Pump sticker wearing state monitoring and early warning device

Abstract

The invention discloses a pump sticker wearing state monitoring and early warning device, which belongs to the technical field of medical equipment monitoring and comprises a data processing module, a baseline calculation module, a feature extraction module, a state evaluation module, an early warning output module, an execution feedback module and a model optimization module. A collaborative method of multi-sensor data fusion, dynamic baseline evaluation and closed-loop self-learning is adopted, and accurate and self-adaptive monitoring and humanized grading early warning of the wearing state of the pump sticker can be achieved.

Description

Technical Field

[0002] This invention relates to the field of medical device monitoring technology, and in particular to a device for monitoring and early warning of the wearing status of a pump patch. Background Technology

[0004] Pump patches, such as insulin pump patches or analgesic pump patches, are external drug delivery devices widely used in modern medicine. They can be attached to the patient's skin for extended periods to achieve continuous and precise drug infusion. These devices are crucial for patients requiring long-term treatment, especially those in hospitals, rehabilitation centers, or home care settings, and the stability of their use directly affects treatment effectiveness and patient safety.

[0005] Currently, existing technologies for monitoring the wearing status of pump patches typically employ a single sensor-based approach. These approaches rely on a pressure sensor installed at the bottom of the pump patch to measure the contact pressure between the patch and the skin in real time and compare it to a pre-set fixed pressure threshold to determine if the patch is at risk of loosening or falling off. When the measured pressure value falls below this threshold, the system triggers an alarm.

[0006] However, due to the inevitable fluctuations in the contact pressure between the pump patch and the skin caused by patients' daily activities such as turning over, walking, or changing body posture, the system struggles to distinguish these harmless pressure changes from pressure drops genuinely caused by the pump patch loosening, leading to numerous false alarms. Furthermore, its alerting methods are often singular and fixed, unable to adapt to the patient's environment, potentially causing unnecessary disturbance during nighttime rest or in quiet settings. Simultaneously, this fixed monitoring standard fails to accommodate individual differences among patients. Summary of the Invention

[0008] To address the aforementioned issues, this invention provides a pump patch wearing status monitoring and early warning device. It employs a collaborative method of multi-sensor data fusion, dynamic baseline assessment, and closed-loop self-learning to achieve accurate, adaptive monitoring and user-friendly tiered early warning of the pump patch wearing status.

[0009] The above objectives can be achieved through the following approach:

[0010] The pump patch wearing status monitoring and early warning device includes a data processing module, which is used to acquire pressure sensor data and multi-axis acceleration sensor data of the pump patch, and perform time alignment and noise filtering on the data to generate a fused data stream;

[0011] The system comprises the following modules: a baseline calculation module for calculating a dynamic pressure baseline based on the fused data stream and preset baseline coefficients; a feature extraction module for extracting pressure motion coupling features from the fused data stream; a state evaluation module for calculating a state deviation index based on the pressure motion coupling features and the dynamic pressure baseline; an early warning output module for acquiring environmental context parameters and performing a balance analysis of early warning effectiveness and intrusion degree based on the state deviation index and the environmental context parameters, thereby generating an early warning control signal; an execution feedback module for executing an early warning output operation according to the early warning control signal and acquiring user feedback data after executing the early warning output operation; and a model optimization module for optimizing the baseline coefficients based on the user feedback data.

[0012] Optionally, the data processing module further includes: a sampling synchronization unit, used to acquire pressure sensor data and multi-axis accelerometer data from the pump patch to form raw multi-source data, and to perform timestamp calibration on the raw multi-source data to obtain time-aligned data; a noise reduction and enhancement unit, used to perform bandpass filtering and outlier suppression processing on the time-aligned data to generate purification sensing data; a time-series fusion unit, used to perform multi-scale time window splicing based on the purification sensing data to obtain a multi-scale sequence; and a weighted fusion unit, used to perform channel weighting and amplitude normalization on the multi-scale sequence to generate a fused data stream.

[0013] Optionally, the baseline calculation module includes: a time period analysis unit, used to analyze multi-axis accelerometer data in the fused data stream to identify stable wearing periods; a pressure extraction unit, used to extract pressure sensor data corresponding to the stable wearing periods to form a stable pressure dataset; and a data calculation unit, used to perform moving average calculation on the stable pressure dataset based on a preset baseline coefficient to generate a dynamic pressure baseline.

[0014] Optionally, the feature extraction module includes: a pattern analysis unit for analyzing multi-axis accelerometer data in the fused data stream to identify the current motion pattern; a change feature extraction unit for extracting change features of pressure sensor data under the current motion pattern to form pressure change features; and a feature coupling unit for combining the current motion pattern with the pressure change features to generate pressure-motion coupling features.

[0015] Optionally, the state assessment module includes: a basic deviation calculation unit, used to obtain the current pressure value from the fused data stream and calculate the statistical deviation of the current pressure value from the dynamic pressure baseline; a coefficient extraction unit, used to obtain adjustment weight coefficients from a preset weight mapping table based on the pressure motion coupling characteristics; and a state deviation calculation unit, used to correct the statistical deviation using the adjustment weight coefficients to generate a state deviation index.

[0016] Optionally, the early warning output module includes: an environmental scenario acquisition unit for acquiring environmental scenario parameters; an early warning effectiveness analysis unit for determining the required early warning effectiveness level based on the state deviation index; an intrusion degree analysis unit for determining the user intrusion sensitivity level based on the environmental scenario parameters; and a signal output unit for selecting and generating an early warning control signal from a preset combination of early warning methods based on the early warning effectiveness level and the user intrusion sensitivity level.

