A method for predicting the work capacity of a worker based on physiological signs

Abstract

The application discloses a kind of underwater operator's work capacity prediction methods based on physiological signs, through integrated sensor module of underwater operation equipment, the physiological sign data of multiple of operator is collected, underwater environmental data is synchronously obtained, the real-time physiological sign data and environmental data collected are preprocessed and feature extraction, integrated to form comprehensive feature vector, the work capacity index of current time is calculated by comprehensive feature vector, the application realizes the prospective, individualized evaluation of work capacity by multi-modal data fusion and machine learning model, can issue early warning before the substantial decline of personnel capacity through trend analysis, overcome the limitations of single parameter evaluation, fusion physiological and environmental multidimensional information, make the evaluation result more comprehensive, reliable, system can adapt to the physiological characteristics of different personnel, realize accurate individual monitoring, and provide scientific data support for optimizing operation process, improve overall operation safety and efficiency.

Description

Technical Field

[0001] This invention relates to the field of underwater operation safety assurance technology, specifically a method for predicting the operational capabilities of underwater workers based on physiological signs. Background Technology

[0002] Underwater operations are critical activities involving diving, underwater construction, rescue, and marine resource development. Their working environments are characterized by high pressure, low temperatures, poor visibility, and limited communication, posing severe challenges to the physical and mental health of personnel. Currently, safety assurance systems for underwater workers primarily rely on a relatively traditional and static management model, which has significant limitations. First, existing assurance measures largely depend on static management and reactive responses. For example, while rigorous medical examinations are conducted before operations, they cannot monitor the dynamic physiological changes underwater; managing risks through fixed operating depths and schedules ignores individual differences and real-time physiological loads among workers; simultaneously, surface monitoring personnel mainly observe and communicate through video and intercom systems, which is inefficient in low visibility or emergency situations and highly dependent on the subjective experience of monitoring personnel, making it difficult to promptly detect potential crises.

[0003] Secondly, existing monitoring technologies and equipment are limited in function and lack intelligent, comprehensive judgment capabilities.

[0004] Currently, while some advanced diving computers or wearable devices on the market can monitor basic parameters such as heart rate, depth, and water temperature, their functions are mostly limited to simple data display and recording. These devices are like isolated "information islands," failing to deeply integrate and analyze multi-source physiological and environmental data. They cannot effectively assess complex conditions such as worker fatigue, nitrogen narcosis risk, hypothermia, or oxygen toxicity, and they lack the ability to predict the downward trend of key indicators such as the worker's overall operational ability, judgment, and reaction speed, resulting in very limited early warning capabilities.

[0005] In summary, the fundamental flaw of existing technologies lies in the lack of a method capable of predicting the operational capabilities of underwater workers in a real-time, dynamic, and comprehensive manner. Current safety measures cannot provide proactive and accurate early warnings before the initial deterioration of personnel's physiological condition or before a substantial decline in capabilities. This "post-event response" rather than "pre-event prevention" model means that underwater operations are always accompanied by high safety risks, failing to meet the urgent need for refined and intelligent safety protection for personnel in modern high-risk operations. Therefore, an improved technology is urgently needed to address this problem in existing technologies. Summary of the Invention

