Sensor data compression method, industrial internet of things system and device

Abstract

This application discloses a sensor data compression method, an industrial Internet of Things (IIoT) system, and equipment, relating to the technical field of IIoT. The method includes: decomposing the original time-domain signal to obtain corresponding low-frequency sub-band signals and high-frequency sub-band signals; selecting at least two low-frequency sub-band signals generated by the same physical state process from different sensors, and choosing one as a reference state component; calculating long-term global correlation parameters between the reference state component and each non-reference state component; detecting and extracting transient waveforms triggered by physical events from the high-frequency sub-band signals of each sensor; determining and associating transient waveforms detected within a preset time tolerance window as originating from the same physical event; selecting one of the at least two associated transient waveforms as a reference event waveform; and calculating event-level correlation parameters between the reference event waveform and each non-reference event waveform. This application improves compression efficiency.

Description

Technical Field

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

[0002] In industrial IoT systems, to achieve accurate state perception and health management of physical systems, various types and quantities of sensors (such as vibration, temperature, pressure, and acoustic emission sensors) are typically deployed in key areas to form a dense sensor network platform. These sensors continuously generate raw time-domain signals with high sampling rates, which are collected by the object platform and transmitted to a remote management platform through edge computing nodes for storage and advanced analysis.

[0003] However, this process faces severe technical challenges: the massive amount of multi-channel sensor data puts enormous pressure on limited edge computing resources, local node storage space, and especially the network communication bandwidth for uploading to the management platform. Existing data compression solutions mainly suffer from two shortcomings: First, compressing the data stream of each sensor independently treats multi-sensor data as independent sources, failing to effectively mine and eliminate the widespread redundancy between data, resulting in limited improvement in compression efficiency, which is particularly prominent in bandwidth-constrained scenarios; Second, although there are some multi-channel joint compression methods that utilize signal statistical correlation, these methods are usually computationally complex, requiring completion in the cloud or high-performance computing units, and have extremely stringent data synchronization requirements. The models are often rigid and difficult to adapt to the real-time processing needs of edge-side resource constraints and dynamically changing operating conditions.

[0004] Therefore, existing technologies lack a compression method that can operate with low complexity at the edge, while deeply mining and utilizing the inherent redundancy of multi-sensor data to achieve a high compression ratio and maintain key data features, so as to improve the large-scale deployment and operational economy of industrial IoT systems in data-intensive application scenarios. Summary of the Invention

[0005] To improve compression efficiency, this application provides a sensor data compression method, an industrial Internet of Things system, and equipment.

[0006] Firstly, this application provides a sensor data compression method, which adopts the following technical solution: A sensor data compression method is applied to an industrial Internet of Things (IIoT) system, which includes a management platform, a sensor network platform, and an object platform connected in sequence via communication. The method is executed by the sensor network platform and includes: The original time-domain signals of each sensor on the target physical system are acquired, and the original time-domain signals are decomposed to obtain the corresponding low-frequency subband signals and high-frequency subband signals. Based on temporal correlation, at least two low-frequency sub-band signals generated by the same physical state process from different sensors are screened out, and one of them is selected as a reference state component. The long-term global correlation parameters between the reference state component and each non-reference state component are calculated, wherein the long-term global correlation parameters include at least a first scaling factor and a first deviation value. Transient waveforms triggered by physical events are detected and extracted from the high-frequency subband signals of various sensors. Based on time proximity, transient waveforms detected within a preset time tolerance window are determined and associated as originating from the same physical event. One of the at least two associated transient waveforms is selected as the reference event waveform. Event-level association parameters between the reference event waveform and each non-reference event waveform are calculated. The event-level association parameters include at least a second scaling factor, a second deviation value, and a time offset used to characterize the causal timing of physical events. The reference state components, the reference event waveform, the long-term global correlation parameters, and the event-level correlation parameters are encoded to generate and output a compressed data stream.

[0007] By adopting the above technical solution, the original time-domain signals of each sensor on the target physical system are acquired, and the original time-domain signals are decomposed to obtain the corresponding low-frequency sub-band signals and high-frequency sub-band signals. Then, based on the temporal correlation, at least two low-frequency sub-band signals generated by the same physical state process from different sensors are selected, and one of them is selected as the reference state component. The long-term global correlation parameters between the reference state component and each non-reference state component are calculated. The long-term global correlation parameters include at least a first scaling factor and a first deviation value. Then, transient waveforms caused by physical events are detected and extracted from the high-frequency sub-band signals of each sensor. Based on time proximity, the transient waveforms detected within a preset time tolerance window are determined and associated as originating from the same physical event. One of the at least two associated transient waveforms is selected as the reference event waveform. The event-level correlation parameters between the reference event waveform and each non-reference event waveform are calculated. The event-level correlation parameters include at least a second scaling factor, a second deviation value, and a time offset used to characterize the causal time sequence of the physical event. Then, the reference state component, the reference event waveform, the long-term global correlation parameters, and the event-level correlation parameters are encoded to generate and output a compressed data stream. This method extracts and utilizes the inherent correlations between different sensor signals to compress multiple data streams into a format of "one baseline data stream + multiple sets of simplified parameters," thereby improving the compression ratio and reducing the demand for edge storage and uplink bandwidth. At the same time, it has low computational complexity, is suitable for real-time processing of resource-constrained edge nodes, and the compression process selectively preserves low-frequency trends reflecting the slow changes in equipment status and high-frequency event characteristics representing sudden failures, ensuring the high availability of compressed data for subsequent task analysis and providing a reliable data processing solution for the Industrial Internet of Things.

