Multi-device edge side cross sensor data cooperative control method and system

Abstract

The invention relates to the technical field of intelligent physiotherapy, in particular to a multi-device edge side-span sensor data cooperative control method and system, and the method comprises the steps: configuring a lightweight edge computing node for each physiotherapy device, enabling the edge computing node to integrate a sensing element, collecting the sensing data of each physiotherapy device, and carrying out the local preprocessing, the local preprocessing comprises real-time detection and compensation of errors of a single sensor. According to the multi-device edge side cross-sensor data cooperative control method and system, local processing of edge nodes, cross-device data sharing, dynamic weight fusion and real-time feedback control logic are carried out; the technical problems that in an existing scheme, data transmission delay is high, errors of a single sensor are not corrected, fusion weights are fixed, and a control mechanism is lacked are effectively solved, and the cooperative control effect of the multi-device physiotherapy system is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent physiotherapy technology, specifically to a multi-device edge-side cross-sensor data collaborative control method and system. Background Technology

[0002] Single-device home therapy systems often only cover localized treatment needs; for example, a massage chair targets back muscles, while an eye massager targets the area around the eyes. This fails to meet users' demands for multi-area, synergistic therapy. Therefore, multi-device collaborative home therapy systems are gradually becoming an industry trend. The core of these systems lies in the coordinated processing of data from multiple device sensors to achieve synchronized adjustments to therapeutic movements. For instance, while a massage chair relaxes the back, an eye massager can increase the intensity of heat therapy to enhance the therapeutic effect.

[0003] Existing multi-device physiotherapy systems mostly adopt a centralized "device-cloud" processing model: each home physiotherapy device is independently equipped with sensors, such as pressure sensors, electromyography sensors, and temperature sensors. The raw data collected by the sensors is uploaded to a cloud server via Wi-Fi or mobile network. The cloud then fuses the multi-source data to generate control commands, which are then sent to each device for execution. However, this model has the following drawbacks: Firstly, centralized cloud processing leads to high data transmission latency (usually 1-3 seconds), while physiotherapy scenarios have extremely high real-time requirements (such as when a user's muscles suddenly tense up, the device needs to respond within 500 milliseconds), and latency will directly affect the physiotherapy effect; Secondly, a single sensor is susceptible to changes in the environment or its own performance. For example, pressure sensors may experience zero drift after long-term use, resulting in deviations in the collected "no-pressure state" data. Existing systems do not perform real-time error correction at the device end, and relying on cloud-based subsequent processing makes it difficult to compensate for front-end data errors. Third, multi-source data fusion often uses fixed weights, such as 50% weight for pressure data and 50% weight for electromyography (EMG) data, without considering the time synchronization of data from multiple devices. For example, the timestamps of pressure data from a massage chair and EMG data from an eye massager differ by 0.5 seconds, and semantic relevance, such as whether pressure data and EMG data reflect the same physiological state, such as "muscle relaxation". This leads to insufficient accuracy of the fusion results. Fourth, the control process is in open-loop mode. After the cloud sends out instructions, it does not collect feedback from the user's physiological signals in real time, and cannot adapt to changes in the user's state in a timely manner. For example, if the user's electromyography signal shows "tension not relieved", the device will still run according to the original instructions, which reduces the self-adaptive ability of the physiotherapy.

[0004] The aforementioned problems point to the following: existing multi-sensor data collaborative control schemes for home physiotherapy devices suffer from high data transmission latency due to reliance on centralized cloud processing; errors of individual sensors are not corrected in real time at the edge; multi-source data fusion does not fully consider time synchronization and semantic association and has fixed weights; and there is a lack of real-time feedback control mechanisms. As a result, these schemes are unable to meet the needs of physiotherapy scenarios for real-time data processing, fusion accuracy, and adaptive control adjustment. Summary of the Invention

[0005] This disclosure proposes a multi-device edge-side cross-sensor data collaborative control method and system, aiming to overcome at least one of the defects existing in the prior art.

[0006] To achieve the above objectives, the technical solution disclosed in this invention is as follows: According to one aspect of this disclosure, a multi-device edge-side cross-sensor data collaborative control method is provided, comprising the steps of: Each physiotherapy device is equipped with a lightweight edge computing node. The edge computing node integrates sensing elements to collect the sensing data of each physiotherapy device and perform local preprocessing. The local preprocessing includes real-time detection and compensation of single sensor errors. The edge computing nodes achieve end-to-end collaborative data sharing through a self-organizing ZigBee edge network or Bluetooth mesh, and construct a time-semantic dual-dimensional cross-device data association matrix based on the timestamps and semantic associations of the sensor data from multiple devices. The dynamic weighted voting method is used to fuse the sensor data of multiple devices. The dynamic weight is determined based on the two-dimensional correlation of the cross-device data association matrix and the historical accuracy of the sensors, and corrects the error of a single sensor. Based on the fusion results, device collaborative control commands are generated to control multiple devices to perform physiotherapy actions synchronously. Subsequent commands are then corrected based on real-time user feedback to meet the real-time needs of the physiotherapy scenario.

[0007] Furthermore, the real-time detection and compensation of the single sensor error includes: the edge node periodically collects pressure sensor data under no-pressure conditions, updates the zero-point value of the pressure sensor, verifies the accuracy of the zero-point value through electromyography (EMG) sensor data, and compensates for the zero-point drift error of the pressure sensor.

[0008] Furthermore, the cross-device data association matrix includes: time synchronization deviation of multi-device sensor data calculated based on timestamps, and semantic correlation coefficient of multi-device sensor data calculated based on Pearson correlation coefficient. The time synchronization deviation is used to ensure the time consistency of data, and the semantic correlation coefficient is used to quantify the degree of data association.

[0009] Furthermore, the weight calculation method of the dynamic weighted voting method is as follows: for each sensor data, its weight is equal to the sum of the product of the semantic relevance coefficient and α and the product of the sensor historical accuracy and β, where α+β=1, α and β are preset weight coefficients, and the sensor historical accuracy is the consistency ratio between the sensor data and the fusion result within the sliding window.

[0010] Furthermore, when the pressure sensor shows "relaxed" but the electromyography (EMG) sensor shows "tense", the semantic relevance coefficient of the EMG sensor data is higher than that of the pressure sensor data, and the historical accuracy of the EMG sensor is higher than that of the pressure sensor. The weight of the EMG data is also higher than that of the pressure data.