[0006] The purpose of this invention is to provide a method for predicting the operational capabilities of underwater workers by integrating multimodal data fusion and machine learning models. This method enables a forward-looking and individualized assessment of operational capabilities, providing early warnings before a substantial decline in personnel capabilities through trend analysis, thus gaining valuable time for intervention and decision-making. It overcomes the limitations of single-parameter assessments and, by integrating multi-dimensional physiological and environmental information, makes the assessment results more comprehensive and reliable. The system can adapt to the physiological characteristics of different personnel, achieving precise individual monitoring and providing scientific data support for optimizing work processes and improving overall operational safety and efficiency. This method aims to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting the operational capabilities of underwater workers based on physiological signs, comprising: a data acquisition and preprocessing module, a feature fusion and processing module, and an intelligent prediction and decision-making module; Multi-source data acquisition: Through sensor modules integrated into underwater operation equipment, multiple physiological data of operators are collected in real time, including at least heart rate, blood oxygen saturation, body temperature, respiratory rate and body movement data; Feature fusion and processing: Simultaneously acquire underwater environmental data, which includes at least the current water depth, water temperature, and operation duration; Work capacity index calculation: The collected real-time physiological and environmental data are preprocessed and feature extracted, and then fused to form a comprehensive feature vector; Status prediction and early warning judgment: The comprehensive feature vector is input into the pre-trained work capability prediction model to calculate the work capability index at the current moment. This index is a normalized value used to quantify the comprehensive work capability level of personnel. Generate tiered early warnings: Based on current and historical operational capability indices, perform trend analysis to predict capability changes over a future period; when the current index or predicted trend falls below a preset threshold, generate and send early warning information of different levels to operational personnel and the surface monitoring platform.

[0008] Preferably, the data acquisition and preprocessing module includes multi-source sensor array data acquisition, signal conditioning and preliminary cleaning, data synchronization and standardization, and generation of standardized data packets; Data collection mainly includes physiological characteristic data and underwater environmental data; The physiological characteristic data include heart rate, blood oxygen, body temperature, and respiration; The heart rate acquisition uses a waterproof patch electrode or a photoplethysmography (PPG) sensor integrated into the chest strap of a diving suit to continuously monitor the heart rate. The blood oxygen saturation data is acquired using a finger clip or wrist-worn blood oxygen probe. The body temperature is monitored in real time using a high-precision digital temperature sensor, which is fixed to the skin surface in the form of a patch. The breathing rate is integrated into the air supply duct of the full-face respirator using a hot-wire flow sensor. The underwater environmental data includes water depth, water temperature, and time. The current operating water depth and ambient water temperature are obtained in real time through the depth sensor and temperature sensor built into the diving computer. The signal conditioning and preliminary cleaning: The analog signal is initially filtered by hardware circuits (such as an RC low-pass filter) to remove high-frequency interference; In the microprocessor, digital signal processing (DSP) algorithms are used for deep cleaning; Set a reasonable range of physiological data, and automatically identify and eliminate obvious abnormal values ​​caused by momentary poor contact of the sensor; Data synchronization and standardization: All sensor data are stamped with a uniform and precise timestamp to ensure that data such as heart rate, respiration, and body movement are strictly aligned on the timeline; The Z-Score standardization method is used to convert data of different dimensions and magnitudes (such as heart rate in "beats / minute" and blood oxygen in "percentage") into dimensionless values. The specific formula is as follows: x′=x μσx′=σx μ Where x is the original data, μ is the mean of the data, and σ is the standard deviation; Generate standardized data packets: The module will package and generate a structured, high-quality "comprehensive data frame," ready to be transmitted to the next feature fusion and processing module for deep feature extraction. Through the meticulous processing of this module, raw and coarse sensor signals are transformed into reliable, clean, and well-organized data streams.

[0009] Preferably, the generated graded early warning mainly includes attention level, warning level, and danger level.

[0010] Preferably, the feature fusion and processing process includes multi-dimensional feature extraction, multi-modal temporal alignment and correlation analysis, feature dimensionality reduction and screening, and construction of a comprehensive feature vector; S1. Extract deeper temporal and nonlinear characteristics from the synchronized standardized data stream; S2. Calculate the cross-correlation function between different signal sequences, find the possible delay relationship between them, and use dynamic time warping or sliding time window techniques. S3. After feature extraction, the feature dimension will expand rapidly, and there may be redundant or irrelevant features. By using the built-in feature importance evaluation function of random forest and tree model, the contribution of each feature to the prediction ability is calculated, and the most critical feature subset is automatically selected and noisy features are removed. S4. The most important features selected are combined into a one-dimensional comprehensive feature vector in a fixed order and format. The comprehensive feature vector is the final output of this module.