[0008] Optionally, the step of performing signal decomposition on the original time-domain signal to obtain the corresponding low-frequency subband signal and high-frequency subband signal includes: For each sensor's original time-domain signal, wavelet transform is performed independently to decompose each original time-domain signal into a set of multi-layer high-frequency detail coefficients and a low-frequency approximation coefficient. The low-frequency approximation coefficients corresponding to each original time-domain signal are used as the low-frequency sub-band signals of the corresponding sensors. For each of the original time-domain signals, based on a preset noise energy threshold, the portion with an amplitude exceeding the preset noise energy threshold is extracted from the corresponding high-frequency detail coefficients to form the high-frequency subband signal of the corresponding sensor.

[0009] By adopting the above technical solution, in order to obtain the corresponding low-frequency subband signal and high-frequency subband signal, wavelet transform is performed independently on the original time domain signal of each sensor to decompose each original time domain signal into a set of multi-layer high-frequency detail coefficients and a low-frequency approximation coefficient. Then, the low-frequency approximation coefficient corresponding to each original time domain signal is used as the low-frequency subband signal of the corresponding sensor. Then, for each original time domain signal, according to the preset noise energy threshold, the part with amplitude exceeding the preset noise energy threshold is extracted from the corresponding high-frequency detail coefficient to form the high-frequency subband signal of the corresponding sensor.

[0010] Optionally, the step of selecting one of them as the reference state component includes: Based on a preset time evaluation window, the signal-to-noise ratio of the low-frequency sub-band signal corresponding to each sensor within the time evaluation window is calculated respectively. Calculate the stability index of the change trend of the low-frequency sub-band signal corresponding to each sensor within the time evaluation window, wherein the stability index is quantified by the variance of the first-order difference sequence of the low-frequency sub-band signal; By comparing the signal-to-noise ratio (SNR) values ​​and stability indices of each low-frequency sub-band signal, the low-frequency sub-band signal with the highest SNR and the most stable trend is selected as the reference state component.

[0011] By adopting the above technical solution, in order to select the reference state component, based on a preset time evaluation window, the signal-to-noise ratio (SNR) of the low-frequency sub-band signal corresponding to each sensor is calculated within the time evaluation window. Then, the stability index of the change trend of the low-frequency sub-band signal corresponding to each sensor within the time evaluation window is calculated. The stability index is quantified by the variance of the first-order difference sequence of the low-frequency sub-band signal. Then, the SNR of each low-frequency sub-band signal is compared with the stability index, and the low-frequency sub-band signal with the highest SNR and the most stable trend is selected as the reference state component.

[0012] Optionally, the step of calculating the long-term global correlation parameters between the reference state component and each non-reference state component includes: For each non-benchmark state component, the time series data of the benchmark state component is used as the independent variable, and the time series data of the non-benchmark state component is used as the dependent variable. The relationship between the dependent variable and the independent variable is fitted by linear regression. The slope obtained from the fitting is used as the first scaling factor of the non-reference state component, and the intercept obtained from the fitting is used as the first deviation value. The update of the long-term global correlation parameter is triggered by a timer or a system condition change event, and the update period is longer than the data sampling period.

[0013] By adopting the above technical solution, in order to obtain long-term global correlation parameters, for each non-baseline state component, the time series data of the baseline state component is used as the independent variable, and the time series data of the non-baseline state component is used as the dependent variable. The relationship between the dependent variable and the independent variable is fitted by linear regression. Then, the slope obtained by fitting is used as the first scaling factor of the non-baseline state component, and the intercept obtained by fitting is used as the first deviation value. The update of the long-term global correlation parameters is triggered by a timer or a system operating condition change event, and the update period is longer than the data sampling period.

[0014] Optionally, the step of detecting and extracting the transient waveform caused by the physical event from the high-frequency subband signals of each sensor includes: For each sensor's high-frequency subband signal, the amplitude of the corresponding signal is detected in real time to see if it exceeds a preset event detection threshold. When the amplitude exceeds the event detection threshold, the preset time width is extended forward and backward based on the point of exceeding the limit in order to capture a complete transient waveform candidate segment. Calculate the signal-to-noise ratio of the transient waveform candidate segment and determine whether the corresponding waveform features conform to the preset physical event template; If the signal-to-noise ratio exceeds the preset signal-to-noise ratio threshold and the waveform feature is successfully matched, the transient waveform candidate segment is determined to be a valid transient waveform caused by a physical event and is extracted.

[0015] By adopting the above technical solution, in order to realize the detection and extraction of transient waveforms, the amplitude of the high-frequency sub-band signal of each sensor is detected in real time to see if it exceeds the preset event detection threshold. When the amplitude exceeds the event detection threshold, the preset time width is extended forward and backward based on the over-limit point to capture a complete transient waveform candidate segment. Then, the signal-to-noise ratio of the transient waveform candidate segment is calculated, and it is determined whether the corresponding waveform features meet the preset physical event template. If the signal-to-noise ratio exceeds the preset signal-to-noise ratio threshold and the waveform features are successfully matched, the transient waveform candidate segment is determined to be a valid transient waveform caused by a physical event and is extracted.

[0016] Optionally, the step of determining and associating transient waveforms detected within a preset time tolerance window based on time proximity as originating from the same physical event includes: When a transient waveform is detected from the high-frequency subband signal of any sensor, a time reference point is determined based on the time-domain characteristics of the transient waveform, and the sensor is taken as the target sensor, and the transient waveform is taken as the target transient waveform. Based on the aforementioned time reference point, and combined with the maximum expected propagation delay of the physical process, a time tolerance window is determined; Determine whether a transient waveform is detected within the time tolerance window in the high-frequency subband signal of the non-target sensor. If so, the transient waveform detected within the time tolerance window is taken as the non-target transient waveform. The target transient waveform and the non-target transient waveform are determined and associated as originating from the same physical event.

