Cognitive load monitoring system based on sweat metabolite analysis

Abstract

The invention discloses a cognitive load monitoring system based on sweat metabolite analysis, and belongs to the technical field of biosensing. Comprising a multi-site microfluidic sweat collection module, a variable temperature regulation type sweat dilution and stabilization module, a feature targeting type metabolite recognition module, a cross-individual dynamic standardization mapping module, a multi-channel fusion calculation module, a self-adaptive cognitive load model training module, an environment state correlation correction module and a wearable integrated output module. All the modules construct a dynamic closed-loop mechanism through flexible substrate structure integration, continuous flow information channel transmission and uniform timestamp and weight feedback. The system realizes high-throughput detection of sweat metabolites (lactic acid, cortisol and glucose) and multi-source physiological signal fusion analysis, supports data standardization expression, adaptive model training and environment disturbance correction among multiple users, finally completes intelligent decision and remote transmission through wearable equipment, and has high integration, intelligence and practicability.

Description

Technical Field

[0001] The present invention relates to the field of biosensor technology, and more particularly to a cognitive load monitoring system based on sweat metabolite analysis. Background Art

[0002] As a non-invasive, physiologically driven monitoring technology, the cognitive load monitoring system based on sweat metabolite analysis has garnered widespread attention in recent years in fields such as smart wearables, biomedical engineering, and human factors engineering. By analyzing changes in the concentrations of metabolites such as lactate, cortisol, electrolytes, and glucose in sweat, the system enables real-time assessment of an individual's cognitive load during work or study. Compared to traditional devices such as electroencephalogram (EEG) and functional near-infrared imaging (fNIRS), sweat sensors offer improved portability, comfort, and sustainable monitoring capabilities. They are suitable for long-term cognitive status tracking in complex environments, particularly in military, aviation, driving, and high-intensity work scenarios. Current technologies have made significant progress in microsensor integration, flexible material development, and multimodal fusion recognition.

[0003] However, there is still a key shortcoming in its development so far, that is, the composition of sweat is significantly affected by individual differences, ambient temperature and fluctuations in sweat volume, which leads to limited data stability and analysis accuracy. It is difficult to achieve highly universal and highly sensitive cognitive load judgment in different populations and changing scenarios. This has also become the biggest technical bottleneck for its large-scale application and promotion. Summary of the Invention

[0004] The purpose of the present invention is to provide a cognitive load monitoring system based on sweat metabolite analysis to solve the problems raised in the above-mentioned background technology: However, its development to date still has a key shortcoming, that is, sweat composition is significantly affected by individual differences, ambient temperature and fluctuations in sweat volume, resulting in limited data stability and analysis accuracy, making it difficult to achieve highly universal and highly sensitive cognitive load determination in different populations and changing scenarios. This has also become the biggest technical bottleneck for its large-scale application and promotion.

[0005] Technical Solution: A cognitive load monitoring system based on sweat metabolite analysis includes a multi-site microfluidic sweat collection module, a temperature-controlled sweat dilution and stabilization module, a feature-targeted metabolite identification module, a cross-individual dynamic normalization mapping module, a multi-channel fusion calculation module, an adaptive cognitive load model training module, an environmental state correlation correction module, and a wearable integrated output module. All modules are structurally integrated through a flexible substrate, data is transmitted through a continuous flow information channel, and algorithmically, a dynamic closed-loop mechanism is formed through a unified timestamp and weight feedback system, forming the following innovative collaborative paths: The multi-site microfluidic sweat collection module constructs multi-source parallel input channels through sweat guiding structures at multiple independent sweat gland sites; the variable temperature control sweat dilution stabilization module stabilizes sweat concentration through a controllable thermoelectric element and micro-osmotic coupling mechanism; the feature-targeted metabolite recognition module achieves high-throughput, multi-electrode, and quantile detection of target metabolites such as lactate, cortisol, and glucose; the cross-individual dynamic standardization mapping module adopts individual feature mapping and normalization correction modeling strategies to achieve unified expression of metabolite data among multiple users; the multi-channel fusion calculation module performs time-frequency fusion processing on sweat metabolic signals with skin electrodermal response, electromyographic signals, and heart rate variability synchronized physiological parameters; the adaptive cognitive load model training module continuously iterates and optimizes the recognition model through a deep pseudo-label learning mechanism; the environmental state association correction module perceives the disturbance of external temperature, humidity, and light intensity variables on sweat indicators in real time, and dynamically feedback-adjusts the model weights; the wearable integrated output module uniformly encapsulates the above processing results in a flexible electronic device to complete local decision-making and remote data transmission.

