A wearable electrochemical sensor online calibration method and system based on pollution state identification

Abstract

This invention relates to an online calibration method and system for wearable electrochemical sensors based on pollution state identification, belonging to the field of electrochemical detection and signal processing technology. By extracting the response distortion characteristics of the target analyte and auxiliary characterization components, a pollution state feature vector is constructed and input into a compensation and reconstruction model to output the equivalent response or correction parameters under a reference state. This allows for online correction of the current measurement results. When the correction parameters or reconstruction error exceed a preset safety threshold, automatic compensation is stopped, and physical recalibration or device maintenance prompts are triggered. This achieves high-precision online correction between two physical calibrations and extends the effective calibration cycle of the sensor.

Description

Technical Field

[0001] This invention relates to an online calibration method and system for wearable electrochemical sensors based on pollution state identification, belonging to the field of electrochemical detection and signal processing technology. Background Technology

[0002] Wearable electrochemical sensors are prone to protein adsorption, biocontamination, and degradation of interfacial active sites during continuous monitoring, which can lead to sensitivity decay, peak shift, baseline drift, and peak distortion in the voltammetric response of the target analyte, thereby reducing the accuracy of quantitative detection and shortening the effective calibration cycle.

[0003] Existing technologies for addressing sensor drift and distortion primarily include periodic physical calibration, mathematical compensation, and machine learning prediction. Periodic physical calibration typically requires regular manual intervention or recalibration, which is insufficient to meet the long-term, low-maintenance operation requirements of wearable continuous monitoring scenarios. Mathematical compensation methods often establish correction relationships based on preset error models or specific drift types, and their adaptability remains limited when facing complex response changes such as sensitivity attenuation, baseline shift, and peak distortion caused by contamination. Existing machine learning-based voltammetric sensor correction methods mostly use the original voltammetric response or preprocessed features as model input, outputting the target analyte concentration or correction result, which can improve some drift and multi-component interference errors. However, most of these methods focus on response-concentration empirical mapping, rather than explicitly modeling the electrode interface contamination state as an independent object. Their characterization of peak position drift, baseline changes, peak distortion, and multi-component differential responses caused by contamination is still insufficient, posing a risk of confusing true concentration changes with contamination response attenuation. Furthermore, electrochemical-physical consistency constraints, compensation failure identification, reference state freezing, and maintenance triggering mechanisms have not yet formed a conventional online closed loop for wearable continuous monitoring. To address the shortcomings of the aforementioned technical approaches, this invention is proposed. By constructing a pollution state feature vector, outputting equivalent response or correction parameters under a reference state, and combining a threshold determination and maintenance triggering mechanism, online calibration and closed-loop maintenance for long-term continuous monitoring scenarios can be achieved. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an online calibration method and system for wearable electrochemical sensors based on pollution state identification. By extracting the response distortion characteristics of the target analyte and auxiliary characterization components, a pollution state feature vector is constructed and input into a compensation and reconstruction model to output the equivalent response or correction parameters under the reference state. This allows for online correction of the current measurement results. When the correction parameters or reconstruction error exceed a preset safety threshold, automatic compensation is stopped and physical recalibration or device maintenance prompts are triggered. This achieves high-precision online correction between two physical calibrations and extends the effective calibration cycle of the sensor.

[0005] The technical solution of the present invention is as follows: An online calibration method for wearable electrochemical sensors based on pollution state identification, comprising the following steps: (1) Under the reference state, control the working electrode, reference electrode and counter electrode to perform a preset potential scanning program, collect response current data and obtain the original volt-ampere response; (2) Filtering, baseline correction, peak identification and pollution state feature vector extraction are performed on the original volt-ampere response; (3) Based on the pollution state feature vector and historical calibration parameters, use the compensation reconstruction model to generate the equivalent response or correction parameters under the reference state; (4) Calculate the estimated concentration of the target analyte based on the compensated equivalent response or correction parameters; (5) When the calibration parameters or reconstruction error exceed the preset safety threshold, output physical recalibration, electrode cleaning, consumable replacement or terminal alarm prompt.

[0006] According to a preferred embodiment of the present invention, in step (1), the original voltammetric response refers to the current-potential response data obtained under a preset potential scanning program. In DPV (differential pulse voltammetry), the original voltammetric response is expressed as a discrete data sequence with the pulse potential as the horizontal axis and the differential current (ΔI) as the vertical axis. It includes the redox peak information (peak potential, peak current, peak area) of the target analyte as well as the baseline drift, peak position shift and peak shape distortion information caused by pollution.

[0007] According to a preferred embodiment of the present invention, in step (1), the reference state includes an initial reference state and an iterative reference state, which respectively correspond to the standard response state formed after the wearable electrochemical sensor is first activated and the initial physical calibration is completed, and after the most recent physical maintenance, cleaning, patch replacement or recalibration. The preset measurement conditions include one or more of the following: scanning parameters, temperature range, ion strength conditions and sampling interval. Under reference conditions, the standard response curve I of the target analyte is obtained through standard solution calibration, standard addition calibration, or preset concentration testing. calib After filtering, baseline correction, and peak identification, the reference state parameter set P is extracted. ref And store it in the reference state library, reference state parameter set P ref Including: reference peak potentials E of each target analyte and auxiliary characterization components. ref Reference peak current I of each component ref_peak Reference peak area A of each component ref_peak The reference baseline slope, reference baseline intercept, and reference half-width parameters for each component in the non-redox region are used as the basis for subsequent online compensation and reference state evolution analysis.