[0017] Optionally, obtaining environmental context parameters includes: obtaining ambient light intensity data and ambient sound decibel data; performing classification mapping processing on the ambient light intensity data and the ambient sound decibel data to determine the illumination state and sound state; and combining the illumination state and the sound state to generate environmental context parameters.

[0018] Optionally, the execution feedback module includes: a warning parsing unit, used to parse the warning control signal to determine the warning level, the warning level including a level one warning and a level two warning; a level one warning output unit, used to call a preset communication unit to send a warning message to a contact when the warning level is a level one warning; a level two warning output unit, used to call a preset local alert unit to activate one or more of vibration, sound, or light alerts when the warning level is a level two warning; and a user feedback unit, used to obtain and analyze the user's response behavior to the warning after sending the warning message and activating one or more of vibration, sound, or light alerts to obtain user feedback data.

[0019] Optionally, the model optimization module includes: a label generation unit, used to acquire the early warning control signal and the user feedback data and associate them to form an early warning feedback label; a sample construction unit, used to time-align and slice the fused data stream, the state deviation index and the early warning feedback label to generate labeled historical data samples; and a coefficient update unit, used to update the baseline coefficient and the adjustment weight coefficient based on the labeled historical data samples.

[0020] Based on the same inventive concept, this invention also provides a method for monitoring and warning of the wearing status of a pump patch. The method includes: acquiring pressure sensor data and multi-axis accelerometer data of the pump patch, and performing time alignment and noise filtering on the data to generate a fused data stream; calculating a dynamic pressure baseline based on the fused data stream and a preset baseline coefficient; extracting pressure-motion coupling features from the fused data stream; calculating a state deviation index based on the pressure-motion coupling features and the dynamic pressure baseline; acquiring environmental context parameters, and performing a balance analysis of warning effectiveness and intrusion degree based on the state deviation index and the environmental context parameters to generate a warning control signal; executing a warning output operation according to the warning control signal, and acquiring user feedback data after executing the warning output operation; and optimizing the baseline coefficient based on the user feedback data.

[0021] Compared with the prior art, the present invention has the following advantages:

[0022] 1. This invention integrates pressure and multi-axis acceleration sensor data and establishes a dynamic pressure baseline and pressure motion coupling characteristics. It can distinguish between pressure changes caused by normal patient activity or positional changes and abnormalities caused by actual loosening or displacement of the pump patch. This improves the accuracy and reliability of monitoring and avoids unnecessary interference to patients or nursing staff due to false alarms.

[0023] 2. This invention introduces the perception and analysis of environmental context, combining early warning decisions with the patient's surrounding environment, such as light and sound conditions, to achieve a dynamic balance between early warning effectiveness and intrusiveness. This allows the device to use a gentle reminder when the patient is resting or in a quiet environment, while using a stronger warning in noisy environments, providing a humanized early warning experience and improving patient comfort and acceptance of the device.

[0024] 3. The device of the present invention can automatically adjust its internal baseline coefficient and weight coefficient and other key parameters according to the actual feedback of the user, such as the response behavior to the alarm, so that the monitoring model can continuously self-evolve and adapt to the physiological characteristics and activity habits of specific patients, thereby achieving deep personalization and ensuring the effectiveness and robustness of the device in long-term use.

[0025] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a framework diagram of the pump patch wearing status monitoring and early warning device according to an embodiment of the present invention.

[0029] Figure 2 This is a schematic diagram of the structure of the pump patch wearing status monitoring and early warning device according to an embodiment of the present invention.

[0030] Figure 3 This is a schematic diagram of the fused data stream and dynamic baseline according to an embodiment of the present invention.

[0031] Figure 4 This is a schematic flowchart of the pump patch wearing status monitoring and early warning device according to an embodiment of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Reference Figure 1 One embodiment of the present invention proposes a pump patch wearing status monitoring and early warning device, which adopts a collaborative method of multi-sensor data fusion, dynamic baseline evaluation and closed-loop self-learning, and can achieve accurate and adaptive monitoring of the pump patch wearing status and humanized hierarchical early warning.

[0035] The apparatus described in this embodiment specifically includes:

[0036] The data processing module is used to acquire pressure sensor data and multi-axis accelerometer data of the pump patch, and to perform time alignment and noise filtering on the data to generate a fused data stream;

[0037] The baseline calculation module is used to calculate the dynamic pressure baseline based on the fused data stream and preset baseline coefficients;

[0038] The feature extraction module is used to extract pressure-motion coupling features from the fused data stream;

[0039] The state assessment module is used to calculate the state deviation index based on the pressure-motion coupling characteristics and the dynamic pressure baseline.

[0040] The early warning output module is used to acquire environmental context parameters, and based on the state deviation index and the environmental context parameters, to perform a balance analysis of early warning effectiveness and intrusion degree, and generate an early warning control signal.

[0041] The execution feedback module is used to execute an early warning output operation based on the early warning control signal, and to obtain user feedback data after the early warning output operation is executed;

[0042] The model optimization module is used to optimize the baseline coefficients based on the user feedback data.

[0043] The device synchronizes and purifies the raw pressure and acceleration sensor data through a data processing module, generating a unified fused data stream. Based on this data, the system establishes a dynamic pressure baseline that slowly changes with the user's wearing status as a stable reference through a baseline calculation module. On the other hand, a feature extraction module couples instantaneous pressure changes with the user's specific movement patterns, generating pressure-motion coupling features that reflect the context. The state assessment module compares the current pressure data with the dynamic pressure baseline and uses the pressure-motion coupling features to weight the degree of deviation, thereby calculating a quantified state deviation index. The warning output module comprehensively considers this risk index and the external environmental context, performing a trade-off analysis between warning effectiveness and user intrusion, generating the most suitable warning control signal. Finally, the execution feedback and model optimization modules form a learning closed loop. After issuing a warning, the device continuously optimizes its internal baseline coefficients and other key parameters based on user feedback, enabling the entire system's judgment logic to continuously improve itself.