[0017] By adopting the above technical solution, in order to determine and associate the same physical event, when a transient waveform is detected in the high-frequency subband signal of any sensor, a time reference point is determined according to the time domain characteristics of the transient waveform, and the sensor is taken as the target sensor, the transient waveform is taken as the target transient waveform, and then a time tolerance window is determined based on the time reference point and the maximum expected propagation delay of the physical process. Then, it is determined whether a transient waveform is detected in the high-frequency subband signal of the non-target sensor within the time tolerance window. If so, the transient waveform detected within the time tolerance window is taken as the non-target transient waveform, and then the target transient waveform and the non-target transient waveform are determined and associated as originating from the same physical event.

[0018] Optionally, the step of selecting one of the at least two associated transient waveforms as the reference event waveform includes: For each associated transient waveform, the corresponding peak amplitude and waveform sharpness are calculated, wherein the waveform sharpness is obtained by calculating the ratio of the average slope of the rising edge of the transient waveform to the local noise energy outside the event window; For each associated transient waveform, the peak amplitude and waveform clarity of each transient waveform are compared, and the transient waveform with the largest peak amplitude and the highest waveform clarity is selected as the reference event waveform.

[0019] By adopting the above technical solution, in order to obtain the reference event waveform, for each associated transient waveform, the corresponding peak amplitude and waveform sharpness are calculated. The waveform sharpness is obtained by calculating the ratio of the average slope of the rising edge of the transient waveform to the local noise energy outside the event window. Then, for each associated transient waveform, the peak amplitude and waveform sharpness of each transient waveform are compared, and the transient waveform with the largest peak amplitude and the highest waveform sharpness is selected as the reference event waveform.

[0020] Optionally, the step of calculating the event-level correlation parameters between the reference event waveform and each non-reference event waveform includes: For each of the non-reference event waveforms that are associated with the same physical event, determine the time offset relative to the reference event waveform; Based on the time offset, the non-reference event waveform is time-series aligned with the reference event waveform, and based on the aligned time series, the amplitude ratio between the non-reference event waveform and the reference event waveform is calculated, and the amplitude ratio is used as the second scaling factor corresponding to the non-reference event waveform. Based on the aligned time series, the mean amplitude difference between the non-reference event waveform and the reference event waveform is calculated, and the mean amplitude difference is used as the second deviation value corresponding to the non-reference event waveform, wherein the time offset, the second scaling factor and the second deviation value constitute event-level correlation parameters; For each independently identified physical event, the corresponding reference event waveform and the event-level correlation parameters of each non-reference event waveform are encapsulated into an independent data group.

[0021] By adopting the above technical solution, in order to obtain event-level correlation parameters, for each non-reference event waveform associated with the same physical event, the time offset relative to the reference event waveform is determined. Then, based on the time offset, the non-reference event waveform is time-series aligned with the reference event waveform. Based on the aligned time series, the amplitude ratio between the non-reference event waveform and the reference event waveform is calculated, and the amplitude ratio is used as the second scaling factor corresponding to the non-reference event waveform. Then, based on the aligned time series, the mean amplitude difference between the non-reference event waveform and the reference event waveform is calculated, and the mean amplitude difference is used as the second deviation value corresponding to the non-reference event waveform. Here, the time offset, the second scaling factor, and the second deviation value constitute the event-level correlation parameters. Then, for each independently identified physical event, the event-level correlation parameters of the corresponding reference event waveform and each non-reference event waveform are encapsulated into an independent data group.

[0022] Secondly, this application also provides an industrial Internet of Things (IoT) system, which adopts the following technical solution: An industrial Internet of Things (IoT) system includes a management platform, a sensor network platform, and an object platform that are sequentially connected in communication. The sensor network platform is configured with: The decomposition module is used to acquire the original time-domain signals of each sensor on the target physical system, and to decompose the original time-domain signals to obtain the corresponding low-frequency subband signals and high-frequency subband signals. The low-frequency compression module is used to filter out at least two low-frequency sub-band signals generated by the same physical state process from different sensors based on time-domain correlation, select one of them as a reference state component, and calculate the long-term global correlation parameters between the reference state component and each non-reference state component. The long-term global correlation parameters include at least a first scaling factor and a first deviation value. A high-frequency compression module is used to detect and extract transient waveforms caused by physical events from the high-frequency subband signals of various sensors. Based on time proximity, it determines and associates the transient waveforms detected within a preset time tolerance window as originating from the same physical event. It selects one of the at least two associated transient waveforms as a reference event waveform and calculates the event-level correlation parameters between the reference event waveform and each non-reference event waveform. The event-level correlation parameters include at least a second scaling factor, a second deviation value, and a time offset used to characterize the causal timing of physical events. The encoding module is used to encode the reference state components, the reference event waveform, the long-term global correlation parameters, and the event-level correlation parameters, and generate and output a compressed data stream.

[0023] 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.