[0008] During subsequent continuous use, the compensation and reconstruction model is based on the measured response I under the current pollution state. obs By combining the pollution state feature vector and historical calibration parameter information, the equivalent response I corresponding to the reference state is reconstructed. ref To make it as close as possible to I calib It can also correct for slow-varying biases caused by long-term aging based on changes in reference state parameters at different points in time.

[0009] When physical recalibration, cleaning and maintenance, sensor patch replacement or electrode assembly replacement is triggered, the reference state parameter set is updated with the latest calibration results and the historical parameter cache is reset, truncated or updated via sliding window.

[0010] Standard solution calibration, standard addition calibration, and preset concentration testing are all conventional methods in the field of electrochemistry. Standard solution calibration involves performing a DPV scan in a standard solution of known concentration to directly obtain the standard voltammetric response of the target analyte. Standard addition calibration involves gradually adding a known amount of standard solution to the sample and obtaining the standard response through extrapolation. Preset concentration testing involves directly measuring and establishing a concentration-response relationship using a standard of known concentration. This invention does not limit the method of obtaining the above standard response curves, but rather uses the obtained I... calib This will serve as a reference benchmark for subsequent compensation and reconstruction.

[0011] According to a preferred embodiment of the present invention, in step (2), the differential pulse voltammetry (DPV) method is used to acquire scanning voltammetric data, and the pollution state feature vector X is extracted from the original voltammetric response. fouling The pollution state feature vector includes peak position shift feature, baseline drift feature, and response attenuation ratio feature; Peak shift characteristics: Extracting the peak shift of the oxidation peak of the target analyte to characterize the pollution state related to changes in interfacial charge transfer kinetics. The specific extraction steps are as follows: Perform peak detection on the preprocessed current DPV differential current sequence (using conventional peak detection methods such as local extremum search, Gaussian fitting, or derivative method) to locate the peak potential E corresponding to the oxidation peak of the target analyte. p_obs Compare it with the reference state parameter set P ref Reference peak potential E in ref Subtracting them, we get the peak position shift ΔEp = E p_obs - E ref The peak detection and parameter extraction described above can be achieved by combining peak candidate region identification, fitting, or derivative analysis. Baseline drift characteristics: Extracting the baseline slope, baseline intercept, or background response change in the non-redox region to characterize the pollution state related to changes in interfacial capacitance and background response. The specific extraction steps are as follows: Select the inactive potential region far from the redox peak in the DPV curve, perform linear or polynomial fitting on the differential current data within this region, and extract the baseline slope k. base and baseline intercept b base Compare them with the reference state parameter set P respectively ref The change in baseline slope Δ is obtained by subtracting the reference baseline slope from the reference baseline intercept. k_base and baseline intercept change Δ b_base ; Response decay ratio characteristics: The response decay ratios of the target analyte channel and the auxiliary characterization component channel relative to the reference state are extracted, and their ratio characteristics are constructed to characterize the differentiated responses of different components to contamination states in a multi-component system. The specific extraction steps are as follows: Extract the peak current I of the target analyte channel in the current scan. DA_obs and the peak current I of the auxiliary characterization component channel AA_obs Compare these values ​​with the reference peak current I in the reference state parameter set Pref. ref_peak_DA and I ref_peak_AA Divide by the ratio to obtain the response retention ratio r of each channel. DA = I DA_obs / I ref_peak_DA and r AA = I AA_obs / I ref_peak_AA Further calculation of the relative response decay ratio R between the two. DA / AA = r DA / r AA The subscripts DA and AA represent two different components, which can be dopamine and ascorbic acid, respectively, in practice.

[0012] According to a preferred embodiment of the present invention, in step (3), the pollution state feature vector X is... fouling The historical calibration parameter set is used as input to the pre-trained compensation reconstruction model; The compensation and reconstruction model employs residual networks, temporal convolutional networks, lightweight feedforward networks, or combinations thereof, for deployment and operation on local processors or edge devices. Its output includes at least one of the following: the equivalent response I in the reference state. ref Response correction factor α, baseline offset correction factor β, and peak distortion correction factor γ; Taking the combined architecture of residual network and one-dimensional temporal convolutional network (TCN) as an example, the workflow of the compensation and reconstruction model is as follows: (1) The input layer receives the contaminated state feature vector X. fouling and historical calibration parameter set P hist(2) The feature extraction layer consists of several residual blocks and / or TCN layers, used to capture the temporal evolution features of the contaminated state and the nonlinear mapping relationship between parameters; (3) The output layer outputs the equivalent response curve I under the reference state according to the task type. ref Alternatively, output a set of correction parameters (including response correction coefficient α, baseline offset correction coefficient β, peak distortion correction coefficient γ, etc.). In the initial activation phase, the reactivation phase after sensor component replacement, or when historical calibration parameters are insufficient to support the stable operation of the compensation and reconstruction model, the system enters cold start mode. Measurement results are directly output based on reference state parameters, or the default compensation parameters are used for preliminary correction. The default compensation parameters are derived from factory calibration values, the statistical average of devices in the same batch, the most recent valid physical calibration parameters, or a combination thereof. When continuous scanning meets the preset validity conditions and the integrity of historical parameters meets the preset requirements, the system switches to the compensation and reconstruction model working mode; otherwise, the cold start mode is maintained.