[0044] This invention is first and foremost about the accuracy of its warnings. By combining pressure data with the background of movement for analysis, it can distinguish between abnormal pressure changes caused by loosening of the device and benign pressure fluctuations caused by normal user activity, thereby reducing the false alarm rate. Secondly, the device enhances the user experience. It can sense the user's environment and, while ensuring effective warning delivery, selects the least intrusive alert method to avoid disturbing users in inappropriate situations, thus solving the alarm fatigue problem of traditional alarm systems. Finally, through continuous model optimization based on user feedback, it can self-evolve, constantly adapting to individual wearing habits and lifestyles. This makes its monitoring and warning performance increasingly reliable and personalized over time, ensuring long-term effectiveness and user trust.

[0045] Optionally, such as Figure 2 As shown, the data processing module further includes:

[0046] The sampling synchronization unit is used to acquire pressure sensor data and multi-axis accelerometer data from the pump patch to form raw multi-source data, and to perform timestamp calibration on the raw multi-source data to obtain time-aligned data;

[0047] The noise reduction and enhancement unit is used to perform bandpass filtering and outlier suppression on the time-aligned data to generate purified sensor data.

[0048] The time-series fusion unit is used to perform multi-scale time window splicing based on the purification sensor data to obtain a multi-scale sequence.

[0049] The weighted fusion unit is used to perform channel weighting and amplitude normalization on the multi-scale sequence to generate a fused data stream.

[0050] Specifically, the sampling synchronization unit first initiates the process. This unit acquires pressure data measured by the pressure sensor on the pump patch and acceleration data along each axis measured by the multi-axis accelerometer. These unprocessed signals constitute the raw multi-source data. Considering the slight differences in sampling clock and response delay that may exist between different sensor hardware, this unit timestamps each data stream. Through linear interpolation or more complex synchronization algorithms, all data points are remapped onto a unified time axis, ensuring that the pressure value at any given moment precisely corresponds to the motion state data at that moment, generating time-aligned data. Subsequently, the time-aligned data is sent to the noise reduction and enhancement unit. To filter out interference components in the signal that are unrelated to the wearing state, this unit first applies bandpass filtering. This processing can effectively remove high-frequency noise introduced by circuit noise or high-frequency vibration, and also filter out slowly changing low-frequency baseline drift caused by factors such as sensor temperature drift, retaining the core frequency band signals related to user activity and pump patch status changes. Next, outlier suppression processing is performed. This process identifies and removes isolated data points that are far beyond the normal range due to sudden strong impacts or data acquisition errors by setting reasonable signal amplitude thresholds or using statistical methods, thereby generating purification sensor data with higher signal-to-noise ratio and purer data.

[0051] Next, the temporal fusion unit performs multi-scale time window stitching based on the purification sensor data. This operation aims to simultaneously capture rapid changes in wearing status within a short timeframe and trends in wearing posture over a long period. This unit sets multiple time windows of varying lengths; for example, short windows are used to identify rapid movements such as walking and turning, while long windows are used to analyze stable states such as sitting and lying down. By sliding these windows across the purification sensor data stream and extracting data segments, these segments at different time scales are then combined to construct a multi-scale sequence containing rich temporal contextual information.

[0052] Finally, the weighted fusion unit integrates the multi-scale sequences to generate the final fused data stream. This unit first weights the channels based on their importance to the wear status assessment. For example, pressure data has a higher weight when determining whether the pump patch fits snugly. Then, to eliminate numerical scale differences caused by different physical dimensions, such as the different ranges of pressure and acceleration values, this unit performs amplitude normalization, mapping all data to a standardized numerical range, typically from 0 to 1. The amplitude normalization calculation can be expressed as:

[0053] ,

[0054] in, These are the normalized data values. This represents the original data value at the current moment. and These represent the maximum and minimum values ​​of the data channel within a preset time period. After this series of processing steps, the module finally outputs a fused data stream that is time-synchronized, noise-suppressed, feature-enhanced, and scale-uniform, laying the foundation for subsequent accurate analysis.

[0055] Optionally, the baseline calculation module includes:

[0056] A time-period analysis unit is used to analyze multi-axis accelerometer data in the fused data stream to identify stable wearing periods;

[0057] The pressure extraction unit is used to extract pressure sensor data corresponding to the stable wearing period to form a stable pressure dataset.

[0058] The data calculation unit is used to perform moving average calculation on the stable pressure dataset based on a preset baseline coefficient to generate a dynamic pressure baseline.

[0059] Specifically, such as Figure 3As shown, the time-period analysis unit continuously analyzes the multi-axis accelerometer data in the fused data stream received from the data processing module. To identify stable wearing periods, this unit calculates the variance or standard deviation of the amplitude of the synthesized acceleration signal vector within a specific time window. When this indicator is below a preset motion stability threshold, it means that the user is currently in a stationary or slightly moving state, such as sitting, lying down, or walking slowly, and this time period is identified as a stable wearing period. Once a stable wearing period is identified, the pressure extraction unit immediately starts. Based on the timestamp information of the stable wearing period, it accurately extracts the pressure sensor data synchronously collected during these periods from the fused data stream. These filtered pressure data points, because they exclude instantaneous pressure fluctuations caused by body compression or pulling during strenuous exercise, can more accurately reflect the pressure level of the pump patch under ideal fit conditions. All these extracted pressure data together constitute a stable pressure dataset.