[0011] Preferably, the state prediction and early warning judgment process includes real-time operation capability index calculation, future state prediction based on time series, early warning judgment based on multi-level thresholds, and instruction generation and transmission. S1. Input the comprehensive feature vector generated by the feature fusion and processing module into a pre-trained machine learning model in real time; The model outputs a normalized value between 0 and 100. S2. Using the LSTM model, the historical index sequence is taken as input, the change pattern is automatically learned, and the index value for a future period of time is calculated to form a prediction trajectory. S3. The system transforms the quantitative assessment and prediction results into clear operation instructions. If the current index is below a certain threshold, a corresponding warning is triggered immediately. If the predicted trajectory will cross a lower level threshold in the near future, a higher level warning is triggered in advance. S4. The module will eventually generate a structured warning instruction package from the decision results and send it to the warning and feedback execution module in real time through the system interface to issue an alarm.

[0012] Compared with the prior art, the beneficial effects of the present invention are: This invention achieves a forward-looking and individualized assessment of operational capabilities through multimodal data fusion and machine learning models. It can issue early warnings before a substantial decline in personnel capabilities through trend analysis, thus gaining valuable time for intervention and decision-making. It overcomes the limitations of single-parameter assessment and, by integrating multi-dimensional physiological and environmental information, makes the assessment results more comprehensive and reliable. The system can adapt to the physiological characteristics of different personnel, achieve precise individual monitoring, and provide scientific data support for optimizing work processes and improving overall operational safety and efficiency. Attached Figure Description

[0013] Figure 1 This is a diagram illustrating the overall framework of the underwater worker capability prediction method of the present invention. Figure 2 This is a flowchart illustrating the core workflow of the data acquisition and preprocessing module of this invention. Figure 3 This is a flowchart of the state assessment-trend prediction-intelligent decision-making process of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] Please see Figures 1-3 The present invention provides a technical solution: a method for predicting the operational capabilities of underwater workers based on physiological signs, comprising: a data acquisition and preprocessing module, a feature fusion and processing module, and an intelligent prediction and decision-making module; Multi-source data acquisition: Through sensor modules integrated into underwater operation equipment, multiple physiological data of operators are collected in real time, including at least heart rate, blood oxygen saturation, body temperature, respiratory rate and body movement data; Feature fusion and processing: Simultaneously acquire underwater environmental data, which includes at least the current water depth, water temperature, and operation duration; Work capacity index calculation: The collected real-time physiological and environmental data are preprocessed and feature extracted, and then fused to form a comprehensive feature vector; Status prediction and early warning judgment: The comprehensive feature vector is input into the pre-trained work capability prediction model to calculate the work capability index at the current moment. This index is a normalized value used to quantify the comprehensive work capability level of personnel. Generate tiered early warnings: Based on current and historical operational capability indices, perform trend analysis to predict capability changes over a future period; when the current index or predicted trend falls below a preset threshold, generate and send early warning information of different levels to operational personnel and the surface monitoring platform.

[0016] The data acquisition and preprocessing module includes multi-source sensor array data acquisition, signal conditioning and preliminary cleaning, data synchronization and standardization, and generation of standardized data packets. The data acquisition and preprocessing module acquires physiological signals such as heart rate and blood oxygen and environmental parameters such as water depth and water temperature in real time through multiple sensors integrated into the diving equipment, and completes data filtering, noise reduction and standardization.

[0017] Data collection mainly includes physiological characteristic data and underwater environmental data; The physiological characteristic data include heart rate, blood oxygen, body temperature, and respiration; The heart rate is collected using waterproof patch electrodes or photoplethysmography (PPG) sensors integrated into the chest strap of a diving suit, continuously monitoring the heart rate, which is a key indicator for assessing physiological load and stress state.