[0006] Preferably, the multi-site microfluidic sweat collection module includes multiple flexible collection units positioned at the forearm, wrist and back of the neck. Each collection unit is composed of a serpentine microchannel array made of polydimethylsiloxane with a channel width of 200 μm and a depth of 150 μm. It is connected to a porous hydrophilic gel guide layer, and the upward speed of sweat is enhanced by capillary force. The collection response time is less than 60 seconds.

[0007] Preferably, the temperature-variable sweat dilution and stabilization module comprises a Bi2Te3 thermoelectric pair array arranged on a flexible PI substrate, with a P / N alternating structure and a density of 160 pairs / cm 2 ; The thermoelectric couple controls the input current through the MOS temperature control unit, with a maximum of ±30mA, to maintain the temperature of the sweat contact area at 33±0.5°C; A graphene / thermal grease composite layer is set under the thermoelectric structure to improve thermal conductivity and temperature control response speed ≤ 3s; The temperature-controlled sweat dilution and stabilization module is provided with a micro-osmotic sustained-release chamber in the center, which is composed of a double-layer PVA nanoporous membrane with an average pore size of 120nm and an ion exchange gel, and the sweat dilution ratio is regulated to be 1:1.07; A one-way pressure relief valve is installed at the tail end of the chamber to ensure flow rate stability, with a maximum allowable pressure difference of 50Pa; The diluted sweat output is directly coupled to the sensor module to avoid evaporation or contamination caused by sample exposure.

[0008] Preferably, the characteristic targeted metabolite recognition module includes three sets of recognition electrode arrays, which respectively detect lactate, cortisol and glucose; each set of arrays includes a graphene / Pt composite conductive skeleton, an electrode spacing of 200 μm, and a flexible PDMS substrate; The lactate recognition region immobilizes lactate oxidase, the glucose region immobilizes GOx enzyme, and the cortisol region immobilizes specific monoclonal antibodies, all of which are grafted via EDC / NHS cross-linking; The three-electrode structure consists of a working electrode, a reference electrode Ag / AgCl, and an auxiliary electrode. The embedded microcontroller collects data in real time and performs differential amplification and low-pass filtering, with an output error of less than 3%.

[0009] Preferably, the cross-individual dynamic normalization mapping module includes a multi-dimensional input normalization model, the inputs of which include pH, sweat rate, conductivity, and skin temperature; the model structure is a four-dimensional LSTM+Transformer hybrid network, and outputs a set of adjustment coefficients (α1, α2, α3) for metabolite normalization mapping; The model adopts a self-supervised learning optimization strategy, and automatically forms an individual weight factor after each user uses it five times in a row; In cross-population validation tests, the cross-individual dynamic normalization mapping module reduced the standard deviation of metabolite concentrations by more than 35%.

[0010] Preferably, the multi-site microfluidic sweat collection module further includes a sweat component stability correction module, which further includes a baseline drift detection unit, which uses an exponentially weighted sliding average algorithm to detect Na in sweat based on a three-point sliding interval. + 2. Whether the lactate concentration fluctuation threshold exceeds ±15%; The sample amplification unit establishes a stability regression model through a local weighted regression fitting curve, and interpolates and reconstructs missing or fluctuating samples; The sweat feature correction network consists of a dual-channel normalizer and a 1D convolution correction network with a convolution kernel size of 3, a stride of 1, and 64 output channels. The output is a normalized feature vector for fusion.