[0060] Finally, the data computation unit calculates and updates the dynamic pressure baseline based on this stable pressure dataset. This unit employs a weighted moving average algorithm, such as an exponentially weighted moving average, to smoothly update the baseline values. This process utilizes a pre-defined baseline coefficient, which determines the weight of the impact of new stable pressure data on the existing baseline. The calculation of the dynamic pressure baseline can be expressed as:

[0061] ,

[0062] in, This represents the currently calculated dynamic pressure baseline value. It is the latest or a range of average pressure values ​​obtained from a stable pressure dataset. This is the preset baseline coefficient, which ranges from 0 to 1 and controls the rate of baseline updates and the sensitivity to new data. This refers to the dynamic pressure baseline value prior to this. In this way, the dynamic pressure baseline is only updated when the user is in a stable state, ensuring its reliability and accuracy as a reference standard.

[0063] Optionally, the feature extraction module includes:

[0064] The pattern analysis unit is used to analyze the multi-axis accelerometer data in the fused data stream to identify the current motion pattern;

[0065] A change feature extraction unit is used to extract the change features of pressure sensor data under the current motion mode to form pressure change features;

[0066] The feature coupling unit is used to combine the current motion mode with the pressure change feature to generate a pressure-motion coupling feature.

[0067] Specifically, the pattern analysis unit first processes the multi-axis accelerometer data portion of the fused data stream. By applying a sliding time window to the continuous data stream, this unit calculates statistical and morphological indices such as amplitude, variance, and frequency domain energy distribution of the acceleration signal within the window. Subsequently, it inputs these calculated indices into a pre-trained motion pattern classifier, such as a decision tree or support vector machine model, to identify and label the user's activity state as a specific current motion pattern in real time, such as standing still, walking, running, or bending over. After determining the current motion pattern, the change feature extraction unit, based on the current motion pattern label provided by the pattern analysis unit, selectively extracts the most relevant change features from the synchronized pressure sensor data. For example, if the current motion mode is "stationary," the unit will focus on calculating the stability and small fluctuation amplitude of the pressure signal, such as its variance or root mean square value. If the mode is "walking," it will use frequency domain analysis methods such as Fourier transform to extract the periodic fluctuation frequency and amplitude of the pressure signal to determine whether it matches the step frequency. If the mode is "bending over," it will focus on the instantaneous rate of change or peak-to-trough values ​​of the pressure signal to capture large-amplitude, non-periodic pressure changes. These quantitative indicators calculated for different motion scenarios collectively constitute the pressure change characteristics.

[0068] Finally, the feature coupling unit integrates the two pieces of information to generate the final pressure-motion coupling feature. This step is not a simple numerical addition, but rather a structured combination of discrete motion pattern labels and continuous pressure change feature values ​​to form an information-rich feature vector or data structure. For example, a generated pressure-motion coupling feature might be described as "pattern walking, pressure fluctuation frequency 1.8 Hz, fluctuation amplitude 0.5 kPa". This combined feature not only describes the pressure change, but more importantly, it explains the motion context in which this change occurs, providing a deep contextual basis for subsequent state assessment.

[0069] This method enables intelligent and contextualized interpretation of pump pressure changes, surpassing the limitations of traditional methods that rely solely on pressure values. By correlating pressure changes with the user's specific movement patterns, it can distinguish between pressure fluctuations caused by regular compression during normal physical activities such as walking, and abnormal pressure changes resulting from actual loosening or displacement of the pump. For example, a sustained drop in pressure at rest is a highly suspicious signal, while a brief, sudden drop in pressure during bending may be considered normal. This coupled analysis improves the accuracy and specificity of monitoring, reduces false alarms caused by daily user activities, and makes the early warning system more precise, reliable, and user-friendly.

[0070] Optionally, the state assessment module includes:

[0071] A basic deviation calculation unit is used to obtain the current pressure value from the fused data stream and calculate the statistical deviation of the current pressure value from the dynamic pressure baseline.

[0072] The coefficient extraction unit is used to obtain the adjustment weight coefficient from the preset weight mapping table based on the pressure motion coupling characteristics.

[0073] The state deviation calculation unit is used to correct the statistical deviation using the adjusted weighting coefficient to generate a state deviation index.

[0074] Specifically, the basic deviation calculation unit first obtains the current pressure value in real time from the fused data stream and compares it with the dynamic pressure baseline provided by the baseline calculation module. To obtain a standardized initial deviation, it calculates the statistical deviation between the two, which can be expressed as:

[0075] ,

[0076] in, It is a dimensionless statistical deviation, representing the relative magnitude of pressure deviating from the baseline. It is the pressure value at the current moment obtained from the fused data stream. This is the current dynamic pressure baseline value. This step yields a preliminary deviation rate that does not consider contextual factors. Next, the coefficient extraction unit assigns contextual weights to the deviation rate. This unit receives the pressure-motion coupling features generated by the feature extraction module. It uses this feature as a query index to look up a value in a pre-defined weight mapping table. This weight mapping table is a pre-calibrated database that maps various possible pressure-motion coupling features, such as "slow pressure decrease in a stationary state" or "periodic pressure fluctuations in a walking state," to a specific adjustment weight coefficient. This adjustment weight coefficient reflects the probability that a certain pressure change feature indicates a wearing abnormality under a specific movement pattern. Once the coefficient extraction unit successfully matches and extracts the corresponding adjustment weight coefficient from the table based on the current pressure-motion coupling features, it completes the quantification of the current contextual risk.