[0011] Furthermore, the step of correcting subsequent instructions based on real-time user feedback includes: after the device executes the collaborative control instruction, the edge node collects the user's electromyographic signal or pressure signal. If the collected signal shows that the user's physiological state has not reached the expected level, the parameters of the subsequent collaborative control instruction are adjusted. The parameters include increasing the massage intensity and / or increasing the massage duration and / or increasing the heat application temperature.

[0012] Furthermore, the self-organizing ZigBee edge network or Bluetooth mesh supports dynamic joining and leaving of edge nodes. When a new device is connected, the network automatically adjusts the communication topology to maintain data transmission. The communication latency of the self-organizing ZigBee edge network or Bluetooth mesh is less than 100ms, enabling real-time sharing of cross-device data.

[0013] Furthermore, the edge computing nodes periodically upload statistical features of the fused data to the cloud. These statistical features include at least the mean, variance, and semantic relevance coefficient of the fused results. The cloud uses an incremental learning algorithm to update the weight model of the dynamic weight voting method to reduce the amount of data transmitted.

[0014] Furthermore, the incremental learning algorithm is random forest incremental learning or neural network incremental learning, used to update the weights in real time to adapt to new data without retraining the weight model.

[0015] According to another aspect of this disclosure, a multi-device edge-side cross-sensor data collaborative control system is provided for implementing the multi-device edge-side cross-sensor data collaborative control method as described above, comprising: An edge node deployment module is used to configure a lightweight edge computing node for each physiotherapy device. The edge computing node integrates sensing elements, collects sensing data and performs local preprocessing. The local preprocessing includes real-time detection and compensation of single sensor errors. The cross-device data sharing module is used to achieve end-to-end collaborative data sharing through a self-organizing ZigBee edge network or Bluetooth mesh. It constructs a time-semantic dual-dimensional cross-device data association matrix based on the timestamps and semantic associations of multi-device sensor data. The cross-validation fusion module is used to fuse multi-source sensor data using a dynamic weighted voting method. The dynamic weights are determined based on the two-dimensional correlation of the cross-device data association matrix and the historical accuracy of the sensors, correcting errors of a single sensor. The collaborative control output module is used to generate collaborative control commands for devices based on the fusion results, control multiple devices to perform physiotherapy actions synchronously, and correct subsequent commands based on real-time user feedback to meet the real-time needs of the physiotherapy scenario.

[0016] The beneficial effects of this invention are: The multi-device edge-side cross-sensor data collaborative control method and system of the present invention effectively solves the technical problems of high data transmission latency, uncorrected single sensor errors, fixed fusion weights, and lack of control mechanisms in existing solutions by using edge node local processing, cross-device data sharing, dynamic weight fusion, and real-time feedback control logic. This significantly improves the collaborative control effect of multi-device physiotherapy systems, enhances the user's physiotherapy experience, and aligns with the industry trend of home physiotherapy equipment developing towards "intelligence, collaboration, and self-adaptation." Its beneficial effects can be summarized as follows: Firstly, by configuring lightweight edge computing nodes for each home physiotherapy device, local preprocessing of sensor data and real-time compensation for single sensor errors, such as dynamic correction of zero-point drift in pressure sensors, were achieved. Local processing at the edge nodes avoids the high latency issues associated with cloud reliance, reducing data transmission latency from 1-3 seconds to less than 500 milliseconds. Simultaneously, cross-device data, such as data from ocular electromyography (EMG) sensors, was used to verify error compensation results, ensuring the accuracy of sensor data and laying a reliable foundation for subsequent multi-source data fusion.

[0017] Secondly, a collaborative data sharing mechanism is built by constructing a self-organizing ZigBee edge network or Bluetooth mesh, and a time-semantic dual-dimensional cross-device data association matrix is ​​constructed based on timestamps and semantic associations, solving the problems of time synchronization and semantic association of multi-device data. The calculation of time synchronization deviation ensures the consistency of multi-device data in the time dimension. For example, the time difference between massage chair pressure data and eye electromyography data is controlled within 100 milliseconds. Semantic correlation coefficients, such as the Pearson coefficient, quantify the degree of correlation between data. For example, when both pressure data and electromyography data reflect a "muscle relaxation" state, the correlation coefficient approaches 1. This provides a structured and highly consistent data foundation for multi-source data fusion, avoiding the fusion errors caused by "simple data splicing" in existing systems.

[0018] Third, a dynamic weighted voting method is used to fuse multi-source sensor data. The weights are dynamically calculated based on two-dimensional correlation (time synchronization deviation and semantic correlation) and the historical accuracy of the sensors (the consistency ratio between sensor data and fusion results within a sliding window). This not only corrects the errors of individual sensors, such as the misjudgment of "relaxation" caused by zero-point drift of pressure sensors, but also corrects them through the high-weight data of electromyography sensors. It can also adapt to changes in sensor performance. For example, when the historical accuracy of a certain sensor decreases, its weight is automatically reduced, thus improving the accuracy and adaptability of multi-source data fusion.

[0019] Fourth, through a real-time feedback closed-loop collaborative control mechanism, after the device executes collaborative control commands, edge nodes collect the user's physiological signals in real time, such as electromyography (EMG) signals and pressure signals, to determine the therapeutic effect. For example, they assess whether the EMG signal has changed from "tension" to "relaxation," and adjust subsequent commands based on the assessment results. If the user's EMG signal is not relieved, the massage intensity of the massage chair is automatically increased. This control mechanism ensures the real-time nature and accuracy of control commands, can adapt to changes in the user's state in a timely manner, and improves the targeting and effectiveness of the therapeutic actions.

[0020] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0021] Figure 1 This is a flowchart of a multi-device edge-side cross-sensor data collaborative control method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of zero-point drift correction of a pressure sensor in one embodiment of the present invention; Figure 3 The diagram shows a heatmap of the time-semantic dual-dimensional correlation matrix, where (a) is a heatmap of cross-device data time synchronization deviation and (b) is a heatmap of cross-device data semantic correlation coefficient. Figure 4 This is a schematic diagram of dynamic weight changes in one embodiment of the present invention; Figure 5 This is a schematic diagram comparing the fusion effect in one embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the real-time feedback control effect in one embodiment of the present invention; Figure 7 A schematic diagram of the integration of statistical features of data and incremental learning, where (a) is a schematic diagram of the changing trends of the mean and variance of the integrated data, and (b) is a schematic diagram of the effect of incremental learning on the accuracy of the weight model; Figure 8 This is a schematic diagram of a radar showing the collaborative effect of multiple devices in one embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0023] In embodiments of the present invention, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Rather, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0024] The present invention provides the following preferred embodiments: Example 1: To address the problems of high data transmission latency, lack of real-time correction of single sensor errors, fixed weights in multi-source data fusion, and lack of closed-loop control in existing multi-device physiotherapy systems that rely on centralized cloud processing, this example discloses a multi-device edge-side cross-sensor data collaborative control method. This method optimizes the real-time performance and accuracy of multi-device collaborative control through local processing at edge nodes, self-organizing network data sharing, dynamic weight fusion, and real-time feedback mechanisms. Figure 1 As shown, the flow of the collaborative control method is as follows: S100: Each physiotherapy device is equipped with a lightweight edge computing node. The edge computing node integrates sensing elements, collects the sensing data of each physiotherapy device and performs local preprocessing. Local preprocessing includes real-time detection and compensation of single sensor errors.