[0011] Preferably, the temperature-variable regulated sweat dilution stabilization module further comprises a signal compression submodule and the following optimization mechanism: the signal compression submodule uses principal component analysis to reduce the dimension of each input channel to a 12-dimensional feature vector and retains more than 95% of the variance information; Asynchronous fusion mechanism, which uses the position encoding strategy in the Transformer structure to time-align data of different frequencies and process asynchronous sampling data; The cross-modal attention mechanism includes four attention channels, which calculate weights for sweat characteristics, physiological electrical signals, skin temperature, and electromyography data respectively, and finally output a unified representation.

[0012] Preferably, the feature-targeted metabolite recognition module further includes the following optimization mechanism: The recognition performance improvement module uses dynamic calibration technology during high-throughput detection, combining external temperature and humidity changes with electrode operating status feedback to dynamically adjust the electrode operating current and recognition threshold to optimize metabolite detection accuracy. A multi-target feedback regulation mechanism, through integrated feedback channels, adjusts the sensitivity and response time of the working electrode in real time to adapt to metabolite concentration fluctuations under different environmental conditions; Multimodal collaborative recognition module combines different recognition methods, including enzyme electrode method and antibody electrochemical method.

[0013] Preferably, the environmental state correction module further includes an environmental drift modeling submodule, which uses a sliding window to dynamically model the ultraviolet intensity, temperature and humidity change trends, and calculate the environmental disturbance risk factor δ; Intervention feedback connection unit, if the value of δ exceeds the set threshold, then send the perturbation vector ζ to the model and introduce it into the middle layer of the model through residual fusion; The high-fluctuation scene discriminator triggers a downgrade of the confidence level of the current prediction result and marks the output when the temperature and humidity change rate exceeds 2% / min for three consecutive minutes.

[0014] Compared with the prior art, the advantages of the present invention are: (1) A sweat component stability correction module is introduced. Through multi-point drift detection and deep convolutional normalization network, data distortion caused by individual differences and sweat volume fluctuations is effectively corrected, greatly improving versatility and accuracy.

[0015] (2) A multi-channel asynchronous fusion mechanism and a cross-modal attention network are used to effectively integrate sweat signals with physiological parameters (such as electrocardiogram, electromyography, and temperature) to improve the ability to distinguish cognitive load in dynamic tasks.

[0016] (3) Construct a cognitive load model with evolvable structure, combining individual style embedding technology with a joint loss function, so that the model can automatically tune its structure according to user differences and task changes, thereby enhancing generalization capabilities.

[0017] (4) Set up an environmental state correction module to model external interference factors such as temperature, humidity, and ultraviolet rays in real time, and correct the model prediction through the disturbance feedback mechanism to ensure the stability of judgment in high-volatility scenarios.

[0018] (5) The modular architecture design facilitates embedding in various wearable terminals, enabling real-time data collection and analysis, meeting the low power consumption and low latency requirements in mobile scenarios, and has good prospects for use.

[0019] (6) Solve the core problems of existing technologies such as unstable sweat analysis results and the inability of the model to adapt to dynamic changes in users and the environment, and realize a truly dynamic intelligent cognitive load monitoring system. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a schematic diagram of the overall system of a cognitive load monitoring system based on sweat metabolite analysis of the present invention. DETAILED DESCRIPTION

[0021] For examples, see Figure 1 A cognitive load monitoring system based on sweat metabolite analysis is characterized in that it includes a multi-site microfluidic sweat collection module, a variable temperature controlled sweat dilution and stabilization module, a feature-targeted metabolite identification module, a cross-individual dynamic standardization mapping module, a multi-channel fusion calculation module, an adaptive cognitive load model training module, an environmental state association correction module, and a wearable integrated output module. All modules are structurally integrated through a flexible substrate, data is transmitted through a continuous flow information channel, and a dynamic closed-loop mechanism is formed through a unified timestamp and weight feedback system in the algorithm, forming the following innovative collaborative path: The multi-site microfluidic sweat collection module constructs multi-source parallel input channels through sweat guiding structures at multiple independent sweat gland sites; the variable temperature control sweat dilution stabilization module stabilizes sweat concentration through a controllable thermoelectric element and micro-osmotic coupling mechanism; the feature-targeted metabolite recognition module achieves high-throughput, multi-electrode, and quantile detection of target metabolites such as lactate, cortisol, and glucose; the cross-individual dynamic standardization mapping module adopts individual feature mapping and normalization correction modeling strategies to achieve unified expression of metabolite data among multiple users; the multi-channel fusion calculation module performs time-frequency fusion processing on sweat metabolic signals with skin electrodermal response, electromyographic signals, and heart rate variability synchronized physiological parameters; the adaptive cognitive load model training module continuously iterates and optimizes the recognition model through a deep pseudo-label learning mechanism; the environmental state association correction module perceives the disturbance of external temperature, humidity, and light intensity variables on sweat indicators in real time, and dynamically feedback-adjusts the model weights; the wearable integrated output module uniformly encapsulates the above processing results in a flexible electronic device to complete local decision-making and remote data transmission.