[0077] Finally, the state deviation calculation unit performs a final correction and calculation of the deviation. It multiplies the statistical deviation obtained from the basic deviation calculation unit with the adjustment weight coefficient obtained from the coefficient extraction unit to generate the final state deviation index. This calculation process can be expressed as:

[0078] ,

[0079] in, It is the state deviation index of the final output. The adjusted weight coefficients are obtained from the weight mapping table. This is the previously calculated statistical deviation. Through this multiplicative correction, if the current situation is considered high-risk, a larger adjustment weighting coefficient will amplify the original statistical deviation; conversely, if the situation is judged to be caused by normal activity, a smaller coefficient will attenuate the deviation, ultimately generating a comprehensive indicator that can accurately reflect the actual wearing risk.

[0080] This method achieves a highly intelligent assessment of wearing status by dynamically weighting pressure deviations based on motion context. It can distinguish between pressure changes caused by real abnormal events such as loosening or detachment of the pump patch, and benign pressure fluctuations caused by normal user movement, such as running or bending over. Compared to traditional methods that rely solely on absolute pressure values ​​or simple deviations, this correction mechanism based on the coupling characteristics of pressure and motion ensures that the assessment results are no longer isolated numerical values, but rather risk judgments containing contextual information. This improves the accuracy and specificity of status monitoring, reduces the incidence of false alarms, and thus enhances the reliability of the entire early warning system and user acceptance.

[0081] Optionally, the early warning output module includes:

[0082] The environmental scenario acquisition unit is used to acquire environmental scenario parameters;

[0083] The early warning effectiveness analysis unit is used to determine the required early warning effectiveness level based on the state deviation index.

[0084] The intrusion analysis unit is used to determine the user's intrusion sensitivity level based on the environmental context parameters.

[0085] The signal output unit is used to select and generate a warning control signal from a preset combination of warning methods based on the warning effectiveness level and the user intrusion sensitivity level.

[0086] Specifically, the environmental scenario acquisition unit acquires raw environmental data, i.e., environmental scenario parameters, by calling built-in or connected sensors, such as light sensors and microphones. Subsequently, the warning effectiveness analysis unit receives a state deviation index from the state assessment module. This unit compares this index with a set of preset thresholds to map it to different warning effectiveness levels. For example, an extremely high state deviation index means the pump sticker may be about to detach, and would be judged as requiring the highest level of warning effectiveness to ensure the user immediately notices the risk. Conversely, a lower index corresponds to a lower warning effectiveness level. Simultaneously, the intrusion analysis unit processes the environmental scenario parameters provided by the environmental scenario acquisition unit. This unit analyzes these parameters to infer the user's current scenario; for example, by analyzing ambient light intensity and sound decibel data, it determines whether the user is in a quiet indoor environment, such as a meeting room or bedroom, or a noisy outdoor environment. Based on this scenario inference, the unit determines a user intrusion sensitivity level. In scenarios where the user may be resting or needs to remain quiet, the sensitivity level is set to high; while in noisy or active scenarios, it is set to low.

[0087] Finally, the signal output unit, acting as the decision-making center, integrates the results of the two analyses. Based on the determined warning effectiveness level and user intrusion sensitivity level, it performs a query and match within a pre-defined warning mode combination matrix. This matrix predefines the most suitable warning strategy under different combinations of effectiveness and sensitivity. For example, when both the warning effectiveness level and the user intrusion sensitivity level are high, the system might choose a silent but strong vibration warning; while when the effectiveness level is high and the sensitivity level is low, it might choose a strong warning that activates sound, vibration, and light simultaneously. Based on the optimal strategy found, the signal output unit generates and outputs a specific warning control signal, which includes instructions on what type of warning to execute and its intensity.

[0088] This method dynamically balances the necessity of early warning with the potential intrusion caused by the warning behavior. The device no longer simply issues an alarm, but transforms into an intelligent assistant capable of selecting the most appropriate notification method based on the user's specific situation. This effectively conveys risk information in emergencies while avoiding unnecessary disturbances in non-critical moments, thereby improving the user's wearing experience and trust in the warning system. This human-centered design solves the alarm fatigue problem commonly found in traditional early warning systems, ensuring that the warning function can exert its maximum effectiveness when truly needed.

[0089] Optionally, obtaining environmental context parameters includes:

[0090] Acquire ambient light intensity data and ambient sound decibel data;

[0091] The ambient light intensity data and the ambient sound decibel data are classified and mapped to determine the lighting and sound conditions.

[0092] The lighting and sound states are combined to generate environmental context parameters.

[0093] Specifically, the process of acquiring environmental context parameters begins with the acquisition of raw environmental signals. The device acquires real-time ambient light intensity data, typically in lux, through its integrated photosensitive element. Simultaneously, it uses a built-in microphone to pick up ambient sound and processes it to obtain quantified ambient sound decibel data. These two types of raw sensor data provide the basic input for subsequent context analysis. Subsequently, the system performs classification and mapping processing on the acquired ambient light intensity data and ambient sound decibel data. This processing is essentially a comparison process of a series of preset thresholds, transforming continuously changing physical quantities into discrete context state labels. For example, the system internally presets multiple light intensity thresholds. When the acquired light intensity data is below a certain extremely low threshold, it is mapped to the lighting state "dark"; when the data is between two intermediate thresholds, it is mapped to "dim"; and when it is above a certain high threshold, it is mapped to "bright". Similarly, the system performs similar processing on the ambient sound decibel data, mapping continuous decibel values ​​to different sound states, such as "quiet," "quiet," or "noisy," based on preset sound decibel thresholds. After determining the lighting and sound states separately, the system combines these two discrete states to generate the final environmental context parameters. This combination forms a two-dimensional scene description. For example, the combination of "dark" and "quiet" states can generate environmental context parameters representing a scenario where the user might be "resting at night"; while the combination of "bright" and "noisy" states might generate parameters representing "daytime outdoor activities." This generated composite parameter is the final output environmental context parameter, providing accurate scene input for subsequent intrusion analysis.