[0025] S200: Edge computing nodes achieve end-to-end collaborative data sharing through self-organizing ZigBee edge networks or Bluetooth mesh, and construct a time-semantic dual-dimensional cross-device data association matrix based on the timestamps and semantic associations of sensor data from multiple devices.

[0026] S300: Employs a dynamic weighted voting method to fuse sensor data from multiple devices. The dynamic weights are determined based on the two-dimensional correlation of the cross-device data association matrix and the historical accuracy of the sensors, correcting errors from individual sensors.

[0027] S400: Generates device collaborative control commands based on the fusion results, controls multiple devices to perform physiotherapy actions synchronously, and corrects subsequent commands based on real-time user feedback to meet the real-time needs of the physiotherapy scenario.

[0028] Specifically, each physiotherapy device, such as a smart massage chair, an eye massager, and a lumbar pressure therapy pad, is equipped with a lightweight edge computing node. This node integrates sensing elements matched to the device's functions; for example, the massage chair integrates a pressure sensor, the eye massager integrates an electromyography (EMG) sensor, and the lumbar pressure therapy pad integrates a temperature sensor. These nodes serve as data processing units on the device side. The edge node handles local preprocessing of the sensor data, with real-time detection and compensation for errors in individual sensors being a key part of this preprocessing. Taking a pressure sensor as an example, prolonged use can lead to zero-point drift due to mechanical fatigue, causing data to deviate from the true value in the absence of pressure. The edge node periodically triggers a no-pressure state detection process, collects the pressure data at this time, calculates the drift, and updates the sensor's zero-point calibration parameters in real time. Simultaneously, it uses sensor data from other devices, such as the stable output of the EMG sensor in the eye massager under no-pressure conditions, to verify the accuracy of the zero-point value, ensuring the reliability of error compensation. Figure 2 As shown, during the zero-point drift correction process of the pressure sensor, the edge nodes track the drift trend in real time and adjust dynamically, avoiding the latency issues associated with subsequent cloud processing. This provides accurate basic data for multi-source data fusion. It should be noted that... Figure 2 The yellow area represents the period during which pressure was applied, and the increase in electromyography (EMG) values ​​during the pressure period verifies the accuracy of the pressure measurement.

[0029] Furthermore, edge nodes achieve end-to-end collaborative data sharing through a self-organizing ZigBee edge network or Bluetooth mesh. This network supports dynamic joining and leaving of nodes. When a new device joins, the network automatically adjusts the communication topology to maintain data transmission continuity. Simultaneously, the low latency of the ZigBee protocol (less than 100ms) ensures real-time data sharing across devices. To address the issues of time synchronization and semantic association of multi-device data, edge nodes construct a time-semantic dual-dimensional cross-device data association matrix based on the timestamps and semantic features of sensor data. In the time dimension, time synchronization deviations are calculated by comparing the timestamps of sensor data from different devices, such as the time difference between massage chair pressure data and eye monitor electromyography (EMG) data, ensuring data consistency in the time dimension. In the semantic dimension, Pearson correlation coefficients are used to calculate the semantic correlation of sensor data from different devices, such as the correlation between the muscle relaxation state reflected by pressure data and the muscle electrical activity state reflected by EMG data. Figure 3 As shown, the time-semantic dual-dimensional correlation matrix visually displays the time synchronization deviation of cross-device data through a heatmap (e.g., Figure 3 As shown in (a), the smaller the value, the better the synchronization) and semantic relevance (such as Figure 3 As shown in (b), the larger the value, the stronger the correlation, providing a structured reference for subsequent multi-source data fusion.

[0030] Furthermore, multi-source data fusion employs a dynamic weighted voting method, where weights are determined by the two-dimensional correlation of the cross-device data association matrix and the historical accuracy of the sensors. Specifically, for each sensor data point, its weight equals the product of the semantic relevance coefficient and a preset weight coefficient α, plus the product of the sensor's historical accuracy and a preset weight coefficient β (α+β=1). Higher semantic relevance coefficients (indicating stronger correlation between the data and the current physiotherapy scenario) and smaller time synchronization deviations (indicating better time consistency of the data) result in higher weights. The historical accuracy of the sensors is calculated using a sliding window, representing the consistency ratio between the sensor data within the window and the fusion result; sensors with higher accuracy have greater weights in the fusion process. For example, when the pressure sensor on the massage chair displays "relaxed" but the electromyography (EMG) sensor on the eye massager displays "tense," if the semantic relevance coefficient of the EMG sensor is higher than that of the pressure sensor, and the historical accuracy of the EMG sensor is also higher, then the weight of the EMG data will be higher than that of the pressure data, ensuring that the fusion result better reflects the user's actual physiological state. Figure 4 As shown, the dynamic weights are adjusted in real time according to changes in semantic relevance and historical accuracy, avoiding fusion errors caused by fixed weights.

[0031] Furthermore, the fused multi-source data generates coordinated control commands for the devices. For example, when the fused results show that the user has back muscle tension and eye muscle fatigue, the control command will instruct the massage chair to increase the intensity of the back massage while the eye massager increases the heat application temperature, achieving synchronized therapeutic actions from multiple devices. To ensure adaptive adjustment of the control commands, the edge node collects real-time feedback of the user's physiological signals, such as electromyography (EMG) signals or pressure signals, after the devices execute the commands to determine whether the therapeutic effect has met expectations. For example, if the EMG signal shows that the user's muscle tension has not been relieved, the edge node will correct subsequent control commands, increasing the massage duration of the massage chair or increasing the heat application temperature of the eye massager; if the pressure signal shows that the user's lumbar pressure has decreased to a comfortable range, the pressure intensity of the lumbar support pad will be appropriately reduced. Figure 6 As shown, the real-time feedback control process forms a closed-loop mechanism, ensuring that control commands can adapt to changes in the user's state in a timely manner and improve the targeting of physiotherapy actions.