[0022] Specifically, the multi-channel fusion computation module is used to integrate and analyze sweat metabolism signals with physiological electrical parameters (such as galvanic skin response, heart rate variability, and electromyographic signals). The signal compression submodule uses PCA to reduce the dimensionality of each channel to 12 dimensions, retaining over 95% of the original information. The fusion mechanism incorporates a Transformer-based positional encoding strategy to process asynchronously sampled data. Combined with a cross-modal attention mechanism (four-head attention channel), it weights each signal individually and outputs a unified physiological state representation.

[0023] Specifically, the adaptive cognitive load model training module uses a pseudo-label supervision mechanism to achieve robust model training in small sample scenarios. It also integrates a structural self-evolution module to adaptively adjust the number of neural network layers and nodes based on the validation set loss. A joint optimization strategy of cross-entropy and confidence regularization is introduced at the loss function layer to improve the model's generalization ability in low-label scenarios. The individual style embedder module maintains a 24-hour feature distribution vector for each user, which is used as a bias vector for the deep model input to enhance the model's individual recognition capabilities.

[0024] Specifically, the wearable integrated output module integrates all sensors, circuits, and algorithms into a wearable electronic device constructed from a flexible substrate, enabling local processing and edge decision-making capabilities. The output module transmits monitoring results to the user terminal or remote server via Bluetooth, LoRa, or NB-IoT, while also visually displaying them on a local interface, including the current cognitive load level, historical fluctuations, and risk warnings.

[0025] Specifically, the workflow of this system constitutes a highly integrated dynamic information processing closed loop. The specific process is as follows: Simultaneous sweat collection at multiple sites → Sweat is guided through PDMS microfluidic channels and quickly sampled into downstream modules; Temperature and dilution control → Maintaining stable sweat temperature and achieving proportional dilution to ensure comparability between different sites; High-throughput metabolite identification → Targeted electrodes detect lactate, cortisol, and glucose, respectively, forming a ternary feature vector; Standardization and correction processing → using individual characteristics, environmental conditions and component correction network to normalize sweat characteristics; Fusion of multimodal physiological signals → combining EDA, EMG, HRV and other signals, and outputting physiological state representation through attention mechanism; Cognitive load state modeling → using deep learning models for identification and prediction, with adaptive and self-evolving capabilities; Result output and data transmission → The processing results are visually output on the wearable device and uploaded to the cloud platform simultaneously.

[0026] Phase 1: Multi-source sensing acquisition and signal preprocessing Input: sweat from multiple parts of the body (lactic acid, glucose, cortisol) + epidermal conductivity (EDA), heart rate variability (HRV), skin temperature and other physiological signals; Processing steps: S1.1 Multi-site PDMS microfluidic unit collects sweat and automatically drains it to the micro-osmotic dilution module; The S1.2Bi2Te3 thermoelectric array controls the temperature in real time, keeping sweat in the optimal metabolic activity temperature range (33±0.5℃); S1.3 Micro-osmotic dilution chamber dilutes sweat from different sources to a standard concentration range (1:1.07); The S1.4 signal synchronizer uniformly aligns the sampling timestamps to improve the accuracy of subsequent multimodal fusion; Output: Temperature-controlled and standard diluted sweat flow + synchronized electrophysiological raw signals.