[0094] This method achieves the perception of the user's macroscopic environment through the quantitative analysis and combination of two key environmental factors: light and sound. It expands the device's perception capabilities from monitoring single physical activities to a comprehensive understanding of social and daily life scenarios. For example, the system can distinguish whether a user is reading in a quiet library or walking on a noisy street, even if their movement patterns may be similar. This scenario judgment based on multimodal environmental perception provides a solid foundation for the intelligent customization of early warning strategies, enabling the device to select the most appropriate warning method without affecting the user's normal social interactions or rest, thereby improving the practicality and user-friendliness of the early warning system.

[0095] Optionally, the execution feedback module includes:

[0096] The early warning analysis unit is used to analyze the early warning control signal to determine the early warning level, which includes a level one early warning and a level two early warning.

[0097] A Level 1 warning output unit is used to send a warning message to a contact person by calling a preset communication unit when the warning level is Level 1.

[0098] A secondary warning output unit is used to call a preset local warning unit to activate one or more of vibration, sound, or light warnings when the warning level is a secondary warning.

[0099] The user feedback unit is used to acquire and analyze the user's response to the warning after sending a warning message and activating one or more of vibration, sound or light alerts, and obtain user feedback data.

[0100] Specifically, the warning parsing unit first decodes the warning control signal to accurately identify the warning level information it contains. The warning levels are clearly divided into Level 1 and Level 2 warnings, corresponding to different risk levels and response strategies. When the warning level is parsed as Level 1, the Level 1 warning output unit is activated. This unit calls a preset communication unit, such as via Bluetooth or a cellular network, to send a structured warning message to a pre-set emergency contact for remote notification. Conversely, if the warning level is Level 2, the Level 2 warning output unit performs its operation. It calls the device's local alert unit and, according to the specific instructions of the warning control signal, activates one or more combinations of the device's own vibration motor, sound generator, or light-emitting element to directly issue a localized, sensor-accessible warning to the wearer. After any warning operation is executed, the user feedback unit begins to monitor and capture the user's response behavior. This includes recording whether the user cancels the alarm via a physical button or a companion application, and monitoring whether the pump sensor data returns to normal after the warning. The unit then analyzes these behaviors and transforms them into structured user feedback data, such as marked as "false alarm confirmed" or "issue resolved," for use in subsequent model optimization.

[0101] This method differentiates between Level 1 and Level 2 warnings, enabling the device to take different intervention measures based on the urgency of the risk. It can notify external contacts for monitoring in non-urgent but necessary situations, or directly and forcefully alert users in emergencies to ensure safety. More importantly, the introduction of a user feedback unit transforms users' natural responses into valuable data for the system to learn from, thus forming a complete intelligent closed loop from "perception, decision-making, and execution" to "feedback and optimization." This transforms the warning system from a static rule executor into a dynamic system that continuously improves and evolves through interaction with users, enhancing long-term accuracy and user experience.

[0102] Optionally, the model optimization module includes:

[0103] A tag generation unit is used to acquire the warning control signal and the user feedback data and associate them to form a warning feedback tag;

[0104] The sample construction unit is used to perform time alignment and slicing of the fused data stream, the state deviation index and the early warning feedback label to generate labeled historical data samples.

[0105] The coefficient update unit is used to update the baseline coefficients and the adjusted weight coefficients based on labeled historical data samples.

[0106] Specifically, the tag generation unit simultaneously acquires the warning control signal generated by the warning output module and the user feedback data captured by the execution feedback module. By correlating these two data points over time, the unit can analyze the actual effect of the warning. For example, after a level-two warning is issued, if user feedback data indicates that the user immediately cancels the alarm and marks it as a false alarm, the system will generate a "false alarm" warning feedback tag. Conversely, if the data shows that the stress value quickly returns to normal after the warning, a "valid warning" tag is generated. Next, the sample construction unit transforms these events with clear conclusions into samples that can be used for machine learning. This unit traces back to a period of time before the warning event occurred, aligns the fused data stream during that period, along with the state deviation index calculated by the system at that time, with the newly generated warning feedback tag and packages them together. Through this entire process, a complete labeled historical data sample is constructed. This sample fully records the entire process of "under what sensor data and risk assessment, what decisions the system made, and what actual results were ultimately obtained." Finally, the coefficient update unit uses one or more labeled historical data samples to optimize the core parameters of the system. This unit uses these samples as input and iteratively adjusts the baseline coefficients and adjustment weights by running an optimization algorithm, such as one based on gradient descent or reinforcement learning. The goal is to enable the model to output a state deviation index that more closely approximates the true result when faced with similar input data in the future. For example, for a sample labeled as a "false positive," the optimization algorithm adjusts the relevant adjustment weights to reduce them, so that even with similar pressure fluctuations in the same motion pattern, the state deviation index will not be too high the next time. Similarly, by analyzing all samples, the algorithm also fine-tunes the baseline coefficients, making the dynamic pressure baseline update strategy more closely match the user's actual wearing habits.

[0107] This method learns continuously from real user feedback, enabling the system to automatically correct its internal judgment logic. This makes its decision-making model increasingly tailored to individual users' specific activity patterns, physical characteristics, and environmental habits. This allows the device to continuously improve the accuracy of its warnings over time, reducing false alarms and missed alarms, thus transforming it from a general-purpose monitoring tool into a highly reliable, personalized intelligent partner. This adaptive optimization mechanism ensures the device's robustness and user satisfaction during long-term use.