[0024] In summary, this application includes at least the following beneficial technical effects: acquiring the original time-domain signals of each sensor on the target physical system, and performing signal decomposition on the original time-domain signals to obtain the corresponding low-frequency sub-band signals and high-frequency sub-band signals; then, based on time-domain correlation, selecting at least two low-frequency sub-band signals generated by the same physical state process from different sensors, and selecting one of them as a reference state component; calculating the long-term global correlation parameters between the reference state component and each non-reference state component, wherein the long-term global correlation parameters include at least a first scaling factor and a first deviation value; and then extracting the high-frequency sub-band signals from each sensor... This method detects and extracts transient waveforms triggered by physical events. Based on time proximity, transient waveforms detected within a preset time tolerance window are identified and associated as originating from the same physical event. One of the at least two associated transient waveforms is selected as the baseline event waveform. Event-level correlation parameters are calculated between the baseline event waveform and each non-baseline event waveform. These parameters include at least a second scaling factor, a second deviation value, and a time offset characterizing the causal sequence of the physical events. The baseline state component, baseline event waveform, long-term global correlation parameters, and event-level correlation parameters are then encoded to generate and output a compressed data stream. This method, by mining and utilizing the inherent correlations between different sensor signals, compresses multiple data streams into a "one baseline data stream + multiple sets of simplified parameters" format, improving the compression ratio and reducing the demand for edge storage and uplink bandwidth. Simultaneously, its computational complexity is low, adaptable to real-time processing of resource-constrained edge nodes. Furthermore, the compression process selectively preserves low-frequency trends reflecting slowly changing equipment states and high-frequency event characteristics characterizing sudden failures, ensuring the high usability of the compressed data for subsequent task analysis. This provides a reliable data processing solution for the Industrial Internet of Things (IIoT). Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the overall process of an embodiment of this application.

[0026] Figure 2 This is a structural diagram of one application scenario of the system in this application embodiment.

[0027] Figure 3 This is a structural diagram of another application scenario of the system according to an embodiment of this application.

[0028] Figure 4 This is a structural block diagram of the computer device described in this application. Detailed Implementation

[0029] 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.

[0030] This application discloses a sensor data compression method.

[0031] Reference Figure 1 A sensor data compression method is applied to an industrial IoT system, which includes a management platform, a sensor network platform, and an object platform that are sequentially connected in communication. The method is executed by the sensor network platform and includes: Step S11: Obtain the original time-domain signals of each sensor on the target physical system, and decompose the original time-domain signals to obtain the corresponding low-frequency subband signals and high-frequency subband signals.

[0032] It should be noted that in step S11, a hardware timestamp or software synchronization mechanism is first used to ensure that the raw time-domain signals collected from multiple sensors on the same physical system have a precise and unified time reference. This is a prerequisite for subsequent cross-sensor correlation analysis. Then, time-frequency domain decomposition is performed on each synchronization signal. This decomposition is not a simple mathematical transformation; its core purpose is to decouple the mixed raw signals into different components based on their physical meaning: the low-frequency subband signal carries the slowly evolving state information of the system (such as overall equipment temperature drift and performance trend degradation); while the high-frequency subband signal mainly contains the instantaneous event information of the system (such as mechanical shock and electrical surge). This step provides the foundation for subsequent differentiated and efficient compression processing of the two different physical processes of "state" and "event".

[0033] Step S12: Based on time-domain correlation, at least two low-frequency sub-band signals generated by the same physical state process from different sensors are selected, and one of them is selected as the reference state component. The long-term global correlation parameters between the reference state component and each non-reference state component are calculated.

[0034] Among them, the long-term global correlation parameters include at least the first scaling factor and the first deviation value.

[0035] It should be noted that the core technology of step S12 lies in utilizing the inherent coupling between low-frequency trends of different sensors caused by the same physical state process (such as system load change). First, by calculating the time-domain cross-correlation coefficients between the low-frequency sub-band signals, highly correlated signal groups that can be determined to originate from the same physical state process are intelligently selected from different sensors. Then, the most stable signal with the highest signal-to-noise ratio is dynamically selected from this group as the reference state component. Then, for each other signal in the group, its long-term global correlation parameters relative to the reference component are calculated using methods such as linear regression. These parameters, with a very small amount of data, characterize the long-term, stable proportional and offset relationship between other signals and the reference signal. Thus, by storing only one reference trend and a small number of parameters, all low-frequency state data can be reconstructed with high fidelity at the decoding end, achieving an extremely high compression ratio.

[0036] Step S13: Detect and extract transient waveforms caused by physical events from the high-frequency subband signals of each sensor. Based on time proximity, determine and associate the transient waveforms detected within the preset time tolerance window as originating from the same physical event. Select one of the at least two associated transient waveforms as the reference event waveform and calculate the event-level association parameters between the reference event waveform and each non-reference event waveform.

[0037] Among them, the event-level correlation parameters include at least a second scaling factor, a second deviation value, and a time offset used to characterize the causal time sequence of physical events.

[0038] It should be noted that the innovation of step S13 lies in realizing cross-sensor event-level correlation and spatiotemporal characterization. First, effective transient waveforms are extracted from the high-frequency sub-band signals of each sensor through threshold detection and waveform matching. Then, based on the principle that physical events have finite time delays in propagation within the system, the concept of a time tolerance window is introduced: when a transient waveform is detected from any sensor, a time window is opened based on the characteristic time point of this waveform (such as the peak moment). Transient waveforms appearing in other sensors within this window are automatically searched, and they are identified and correlated as collaborative observations of the same physical event. The optimal waveform is selected from this set of correlated waveforms as the reference event waveform. Finally, event-level correlation parameters are calculated. This set of parameters accurately describes the amplitude differences, baseline differences, and key causal time relationships observed at different spatial locations for the same event, achieving an extremely concise representation of multi-dimensional observation information of the event.

[0039] Step S14: Encode the reference state components, reference event waveforms, long-term global correlation parameters, and event-level correlation parameters to generate and output a compressed data stream.