[0013] According to a further preferred embodiment of the present invention, the historical calibration parameter set refers to the set of parameter records accumulated during each physical calibration and maintenance cycle, including but not limited to: gain parameters (sensitivity coefficients of each channel), baseline parameters (baseline slope, baseline intercept), peak shape parameters (half-width at half maximum, symmetry index), timestamps of each calibration, environmental conditions (temperature, ion strength, etc.), and maintenance records obtained from the most recent valid physical calibration. hist A set of reference state parameters P at one or more time points ref It consists of its corresponding timestamp, environmental conditions, and maintenance records, i.e., P hist For P ref A collection of time series data.

[0014] According to a preferred embodiment of the present invention, in step (3), the training data of the compensation reconstruction model comes from at least one of the following: in vitro accelerated pollution or aging experimental data under standard concentration gradient conditions, multi-cycle control data collected during in vivo continuous monitoring by the sensor, and periodic physical calibration data under different pollution levels. The training process involves first establishing a reference state under fixed component concentration conditions and obtaining the standard response I. calib Subsequently, the wearable electrochemical sensor was subjected to several rounds of contamination or aging treatment, and the measured response curves under the current contamination state were acquired under the same measurement conditions. obs Then, the pollution state feature vector X is extracted from the current response. fouling and historical calibration parameter set P hist This allows for the construction of training samples, where the input to the training samples includes at least X. fouling and P hist and selectively include Iobs The supervision label is the reference state standard response I. calib , or the response correction parameters, baseline correction parameters and peak shape correction parameters obtained by fitting them; During sample pairing, the current response of the same wearable electrochemical sensor under contaminated conditions is paired with its reference state response under the same or equivalent target concentration conditions after its most recent effective physical calibration. The equivalent target concentration is confirmed through the standard addition method, reference response inversion, or a preset calibration relationship. The preset tolerance range is predetermined based on the statistical fluctuation range during the reference state establishment phase, the instrument repeatability interval, or the upper limit of the standard solution calibration error. The training set, validation set, and test set are divided according to the wearable electrochemical sensor number, the acquisition time period, or a combination of both to avoid the same wearable electrochemical sensor number appearing simultaneously in the training set and the test set. One of the goals of model training is to make the output equivalent response I... ref Approximating the reference state standard response I calib ; To improve the accuracy and physical consistency of the compensation results, the model training uses a joint loss function consisting of a data reconstruction error term and a physical consistency constraint term: L total = L recon + λL phys Among them, L recon The reconstruction error used to characterize the difference between the model output and the standard response of the reference state can be the mean square error, the mean absolute error, or a combination thereof. It reflects the data reconstruction deviation between the compensation result and the target response of the reference state, and is mainly used to measure the numerical accuracy of the compensation result; L phys The peak position, peak width, baseline, or peak shape parameters used to constrain the compensated response to be within the preset reasonable electrochemical range reflect whether the compensation result meets the preset electrochemical response law and parameter reasonable boundary, and are used to measure the physical rationality and interpretability of the compensation result. Joint loss function L total From the reconstruction error term L recon Physical consistency constraint term λL phys Composition, L recon The smaller the value, the closer the compensation result is to the target reference response at the data level, reflecting the accuracy of the compensation result; L phys The smaller the value, the more the compensation result conforms to the electrochemical response law, reflecting the physical consistency of the compensation result. During model training and validation, L is jointly monitored on the validation set. recon With L phys The changing trend, combined with L total The overall convergence of the model is used to comprehensively judge whether the model output simultaneously possesses good compensation accuracy and physical rationality. When L reconConverging to the preset accuracy range and L phys When all components are stably within a reasonable range, the model is considered to be in a usable state. The peak position constraint is the peak position residual between the target analyte and the auxiliary characterization component; the peak width constraint is the corresponding half-peak width or the change in peak width; the baseline constraint is the baseline slope and baseline intercept of the background interval; the reasonable range is derived from the reference state statistical interval, physical calibration interval or empirical boundary; the parameter λ is determined comprehensively based on the reconstruction error on the validation set and the parameter out-of-bounds ratio. In a preferred training implementation, the compensation reconstruction model uses a one-dimensional temporal convolutional network as the main model, which may contain multiple one-dimensional convolutional layers with kernel sizes of 7, 5 and 3 respectively. After each convolutional layer, a nonlinear activation function and a regularization module are sequentially connected, and a curve reconstruction output head or a parametric regression output head is set according to the output type. The model training is preferably completed offline, and the trained model is deployed on a local processor or edge device to perform online inference.