[0108] Based on the same inventive concept, such as Figure 4 As shown, the present invention also provides a method for monitoring and early warning of the wearing status of the pump patch, the method comprising:

[0109] Acquire pressure sensor data and multi-axis accelerometer data from the pump patch, and perform time alignment and noise filtering on the data to generate a fused data stream;

[0110] Based on the fused data stream and the preset baseline coefficients, the dynamic pressure baseline is calculated;

[0111] Extract pressure-motion coupling features from the fused data stream;

[0112] Based on the pressure-motion coupling characteristics and the dynamic pressure baseline, the state deviation index is calculated.

[0113] Obtain environmental context parameters, and based on the state deviation index and the environmental context parameters, perform a balance analysis of early warning effectiveness and intrusion degree to generate an early warning control signal;

[0114] Execute an early warning output operation based on the early warning control signal, and obtain user feedback data after the early warning output operation is executed;

[0115] The baseline coefficients are optimized based on the user feedback data.

[0116] To verify the feasibility of this invention in practice, it was applied to the daily monitoring of a diabetic patient who wears insulin pump patches long-term. In daily life, the patient's pump patch is frequently squeezed or stretched due to movement and changes in body posture. Traditional pump patches only trigger alarms based on fixed pressure thresholds, leading to frequent false alarms during normal activities. Furthermore, during resting states such as sleep at night, the gradual loosening of the pump patch due to sweat or friction is difficult to detect in time, posing a safety hazard.

[0117] In this embodiment, the patient wears a pump patch integrated with the device of this invention. The device continuously collects pressure data between the pump patch and the skin, as well as the user's motion data, through a built-in pressure sensor and a multi-axis accelerometer. The collected raw multi-source data undergoes timestamp calibration, bandpass filtering, and weighted fusion via a data processing module to generate a high-quality fused data stream. This test lasted for four weeks, recording the patient's wear-and-wear data and warning response in different daily life scenarios.

[0118] In this embodiment, the baseline calculation module plays its primary role. During the initial period of patient use, the system analyzes data from the patient's multi-axis accelerometer to identify stable wearing periods, such as when the patient is sitting or lying down. Within these periods, the system extracts corresponding pressure sensor data to form a stable pressure dataset and calculates an exponentially weighted moving average based on a preset baseline coefficient (α=0.1) to generate a personalized dynamic pressure baseline for the patient. For example, on the first day, the system identifies the patient's one-hour afternoon period of sitting at work as a stable wearing period and maintains the dynamic pressure baseline at approximately 3.5 kPa during this time.

[0119] In subsequent monitoring, the feature extraction module and the state assessment module worked together to achieve contextualized analysis of the wearing status. For example, on a certain morning, the patient was jogging. The pattern analysis unit identified the current movement mode as "running," while the change feature extraction unit extracted pressure data showing a periodic fluctuation of approximately 1.5 Hz synchronized with the stride frequency. Although the instantaneous pressure value deviated from the dynamic pressure baseline at this time, the coefficient extraction unit obtained a small adjustment weight coefficient (e.g., W_adj=0.2) from the preset weight mapping table based on the pressure-motion coupling feature of "periodic pressure fluctuation in running mode." Therefore, the final state deviation index calculated by the state deviation calculation unit was very low, and the system determined this to be normal activity, without triggering an alarm, effectively avoiding a false alarm.

[0120] Conversely, on another night, the patient was asleep, and the pattern analysis unit identified the current motion pattern as "stationary." At this time, a corner of the pump patch curled slightly due to sweat, causing the pressure sensor data to slowly decrease from 3.5 kPa to 2.8 kPa within 30 minutes. For the pressure-motion coupling characteristic of "continuous pressure decrease in the stationary mode," the coefficient extraction unit obtained a relatively large adjustment weighting coefficient (e.g., This caused the final calculated state deviation index to rise, exceeding the warning threshold.

[0121] At this point, the early warning output module intervenes. The environmental scenario acquisition unit uses a light sensor and microphone to detect ambient light intensity below 10 lux and ambient sound below 30 decibels, thus determining the environmental scenario parameters as "nighttime rest." The early warning effectiveness analysis unit determines that a high-level early warning is needed based on a high state deviation index, while the intrusion analysis unit determines the user's intrusion sensitivity level to be high based on the "nighttime rest" scenario. Finally, the signal output unit balances the needs of high early warning effectiveness and high intrusion sensitivity, selecting a "strong vibration plus silence" strategy from the preset early warning method combinations to generate the corresponding early warning control signal.

[0122] The execution feedback module initiated a level-two warning based on the signal, activating the device's vibration alert. After being awakened by the vibration, the patient checked and re-secured the pump patch, subsequently confirming the warning as "valid" via the accompanying mobile application. The user feedback unit captured this response and generated user feedback data. Based on this feedback, the model optimization module constructed a labeled historical data sample from the fused data stream prior to the event, the state deviation index, and the "valid warning" feedback label. This sample was then used to reinforce and update the relevant adjustment weight coefficients in the weight mapping table, making the system's judgments on similar situations more accurate in the future.

[0123] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method can be applied to the embodiments of the present invention as long as it achieves the purpose of the present invention. The above descriptions are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the present invention.