[0040] It should be noted that step S14 is responsible for encapsulating the heterogeneous feature data generated in the preceding steps into an efficient compressed bitstream. Its inputs include: a baseline state component characterizing the overall system state, a baseline event waveform characterizing a specific event, a long-term global correlation parameter set describing low-frequency trend relationships, and an event-level correlation parameter set describing the response characteristics of high-frequency events. The encoding process is not a simple concatenation, but rather differentiated processing based on the characteristics of the data: efficient entropy coding or differential coding is used for continuous trend data; compact sparse representation is used for sparse event waveforms; and low-bit-rate parameters are directly encapsulated. Finally, all encoded data units are packaged together with necessary metadata (such as timestamps, sensor IDs, and parameter versions) to generate a standardized compressed data stream for output or transmission. This data stream significantly reduces the amount of original data, and because it strictly preserves the physical correlation structure established in steps S12 and S13, it provides a complete and efficient information carrier for high-fidelity reconstruction at the decoding end and subsequent intelligent analysis.

[0041] In the above implementation, the original time-domain signals of each sensor on the target physical system are acquired, and the original time-domain signals are decomposed to obtain the corresponding low-frequency sub-band signals and high-frequency sub-band signals. Then, based on time-domain correlation, at least two low-frequency sub-band signals generated by the same physical state process from different sensors are selected, and one of them is selected as the reference state component. The long-term global correlation parameters between the reference state component and each non-reference state component are calculated. The long-term global correlation parameters include at least a first scaling factor and a first deviation value. Then, transient waveforms caused by physical events are detected and extracted from the high-frequency sub-band signals of each sensor. Based on time proximity, the transient waveforms detected within a preset time tolerance window are determined and associated as originating from the same physical event. One of the at least two associated transient waveforms is selected as the reference event waveform. The event-level correlation parameters between the reference event waveform and each non-reference event waveform are calculated. The event-level correlation parameters include at least a second scaling factor, a second deviation value, and a time offset used to characterize the causal time sequence of the physical event. Then, the reference state component, the reference event waveform, the long-term global correlation parameters, and the event-level correlation parameters are encoded to generate and output a compressed data stream. This method extracts and utilizes the inherent correlations between different sensor signals to compress multiple data streams into a format of "one baseline data stream + multiple sets of simplified parameters," thereby improving the compression ratio and reducing the demand for edge storage and uplink bandwidth. At the same time, it has low computational complexity, is suitable for real-time processing of resource-constrained edge nodes, and the compression process selectively preserves low-frequency trends reflecting the slow changes in equipment status and high-frequency event characteristics representing sudden failures, ensuring the high availability of compressed data for subsequent task analysis and providing a reliable data processing solution for the Industrial Internet of Things.

[0042] As a further implementation of the method, the step of performing signal decomposition on the original time-domain signal to obtain the corresponding low-frequency sub-band signal and high-frequency sub-band signal includes: Step S21: Perform wavelet transform independently on the original time-domain signal of each sensor to decompose each original time-domain signal into a set of multi-layer high-frequency detail coefficients and a low-frequency approximation coefficient.

[0043] Step S22: Use the low-frequency approximation coefficients corresponding to each original time-domain signal as the low-frequency sub-band signal of the corresponding sensor.

[0044] Step S23: For each original time-domain signal, extract the portion with amplitude exceeding the preset noise energy threshold from the corresponding high-frequency detail coefficients according to the preset noise energy threshold, so as to form the high-frequency subband signal of the corresponding sensor.

[0045] In the above implementation, in order to obtain the corresponding low-frequency subband signal and high-frequency subband signal, wavelet transform is performed independently for the original time domain signal of each sensor to decompose each original time domain signal into a set of multi-layer high-frequency detail coefficients and a low-frequency approximation coefficient. Then, the low-frequency approximation coefficient corresponding to each original time domain signal is used as the low-frequency subband signal of the corresponding sensor. Then, for each original time domain signal, according to a preset noise energy threshold, the part with amplitude exceeding the preset noise energy threshold is extracted from the corresponding high-frequency detail coefficient to form the high-frequency subband signal of the corresponding sensor.

[0046] As a further implementation of the method, the step of selecting one as a reference state component includes: Step S31: Based on the preset time evaluation window, calculate the signal-to-noise ratio of the low-frequency sub-band signal corresponding to each sensor within the time evaluation window.

[0047] Step S32: Calculate the stability index of the change trend of the low-frequency subband signal corresponding to each sensor within the time evaluation window. The stability index is quantified by the variance of the first-order difference sequence of the low-frequency subband signal.

[0048] Step S33: Compare the signal-to-noise ratio (SNR) values ​​and stability indicators of each low-frequency sub-band signal, and select the low-frequency sub-band signal with the highest SNR and the most stable trend as the reference state component.

[0049] In the above implementation, in order to select the reference state component, based on a preset time evaluation window, the signal-to-noise ratio (SNR) of the low-frequency sub-band signal corresponding to each sensor within the time evaluation window is calculated. Then, the stability index of the change trend of the low-frequency sub-band signal corresponding to each sensor within the time evaluation window is calculated. The stability index is quantified by the variance of the first-order difference sequence of the low-frequency sub-band signal. Then, the SNR of each low-frequency sub-band signal is compared with the stability index, and the low-frequency sub-band signal with the highest SNR and the most stable trend is selected as the reference state component.

[0050] As a further implementation of the method, the step of calculating the long-term global correlation parameters between the reference state component and each non-reference state component includes: Step S41: For each non-baseline state component, the time series data of the baseline state component is used as the independent variable, and the time series data of the non-baseline state component is used as the dependent variable. The relationship between the dependent variable and the independent variable is fitted by linear regression.

[0051] Step S42: The slope obtained by fitting is used as the first scaling factor of the non-reference state component, and the intercept obtained by fitting is used as the first deviation value. The update of the long-term global correlation parameter is triggered by a timer or a system operating condition change event, and the update period is longer than the data sampling period.