[0032] This embodiment achieves local preprocessing by configuring edge computing nodes for each physiotherapy device, constructs a two-dimensional correlation matrix using a self-organizing ZigBee network, fuses data using dynamic weights, and combines real-time feedback correction instructions to optimize the real-time performance, accuracy, and adaptability of multi-device collaborative control, providing an effective technical solution for multi-device collaboration in home physiotherapy systems.

[0033] Example 2: To address the issue of zero-point drift caused by mechanical fatigue during long-term use of pressure sensors, which in turn affects the accuracy of multi-source data fusion, this example further optimizes the real-time detection and compensation process for single sensor errors. It focuses on the dynamic correction and verification mechanism for zero-point drift of pressure sensors, and achieves accurate compensation by relying on the local processing capabilities of edge nodes.

[0034] As the foundation for device-side data processing, edge nodes dynamically adjust the no-pressure state detection cycle based on the usage frequency of the pressure sensor. For example, after the massage chair has been used five times consecutively, or during the device's idle period in the early morning, the no-pressure state detection process is automatically triggered. At this time, the edge node controls the pressure sensor to stop receiving external pressure input, collects its raw output data, and compares this data with the sensor's initial zero-point value (the no-pressure output value calibrated at the device's factory) to calculate the current zero-point drift. Subsequently, the edge node updates the pressure sensor's zero-point calibration parameters based on the drift, subtracting the drift from subsequent pressure data to achieve initial zero-point drift compensation. This process is as follows: Figure 2 As shown in the figure, the original pressure data shows a slow upward drift trend over time. After real-time correction by the edge nodes, the data returns to the zero point value under no pressure, which intuitively demonstrates the dynamic correction effect of zero point drift.

[0035] Furthermore, to ensure the reliability of zero-point compensation, edge nodes acquire electromyography (EMG) sensor data from the eye massager via a self-organizing ZigBee network, using the EMG signal as a verification reference. The output characteristics of the eye EMG sensor are: in a pressure-free state (when the user is not wearing the eye massager), the EMG signal amplitude is extremely low and fluctuates little, exhibiting stability far exceeding that of the mechanical output of the pressure sensor. The edge node compares the output data corresponding to the zero-point value corrected by the pressure sensor with the pressure-free output data of the eye EMG sensor. If the difference is within a preset threshold range, for example, less than 0.1V, the zero-point compensation of the pressure sensor is considered accurate; if the difference exceeds the threshold, the edge node re-triggers the pressure-free state detection process, re-collects pressure data, and calculates the drift until the compensated zero-point value passes EMG verification. The logic of this verification step is that the EMG sensor is unaffected by mechanical fatigue, and its pressure-free output serves as an independent and reliable reference, effectively avoiding miscompensation caused by changes in the mechanical characteristics of the pressure sensor.

[0036] It's important to understand that the local processing capabilities of the edge nodes are the technical support for this process. All detection, calculation, and verification steps are completed on the device itself, without relying on cloud transmission, ensuring real-time zero-point compensation. For example, when a pressure sensor experiences sudden drift during use, the edge node can complete detection and compensation within 100ms, avoiding potential delays caused by cloud processing. Furthermore, the verification process for the electromyography (EMG) sensor is not a simple numerical comparison, but rather based on semantic correlation across device data. Both EMG and pressure signals reflect the user's physiological state (such as relaxation or tension), and in a stress-free state, their outputs should be consistent. This semantic correlation further enhances the validity of the verification.

[0037] This embodiment updates the zero-point value by periodically collecting data under no-pressure conditions and performs dual verification by combining data from an ocular electromyography (EMG) sensor, achieving real-time and reliable compensation for zero-point drift of the pressure sensor. This process not only solves the problem of drift of a single sensor but also ensures the accuracy of compensation through cross-device data correlation verification, providing stable and accurate basic data for subsequent multi-source data fusion and supporting the precision of multi-device collaborative control.

[0038] Example 3: To address the issues of time inconsistency caused by timestamp differences in multi-device sensor data and the impact of unclear semantic associations on the effectiveness of data fusion, this example further refines the construction logic of the cross-device data association matrix, focusing on clarifying the calculation methods and synergistic effects of time synchronization deviation and semantic correlation coefficients, and relying on the local computing capabilities of edge nodes to achieve accurate association of multi-device data.

[0039] Specifically, the construction of the cross-device data association matrix is ​​based on the calculation of time synchronization deviation. Edge nodes acquire sensor data from the massage chair pressure sensor, the eye monitor's electromyography (EMG) sensor, and the lumbar support pressure sensor via a self-organizing ZigBee network. They extract the millisecond-level timestamps of each data point and calculate the time difference between each pair of devices. For example, if the massage chair pressure data timestamp is T1 and the eye monitor's EMG data timestamp is T2, the time synchronization deviation between them is |T1-T2|. The edge nodes then organize the time synchronization deviations of all devices into a matrix, where rows and columns represent the source and target devices, respectively, and the matrix elements are the time differences between the corresponding devices. This process is as follows: Figure 3As shown in the figure, the massage chair and lumbar support cushion have the smallest time synchronization deviation, indicating the best time consistency between their data. The massage chair and eye massager have a slightly larger time deviation. The edge nodes adjust the timestamps of the eye massager data according to the deviation value, aligning them with the time axis of the massage chair to ensure data consistency in the time dimension. It is important to understand that the calculation of time synchronization deviation is not a simple statistical analysis of time differences, but a dynamic correction based on device clock drift. The edge nodes periodically calibrate the clocks of each device, incorporating clock drift into the time deviation calculation to avoid cumulative time errors caused by long-term use.

[0040] Furthermore, based on time synchronization, edge nodes calculate the semantic correlation coefficient of multi-device sensor data using the Pearson correlation coefficient. For example, pressure data from a massage chair reflects the user's back pressure sensation, while electromyography (EMG) data from an eye monitor reflects the user's eye muscle tension; both are related to the user's relaxation state. Edge nodes extract pressure and EMG data within the same time window and calculate their Pearson correlation coefficient; a larger coefficient value indicates a closer semantic relationship between the two. This calculation result is as follows: Figure 3 As shown in the figure, the semantic correlation coefficient between the massage chair pressure data and the eye monitor's electromyography (EMG) data is the highest, indicating that the two are most closely related in reflecting the user's physiological state. The coefficients for the lumbar support cushion pressure data and the eye monitor's EMG data are lower, indicating a weaker semantic correlation. It is understandable that the calculation of the semantic correlation coefficient is not limited to a single time window, but is based on a dynamic calculation using a sliding window. That is, edge nodes use a 10-second sliding window, with the coefficient updated every 5 seconds to ensure that the coefficient reflects the real-time correlation status of the data.