[0015] According to a preferred embodiment of the present invention, in step (4), based on the equivalent response I ref Alternatively, correction parameters α, β, and γ can be used to correct the current measured response and further calculate the estimated concentration of the target analyte. The specific implementation process involves two scenarios: Scenario 1, where the output of the compensation and reconstruction model is the equivalent response I... ref At that time, directly from the equivalent response I ref The peak current or peak area of ​​the target analyte is extracted, and the estimated concentration is calculated based on the concentration-response calibration relationship established under the reference state. In scenario two, when the model output is the correction parameters α, β, and γ, the current response curve is adjusted for gain (i.e., the peak current is scaled by α) based on the response correction coefficient α, the current response curve is adjusted for baseline offset (i.e., the baseline offset component represented by β is subtracted from the current response curve) based on the baseline offset correction coefficient β, and the peak shape distortion of the current response is corrected based on the peak shape distortion correction coefficient γ (γ is used to characterize the degree of distortion of the current peak shape relative to the reference peak shape; the post-processing module can use this parameter to correct the peak width, symmetry, peak area, or related peak shape characterization quantities). After correction, quantitative parameters are extracted from the corrected response and the estimated concentration is calculated.

[0016] According to a preferred embodiment of the present invention, in step (5), the parameter overflow or reconstruction error during the compensation process is continuously monitored. When any one of the indicators, such as the response correction coefficient, peak distortion correction coefficient, baseline offset, or reconstruction error, exceeds the preset safety threshold, it is determined that the current automatic compensation has exceeded the effective working range and the automatic correction is stopped. At the same time, physical recalibration, electrode cleaning, consumable replacement, or terminal alarm prompts are output, thereby forming a closed loop of equipment operation of online correction - threshold monitoring - maintenance triggering. The specific process is shown in Figure 2. The safety threshold is determined based on the results of the in vitro accelerated contamination experiment, the statistical results of periodic physical calibration, the distribution of historical operating data, or a combination thereof. A fixed threshold or an adaptive threshold may be used to maintain online calibration when compensation is effective and to trigger maintenance operations in a timely manner when the compensation capability is close to the upper limit.

[0017] An online calibration system for wearable electrochemical sensors based on pollution state estimation and response compensation reconstruction includes: The electrochemical signal acquisition unit is used to control the working electrode, reference electrode and counter electrode to perform a preset potential scanning program under reference conditions, acquire response current data and obtain the raw volt-ampere response; The preprocessing and feature extraction unit is used to filter, correct baselines, identify peaks, and extract pollution state feature vectors from the raw volt-ampere response. The compensation and reconstruction unit generates equivalent response or correction parameters under the reference state based on the pollution state feature vector and historical calibration parameters using the compensation and reconstruction model. The concentration estimation unit calculates the estimated concentration of the target analyte based on the compensated equivalent response or correction parameters. The threshold determination and maintenance triggering unit outputs physical recalibration, electrode cleaning, consumable replacement, or terminal alarm prompts when the calibration parameters or reconstruction error exceed the preset safety threshold.

[0018] The beneficial effects of this invention are as follows: 1. This invention does not rely directly on the original response waveform for black-box regression, but first extracts the pollution state characteristic parameters with physical meaning, which improves the stability and interpretability of the compensation process.

[0019] 2. This invention not only outputs the concentration of the target analyte, but also the equivalent response or correction parameters under the reference state, transforming the scheme from simple prediction to online calibration and compensation.

[0020] 3. This invention combines pollution state estimation, response compensation reconstruction, and threshold triggering mechanism to form an online maintenance closed loop, which helps to maintain the stability and availability of detection results between two physical calibrations.

[0021] 4. By setting a compensation failure threshold, this invention can promptly trigger recalibration or hardware maintenance when automatic compensation exceeds the effective working range, thereby improving the long-term reliability and security of the system.

[0022] 5. The model of this invention adopts a lightweight structure and is deployed on the edge or device side, making it suitable for wearable continuous monitoring scenarios.

[0023] 6. By simultaneously introducing the response characteristics of the target analyte, the response characteristics of the auxiliary characterizing components, and the relative response decay ratio of the two, and combining them with the physical consistency constraint to train the compensation and reconstruction model, this invention can, to a certain extent, distinguish between the actual concentration change and the response distortion caused by interface pollution, thereby reducing the risk of misjudging the actual concentration change as pollution decay or misjudging the pollution decay as concentration change. Attached Figure Description

[0024] Figure 1 This is a system architecture diagram of the present invention; Figure 2 This is a flowchart of the closed-loop maintenance process of the present invention. Detailed Implementation

[0025] The present invention will be further described below with reference to the embodiments and accompanying drawings, but is not limited thereto.

[0026] Example 1: This embodiment provides an online calibration method for a wearable electrochemical sensor based on pollution state identification. The following description uses dopamine (DA) as the target analyte and ascorbic acid (AA) as an auxiliary characterization component. This embodiment is for illustrative purposes only and does not constitute a limitation on the scope of protection. The steps are as follows: 1.1 Detection objects and application scenarios; Wearable electrochemical sensors are used to continuously monitor changes in dopamine concentration in the test system. Since DA and AA may coexist in physiological systems, and the sensor is susceptible to protein adsorption, interface contamination and active site decay during long-term use, the target peak response may exhibit reduced sensitivity, peak position drift, baseline drift and peak shape distortion. Therefore, the AA response is introduced as an auxiliary characterization information to enhance the ability to identify contamination status.