[0018] The blood oxygen saturation data acquisition utilizes a finger clip or wrist-worn blood oxygen probe, which non-invasively measures arterial blood oxygen saturation (SpO2) using red light and infrared light spectroscopy to directly determine oxygenation status and provide early warning of the risk of hypoxia or oxygen toxicity.

[0019] The body temperature sensor uses a high-precision digital temperature sensor, which is fixed to the skin surface in the form of a patch to monitor changes in core body temperature or body surface temperature in real time, preventing the body temperature from being too low or too high.

[0020] The breathing rate is measured by a hot-wire flow sensor integrated into the air supply duct of the full-face respirator, which accurately measures the cycle and depth of each breath.

[0021] The underwater environmental data includes water depth, water temperature, and time. The current operating water depth and ambient water temperature are obtained in real time through the depth sensor and temperature sensor built into the diving computer.

[0022] The signal conditioning and preliminary cleaning: The analog signal is initially filtered by hardware circuits (such as an RC low-pass filter) to remove high-frequency interference; In the microprocessor, digital signal processing (DSP) algorithms are used for deep cleaning. For example, moving average filtering is used to remove motion artifacts from PPG signals, and wavelet transform is used to remove power frequency interference and baseline drift caused by water flow impact from ECG signals.

[0023] Set a reasonable range of physiological data, and automatically identify and eliminate obvious abnormal values ​​caused by momentary poor contact of the sensor.

[0024] Data synchronization and standardization: All sensor data are stamped with a unified and precise timestamp to ensure that data such as heart rate, respiration, and body movement are strictly aligned on the timeline, laying the foundation for analyzing their coordinated changes.

[0025] The Z-Score standardization method is used to convert data of different dimensions and magnitudes (such as heart rate in "beats / minute" and blood oxygen in "percentage") into dimensionless values. The specific formula is as follows: x′=x μσx′=σx μ Where x is the original data, μ is the mean of the data, and σ is the standard deviation. All features are on the same comparable scale, which greatly improves the convergence speed and performance of subsequent machine learning models.

[0026] Generate standardized data packets: The module will package and generate a structured, high-quality "comprehensive data frame," ready to be transmitted to the next feature fusion and processing module for deep feature extraction. Through the meticulous processing of this module, raw and coarse sensor signals are transformed into reliable, clean, and well-organized data streams.

[0027] The generation of graded early warnings mainly includes the attention level, warning level, and danger level.

[0028] The feature fusion and processing module includes multi-dimensional feature extraction, multi-modal temporal alignment and correlation analysis, feature dimensionality reduction and filtering, and construction of a comprehensive feature vector. This module is the intelligent hub of the system, and its core task is to transform the preprocessed multi-source heterogeneous data into a comprehensive feature vector that can fully and deeply represent the current physiological state of the workers. Through temporal alignment, feature extraction, and information fusion, the feature fusion and processing module provides high-value, structured input for subsequent prediction models. Its core workflow is as follows: We can extract deeper temporal and nonlinear features from the synchronized, standardized data stream.

[0029] Calculate the cross-correlation function between different signal sequences, look for possible delay relationships between them, and use dynamic time warping or sliding time window techniques to ensure that different features with causal or synergistic relationships are truly "aligned" in the same analysis unit in a physiological sense during feature fusion.

[0030] After feature extraction, the feature dimension will expand rapidly, and there may be redundant or irrelevant features. By using the built-in feature importance evaluation function of random forest and tree models, the contribution of each feature to the prediction ability is calculated, the most critical feature subset is automatically selected, and noisy features are removed.

[0031] The most important features selected are combined into a one-dimensional comprehensive feature vector in a fixed order and format. The comprehensive feature vector is the final output of this module. It retains the effective components of multi-source information to the maximum extent, eliminates redundancy and noise, and provides the best "fuel" for the downstream intelligent prediction and decision-making module. The quality of this vector directly determines the upper limit of the prediction performance of the entire system.