[0027] Phase 2: Targeted Identification and Physiological Characterization Input: processed sweat sample + electrophysiological raw data; Processing steps: S2.1 Targeted identification of three types of metabolites (lactate, glucose, and cortisol) using specific immobilized electrodes; S2.2 electrochemical readout is amplified by a differential amplifier and digitized by a 12-bit ADC module; S2.3 Extract electrophysiological signals in the time domain (mean, variance, frequency) and frequency domain (wavelet packet energy spectrum); S2.4 construct the initial physiological feature vector V0 = [Lac, Glu, Cort, HRV, EDA, Temp, EMG...]; Output: Normalized physiological feature vector V0, which enters the normalization mapping and disturbance correction module.

[0028] Stage 3: Standardization and multi-factor disturbance correction Input: original physiological feature vector V0 + current environmental parameters (temperature, humidity, light); Processing steps: S3.1 construct an individual feature mapping function F_ind based on individual sweat rate, conductivity, pH, body temperature, etc.; The S3.2LSTM network generates an individualized mapping weight group (α1, α2, α3) and maps the original features into the individual standard space V1; S3.3 The sliding window tracks the environmental disturbance factor δ(t) and introduces the disturbance vector ζ(t) for dynamic correction; S3.4 local weighted regression algorithm and 1D convolution correction module fit and smooth the residual part of the feature; Output: Stable and unified feature vector V2, representing the true physiological state at that moment.

[0029] Phase 4: Multimodal Fusion and Cognitive Load Modeling Input: multimodal unified feature vector V2; Processing steps: S4.1 fusion submodule performs linear projection and multi-head attention weighting on each modality information; S4.2 Constructing a multi-level representation matrix of cognitive load H = [h1,h2,…h n ]; S4.3 uses Transformer + adaptive structural evolution network to distinguish load levels (low / medium / high / very high); The S4.4 system automatically generates a local embedded model Mini-Net, embedding individual bias vectors to achieve online fast prediction; Output: Current cognitive load prediction result R(t)∈{L,M,H,VH}, and output confidence index P(r).

[0030] Phase 5: Result Feedback and Dynamic Self-Optimization Mechanism Input: predicted output R(t) + user historical data + environmental trends; Processing steps: S5.1 output is presented through OLED or APP interface for visual analysis (curves, trends, load hot spots); S5.2 Users can provide feedback on "whether it is accurate" through the subjective annotation interface to form pseudo-labeled data; The S5.3 system performs online fine-tuning on individual models once every 24 hours and uses transfer learning to tune the Mini-Net; S5.4 Anomaly Detection Module (IsolationForest) detects outliers and activates the alarm mechanism; S5.5 If abnormalities are detected three times in a row, the high-precision model will be awakened for secondary discrimination; Output: Visual analysis results + user adaptation model update + database reverse update weight.

[0031] Summary of the information closed-loop structure: Through the structure of "perception → preprocessing → standardization → fusion modeling → output feedback → self-learning iteration", the system forms an adaptive learning closed-loop and dynamic feedback regulation loop with the following characteristics: Closed loop 1: Feature feedback regulation The prediction results output by the model serve as feedback input to influence the feature interpretation method at the next moment (used for sliding window parameters and environmental drift factor correction); Closed Loop 2: Individual Model Optimization The difference between user annotation and system prediction forms a pseudo-supervision signal, which is used to fine-tune the personalized model and improve long-term prediction accuracy. Closed loop three: self-evolution of model structure If the model's stability decreases over a specific period of time, the system will dynamically adjust the network depth and attention structure to maintain robustness; Closed Loop 4: Risk Event Response Mechanism The continuous occurrence of cognitive overload risks or mutation indicators will trigger local alarms and cloud model activation, starting the expert-level discrimination network.

[0032] The multi-site microfluidic sweat collection module includes multiple flexible collection units located on the forearm, wrist and back of the neck. Each collection unit consists of a serpentine microchannel array made of polydimethylsiloxane with a channel width of 200μm and a depth of 150μm. It is connected to a porous hydrophilic gel guide layer, which enhances the upward speed of sweat through capillary force, and the collection response time is less than 60 seconds.

