Old people balance training system integrating cardiopulmonary monitoring and anti-falling functions

Abstract

The invention relates to the technical field of biomedical engineering and intelligent rehabilitation, in particular to an old people balance training system integrating cardiopulmonary monitoring and anti-falling functions. The system comprises a multi-modal data acquisition unit, a physiological load evaluation unit, a neuromuscular fatigue evaluation unit, a dynamic stability evaluation unit, a coupling risk fusion prediction unit and an adaptive training regulation unit. The system synchronously collects physiological, biomechanical and electromyographic signals of a user; the core of the method is to respectively calculate cardiopulmonary reserve consumption, track a myoelectricity median frequency descending trend and evaluate a dynamic stability boundary, and establish a fusion safety model based on machine learning to generate a prospective comprehensive risk index; the system dynamically regulates the training difficulty according to the index; according to the method, the defect of dependence on apparent indexes is overcome, compensation behaviors of the user can be penetrated, and the real internal physical state of the user can be accurately obtained.

Description

Technical Field

[0001] This invention relates to the fields of biomedical engineering and intelligent rehabilitation technology, specifically to a balance training system for the elderly that integrates cardiopulmonary monitoring and fall prevention functions. Background Technology

[0002] With the increasing aging of the population, balance training for the elderly is crucial for fall prevention. However, ensuring both safety and effectiveness during training is a major challenge in the rehabilitation field. Currently, most existing training systems rely on apparent indicators such as swing speed or task completion for evaluation. However, these apparent indicators often fail to reflect the user's true internal physical state. More importantly, existing technologies generally neglect the nonlinear coupling relationship between physiological load and biomechanical stability. This coupling can amplify risks, and traditional systems only perform passive monitoring, failing to provide predictive intervention before instability or physiological overload events occur. Therefore, how to integrate multimodal data to construct a model that can proactively predict physiological-balance coupling risks and thereby achieve adaptive closed-loop regulation has become an urgent problem to be solved in this field. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a balance training system for the elderly that integrates cardiopulmonary monitoring and fall prevention functions. Specifically, the technical solution of this invention includes:

[0004] A multimodal data acquisition layer is used to simultaneously acquire users' physiological data, biomechanical data, and electromyographic signals;

[0005] The physiological load assessment unit is used to calculate the percentage of cardiopulmonary reserve depletion based on the physiological data and obtain a physiological load status signal.

[0006] The neuromuscular fatigue assessment unit is used to track the downward trend of the median frequency based on electromyographic signals; the neuromuscular fatigue assessment unit is also used to qualitatively transform the downward trend of the median frequency into a muscle fatigue index to obtain a fatigue state signal.

[0007] The dynamic stability assessment unit is used to calculate the user's stability boundary based on the biomechanical data and obtain an instantaneous stability signal.

[0008] The coupled risk fusion prediction unit is used to receive the physiological load state signal, the fatigue state signal and the instantaneous stability signal, and calculate and generate a forward-looking comprehensive risk index through a preset machine learning-based physiological-balance fusion safety model.

[0009] An adaptive training control unit is used to dynamically adjust the training difficulty based on the comprehensive risk index and generate immediate decision instructions.

[0010] Preferably, the physiological data includes heart rate, heart rate variability, and blood oxygen saturation obtained through a continuous ECG sensor and a SpO2 sensor; the biomechanical data includes pressure center data obtained through a force-sensitive sensor array, and posture and angular velocity data obtained through an inertial sensor; and the electromyographic signals are obtained through a surface electromyography sensor.

[0011] Preferably, the physiological load assessment unit is used to process data based on a physiological control theory model to calculate the degree of proximity of the user's current cardiopulmonary load to a preset maximum safe load threshold, so as to obtain the percentage of cardiopulmonary reserve depletion.

[0012] Preferably, the neuromuscular fatigue assessment unit is used to continuously track the decreasing trend of the median frequency of the electromyographic signal using a time-frequency analysis method.

[0013] Preferably, the dynamic stability assessment unit is used to calculate, in real time, dynamic physical quantities characterizing the robustness of the user's current posture based on biomechanics and dynamic system stability theory, so as to obtain the stability boundary.

[0014] Preferably, the machine learning-based physiological-equilibrium fusion safety model is a pre-trained model; the pre-trained model is trained on offline training data of near-instability and physiological overload events to learn coupling feature patterns before the risk occurs.

[0015] Preferably, the coupled risk fusion prediction unit is used to specifically generate a future instability probability signal and a physiological overload risk signal; the coupled risk fusion prediction unit is also used to integrate the future instability probability signal and the physiological overload risk signal into the comprehensive risk index.

[0016] Preferably, the adaptive training control unit is used for:

[0017] The comprehensive risk index is compared with the preset safety training threshold and the preset critical risk threshold.

