value corresponding to each sensor node; Based on the reference steady-state vector, the steady-state time base matrix is ​​generated, and based on the first scaling factor and the first deviation value, the steady-state space coefficient matrix is ​​generated.

7. The sensor data transmission method based on the Internet of Things according to claim 4, characterized in that, The step of performing a second decomposition on the transient component matrix to obtain at least one transient time basis matrix and at least one transient spatial coefficient matrix includes: Transient waveform detection is performed on each row vector in the transient component matrix to extract at least one transient waveform, wherein the transient waveform is a signal segment whose amplitude exceeds a preset event detection threshold; If the transient waveforms detected on different sensor nodes overlap in time, or the interval between timestamps is less than a preset time tolerance window, then the corresponding transient waveforms are determined to originate from the same physical event, and the corresponding transient waveforms are assigned to the same set of associated transient waveforms. For each of the associated transient waveform sets, after length normalization of the corresponding transient waveform, at least one principal component is extracted as a reference event waveform vector, wherein the reference event waveform vector is used to characterize the main common waveform shape of each transient waveform in the associated transient waveform set; Using the reference event waveform vector as a reference, calculate the second scaling factor, the second deviation value, and the time offset of each transient waveform in the associated transient waveform set relative to the reference event waveform vector; The reference event waveform vectors corresponding to each of the associated transient waveform sets are combined to generate a transient time base matrix, and the second scaling factor, the second deviation value, and the time offset corresponding to each of the sensor nodes are combined to generate a transient spatial coefficient matrix.

8. The sensor data transmission method based on the Internet of Things according to claim 1, characterized in that, The step of encoding the time base matrix and the spatial coefficient matrix to generate an encoded data stream includes: Each row vector in the time base matrix is ​​treated as a data sequence. The sampling time sequence within each data sequence is kept in chronological order, and timestamp metadata is appended to obtain time base encoded data. The timestamp metadata includes the start time, sampling interval, and sequence length. The elements in the spatial coefficient matrix are associated and encapsulated with the corresponding sensor node identifiers to obtain spatial coefficient encoded data. The spatial extent information of the physically associated clusters is encoded into geometric description parameters to obtain geometric description parameter data, wherein the geometric description parameters include at least one of cluster boundary coordinates, cluster center coordinates, and cluster coverage radius; The time-based encoded data, the spatial coefficient encoded data, and the geometric description parameter data are combined and packaged to generate the encoded data stream.

9. A sensor data transmission system based on the Internet of Things, characterized in that, This is applied to an industrial Internet of Things (IIoT) system, which includes a management platform, a sensor network platform, and an object platform that are sequentially and communicatively connected. The edge computing nodes of the sensor network platform are configured with: The data acquisition module is used to acquire the spatial location information and timing signals of each sensor node within the target space; The clustering module is used to divide the sensor nodes based on the spatial location information and a preset physical field propagation model to obtain at least one physical association cluster. The sensor nodes in the same physical association cluster meet the preset proximity condition in spatial location, and the change of the time series signal meets the continuity constraint described by the physical field propagation model. The matrix construction module is used to construct a corresponding spatiotemporal matrix for each of the physical association clusters based on the time-series signals of each of the sensor nodes in the cluster. The matrix decomposition module is used to decompose the spatiotemporal matrix to obtain at least one time base matrix and at least one spatial coefficient matrix. The time base matrix is ​​used to characterize the change pattern of the physical field within the cluster over time, and the spatial coefficient matrix is ​​used to characterize the response relationship of each sensor node within the cluster relative to the time base matrix. The total number of elements in the time base matrix and the spatial coefficient matrix is ​​less than the total number of elements in the spatiotemporal matrix. The encoding and transmission module is used to encode the time base matrix and the spatial coefficient matrix to generate an encoded data stream and transmit it.

10. A computer device, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the method of any one of claims 1 to 8.

Images