[0124] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A device for monitoring and early warning of the wearing status of a pump patch, characterized in that, The device includes: The data processing module is used to acquire pressure sensor data and multi-axis accelerometer data of the pump patch, and to perform time alignment and noise filtering on the data to generate a fused data stream; The baseline calculation module is used to calculate the dynamic pressure baseline based on the fused data stream and preset baseline coefficients; The feature extraction module is used to extract pressure-motion coupling features from the fused data stream; The state assessment module is used to calculate the state deviation index based on the pressure-motion coupling characteristics and the dynamic pressure baseline. The early warning output module is used to acquire environmental context parameters, and based on the state deviation index and the environmental context parameters, to perform a balance analysis of early warning effectiveness and intrusion degree, and generate an early warning control signal. The execution feedback module is used to execute an early warning output operation based on the early warning control signal, and to obtain user feedback data after the early warning output operation is executed; The model optimization module is used to optimize the baseline coefficients based on the user feedback data.

2. The pump patch wearing status monitoring and early warning device according to claim 1, characterized in that, The data processing module further includes: The sampling synchronization unit is used to acquire pressure sensor data and multi-axis accelerometer data from the pump patch to form raw multi-source data, and to perform timestamp calibration on the raw multi-source data to obtain time-aligned data; The noise reduction and enhancement unit is used to perform bandpass filtering and outlier suppression on the time-aligned data to generate purified sensor data. The time-series fusion unit is used to perform multi-scale time window splicing based on the purification sensor data to obtain a multi-scale sequence. The weighted fusion unit is used to perform channel weighting and amplitude normalization on the multi-scale sequence to generate a fused data stream.

3. The pump patch wearing status monitoring and early warning device according to claim 2, characterized in that, The baseline calculation module includes: A time-period analysis unit is used to analyze multi-axis accelerometer data in the fused data stream to identify stable wearing periods; The pressure extraction unit is used to extract pressure sensor data corresponding to the stable wearing period to form a stable pressure dataset. The data calculation unit is used to perform moving average calculation on the stable pressure dataset based on a preset baseline coefficient to generate a dynamic pressure baseline.

4. The pump patch wearing status monitoring and early warning device according to claim 3, characterized in that, The feature extraction module includes: The pattern analysis unit is used to analyze the multi-axis accelerometer data in the fused data stream to identify the current motion pattern; A change feature extraction unit is used to extract the change features of pressure sensor data under the current motion mode to form pressure change features; The feature coupling unit is used to combine the current motion mode with the pressure change feature to generate a pressure-motion coupling feature.

5. The pump patch wearing status monitoring and early warning device according to claim 4, characterized in that, The status assessment module includes: The basic deviation calculation unit is used to obtain the current pressure value from the fused data stream and calculate the statistical deviation between the current pressure value and the dynamic pressure baseline; The coefficient extraction unit is used to obtain the adjustment weight coefficient from the preset weight mapping table based on the pressure motion coupling characteristics. The state deviation calculation unit is used to correct the statistical deviation using the adjusted weighting coefficient to generate a state deviation index.

6. The pump patch wearing status monitoring and early warning device according to claim 5, characterized in that, The early warning output module includes: The environmental scenario acquisition unit is used to acquire environmental scenario parameters; The early warning effectiveness analysis unit is used to determine the required early warning effectiveness level based on the state deviation index. The intrusion analysis unit is used to determine the user's intrusion sensitivity level based on the environmental context parameters. The signal output unit is used to select and generate a warning control signal from a preset combination of warning methods based on the warning effectiveness level and the user intrusion sensitivity level.

7. The pump patch wearing status monitoring and early warning device according to claim 6, characterized in that, The acquisition of environmental context parameters includes: Acquire ambient light intensity data and ambient sound decibel data; The ambient light intensity data and the ambient sound decibel data are classified and mapped to determine the lighting and sound conditions. The lighting and sound states are combined to generate environmental context parameters.

8. The pump patch wearing status monitoring and early warning device according to claim 6, characterized in that, The execution feedback module includes: The early warning analysis unit is used to analyze the early warning control signal to determine the early warning level, which includes a level one early warning and a level two early warning. A Level 1 warning output unit is used to send a warning message to a contact person by calling a preset communication unit when the warning level is Level 1. A secondary warning output unit is used to call a preset local warning unit to activate one or more of vibration, sound, or light warnings when the warning level is a secondary warning. The user feedback unit is used to acquire and analyze the user's response to the warning after sending a warning message and activating one or more of vibration, sound or light alerts, and obtain user feedback data.

9. The pump patch wearing status monitoring and early warning device according to claim 8, characterized in that, The model optimization module includes: A tag generation unit is used to acquire the warning control signal and the user feedback data and associate them to form a warning feedback tag; The sample construction unit is used to perform time alignment and slicing of the fused data stream, the state deviation index and the early warning feedback label to generate labeled historical data samples. The coefficient update unit is used to update the baseline coefficients and the adjusted weight coefficients based on labeled historical data samples.

10. A method for monitoring and early warning of the wearing status of a pump patch, characterized in that, The method includes: Acquire pressure sensor data and multi-axis accelerometer data from the pump patch, and perform time alignment and noise filtering on the data to generate a fused data stream; Based on the fused data stream and the preset baseline coefficients, the dynamic pressure baseline is calculated; Extract pressure-motion coupling features from the fused data stream; Based on the pressure-motion coupling characteristics and the dynamic pressure baseline, the state deviation index is calculated. Obtain environmental context parameters, and based on the state deviation index and the environmental context parameters, perform a balance analysis of early warning effectiveness and intrusion degree to generate an early warning control signal; Execute an early warning output operation based on the early warning control signal, and obtain user feedback data after the early warning output operation is executed; The baseline coefficients are optimized based on the user feedback data.

Images