[0033] The temperature-controlled sweat dilution stabilization module includes a Bi2Te3 thermoelectric pair array arranged on a flexible PI substrate, with a P / N alternating structure and a density of 160 pairs / cm 2 ; The thermoelectric couple controls the input current through the MOS temperature control unit, with a maximum of ±30mA, maintaining the temperature of the sweat contact area at 33±0.5℃; A graphene / thermal grease composite layer is set under the thermoelectric structure to improve thermal conductivity and temperature control response speed ≤ 3s; The temperature-controlled sweat dilution and stabilization module has a micro-osmotic sustained-release chamber in the center, which is composed of a double-layer PVA nanoporous membrane with an average pore size of 120nm and an ion exchange gel, and can regulate the sweat dilution ratio to 1:1.07. A one-way pressure relief valve is installed at the tail end of the chamber to ensure flow rate stability, with a maximum allowable pressure difference of 50Pa; The diluted sweat output is directly coupled to the sensor module to avoid evaporation or contamination caused by sample exposure.

[0034] The characteristic targeted metabolite recognition module includes three sets of recognition electrode arrays, one for detecting lactate, one for cortisol, and one for glucose. Each array consists of a graphene / Pt composite conductive skeleton, with an electrode spacing of 200 μm and a flexible PDMS substrate. The lactate recognition region immobilizes lactate oxidase, the glucose region immobilizes GOx enzyme, and the cortisol region immobilizes specific monoclonal antibodies, all of which are grafted via EDC / NHS cross-linking; The three-electrode structure consists of a working electrode, a reference electrode Ag / AgCl, and an auxiliary electrode. The embedded microcontroller collects data in real time and performs differential amplification and low-pass filtering, with an output error of less than 3%.

[0035] The cross-individual dynamic normalization mapping module includes a multi-dimensional input normalization model with inputs including pH, sweat rate, conductivity, and skin temperature. The model structure is a four-dimensional LSTM+Transformer hybrid network, which outputs a set of adjustment coefficients (α1, α2, α3) for metabolite normalization mapping. The model adopts a self-supervised learning optimization strategy, and automatically forms an individual weight factor after each user uses it five times in a row; In cross-population validation tests, the above-mentioned cross-individual dynamic normalization mapping module reduced the standard deviation of metabolite concentration by more than 35%.

[0036] The multi-site microfluidic sweat collection module also includes a sweat component stability correction module, which further includes a baseline drift detection unit. Based on the three-point sliding interval, an exponentially weighted sliding average algorithm is used to detect Na in sweat. + 2. Whether the lactate concentration fluctuation threshold exceeds ±15%; The sample amplification unit establishes a stability regression model through a local weighted regression fitting curve, and interpolates and reconstructs missing or fluctuating samples; The sweat feature correction network consists of a dual-channel normalizer and a 1D convolution correction network with a convolution kernel size of 3, a stride of 1, and 64 output channels. The output is a normalized feature vector for fusion.

[0037] The thermoregulatory sweat dilution stabilization module further includes a signal compression submodule and the following optimization mechanism: the signal compression submodule uses principal component analysis to reduce the dimensionality of each input channel to a 12-dimensional feature vector and retains more than 95% of the variance information; Asynchronous fusion mechanism, which uses the position encoding strategy in the Transformer structure to time-align data of different frequencies and process asynchronous sampling data; The cross-modal attention mechanism includes four attention channels, which calculate weights for sweat characteristics, physiological electrical signals, skin temperature, and electromyography data respectively, and finally output a unified representation.

[0038] The feature-targeted metabolite identification module further includes the following optimization mechanisms: The recognition performance improvement module uses dynamic calibration technology during high-throughput detection, combining external temperature and humidity changes with electrode operating status feedback to dynamically adjust the electrode operating current and recognition threshold to optimize metabolite detection accuracy. A multi-target feedback regulation mechanism, through integrated feedback channels, adjusts the sensitivity and response time of the working electrode in real time to adapt to metabolite concentration fluctuations under different environmental conditions; Multimodal collaborative recognition module combines different recognition methods, including enzyme electrode method and antibody electrochemical method.