[0027] Since the effects of interfacial contamination or adsorption layer formation on different molecular sizes, charge distributions and mass transfer behaviors vary, the response decay of DA and AA under the same contamination state usually exhibits different degrees of asymmetric changes. The characteristic parameters constructed based on the relative decay relationship between the two can help improve the accuracy of identifying the interfacial contamination state and the effective compensation range.

[0028] 1.2 Establishing a reference state; After the sensor is first activated or the electrode patch is replaced, a physical calibration is performed to establish a reference state. The preset scanning conditions are preferably differential pulse voltammetry (DPV) scanning conditions, which include at least the parameters such as scan start potential, end potential, pulse amplitude, pulse width, step potential and sampling interval. The scanning window covers the characteristic oxidation response range of DA and AA, and includes at least a background range without obvious redox peaks.

[0029] The working electrode can be a carbon-based modified electrode, such as a screen-printed electrode modified with carbon nanomaterials or conductive polymers. Under one example condition, the DPV scanning parameters can be set as follows: starting potential -0.2 V, ending potential 0.6 V, pulse amplitude 50 mV, pulse width 0.05 s, and potential step 4 mV. The above parameters are only examples, and those skilled in the art can make equivalent adjustments according to the electrode material, the range of target analytes, and the application scenario.

[0030] When establishing the reference state, record the dopamine reference peak position E. DA_ref Ascorbic acid reference peak position E AA_ref Dopamine reference peak current I DA_ref Ascorbic acid reference peak current I AA_ref Background interval reference baseline slope k base_ref Background interval reference baseline intercept b base_ref And the reference full width at half maximum (FWHM) parameters of the two components, the above parameters and the corresponding reference response curves I calib Stored in the local processor or associated memory.

[0031] 1.3 Online scanning and feature construction; During continuous monitoring, DPV scans are performed at preset time intervals to obtain the measured response curve I at the current moment. obs After each scan, the response curve is smoothed, abnormal noise is removed, preliminary background is subtracted, peak candidate regions are identified, peak position and peak current are estimated, background intervals are truncated, and peak width or peak area is extracted.

[0032] Subsequently, a DA / AA two-component pollution state feature vector X_fouling(DA / AA) is constructed, which includes: dopamine peak offset ΔE. DA Ascorbic acid peak position shift ΔE AA Baseline slope change Δk base Baseline intercept change Δb base Dopamine response retention ratio r DA Ascorbic acid response ratio r AA And the relative response decay ratio R between the two DA / AA If necessary, the changes in half-width, peak area, peak current change rate in two adjacent scans, background noise mean square value, or temperature compensation parameters of dopamine and ascorbic acid can also be added.

[0033] 1.4 Model Input and Output; The pollution state feature vector X fouling (DA / AA) and historical calibration parameter set P histInput the compensation and reconstruction model; in some training or enhancement implementations, the current measured response curve I can also be selectively introduced. obs As an auxiliary input, the model can be deployed on a local processor or edge device, preferably using a lightweight residual network, a temporal convolutional network, or a combination of both.

[0034] It should be noted that the core input in this embodiment is the pollution state feature vector X. fouling and historical calibration parameter set P hist Instead of directly relying on the original measured response curve for end-to-end regression, during the model inference phase, the input is X... fouling and P hist The primary focus is on model training; during the model training phase, I can be selectively introduced. obs This serves as an auxiliary input to enhance feature learning capabilities. This design distinguishes the scheme in this embodiment from existing methods that rely solely on black-box regression of the original waveform, improving the interpretability and physical consistency of the compensation process.

[0035] In one specific implementation, the compensation reconstruction model can employ a one-dimensional temporal convolutional network or a lightweight residual network structure. Taking a one-dimensional temporal convolutional network as an example, the model can include multiple convolutional layers with kernel sizes of 7, 5, and 3, respectively. A nonlinear activation function and a regularization mechanism are then applied after the convolutional layers to improve the ability to represent temporal response distortions. During training, the weight λ of the physical consistency constraint term in the joint loss function can be set as an adjustable parameter within a preset range to balance the data reconstruction accuracy with the constraints of the reasonableness of the electrochemical response.

[0036] The model output includes at least one of the following: the equivalent response curve under the reference state I ref Dopamine response correction coefficient α DA Ascorbic acid response correction coefficient α AA Baseline offset correction factor β and peak distortion correction factor γ. When the output is a set of parameters, the post-processing module can perform gain correction, baseline shift correction, and peak distortion correction on the current response.

[0037] 1.5 Dopamine concentration estimation and threshold triggering; In obtaining the equivalent reference response I ref After obtaining the corrected DA response parameters, extract the peak current, peak area, or other quantitative parameters corresponding to DA, and calculate the estimated concentration of DA based on the concentration-response relationship established under the reference state.

[0038] After each round of compensation, at least one or more of the following metrics should be monitored: DA gain correction coefficient α DA AA gain correction factor α AA Relative response decay ratio R DA / AABaseline drift change, current reconstruction error e recon One or more of the following: peak position residual and model output confidence. The model output confidence may be composed of reconstruction residual, fluctuation of continuous inference results, parameter regression consistency score, multi-model output variance or a combination thereof. When any indicator exceeds the preset safety threshold, it is determined that the current pollution state has approached or exceeded the effective working range of automatic compensation, and the output of the automatically corrected DA concentration value is stopped. At the same time, a prompt for physical recalibration, cleaning of electrode interface or replacement of sensor patch is output.