[0052] In the above implementation, in order to obtain the long-term global correlation parameters, for each non-baseline state component, the time series data of the baseline state component is used as the independent variable, and the time series data of the non-baseline state component is used as the dependent variable. The relationship between the dependent variable and the independent variable is fitted by linear regression. Then, the slope obtained by fitting is used as the first scaling factor of the non-baseline state component, and the intercept obtained by fitting is used as the first deviation value. The update of the long-term global correlation parameters is triggered by a timer or a system operating condition change event, and the update period is longer than the data sampling period.

[0053] As a further implementation of the method, the step of detecting and extracting transient waveforms caused by physical events from the high-frequency subband signals of each sensor includes: Step S51: For each sensor's high-frequency subband signal, detect in real time whether the corresponding amplitude exceeds the preset event detection threshold.

[0054] Step S52: When the amplitude exceeds the event detection threshold, the preset time width is extended forward and backward based on the over-limit point to capture a complete transient waveform candidate segment.

[0055] Step S53: Calculate the signal-to-noise ratio of the transient waveform candidate segment and determine whether the corresponding waveform features conform to the preset physical event template.

[0056] Step S54: If the signal-to-noise ratio exceeds the preset signal-to-noise ratio threshold and the waveform feature is successfully matched, the transient waveform candidate segment is determined as a valid transient waveform caused by a physical event and extracted.

[0057] In the above implementation, in order to detect and extract transient waveforms, the amplitude of the high-frequency subband signal of each sensor is detected in real time to see if it exceeds a preset event detection threshold. When the amplitude exceeds the event detection threshold, the preset time width is extended forward and backward based on the over-limit point to capture a complete transient waveform candidate segment. Then, the signal-to-noise ratio of the transient waveform candidate segment is calculated, and it is determined whether the corresponding waveform features conform to a preset physical event template. If the signal-to-noise ratio exceeds the preset signal-to-noise ratio threshold and the waveform features are successfully matched, the transient waveform candidate segment is determined to be a valid transient waveform caused by a physical event and is extracted.

[0058] As a further implementation of the method, the step of determining and associating transient waveforms detected within a preset time tolerance window as originating from the same physical event based on time proximity includes: Step S61: When a transient waveform is detected from the high-frequency subband signal of any sensor, a time reference point is determined based on the time-domain characteristics of the transient waveform, and the sensor is taken as the target sensor and the transient waveform is taken as the target transient waveform.

[0059] Step S62: Using the time reference point as a benchmark, and combining it with the maximum expected propagation delay of the physical process, determine the time tolerance window.

[0060] Step S63: Determine whether a transient waveform is detected within the time tolerance window in the high-frequency subband signal of the non-target sensor. If so, the transient waveform detected within the time tolerance window will be taken as the non-target transient waveform.

[0061] Step S64: Determine and associate the target transient waveform and the non-target transient waveform as originating from the same physical event.

[0062] In the above embodiments, in order to determine and associate the same physical event, when a transient waveform is detected in the high-frequency subband signal of any sensor, a time reference point is determined based on the time-domain characteristics of the transient waveform, and the sensor is taken as the target sensor, the transient waveform is taken as the target transient waveform, and then a time tolerance window is determined based on the time reference point and the maximum expected propagation delay of the physical process. Then, it is determined whether a transient waveform is detected in the high-frequency subband signal of the non-target sensor within the time tolerance window. If so, the transient waveform detected within the time tolerance window is taken as the non-target transient waveform, and then the target transient waveform and the non-target transient waveform are determined and associated as originating from the same physical event.

[0063] As a further implementation of the method, the step of selecting one of the at least two associated transient waveforms as a reference event waveform includes: Step S71: For each associated transient waveform, calculate the corresponding peak amplitude and waveform sharpness. The waveform sharpness is obtained by calculating the ratio of the average slope of the rising edge of the transient waveform to the local noise energy outside the event window.

[0064] Step S72: For each associated transient waveform, compare the peak amplitude and waveform clarity of each transient waveform, and select the transient waveform with the largest peak amplitude and the highest waveform clarity as the reference event waveform.

[0065] In the above implementation, in order to obtain the reference event waveform, for each associated transient waveform, the corresponding peak amplitude and waveform sharpness are calculated. The waveform sharpness is obtained by calculating the ratio of the average slope of the rising edge of the transient waveform to the local noise energy outside the event window. Then, for each associated transient waveform, the peak amplitude and waveform sharpness of each transient waveform are compared, and the transient waveform with the largest peak amplitude and the highest waveform sharpness is selected as the reference event waveform.

[0066] As a further implementation of the method, the step of calculating the event-level correlation parameters between the reference event waveform and each non-reference event waveform includes: Step S81: For each non-reference event waveform that is associated with the same physical event, determine the time offset relative to the reference event waveform.

[0067] Step S82: Based on the time offset, align the non-reference event waveform with the reference event waveform in terms of time series, and calculate the amplitude ratio between the non-reference event waveform and the reference event waveform based on the aligned time series. Use the amplitude ratio as the second scaling factor corresponding to the non-reference event waveform.

[0068] Step S83: Based on the aligned time series, calculate the mean amplitude difference between the non-reference event waveform and the reference event waveform, and use the mean amplitude difference as the second deviation value corresponding to the non-reference event waveform. Here, the time offset, the second scaling factor, and the second deviation value constitute the event-level correlation parameters.

[0069] Step S84: For each independently identified physical event, the corresponding reference event waveform and the event-level correlation parameters of each non-reference event waveform are encapsulated into an independent data group.