[0041] Furthermore, the time synchronization deviation and semantic relevance coefficient together constitute a cross-device data association matrix. The time synchronization deviation ensures the consistency of data over time and is the foundation for subsequent data fusion. The semantic relevance coefficient quantifies the degree of data correlation and provides a basis for allocating fusion weights. For example, when the time synchronization deviation between massage chair pressure data and eye monitor EMG data is small and the semantic relevance coefficient is high, the edge node will treat them as strongly correlated data and assign them higher weights during fusion. When the time deviation between lumbar pad pressure data and eye monitor EMG data is large or the semantic relevance coefficient is low, the edge node will reduce its weight in the fusion or even exclude the data to avoid interference.

[0042] This embodiment constructs a cross-device data association matrix that includes time synchronization deviation and semantic relevance coefficients, providing a foundation for time consistency and a quantitative basis for semantic association in multi-source data fusion. The calculation of time synchronization deviation ensures data alignment in the time dimension, while the calculation of semantic relevance coefficients quantifies the degree of data association. The synergistic effect of the two enables edge nodes to accurately identify effective associations in multi-device data, providing support for the effectiveness of subsequent fusion processes.

[0043] Example 4: To address the issue that a single semantic relevance coefficient cannot reflect long-term reliability changes of the sensor, leading to a disconnect between weight allocation and actual sensor performance, this example further refines the weight calculation logic of the dynamic weight voting method. It incorporates the semantic relevance coefficient and the sensor's historical accuracy into the weight calculation, and achieves weighted fusion of the two through preset coefficients α and β. It also relies on the real-time computing capabilities of edge nodes to achieve dynamic adjustment of weights.

[0044] Specifically, the calculation of dynamic weights is based on a semantic relevance coefficient, which quantifies the degree of association between the current sensor data and the fusion target using the Pearson correlation coefficient. Examples include the semantic association between massage chair pressure sensor data and the user's back pressure sensation, and the semantic association between eye monitor electromyography (EMG) sensor data and the user's eye muscle tension state. Its value changes in real time with the data, such as... Figure 4 As shown by the blue solid line, the semantic relevance coefficient of the pressure sensor fluctuates between 0.2 and 0.5, reflecting the correlation strength between the current data and the fusion target. Based on this, this embodiment introduces the historical accuracy rate of the sensor, which is the consistency ratio between the sensor data and the fusion result within a sliding window. Specifically, a 10-second sliding window is set for the edge nodes, updated every 5 seconds. The deviation between the sensor data and the fusion result within the window is statistically analyzed to see if it is within a preset threshold; the consistency ratio is the historical accuracy rate. Figure 4 As shown by the blue dashed line, the historical accuracy of the pressure sensor gradually decreased from 0.7 to 0.6 over time, reflecting a decline in its recent reliability.

[0045] Furthermore, the specific calculation method for the weights is as follows: the weight of the sensor data equals the product of the semantic relevance coefficient and α, plus the product of the historical accuracy and β, where α and β are preset coefficients, and their sum is 1. The value of α reflects the importance of the semantic relevance coefficient, and the value of β reflects the importance of the historical accuracy. For example, when α is set to 0.6 and β is set to 0.4, the impact of the semantic relevance coefficient is slightly greater than that of the historical accuracy, but the latter can effectively correct the bias of the single semantic relevance coefficient. Figure 4As shown in the blue-filled area, when the semantic relevance coefficient of the pressure sensor (solid blue line) fluctuates and the historical accuracy (dashed blue line) continues to decline, its weight gradually decreases with the weighted fusion result of the two; while when the semantic relevance coefficient of the electromyography sensor (solid red line) remains at a high level and the historical accuracy (dashed red line) steadily increases, its weight (red-filled area) gradually increases with the synergistic effect of the two, realizing dynamic adjustment of the weight.

[0046] It's important to understand that the semantic relevance coefficient ensures the semantic connection between the current data and the fusion target, while historical accuracy ensures the long-term reliability of the sensor. Their synergistic effect allows the weight allocation to both align with the usefulness of the current data and reflect the actual performance of the sensor. For example, if a sensor has a high current semantic relevance coefficient but a low historical accuracy, it indicates that its current data is closely related to the fusion target but lacks long-term reliability. In this case, the presence of β will lower its weight, preventing abnormal data from the sensor from having an excessive impact on the fusion result. Conversely, if a sensor has a low current semantic relevance coefficient but a high historical accuracy, it indicates that its current data has low relevance but is stable in the long term. The presence of α will limit its weight, preventing irrelevant data from interfering with the fusion result.

[0047] The advantage of this embodiment is that by weighting and fusing the semantic relevance coefficient with the sensor’s historical accuracy, the weight allocation of the dynamic weighted voting method is more in line with the actual performance of the sensor. It not only preserves the semantic relevance characteristics of the current data, but also incorporates the long-term reliability of the sensor, thereby improving the stability and accuracy of the fusion results.

[0048] Example 5: To address the issue of weight allocation being disconnected from the actual user state when different sensor signals conflict, this example further refines the specific application logic of the dynamic weight voting method in conflict scenarios, clarifies the weight adjustment rules under the synergistic effect of semantic relevance coefficient and historical accuracy, and ensures that the weight allocation in conflict scenarios is more in line with the true correlation between sensor performance and user state.

[0049] This embodiment uses a scenario where a user is using a massage chair and an eye massager in tandem: When the user's back is against the massage chair, the pressure sensor detects that the back pressure is below a preset relaxation threshold, displaying a "relaxed" state; however, the eye electromyography (EMG) sensor detects that the EMG signal amplitude of the orbicularis oculi muscle is above a tension threshold, displaying a "tension" state, showing a clear contradiction between the two signals. In this situation, the dynamic weighted voting method needs to determine which sensor data better reflects the user's true state through a combined judgment of semantic relevance and historical accuracy.

[0050] Furthermore, the semantic relevance coefficient is the fundamental logic for weight allocation: EMG sensor data directly originates from the electrophysiological activity of eye muscles, and its semantic association with "eye tension state" is more direct and closer; while pressure sensor data is an indirect measurement of back pressure, and its semantic association with "eye tension state" is weaker. After quantification using the Pearson correlation coefficient, the semantic relevance coefficient between EMG sensor data and "eye tension state" is higher than that between pressure sensor data and "relaxation state," as shown below. Figure 4 The red solid line (semantic relevance) of the electromyography sensor is always above the blue solid line of the pressure sensor, clearly reflecting the difference in the strength of the semantic association between the two.