[0039] During the initial startup phase, the restart phase after sensor component replacement, or when historical calibration parameters are insufficient to support stable model operation, the system enters a cold start mode. In this mode, measurement results are directly output based on reference state parameters, or preliminary corrections are made using default compensation parameters. These default compensation parameters can be derived from factory calibration values, the statistical average of devices in the same batch, the most recent valid physical calibration parameters, or a combination thereof. Once the preset number of valid scans is reached and the integrity of historical parameters meets preset conditions, the system switches to the compensation and reconstruction model working mode; otherwise, it continues to maintain the reference state direct output mode or the default compensation mode.

[0040] During the automatic compensation phase, respectively targeting e recon α DA α AA Independent thresholds are set for peak position residual, baseline correction parameters, and peak shape correction parameters. When the reconstruction error of several consecutive scans exceeds the threshold, or when at least two of the gain correction parameters, baseline correction parameters, and peak shape correction parameters exceed the allowable range, automatic correction stops, and the current historical parameter cache is frozen. Abnormal phase data is no longer written to the reference state library. During the freeze period, only the most recent valid reference state can be read, and the reference parameters cannot be updated in reverse with the compensation result. The reference state parameters are preferably updated only after physical recalibration, cleaning and maintenance, sensor patch replacement, or electrode assembly replacement is completed. During the automatic compensation phase, the reconstruction results can only be temporarily stored as candidate states and written to the reference state library after subsequent physical recalibration confirmation.

[0041] To verify the effectiveness of this embodiment in pollution identification and online compensation, a minimal verification scheme was designed without pre-defining the actual performance values. The verification scheme includes at least the following three sets of comparisons: The first group involved fixing the target concentration and changing the pollution level to observe the trends of peak shift, baseline drift, and relative response decay ratio with the pollution level. The second group involved fixing the pollution level and changing the target concentration to verify the model's ability to distinguish between changes in actual concentration and changes in pollution. The third group compares the three compensation methods: uncompensated, linearly compensated, and the compensation method of this invention, to illustrate the changes in target analyte estimation error, reconstruction residual, and threshold triggering behavior before and after compensation.

[0042] In one exemplary verification design, an in vitro accelerated contamination experiment is used to construct 3 to 5 contamination states, and Ig is collected under each contamination state. obs Extract X fouling 1. Perform compensation inference and record I ref Or the trend of change of correction parameters; and based on I obtained from the most recent physical calibration. calib As a reference, the above verification scheme can be used to illustrate the technical effect evaluation path of the present invention by comparing the peak position residual, baseline drift, target analyte estimation error and maintenance triggering time before and after compensation, without constituting a limitation on the scope of protection.

[0043] Example 2: This embodiment provides an online calibration system for wearable electrochemical sensors based on pollution state estimation and response compensation reconstruction, such as... Figure 1 As shown, it includes: The electrochemical signal acquisition unit is used to control the working electrode, reference electrode and counter electrode to perform a preset potential scanning program under reference conditions, acquire response current data and obtain the raw volt-ampere response; The preprocessing and feature extraction unit is used to filter, correct baselines, identify peaks, and extract pollution state feature vectors from the raw volt-ampere response. The compensation and reconstruction unit generates equivalent response or correction parameters under the reference state based on the pollution state feature vector and historical calibration parameters using the compensation and reconstruction model. The concentration estimation unit calculates the estimated concentration of the target analyte based on the compensated equivalent response or correction parameters. The threshold determination and maintenance triggering unit outputs physical recalibration, electrode cleaning, consumable replacement, or terminal alarm prompts when the calibration parameters or reconstruction error exceed the preset safety threshold.

Claims

1. An online calibration method for wearable electrochemical sensors based on pollution state identification, characterized in that, The steps are as follows: (1) Under the reference state, control the working electrode, reference electrode and counter electrode to perform a preset potential scanning program, collect response current data and obtain the original volt-ampere response; (2) Filtering, baseline correction, peak identification and pollution state feature vector extraction are performed on the original volt-ampere response; (3) Based on the pollution state feature vector and historical calibration parameters, use the compensation reconstruction model to generate the equivalent response or correction parameters under the reference state; (4) Calculate the estimated concentration of the target analyte based on the compensated equivalent response or correction parameters; (5) When the calibration parameters or reconstruction error exceed the preset safety threshold, output physical recalibration, electrode cleaning, consumable replacement or terminal alarm prompt.

2. The online calibration method for wearable electrochemical sensors based on pollution state identification as described in claim 1, characterized in that, In step (1), the raw volt-ampere response refers to the current-potential response data obtained under the preset potential scanning program. In DPV, the raw volt-ampere response is a discrete data sequence with pulse potential as the horizontal axis and differential current as the vertical axis, which includes the redox peak information of the target analyte as well as the baseline drift, peak position shift and peak shape distortion information caused by pollution.