[0018] When the comprehensive risk index is lower than the safety training threshold, a training difficulty increase instruction is generated;

[0019] When the comprehensive risk index approaches or exceeds the critical risk threshold, an instruction to immediately reduce the training difficulty or an active safety intervention instruction is generated.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] 1. This system simultaneously collects multimodal data such as physiological, biomechanical, and electromyographic data, and assesses cardiopulmonary reserve depletion, neuromuscular fatigue trend, and dynamic stability boundary respectively. This invention overcomes the shortcomings of existing technologies that rely on apparent indicators, and can penetrate user compensatory behavior to accurately obtain their true internal physical state.

[0022] 2. This system constructs a physiological-equilibrium fusion safety model based on machine learning. This model learns the coupling feature patterns between physiological load, muscle fatigue and dynamic stability through offline training, which solves the problem that existing technologies ignore the nonlinear coupling relationship between physiological and equilibrium, leading to the amplification of risks.

[0023] 3. The coupled risk fusion prediction unit of this system can calculate and generate a forward-looking comprehensive risk index, which includes predictions of the probability of future instability and the risk of physiological overload. This realizes the transformation from passive monitoring to active prediction and enables intervention before adverse events such as instability or physiological overload occur.

[0024] 4. This system uses an adaptive training control unit to compare a forward-looking comprehensive risk index with preset safety and critical thresholds, thereby achieving dynamic closed-loop control of training difficulty. When the risk is low, the difficulty is increased to ensure effectiveness; when the risk is high, immediate intervention is provided to ensure safety, thus improving the effectiveness of rehabilitation training while ensuring safety. Attached Figure Description

[0025] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0026] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0028] Example 1:

[0029] Please see Figure 1 A balance training system for the elderly integrating cardiopulmonary monitoring and fall prevention functions, comprising:

[0030] A multimodal data acquisition layer is used to simultaneously acquire users' physiological data, biomechanical data, and electromyographic signals;

[0031] The physiological load assessment unit is used to calculate the percentage of cardiopulmonary reserve depletion based on physiological data and obtain a physiological load status signal.

[0032] The neuromuscular fatigue assessment unit is used to track the downward trend of the median frequency based on electromyographic signals; the neuromuscular fatigue assessment unit is also used to qualitatively transform the downward trend of the median frequency into a muscle fatigue index to obtain a fatigue state signal.

[0033] The dynamic stability assessment unit is used to calculate the user's stability boundary based on biomechanical data and obtain an instantaneous stability signal;

[0034] The coupled risk fusion prediction unit is used to receive physiological load state signals, fatigue state signals and instantaneous stability signals, and calculate them through a preset machine learning-based physiological-balance fusion safety model to generate a forward-looking comprehensive risk index.

[0035] The adaptive training control unit is used to dynamically adjust the training difficulty based on the comprehensive risk index and generate real-time decision instructions.

[0036] This invention provides a balance training system for the elderly that integrates cardiopulmonary monitoring and fall prevention functions. When the system is running, a multimodal data acquisition layer is activated to acquire multi-dimensional raw data required to assess the user's hidden physical layer status. This layer is configured to synchronously and frequently acquire three-dimensional raw data streams from the user: physiological data, biomechanical data, and electromyographic signals.

[0037] These three data streams are distributed in parallel to three independent assessment units. The physiological load assessment unit receives the physiological data, and its core purpose is not simply to list values ​​such as heart rate, but to transform it into an indicator with safety warning significance. In this embodiment, the unit processes the data based on a built-in physiological control theory model to calculate a quantified percentage of cardiopulmonary reserve depletion. The percentage of cardiopulmonary reserve depletion is a quantitative indicator used to characterize how close the user's current cardiopulmonary load is to their individual maximum safe load threshold. The maximum safe load threshold is a configurable safety baseline, which is preset by those skilled in the art based on conventional rehabilitation medicine consensus and combined with the user's individual health condition. The unit finally outputs a physiological load status signal.

[0038] Meanwhile, the neuromuscular fatigue assessment unit receives electromyographic signals to quantify and assess muscle fatigue as a key risk amplifier. The unit continuously tracks the downward trend of the median frequency of the electromyographic signals using time-frequency analysis. The unit further qualitatively processes this physical trend, for example, by mapping the slope or amplitude of the downward trend to a normalized fatigue scale, thereby generating a fatigue state signal, namely the muscle fatigue index.

[0039] Meanwhile, the dynamic stability assessment unit receives biomechanical data to assess the user's immediate, physical instability risk. Based on biomechanics and dynamic system stability theory, this unit calculates the user's stability boundary in real time. The stability boundary is a dynamic physical quantity, such as quantified based on the instantaneous distance between the extrapolated center of mass and the support base boundary, to characterize the robustness of the user's current posture, rather than a static equilibrium fraction. The unit ultimately outputs an instantaneous stability signal.