[0051] Furthermore, historical accuracy serves as a guarantee of the reliability of weight allocation: edge nodes use a 10-second sliding window to statistically analyze the consistency ratio between sensor data and the fusion result. The electromyography (EMG) sensor has a higher recent consistency ratio than the pressure sensor, for example... Figure 4 The red dashed line (historical accuracy) of the electromyography (EMG) sensor gradually increases over time, while the blue dashed line of the pressure sensor decreases slightly, indicating that the recent data of the EMG sensor has a smaller deviation from the fusion results and its performance is more stable.

[0052] Furthermore, based on the weighting calculation logic, the product of the semantic relevance coefficient of the electromyography (EMG) sensor and α, plus the product of the historical accuracy and β, results in a higher value than the corresponding calculated value for the pressure sensor, leading to a higher weighting of EMG data compared to pressure data. For example... Figure 4 The gray-filled area (weight difference) shows that when the semantic relevance and historical accuracy of the electromyography (EMG) sensor are both higher than those of the pressure sensor, the weight difference is always positive, clearly reflecting the dominant position of EMG weight.

[0053] It's important to understand that this weighting allocation doesn't negate the value of pressure sensors, but rather prioritizes sensor data with more direct semantic relevance and more stable historical performance in conflict scenarios. Electromyography (EMG) sensors exhibit stronger semantic relevance in detecting muscle tension, and their historical accuracy further confirms the reliability of the data. Therefore, in conflicting scenarios, the weighting shifts towards EMG sensors to ensure that the fusion result more accurately reflects the user's actual eye muscle state.

[0054] The advantage of this embodiment is that by dynamically adjusting the weights in conflict scenarios, it avoids the impact of misjudgment of a single sensor signal on the fusion result, ensuring that the multi-device collaborative control strategy is more in line with the user's actual needs. When the pressure sensor shows "relaxed" but the electromyography (EMG) sensor shows "tense", the higher EMG weight will guide the fusion result to lean towards the "tense" state, pushing the eye massager to increase the massage intensity, while the massage chair maintains the back relaxation setting, achieving more precise collaborative control.

[0055] Example 6: To achieve real-time linkage between the user's physiological state and device control commands, this example further refines the specific process of correcting subsequent commands based on real-time user feedback. After the device executes the collaborative control command, the edge node collects electromyographic or pressure signals in real time through sensors deployed at the user's contact points. These signals directly reflect the user's current muscle tension or force state.

[0056] For example, when a user receives massage therapy, the amplitude changes of electromyography (EMG) signals can accurately reflect the relaxation status of muscles. If the collected EMG signals are consistently higher than the preset tension threshold, it indicates that the user's muscles are still tense and the expected relaxation effect has not been achieved. At this time, the edge node will adjust the parameters of subsequent collaborative control commands based on the signal analysis results. If the EMG signal shows high muscle tension, the output intensity of the massage module can be increased to enhance the pressing effect; if the signal duration is long, the massage duration can be extended to ensure that the muscles are fully relaxed; if used in conjunction with a heat therapy device, the heat therapy temperature can also be increased to promote local blood circulation and accelerate muscle relaxation.

[0057] Furthermore, such as Figure 6 As shown, during the real-time feedback control process, the fluctuation of electromyographic signals and the adjustment of massage intensity show a clear linkage: when the electromyographic signal increases, the massage intensity increases linearly; when the electromyographic signal decreases to below the relaxation threshold, the massage intensity decreases accordingly. This dynamic adjustment mechanism ensures a precise match between equipment control and user needs.

[0058] It's important to understand that the timeliness of feedback correction directly impacts user experience. The local processing capabilities of edge nodes avoid the latency of cloud transmission, enabling instruction adjustments to be completed as soon as the user's physiological state changes. This embodiment achieves feedback control by directly linking the user's physiological signals with device control parameters, ensuring that the device output always matches the user's current physiological state.

[0059] Example 7: To ensure the stability and real-time performance of multi-device collaborative control, this example further optimizes the dynamic adaptation and low-latency communication mechanism of the self-organizing ZigBee edge network or Bluetooth mesh. The self-organizing ZigBee edge network or Bluetooth mesh supports the dynamic addition and removal of edge nodes. When a new device is connected, such as a newly added heating module or massage head, the network automatically initiates a neighbor discovery protocol. It identifies the physical location, communication capabilities, and functional attributes of the new device through broadcast signals. Subsequently, it recalculates the optimal communication topology based on graph theory algorithms, incorporating the new device into the existing network architecture and allocating a dedicated communication channel to avoid signal conflicts with other devices. For example, when a new heating device is added, the network adjusts the communication path of adjacent massage devices to ensure that the heating device can receive the pressure signal from the massage module in a timely manner, achieving synchronized output from both. Simultaneously, the network's communication latency is strictly controlled within 100ms. This low-latency characteristic ensures real-time data sharing across devices; that is, the pressure data from the massage device can be transmitted to the heating device within 50ms, and the temperature data from the heating device can also be synchronously fed back to the massage module, ensuring a high degree of coordination between the control commands of both.

[0060] Furthermore, such as Figure 8 As shown in the radar chart illustrating the multi-device collaboration effect, with the support of a self-organizing ZigBee network, the collaboration efficiency, response speed, and data synchronization between devices remain at a high level. This is directly related to the network's dynamic adaptability and low-latency communication. It is understandable that dynamic topology adjustment avoids the impact of device addition and removal on the collaboration process, while low-latency communication ensures the real-time nature of data sharing; both jointly support efficient multi-device collaboration. This embodiment solves the network adaptation and latency problems in multi-device collaboration through the design of a self-organizing ZigBee edge network or Bluetooth mesh, providing a reliable network foundation for collaborative control.

[0061] Example 8: To achieve continuous model optimization while reducing data transmission volume, this example further refines the collaborative update mechanism between edge nodes and the cloud. After completing cross-sensor data fusion, the edge computing nodes periodically extract statistical features of the fused data. These features include the mean, variance, and semantic correlation coefficient of the fusion results. The mean reflects the overall trend of the fused data, the variance reflects the degree of data fluctuation, and the semantic correlation coefficient reflects the degree of correlation between different sensor data (such as pressure and electromyography). The combination of these three is sufficient to cover the key information of the fused data.