3. The online calibration method for wearable electrochemical sensors based on pollution state identification as described in claim 2, characterized in that, In step (1), the reference state includes the initial reference state and the iterative reference state, which correspond to the standard response state formed after the wearable electrochemical sensor is first activated and the initial physical calibration is completed, and after the most recent physical maintenance, cleaning, patch replacement or recalibration. The preset measurement conditions include one or more of the following: scanning parameters, temperature range, ion strength conditions and sampling interval. Under reference conditions, the standard response curve I of the target analyte is obtained through standard solution calibration, standard addition calibration, or preset concentration testing. calib After filtering, baseline correction, and peak identification, the reference state parameter set P is extracted. ref And store it in the reference state library, reference state parameter set P ref Including: reference peak potentials E of each target analyte and auxiliary characterization components. ref Reference peak current I of each component ref_peak Reference peak area A of each component ref_peak Reference baseline slope and reference baseline intercept in the non-redox region, and reference half-width parameters for each component; When physical recalibration, cleaning and maintenance, sensor patch replacement or electrode assembly replacement is triggered, the reference state parameter set is updated with the latest calibration results and the historical parameter cache is reset, truncated or updated via sliding window.

4. The online calibration method for wearable electrochemical sensors based on pollution state identification as described in claim 3, characterized in that, In step (2), the differential pulse voltammetry (DPV) method is used to acquire scanning voltammetric data, and the pollution state feature vector X is extracted from the original voltammetric response. fouling The pollution state feature vector includes peak position shift feature, baseline drift feature, and response attenuation ratio feature; Peak shift characteristics: The peak shift of the oxidation peak of the target analyte is extracted to characterize the contamination state related to changes in interfacial charge transfer kinetics. The specific extraction steps are as follows: Peak detection is performed on the preprocessed current DPV differential current sequence to locate the peak potential E corresponding to the oxidation peak of the target analyte. p_obs Compare it with the reference state parameter set P ref Reference peak potential E in ref Subtracting them, we get the peak position shift ΔEp = E p_obs - E ref ; Baseline drift characteristics: Extracting the baseline slope, baseline intercept, or background response change in the non-redox region to characterize the pollution state related to changes in interfacial capacitance and background response. The specific extraction steps are as follows: Select the inactive potential region in the DPV curve far from the redox peak, perform linear or polynomial fitting on the differential current data within this region, and extract the baseline slope k. base and baseline intercept b base Compare them with the reference state parameter set P respectively ref The change in baseline slope Δ is obtained by subtracting the reference baseline slope from the reference baseline intercept. k_base and baseline intercept change Δ b_base ; Response decay ratio characteristics: The response decay ratios of the target analyte channel and the auxiliary characterization component channel relative to the reference state are extracted, and their ratio characteristics are constructed to characterize the differentiated responses of different components to contamination states in a multi-component system. The specific extraction steps are as follows: Extract the peak current I of the target analyte channel in the current scan. DA_obs and the peak current I of the auxiliary characterization component channel AA_obs Compare these values ​​with the reference peak current I in the reference state parameter set Pref. ref_peak_DA and I ref_peak_AA Divide by the ratio to obtain the response retention ratio r of each channel. DA = I DA_obs / I ref_peak_DA and r AA = I AA_obs / I ref_peak_AA Further calculation of the relative response decay ratio R between the two. DA / AA = r DA / r AA The subscripts DA and AA represent two different components.

5. The online calibration method for wearable electrochemical sensors based on pollution state identification as described in claim 4, characterized in that, In step (3), the pollution state feature vector X is... fouling The historical calibration parameter set is used as input to the pre-trained compensation reconstruction model; The compensation and reconstruction model employs residual networks, temporal convolutional networks, lightweight feedforward networks, or combinations thereof, and its output includes at least one of the following: the equivalent response I under the reference state. ref Response correction factor α, baseline offset correction factor β, and peak distortion correction factor γ; In the initial activation phase, the reactivation phase after sensor component replacement, or when historical calibration parameters are insufficient to support the stable operation of the compensation and reconstruction model, the system enters cold start mode. Measurement results are directly output based on reference state parameters, or the default compensation parameters are used for preliminary correction. The default compensation parameters are derived from factory calibration values, the statistical average of devices in the same batch, the most recent valid physical calibration parameters, or a combination thereof. When continuous scanning meets the preset validity conditions and the integrity of historical parameters meets the preset requirements, the system switches to the compensation and reconstruction model working mode; otherwise, the cold start mode is maintained.

6. The online calibration method for wearable electrochemical sensors based on pollution state identification as described in claim 5, characterized in that, The historical calibration parameter set refers to the collection of parameter records accumulated throughout the various physical calibration and maintenance cycles, including but not limited to: gain parameters, baseline parameters, peak shape parameters obtained from the most recent valid physical calibration, timestamps of each calibration, environmental conditions, and maintenance records. The historical calibration parameter set P hist A set of reference state parameters P at one or more time points ref It consists of its corresponding timestamp, environmental conditions, and maintenance records, i.e., P hist For P ref A collection of time series data.