[0039] The environmental state correction module further includes an environmental drift modeling submodule, which uses a sliding window to dynamically model the trend of UV intensity, temperature and humidity, and calculates the environmental disturbance risk factor δ; Intervention feedback connection unit, if the value of δ exceeds the set threshold, then send the perturbation vector ζ to the model and introduce it into the middle layer of the model through residual fusion; The high-fluctuation scene discriminator triggers a downgrade of the confidence level of the current prediction result and marks the output when the temperature and humidity change rate exceeds 2% / min for three consecutive minutes.

[0040] The above shows and describes the basic principles, main features and advantages of the present invention; those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected; the scope of protection claimed in the present invention is defined by the attached claims and their equivalents.

Claims

1. A cognitive load monitoring system based on sweat metabolite analysis, characterized in that: The cognitive load monitoring system based on sweat metabolite analysis includes a multi-site microfluidic sweat collection module, a variable temperature controlled sweat dilution and stabilization module, a feature-targeted metabolite identification module, a cross-individual dynamic normalization mapping module, a multi-channel fusion calculation module, an adaptive cognitive load model training module, an environmental state association correction module, and a wearable integrated output module. All modules are structurally integrated through a flexible substrate, data is transmitted through a continuous flow information channel, and a dynamic closed-loop mechanism is formed through a unified timestamp and weight feedback system in the algorithm, forming the following innovative collaborative path: The multi-site microfluidic sweat collection module constructs multi-source parallel input channels through sweat guiding structures at multiple independent sweat gland sites; the variable temperature control sweat dilution stabilization module stabilizes sweat concentration through a controllable thermoelectric element and micro-osmotic coupling mechanism; the feature-targeted metabolite recognition module achieves high-throughput, multi-electrode, and quantile detection of target metabolites such as lactate, cortisol, and glucose; the cross-individual dynamic standardization mapping module adopts individual feature mapping and normalization correction modeling strategies to achieve unified expression of metabolite data among multiple users; the multi-channel fusion calculation module performs time-frequency fusion processing on sweat metabolic signals with skin electrodermal response, electromyographic signals, and heart rate variability synchronized physiological parameters; the adaptive cognitive load model training module continuously iterates and optimizes the recognition model through a deep pseudo-label learning mechanism; the environmental state association correction module perceives the disturbance of external temperature, humidity, and light intensity variables on sweat indicators in real time, and dynamically feedback-adjusts the model weights; the wearable integrated output module uniformly encapsulates the above processing results in a flexible electronic device to complete local decision-making and remote data transmission.

2. A cognitive load monitoring system based on sweat metabolite analysis according to claim 1, characterized in that: The multi-site microfluidic sweat collection module includes multiple flexible collection units positioned on the forearm, wrist and back of the neck. Each flexible collection unit is composed of a serpentine microchannel array made of polydimethylsiloxane, with a channel width of 200 μm and a depth of 150 μm. It is connected to a porous hydrophilic gel guide layer, which enhances the upward speed of sweat through capillary force, and the collection response time is less than 60 seconds.

3. The cognitive load monitoring system based on sweat metabolite analysis according to claim 1, characterized in that: The variable temperature control sweat dilution stabilization module includes a Bi2Te3 thermoelectric pair array arranged on a flexible PI substrate, with a P / N alternating structure and a density of 160 pairs / cm 2 ; The thermoelectric couple controls the input current through the MOS temperature control unit, with a maximum of ±30mA, to maintain the temperature of the sweat contact area at 33±0.5°C; A graphene / thermal grease composite layer is set under the thermoelectric structure to improve thermal conductivity and temperature control response speed ≤ 3s; The temperature-controlled sweat dilution and stabilization module is provided with a micro-osmotic sustained-release chamber in the center, which is composed of a double-layer PVA nanoporous membrane with an average pore size of 120nm and an ion exchange gel, and the sweat dilution ratio is regulated to be 1:1.07; A one-way pressure release valve is provided at the tail end of the micro-osmotic sustained-release chamber to ensure flow rate stability, with a maximum allowable pressure difference of 50 Pa; The diluted sweat output end is directly coupled to the sensor module to avoid evaporation contamination caused by sample exposure.