[0062] Furthermore, instead of transmitting the raw fused data, edge nodes package these statistical features and upload them to the cloud, significantly reducing the amount of data transmitted between the edge and the cloud. After receiving the statistical features, the cloud uses an incremental learning algorithm to update the weight model of the dynamic weight voting method. This means the incremental learning algorithm does not require retraining the entire model; it only needs to adjust the weight parameters in the model based on the new statistical features, such as the allocation ratio between semantic relevance weights and historical accuracy weights. This reduces the computational burden on the cloud while ensuring the model's timeliness.

[0063] like Figure 7 (a) and Figure 7 As shown in (b), changes in the statistical features of the fused data trigger updates to the incremental learning model: when the variance of the fusion result increases, the model will appropriately increase the weight of the pressure sensor data to reduce the impact of fluctuations on the fusion result; when the semantic relevance coefficient increases, the model will increase the weight of the electromyographic signal to strengthen the correlation between the data.

[0064] It's important to understand that uploading statistical features instead of raw data avoids bandwidth waste, while incremental learning ensures continuous model optimization, keeping the fusion effect of the dynamic weight voting method consistently at a high level. This embodiment achieves a balance between model optimization and data transmission efficiency through a collaborative mechanism of uploading statistical features from edge nodes and incremental learning in the cloud.

[0065] Example 9: To achieve real-time model updates to adapt to new data without retraining the weight model, this example further refines the specific implementation of the incremental learning algorithm and its collaborative mechanism with the weight model. After completing cross-sensor data fusion, the edge nodes periodically extract statistical features of the fused data and upload them to the cloud. These statistical features include the mean, variance, and semantic relevance coefficient of the fusion results. Based on these statistical features, the cloud uses random forest incremental learning or neural network incremental learning to update the weight model of the dynamic weight voting method.

[0066] For incremental learning of random forests, the cloud-based random forest model consists of multiple decision trees, each corresponding to a weight allocation rule for different sensor data. When receiving statistical features uploaded from edge nodes, the random forest model does not need to retrain all decision trees; it only needs to adjust the node splitting rules of some trees or add a few new trees based on the new statistical features. For example, when the variance of the fusion result increases, the random forest model will identify the increased data fluctuation and reduce the impact of fluctuation on the fusion result by adjusting the decision tree nodes corresponding to the pressure sensor data and increasing the weight parameters of the pressure sensor data. When the semantic relevance coefficient increases, the model will adjust the decision tree structure corresponding to the electromyography (EMG) signal, increase the weight of the EMG signal, and strengthen the correlation between different sensor data. This approach retains the knowledge of the original model while gradually absorbing information from new data, avoiding the time and computational costs of retraining.

[0067] For incremental learning of neural networks, cloud-based neural network models employ momentum gradient descent or adaptive learning rate algorithms (such as Adam) to adjust the network's weight layer parameters in response to new statistical features. For example, when the mean of the fused data changes, the neural network calculates the error between the statistical features and the original model's output, and uses backpropagation to adjust the allocation ratio of semantic relevance weights and historical accuracy weights, allowing the weight model to adapt to the new trends in the fused data. It's important to understand that incremental learning of neural networks maintains the continuity and stability of the model by adjusting the existing network's weight parameters, avoiding performance fluctuations caused by re-initializing the network.

[0068] like Figure 7 (a) and Figure 7 As shown in (b), changes in the statistical features of the fused data trigger updates to the incremental learning model: when the variance of the fusion result increases, the model appropriately increases the weight of the pressure sensor data; when the semantic relevance coefficient increases, the model increases the weight of the electromyographic signal. This update method does not require retraining the entire model, but only adjusts some parameters or structures, which reduces the computational burden on the cloud and ensures the real-time nature of weight updates.

[0069] This embodiment achieves a balance between real-time updates of the weight model and computational efficiency by clearly defining the specific types and implementation methods of the incremental learning algorithm. It ensures that the fusion effect of the dynamic weight voting method always adapts to the changes in new data, providing a continuously optimized model foundation for cross-sensor data fusion.

[0070] Example 10: To address the issue of collaborative interaction and real-time control of sensor data from distributed physiotherapy devices, this example discloses a multi-device edge-side cross-sensor data collaborative control system. It further refines the module composition and interaction mechanism of the multi-device edge-side cross-sensor data collaborative control system, and achieves full-process optimization of data acquisition, sharing, fusion and control through modular design.

[0071] The edge node deployment module configures a lightweight edge computing node for each physiotherapy device. These nodes integrate sensing elements such as pressure sensors and electromyography (EMG) sensors to collect pressure data and EMG data of the user's muscle state in real time during device operation. The edge nodes perform local preprocessing of the raw data. Real-time detection and compensation of single sensor errors are fundamental operations. Specifically, a sliding window algorithm monitors the zero-point drift of the pressure sensor; when the drift exceeds a preset threshold, a linear compensation model corrects the raw pressure data. High-frequency noise from the EMG sensor is low-pass filtered to reduce the impact of environmental interference on the data, ensuring the basic accuracy of the raw data uploaded to the network.

[0072] Furthermore, the cross-device data sharing module employs a self-organizing ZigBee edge network or Bluetooth mesh to achieve end-to-end collaborative data transmission. Each edge node appends a precise timestamp to the pre-processed data and transmits it to the edge nodes of adjacent physiotherapy devices via the ZigBee network. Nodes calibrate the temporal consistency of data through a timestamp synchronization mechanism, calculating the time synchronization deviation of cross-device data. Simultaneously, based on the semantic attributes of sensor data, such as pressure data corresponding to massage intensity and electromyography data corresponding to muscle tension, the semantic correlation coefficient of sensor data from different devices is calculated using the Pearson correlation coefficient. Based on the time synchronization deviation and semantic correlation coefficient, the system constructs a time-semantic dual-dimensional cross-device data association matrix, providing a data association basis for the subsequent fusion process and ensuring the temporal and semantic consistency of multi-source data.

[0073] Furthermore, the cross-validation fusion module receives time-aligned data transmitted from each edge node and determines the dynamic weights of each sensor's data based on a time-semantic dual-dimensional correlation matrix. For example... Figure 4 As shown, a higher semantic relevance coefficient indicates a stronger correlation between the sensor data and the target physical therapy state, and its weight value increases accordingly. Simultaneously, the historical accuracy of the sensor is considered; this historical accuracy is calculated by comparing the fusion results with the actual values ​​over a past period, thus correcting the semantic weight and forming the final dynamic weight. The fusion module uses a dynamic weighted voting method to weight and fuse multi-source data. For example, the weight of pressure sensor data is calculated by combining the semantic relevance coefficient and historical accuracy; the same applies to electromyography (EMG) sensor data. The fusion result effectively corrects the errors of individual sensors, improving data reliability.