[0032] This module is the intelligent core and decision-making center of this invention. Its core task is to simulate expert judgment based on the comprehensive feature vectors provided by the preceding modules, to achieve a quantitative assessment of the current capabilities of the operators and an accurate prediction of future risks, and to make early warning decisions accordingly. Its workflow is a typical closed loop of "state assessment - trend prediction - intelligent decision-making".

[0033] The state prediction and early warning judgment process includes real-time operational capability index calculation, future state prediction based on time series, early warning judgment based on multi-level thresholds, and instruction generation and transmission.

[0034] The comprehensive feature vector generated by the feature fusion and processing module is input into a pre-trained machine learning model in real time. This model, with a Long Short-Term Memory (LSTM) network at its core, can effectively learn the dependence and dynamic patterns of physiological parameters over time.

[0035] The model outputs a normalized value between 0 and 100, known as the work performance index. This index is a comprehensive scale; a higher score indicates a better physiological state for the worker and a greater potential to maintain high-intensity, high-attention work. For example, an index of 85 represents a good state, while an index of 55 indicates a significant decline in ability, requiring attention.

[0036] By using the LSTM model as input, historical index sequences are automatically learned to predict their patterns of change and to calculate the index value for a future period, thus forming a predictive trajectory.

[0037] The system translates quantitative assessment and prediction results into explicit operational instructions. The system has a pre-defined tiered warning threshold library, typically containing at least three levels. Attention level (threshold=75): Indicates that the state begins to deviate from the optimal range.

[0038] Warning level (threshold=60): Indicates that the capability has substantially declined and the risk of an accident has increased.

[0039] Danger level (threshold=40): Indicates a highly dangerous situation requiring immediate intervention.

[0040] The early warning decision is triggered by both the current state and the predicted trend. If the current index is below a certain threshold, the corresponding early warning is triggered immediately. If the predicted trajectory will cross a lower level threshold in the near future, a higher level early warning is triggered in advance.

[0041] The module ultimately generates a structured warning instruction package from the decision results and sends it to the warning and feedback execution module in real time through the system interface to issue an alarm.

[0042] This invention achieves a forward-looking and individualized assessment of operational capabilities through multimodal data fusion and machine learning models. It can issue early warnings before a substantial decline in personnel capabilities through trend analysis, thus gaining valuable time for intervention and decision-making. It overcomes the limitations of single-parameter assessment and, by integrating multi-dimensional physiological and environmental information, makes the assessment results more comprehensive and reliable. The system can adapt to the physiological characteristics of different personnel, achieve precise individual monitoring, and provide scientific data support for optimizing work processes and improving overall operational safety and efficiency.

[0043] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the work capacity of underwater workers based on physiological signs, comprising: Data acquisition and preprocessing module, feature fusion and processing module, intelligent prediction and decision-making module; Multi-source data acquisition: Through sensor modules integrated into underwater operation equipment, multiple physiological data of operators are collected in real time, including at least heart rate, blood oxygen saturation, body temperature, respiratory rate and body movement data; Feature fusion and processing: Simultaneously acquire underwater environmental data, which includes at least the current water depth, water temperature, and operation duration; Work capacity index calculation: The collected real-time physiological and environmental data are preprocessed and feature extracted, and then fused to form a comprehensive feature vector; Status prediction and early warning judgment: The comprehensive feature vector is input into the pre-trained work capability prediction model to calculate the work capability index at the current moment. This index is a normalized value used to quantify the comprehensive work capability level of personnel. Generate tiered early warnings: Based on current and historical operational capability indices, perform trend analysis to predict capability changes over a future period; when the current index or predicted trend falls below a preset threshold, generate and send early warning information of different levels to operational personnel and the surface monitoring platform.