[0070] In the above implementation, in order to obtain the event-level correlation parameters, for each non-reference event waveform associated with the same physical event, a time offset relative to the reference event waveform is determined. Then, based on the time offset, the non-reference event waveform is time-series aligned with the reference event waveform. Based on the aligned time series, the amplitude ratio between the non-reference event waveform and the reference event waveform is calculated, and the amplitude ratio is used as the second scaling factor corresponding to the non-reference event waveform. Then, based on the aligned time series, the mean amplitude difference between the non-reference event waveform and the reference event waveform is calculated, and the mean amplitude difference is used as the second deviation value corresponding to the non-reference event waveform. Here, the time offset, the second scaling factor, and the second deviation value constitute the event-level correlation parameters. Then, for each independently identified physical event, the event-level correlation parameters of the corresponding reference event waveform and each non-reference event waveform are encapsulated into an independent data group.

[0071] This application also discloses an industrial Internet of Things (IoT) system.

[0072] refer to Figure 2 An industrial Internet of Things (IoT) system includes a management platform, a sensor network platform, and an object platform that are connected in sequence via communication. The sensor network platform is configured with: The decomposition module is used to acquire the original time-domain signals of each sensor on the target physical system, and to decompose the original time-domain signals to obtain the corresponding low-frequency subband signals and high-frequency subband signals. The low-frequency compression module is used to filter out at least two low-frequency sub-band signals generated by the same physical state process from different sensors based on time-domain correlation, select one of them as a reference state component, and calculate the long-term global correlation parameters between the reference state component and each non-reference state component. The long-term global correlation parameters include at least a first scaling factor and a first deviation value. The high-frequency compression module is used to detect and extract transient waveforms caused by physical events from the high-frequency subband signals of various sensors. Based on time proximity, it determines and associates the transient waveforms detected within a preset time tolerance window as originating from the same physical event. It selects one of the at least two associated transient waveforms as a reference event waveform and calculates the event-level correlation parameters between the reference event waveform and each non-reference event waveform. The event-level correlation parameters include at least a second scaling factor, a second deviation value, and a time offset used to characterize the causal timing of the physical event. The encoding module is used to encode the reference state components, reference event waveforms, long-term global correlation parameters, and event-level correlation parameters, and generate and output a compressed data stream.

[0073] The overall framework of another application scenario of the industrial IoT system in this application 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 Industrial Internet of Things. The sensor network platform includes a decomposition module, a low-frequency compression module, a high-frequency compression module, and an encoding module.

[0074] Specifically, in another application scenario mentioned above, the industrial IoT system includes a sensor network platform. This platform is configured to: acquire the raw time-domain signals from various sensors on the target physical system; decompose the raw time-domain signals to obtain corresponding low-frequency sub-band signals and high-frequency sub-band signals; based on time-domain correlation, select at least two low-frequency sub-band signals generated by the same physical state process from different sensors, and choose one as a reference state component; calculate the long-term global correlation parameters between the reference state component and each non-reference state component, wherein the long-term global correlation parameters include at least a first scaling factor and a first deviation value; and obtain signals from various sensors... Transient waveforms triggered by physical events are detected and extracted from the high-frequency subband signal of the sensor. Based on time proximity, transient waveforms detected within a preset time tolerance window are determined and associated as originating from the same physical event. One of the at least two associated transient waveforms is selected as the reference event waveform. Event-level correlation parameters between the reference event waveform and each non-reference event waveform are calculated. The event-level correlation parameters include at least a second scaling factor, a second deviation value, and a time offset used to characterize the causal timing of the physical event. The reference state component, the reference event waveform, the long-term global correlation parameters, and the event-level correlation parameters are encoded to generate and output a compressed data stream.

[0075] Through the interaction between the various functional platforms of the industrial IoT system based on the above three or five platforms, a complete closed-loop information operation logic is established, ensuring the orderly operation of sensing and control information and realizing intelligent equipment management.

[0076] The industrial IoT system of the present invention can implement any of the sensor data compression methods, and the specific working process of the industrial IoT system of the present invention can refer to the corresponding process in the above-mentioned sensor data compression methods.

[0077] This application also discloses a computer device.

[0078] 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 sensor data compression methods described above.

[0079] 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 compression method, 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 the sensor network platform and includes: The original time-domain signals of each sensor on the target physical system are acquired, and the original time-domain signals are decomposed to obtain the corresponding low-frequency subband signals and high-frequency subband signals. Based on temporal correlation, at least two low-frequency sub-band signals generated by the same physical state process from different sensors are screened out, and one of them is selected as a reference state component. The long-term global correlation parameters between the reference state component and each non-reference state component are calculated, wherein the long-term global correlation parameters include at least a first scaling factor and a first deviation value. Transient waveforms triggered by physical events are detected and extracted from the high-frequency subband signals of various sensors. Based on time proximity, transient waveforms detected within a preset time tolerance window are determined and associated as originating from the same physical event. One of the at least two associated transient waveforms is selected as the reference event waveform. Event-level association parameters between the reference event waveform and each non-reference event waveform are calculated. The event-level association parameters include at least a second scaling factor, a second deviation value, and a time offset used to characterize the causal timing of physical events. The reference state components, the reference event waveform, the long-term global correlation parameters, and the event-level correlation parameters are encoded to generate and output a compressed data stream.

2. The sensor data compression method according to claim 1, characterized in that, The step of performing signal decomposition on the original time-domain signal to obtain the corresponding low-frequency sub-band signal and high-frequency sub-band signal includes: For each sensor's original time-domain signal, wavelet transform is performed independently to decompose each original time-domain signal into a set of multi-layer high-frequency detail coefficients and a low-frequency approximation coefficient. The low-frequency approximation coefficients corresponding to each original time-domain signal are used as the low-frequency sub-band signals of the corresponding sensors. For each of the original time-domain signals, based on a preset noise energy threshold, the portion with an amplitude exceeding the preset noise energy threshold is extracted from the corresponding high-frequency detail coefficients to form the high-frequency subband signal of the corresponding sensor.