[0074] Furthermore, the collaborative control output module generates multi-device collaborative control commands based on the fusion results. For example, when the fusion results indicate high muscle tension in the user, a command to increase the massage intensity is generated and synchronously sent to all physiotherapy devices via a ZigBee network to ensure consistent device actions. Simultaneously, the system receives user feedback data in real time, such as changes in electromyography (EMG) signals. When EMG signals indicate decreased muscle tension, the collaborative control module adjusts subsequent commands, reducing the massage intensity to achieve real-time optimization of the physiotherapy effect. This feedback mechanism ensures dynamic matching between control commands and the user's state, meeting the real-time needs of physiotherapy scenarios.

[0075] This embodiment achieves sensor data collaboration and real-time control of distributed physiotherapy devices through modular design. Local preprocessing at edge nodes reduces the data transmission burden, cross-device data sharing ensures the temporal and semantic consistency of data, cross-validation fusion improves data reliability, and collaborative control output enables the synchronization and feedback adjustment of device actions, providing system architecture support for cross-sensor data collaboration at the edge of multiple devices.

[0076] Although the present invention has been specifically described above with reference to preferred embodiments, it should be understood that the present invention is not limited to the embodiments described above. Various modifications and variations can be made by those skilled in the art without departing from the spirit of the present invention, and such modifications and variations should fall within the scope defined by the appended claims and their equivalents.

Claims

1. A multi-device edge side-cross sensor data collaborative control method, characterized by the steps of The application relates to a multi-device collaborative control method for a physiotherapy system. A lightweight edge computing node is configured for each physiotherapy device, the edge computing node integrates a sensing element, collects sensing data of each physiotherapy device and performs local preprocessing, the local preprocessing includes real-time detection and compensation of single sensor error; The edge computing node realizes end-to-end collaborative data sharing through a self-organizing ZigBee edge network or a Bluetooth mesh, and constructs a time-semantic two-dimensional cross-device data association matrix based on the time stamp and semantic association of the sensing data of multiple devices; A dynamic weight voting method is used to fuse the sensing data of multiple devices, the dynamic weight is determined based on the two-dimensional correlation of the cross-device data association matrix and the historical accuracy of the sensor, and the single sensor error is corrected; According to the fusion result, a device collaborative control instruction is generated to control multiple devices to synchronously perform physiotherapy actions, and subsequent instructions are corrected based on real-time user feedback to meet the real-time needs of a physiotherapy scene. 2.The multi-device edge side-crossing sensor data collaborative control method of claim 1, wherein, The real-time detection and compensation of the single sensor error includes that the edge node regularly collects pressure sensor data in a pressure-free state, updates the zero point value of the pressure sensor, and verifies the accuracy of the zero point value through eye muscle electric sensor data to compensate the zero point drift error of the pressure sensor. 3.The multi-device edge side-passing sensor data collaborative control method of claim 2, wherein, The cross-device data association matrix includes a multi-device sensing data time synchronization deviation calculated based on a time stamp and a multi-device sensing data semantic correlation coefficient calculated based on a Pearson correlation coefficient, the time synchronization deviation is used to ensure the time consistency of the data, and the semantic correlation coefficient is used to quantify the correlation degree of the data. 4.The multi-device edge side-crossing sensor data collaborative control method of claim 3, wherein, The weight calculation mode of the dynamic weight voting method is that for each sensor data, the weight is equal to the sum of the product of the semantic correlation coefficient and alpha and the product of the historical accuracy of the sensor and beta, wherein alpha+beta=1, alpha and beta are preset weight coefficients, and the historical accuracy of the sensor is the consistency proportion of the sensor data and the fusion result in a sliding window. 5.The multi-device edge side-crossing sensor data collaborative control method of claim 4, wherein, When the pressure sensor displays "relaxation" but the eye muscle electric sensor displays "tension", the semantic correlation coefficient of the muscle electric sensor data is higher than that of the pressure sensor data, and the historical accuracy of the muscle electric sensor is higher than that of the pressure sensor, so the weight proportion of the muscle electric data is higher than that of the pressure data. 6.The multi-device edge side-crossing sensor data collaborative control method of claim 1, wherein, The step of correcting subsequent instructions based on real-time user feedback includes that after the devices execute the collaborative control instruction, the edge node collects the muscle electric signal or the pressure signal of the user, if the collected signal shows that the physiological state of the user does not reach the expectation, the parameters of the subsequent collaborative control instruction are adjusted, the parameters include increasing the massage intensity or / and increasing the massage time or / and increasing the hot compress temperature. 7.The multi-device edge side-crossing sensor data collaborative control method of claim 1, wherein, The self-organizing ZigBee edge network or the Bluetooth mesh supports dynamic joining and exiting of the edge node, when a new device accesses, the network automatically adjusts the communication topology to maintain data transmission; the communication delay of the self-organizing ZigBee edge network or the Bluetooth mesh is less than 100 ms, so that the cross-device data can be shared in real time. 8.The multi-device edge side-crossing sensor data collaborative control method of claim 1, wherein, The edge computing node periodically uploads statistical features of the fused data to the cloud, the statistical features at least including mean, variance and semantic correlation coefficient of the fusion result, and the cloud updates a weight model of a dynamic weight voting method using an incremental learning algorithm to reduce data transmission volume. 9.The multi-device edge side-crossing sensor data collaborative control method of claim 8, wherein, The incremental learning algorithm is random forest incremental learning or neural network incremental learning, which is used to update weights in real time without retraining the weight model to adapt to new data.

10. A multi-device edge side-crossing sensor data collaborative control system for implementing the multi-device edge side-crossing sensor data collaborative control method of any one of claims 1-9, wherein, The method comprises the following steps: An edge node deployment module is configured to configure a lightweight edge computing node for each physiotherapy device, the edge computing node integrating a sensing element, collecting sensing data and performing local preprocessing, the local preprocessing including real-time detection and compensation of single sensor error; A cross-device data sharing module is configured to realize end-to-end collaborative data sharing through a self-organizing ZigBee edge network or a Bluetooth mesh, and to construct a time-semantic two-dimensional cross-device data association matrix based on timestamps and semantic association of multi-device sensing data; A cross-validation fusion module is configured to fuse multi-source sensing data using a dynamic weight voting method, the dynamic weight being determined based on two-dimensional correlation of the cross-device data association matrix and historical accuracy of the sensor to correct single sensor error; A collaborative control output module is configured to generate device collaborative control instructions according to the fusion result, control multiple devices to perform physiotherapy actions synchronously, and correct subsequent instructions based on real-time user feedback to meet real-time requirements of the physiotherapy scene.

Images