2. The method for predicting the work capacity of underwater workers based on physiological signs according to claim 1, characterized in that: The data acquisition and preprocessing module includes multi-source sensor array data acquisition, signal conditioning and preliminary cleaning, data synchronization and standardization, and generation of standardized data packets; Data collection mainly includes physiological characteristic data and underwater environmental data; The physiological characteristic data include heart rate, blood oxygen, body temperature, and respiration; The heart rate acquisition uses a waterproof patch electrode or a photoplethysmography (PPG) sensor integrated into the chest strap of a diving suit to continuously monitor the heart rate. The blood oxygen saturation data is acquired using a finger clip or wrist-worn blood oxygen probe. The body temperature is monitored in real time using a high-precision digital temperature sensor, which is fixed to the skin surface in the form of a patch. The breathing rate is integrated into the air supply duct of the full-face respirator using a hot-wire flow sensor. The underwater environmental data includes water depth, water temperature, and time. The current operating water depth and ambient water temperature are obtained in real time through the depth sensor and temperature sensor built into the diving computer. The signal conditioning and preliminary cleaning: The analog signal is initially filtered by hardware circuits (such as an RC low-pass filter) to remove high-frequency interference; In the microprocessor, digital signal processing (DSP) algorithms are used for deep cleaning; Set a reasonable range of physiological data, and automatically identify and eliminate obvious abnormal values ​​caused by momentary poor contact of the sensor; Data synchronization and standardization: All sensor data are stamped with a uniform and precise timestamp to ensure that data such as heart rate, respiration, and body movement are strictly aligned on the timeline; The Z-Score standardization method is used to convert data of different dimensions and magnitudes (such as heart rate in "beats / minute" and blood oxygen in "percentage") into dimensionless values. The specific formula is as follows: x′=x μσx′=σx m Where x is the original data, μ is the mean of the data, and σ is the standard deviation; Generate standardized data packets: The module will package and generate a structured, high-quality "comprehensive data frame", ready to be transmitted to the next feature fusion and processing module for deep feature extraction; Through the meticulous processing of this module, raw and coarse sensor signals are transformed into reliable, clean, and well-organized data streams.

3. The method for predicting the work capacity of underwater workers based on physiological signs according to claim 1, characterized in that: The generated graded early warning mainly includes the attention level, warning level, and danger level.

4. The method for predicting the work capacity of underwater workers based on physiological signs according to claim 1, characterized in that: The feature fusion and processing process includes multi-dimensional feature extraction, multi-modal temporal alignment and correlation analysis, feature dimensionality reduction and filtering, and construction of a comprehensive feature vector; S1. Extract deeper temporal and nonlinear characteristics from the synchronized standardized data stream; S2. Calculate the cross-correlation function between different signal sequences, find the possible delay relationship between them, and use dynamic time warping or sliding time window techniques. S3. After feature extraction, the feature dimension will expand rapidly, and there may be redundant or irrelevant features. By using the built-in feature importance evaluation function of random forest and tree model, the contribution of each feature to the prediction ability is calculated, and the most critical feature subset is automatically selected and noisy features are removed. S4. The most important features selected are combined into a one-dimensional comprehensive feature vector in a fixed order and format. The comprehensive feature vector is the final output of this module.

5. The method for predicting the work capacity of underwater workers based on physiological signs according to claim 1, characterized in that: The state prediction and early warning judgment process includes real-time operation capability index calculation, future state prediction based on time series, early warning judgment based on multi-level thresholds, and instruction generation and transmission. S1. Input the comprehensive feature vector generated by the feature fusion and processing module into a pre-trained machine learning model in real time; The model outputs a normalized value between 0 and 100. S2. Using the LSTM model, the historical index sequence is taken as input, the change pattern is automatically learned, and the index value for a future period of time is calculated to form a prediction trajectory. S3. The system transforms the quantitative assessment and prediction results into clear operation instructions. If the current index is below a certain threshold, a corresponding warning is triggered immediately. If the predicted trajectory will cross a lower level threshold in the near future, a higher level warning is triggered in advance. S4. The module will eventually generate a structured warning instruction package from the decision results and send it to the warning and feedback execution module in real time through the system interface to issue an alarm.

Images