7. The online calibration method for wearable electrochemical sensors based on pollution state identification as described in claim 6, characterized in that, In step (3), the training data for the compensation reconstruction model comes from at least one of the following: in vitro accelerated pollution or aging experimental data under standard concentration gradient conditions, multi-cycle control data collected during continuous in vivo monitoring by the sensor, and periodic physical calibration data under different pollution levels. The training process involves first establishing a reference state under fixed component concentration conditions and obtaining the standard response I. calib Subsequently, the wearable electrochemical sensor was subjected to several rounds of contamination or aging treatment, and the measured response curves under the current contamination state were acquired under the same measurement conditions. obs Then, the pollution state feature vector X is extracted from the current response. fouling and historical calibration parameter set P hist This allows for the construction of training samples, where the input to the training samples includes at least X. fouling and P hist and selectively include I obs The supervision label is the reference state standard response I. calib , or the response correction parameters, baseline correction parameters and peak shape correction parameters obtained by fitting them; During sample pairing, the current response of the same wearable electrochemical sensor under contaminated conditions is paired with its reference state response under the same or equivalent target concentration conditions after its most recent effective physical calibration. The equivalent target concentration is confirmed through the standard addition method, reference response inversion, or a preset calibration relationship. The training set, validation set, and test set are divided according to the wearable electrochemical sensor number, the data acquisition time period, or a combination of both to avoid the same wearable electrochemical sensor number appearing in both the training set and the test set simultaneously. One of the goals of model training is to make the output equivalent response I... ref Approximating the reference state standard response I calib ; To improve the accuracy and physical consistency of the compensation results, the model training uses a joint loss function consisting of a data reconstruction error term and a physical consistency constraint term: THE total = L recon + λL phys Among them, L recon Used to characterize the reconstruction error between the model output and the reference state standard response, reflecting the data reconstruction deviation between the compensation result and the reference state target response; L phys The peak position, peak width, baseline or peak shape parameters used to constrain the compensated response are within the preset reasonable range of electrochemical parameters, reflecting whether the compensation result meets the preset electrochemical response law and the reasonable boundary of parameters. Joint loss function L total From the reconstruction error term L recon Physical consistency constraint term λL phys Composition, L recon The smaller the value, the closer the compensation result is to the target reference response at the data level, reflecting the accuracy of the compensation result; L phys The smaller the value, the more the compensation result conforms to the electrochemical response law, reflecting the physical consistency of the compensation result. During model training and validation, L is jointly monitored on the validation set. recon With L phys The changing trend, combined with L total The overall convergence of the model is used to comprehensively judge whether the model output simultaneously possesses compensation accuracy and physical rationality. When L recon Converging to the preset accuracy range and L phys When all components are stably within a reasonable range, the model is considered to be in a usable state. The peak position constraint is the peak position residual between the target analyte and the auxiliary characterization component; the peak width constraint is the corresponding half-peak width or the change in peak width; and the baseline constraint is the baseline slope and baseline intercept of the background interval.

8. The online calibration method for wearable electrochemical sensors based on pollution state identification as described in claim 7, characterized in that, In step (4), based on the equivalent response I ref Alternatively, correction parameters α, β, and γ can be used to correct the current measured response and further calculate the estimated concentration of the target analyte. The specific implementation process involves two scenarios: Scenario 1, where the output of the compensation and reconstruction model is the equivalent response I... ref At that time, directly from the equivalent response I ref The peak current or peak area of ​​the target analyte is extracted, and the estimated concentration is calculated based on the concentration-response calibration relationship established under the reference state. In scenario two, when the model output is the correction parameters α, β, and γ, the current response curve is corrected for gain based on the response correction coefficient α, the current response curve is corrected for baseline offset based on the baseline offset correction coefficient β, and the peak distortion of the current response is corrected based on the peak distortion correction coefficient γ. After correction, quantitative parameters are extracted from the corrected response and the estimated concentration is calculated.

9. The online calibration method for wearable electrochemical sensors based on pollution state identification as described in claim 8, characterized in that, In step (5), the parameter overflow or reconstruction error during the compensation process is continuously monitored. When any of the indicators, such as the response correction coefficient, peak distortion correction coefficient, baseline offset, or reconstruction error, exceeds the preset safety threshold, it is determined that the current automatic compensation has exceeded the effective working range and the automatic correction is stopped. At the same time, physical recalibration, electrode cleaning, consumable replacement, or terminal alarm prompts are output, thereby forming a closed loop of equipment operation: online correction - threshold monitoring - maintenance triggering.

10. An online calibration system for wearable electrochemical sensors based on pollution state identification, characterized in that, include: The electrochemical signal acquisition unit is used to control the working electrode, reference electrode and counter electrode to perform a preset potential scanning program under reference conditions, acquire response current data and obtain the raw volt-ampere response; The preprocessing and feature extraction unit is used to filter, correct baselines, identify peaks, and extract pollution state feature vectors from the raw volt-ampere response. The compensation and reconstruction unit generates equivalent response or correction parameters under the reference state based on the pollution state feature vector and historical calibration parameters using the compensation and reconstruction model. The concentration estimation unit calculates the estimated concentration of the target analyte based on the compensated equivalent response or correction parameters. The threshold determination and maintenance triggering unit outputs physical recalibration, electrode cleaning, consumable replacement, or terminal alarm prompts when the calibration parameters or reconstruction error exceed the preset safety threshold.

Images