4. The cognitive load monitoring system based on sweat metabolite analysis according to claim 1, characterized in that: The characteristic targeted metabolite recognition module includes three sets of recognition electrode arrays, which detect lactate, cortisol and glucose respectively; each set of arrays includes a graphene / Pt composite conductive skeleton, an electrode spacing of 200 μm, and a flexible PDMS substrate; The lactate recognition region immobilizes lactate oxidase, the glucose region immobilizes GOx enzyme, and the cortisol region immobilizes specific monoclonal antibodies, all of which are grafted via EDC / NHS cross-linking; The three-electrode structure consists of a working electrode, a reference electrode Ag / AgCl, and an auxiliary electrode. The embedded microcontroller collects data in real time and performs differential amplification and low-pass filtering, with an output error of less than 3%.

5. The cognitive load monitoring system based on sweat metabolite analysis according to claim 1, characterized in that: The cross-individual dynamic normalization mapping module includes a multi-dimensional input normalization model, whose inputs include pH, sweat rate, conductivity, and skin temperature. The model structure is a four-dimensional LSTM+Transformer hybrid network, which outputs a set of adjustment coefficients (α1, α2, α3) for metabolite normalization mapping. The multi-dimensional input normalization model adopts a self-supervised learning optimization strategy, and automatically forms an individual weight factor after each user uses it five times continuously; In cross-population validation tests, the cross-individual dynamic normalization mapping module reduced the standard deviation of metabolite concentrations by more than 35%.

6. The cognitive load monitoring system based on sweat metabolite analysis according to claim 2, characterized in that: The multi-site microfluidic sweat collection module further includes a sweat component stability correction module, and the sweat component stability correction module further includes: The baseline drift detection unit uses an exponentially weighted sliding average algorithm to detect Na in sweat based on a three-point sliding interval. + 2. Whether the lactate concentration fluctuation threshold exceeds ±15%; The sample amplification unit establishes a stability regression model through local weighted regression fitting curve, and interpolates and reconstructs missing or fluctuating samples; The sweat feature correction network consists of a dual-channel normalizer and a 1D convolution correction network with a convolution kernel size of 3, a stride of 1, and 64 output channels. The output is a normalized feature vector for fusion.

7. The cognitive load monitoring system based on sweat metabolite analysis according to claim 3, characterized in that: The temperature-controlled sweat dilution stabilization module further includes a signal compression submodule and the following optimization mechanism: The signal compression submodule uses principal component analysis to reduce the dimension of each input channel to a 12-dimensional feature vector and retains more than 95% of the variance information; Asynchronous fusion mechanism, which uses the position encoding strategy in the Transformer structure to time-align data of different frequencies and process asynchronous sampling data; The cross-modal attention mechanism includes four attention channels, which calculate weights for sweat characteristics, physiological electrical signals, skin temperature, and electromyography data respectively, and finally output a unified representation.

8. The cognitive load monitoring system based on sweat metabolite analysis according to claim 4, characterized in that: The feature-targeted metabolite recognition module further includes the following optimization mechanisms: The recognition performance improvement module uses dynamic calibration technology during high-throughput detection, combining external temperature and humidity changes with electrode operating status feedback to dynamically adjust the electrode operating current and recognition threshold to optimize metabolite detection accuracy. A multi-target feedback regulation mechanism, through integrated feedback channels, adjusts the sensitivity and response time of the working electrode in real time to adapt to metabolite concentration fluctuations under different environmental conditions; Multimodal collaborative recognition module combines different recognition methods, including enzyme electrode method and antibody electrochemical method.

9. The cognitive load monitoring system based on sweat metabolite analysis according to claim 1, characterized in that: The environmental state correction module further includes an environmental drift modeling submodule, which uses a sliding window to dynamically model the ultraviolet intensity, temperature and humidity change trends, and calculates the environmental disturbance risk factor δ; Intervention feedback connection unit, if the value of δ exceeds the set threshold, then send the perturbation vector ζ to the model and introduce it into the middle layer of the model through residual fusion; The high-fluctuation scene discriminator triggers a downgrade of the confidence level of the current prediction result and marks the output when the temperature and humidity change rate exceeds 2% / min for three consecutive minutes.

Images