3. The sensor data compression method according to claim 1, characterized in that, The step of selecting one of them as the reference state component includes: Based on a preset time evaluation window, the signal-to-noise ratio of the low-frequency sub-band signal corresponding to each sensor within the time evaluation window is calculated respectively. Calculate the stability index of the change trend of the low-frequency sub-band signal corresponding to each sensor within the time evaluation window, wherein the stability index is quantified by the variance of the first-order difference sequence of the low-frequency sub-band signal; By comparing the signal-to-noise ratio (SNR) values ​​and stability indices of each low-frequency sub-band signal, the low-frequency sub-band signal with the highest SNR and the most stable trend is selected as the reference state component.

4. The sensor data compression method according to claim 1, characterized in that, The step of calculating the long-term global correlation parameters between the reference state component and each non-reference state component includes: For each non-benchmark state component, the time series data of the benchmark state component is used as the independent variable, and the time series data of the non-benchmark state component is used as the dependent variable. The relationship between the dependent variable and the independent variable is fitted by linear regression. The slope obtained from the fitting is used as the first scaling factor of the non-reference state component, and the intercept obtained from the fitting is used as the first deviation value. The update of the long-term global correlation parameter is triggered by a timer or a system condition change event, and the update period is longer than the data sampling period.

5. The sensor data compression method according to claim 1, characterized in that, The step of detecting and extracting transient waveforms caused by physical events from the high-frequency subband signals of each sensor includes: For each sensor's high-frequency subband signal, the amplitude of the corresponding signal is detected in real time to see if it exceeds a preset event detection threshold. When the amplitude exceeds the event detection threshold, the preset time width is extended forward and backward based on the point of exceeding the limit in order to capture a complete transient waveform candidate segment. Calculate the signal-to-noise ratio of the transient waveform candidate segment and determine whether the corresponding waveform features conform to the preset physical event template; If the signal-to-noise ratio exceeds the preset signal-to-noise ratio threshold and the waveform feature is successfully matched, the transient waveform candidate segment is determined to be a valid transient waveform caused by a physical event and is extracted.

6. The sensor data compression method according to claim 1, characterized in that, The step of determining and associating transient waveforms detected within a preset time tolerance window based on time proximity as originating from the same physical event includes: When a transient waveform is detected from the high-frequency subband signal of any sensor, a time reference point is determined based on the time-domain characteristics of the transient waveform, and the sensor is taken as the target sensor, and the transient waveform is taken as the target transient waveform. Based on the aforementioned time reference point, and combined with the maximum expected propagation delay of the physical process, a time tolerance window is determined; Determine whether a transient waveform is detected within the time tolerance window in the high-frequency subband signal of the non-target sensor. If so, the transient waveform detected within the time tolerance window is taken as the non-target transient waveform. The target transient waveform and the non-target transient waveform are determined and associated as originating from the same physical event.

7. The sensor data compression method according to claim 1, characterized in that, The step of selecting one of the at least two associated transient waveforms as the reference event waveform includes: For each associated transient waveform, the corresponding peak amplitude and waveform sharpness are calculated, wherein the waveform sharpness is obtained by calculating the ratio of the average slope of the rising edge of the transient waveform to the local noise energy outside the event window; For each associated transient waveform, the peak amplitude and waveform clarity of each transient waveform are compared, and the transient waveform with the largest peak amplitude and the highest waveform clarity is selected as the reference event waveform.

8. The sensor data compression method according to claim 1, characterized in that, The step of calculating the event-level correlation parameters between the reference event waveform and each non-reference event waveform includes: For each of the non-reference event waveforms that are associated with the same physical event, determine the time offset relative to the reference event waveform; Based on the time offset, the non-reference event waveform is time-series aligned with the reference event waveform, and based on the aligned time series, the amplitude ratio between the non-reference event waveform and the reference event waveform is calculated, and the amplitude ratio is used as the second scaling factor corresponding to the non-reference event waveform. Based on the aligned time series, the mean amplitude difference between the non-reference event waveform and the reference event waveform is calculated, and the mean amplitude difference is used as the second deviation value corresponding to the non-reference event waveform, wherein the time offset, the second scaling factor and the second deviation value constitute event-level correlation parameters; For each independently identified physical event, the corresponding reference event waveform and the event-level correlation parameters of each non-reference event waveform are encapsulated into an independent data group.

9. An industrial Internet of Things (IoT) system, characterized in that, It includes a management platform, a sensor network platform, and an object platform that are connected in sequence, and the sensor network platform is configured with: The decomposition module is used to acquire the original time-domain signals of each sensor on the target physical system, and to decompose the original time-domain signals to obtain the corresponding low-frequency subband signals and high-frequency subband signals. The low-frequency compression module is used to filter out at least two low-frequency sub-band signals generated by the same physical state process from different sensors based on time-domain correlation, select one of them as a reference state component, and calculate the long-term global correlation parameters between the reference state component and each non-reference state component. The long-term global correlation parameters include at least a first scaling factor and a first deviation value. A high-frequency compression module is used to detect and extract transient waveforms caused by physical events from the high-frequency subband signals of various sensors. Based on time proximity, it determines and associates the transient waveforms detected within a preset time tolerance window as originating from the same physical event. It selects one of the at least two associated transient waveforms as a reference event waveform and calculates the event-level correlation parameters between the reference event waveform and each non-reference event waveform. The event-level correlation parameters include at least a second scaling factor, a second deviation value, and a time offset used to characterize the causal timing of physical events. The encoding module is used to encode the reference state components, the reference event waveform, the long-term global correlation parameters, and the event-level correlation parameters, and generate and output a compressed data stream.

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