[0040] The core inventive information flow of this invention lies in the coupled risk fusion prediction unit. Its purpose is not to simply superimpose or isolate the three signals mentioned above, but rather to use the outputs of these evaluation units as input to a prediction model, proactively predicting the instantaneous collapse risk caused by the nonlinear coupling of these three factors. This unit receives the aforementioned physiological load state signal, fatigue state signal, and instantaneous stability signal as its multi-channel input. It performs calculations through a pre-defined machine learning-based physiological-equilibrium fusion safety model. This model is constructed by learning the coupling characteristic patterns of near-instability and physiological overload events, enabling it to anticipate risks. Ultimately, this unit generates a forward-looking comprehensive risk index.

[0041] The closed loop of the system is reflected in the adaptive training control unit, which receives the comprehensive risk index as its core control input. Its purpose is to realize the dynamic closed loop of predictive risk-training load. The unit compares the comprehensive risk index with the preset threshold, dynamically adjusts the training difficulty, and generates real-time decision instructions to control the execution mechanism of the training platform in reverse.

[0042] Through the collaborative work of the aforementioned units, this invention constructs a complete technical chain from multimodal data acquisition to independent state assessment, then to coupled risk prediction, and finally to adaptive closed-loop regulation. The essential difference and core advantage of this invention are that it solves the problems of apparent indicators being disconnected from fundamental value and the risk amplification caused by neglecting physiological-balance coupling in existing technologies. With its predictive fusion model and closed-loop regulation mechanism, this system realizes the transformation from passive monitoring to active prediction, and can intervene before the user experiences instability or physiological overload, greatly improving the safety and effectiveness of balance training for the elderly.

[0043] Example 2:

[0044] Physiological data include heart rate, heart rate variability, and blood oxygen saturation obtained through continuous ECG and SpO2 sensors; biomechanical data include center of pressure data obtained through a force-sensitive sensor array, and posture and angular velocity data obtained through inertial sensors; electromyographic signals are obtained through surface electromyography sensors.

[0045] In a further implementation of the system of Example 1, the specific configuration of the multimodal data acquisition layer is defined; the purpose of this layer is to acquire multidimensional raw data required to evaluate the user's hidden physical layer state.

[0046] To achieve this, the layer is configured to use a specific combination of sensors and ensure synchronized data acquisition:

[0047] Physiological data is acquired through continuous ECG and SpO2 sensors placed at appropriate locations on the user's body. The ECG sensor continuously captures electrocardiogram signals to extract heart rate and heart rate variability, while the SpO2 sensor acquires blood oxygen saturation. These three sets of data together form the basis for assessing cardiopulmonary load.

[0048] Biomechanical data acquisition is accomplished through the collaboration of two types of sensors: a force-sensitive sensor array placed below the training platform to collect real-time data on the user's center of pressure; and inertial sensors worn by the user at key points to acquire attitude and angular velocity data. These two sets of data together form the basis for evaluating dynamic stability.

[0049] Electromyographic signals are acquired by attaching surface electromyography sensors to key lower limb muscle groups of the user; these signals form the basis for assessing neuromuscular fatigue.

[0050] By employing a specific and complementary combination of sensors, including ECG, SpO2, force-sensitive arrays, IMU, and sEMG, this invention ensures that the collected data fully covers the three key dimensions of cardiopulmonary, biomechanical, and muscular. This simultaneous acquisition of multimodal data is a necessary prerequisite for subsequent physiological-balance coupling risk prediction, guaranteeing the comprehensiveness and high quality of the input data, thereby improving the accuracy of the entire system assessment.

[0051] Example 3:

[0052] The physiological load assessment unit is used to process data based on a physiological control theory model, calculate the degree to which the user's current cardiopulmonary load is close to the preset maximum safe load threshold, and obtain the percentage of cardiopulmonary reserve depletion.

[0053] In a further implementation of the system of Example 1, the internal processing logic of the physiological load assessment unit was defined; as before, the purpose of this unit is to transform the raw physiological signal into a load index with safety boundary significance.

[0054] To achieve this, the unit does not simply display raw values, but is configured to process data based on a physiological control theory model. As a specific, non-limiting implementation, the physiological control theory model is a dynamic load calculation model based on the heart rate reserve method. This model takes a user-preset resting heart rate and maximum safe heart rate as input and receives real-time heart rate data. Based on this, the model calculates the percentage of cardiopulmonary reserve depletion. This percentage serves as a physiological load status signal; the model receives physiological data streams from the acquisition layer as input; the core of this calculation lies in determining how close the user's current cardiopulmonary load is to a preset maximum safe load threshold.

[0055] Here, the maximum safe load threshold must be explicitly defined and set: it is not a fixed physiological value, but a configurable safety baseline pre-set by those skilled in the art based on conventional rehabilitation medicine consensus and in combination with the user's individual health condition; for example, the threshold can be initially set as a target heart rate range based on the Karvonen formula or heart rate reserve method recognized in the art, combined with the user's resting heart rate and age-estimated maximum heart rate, and then adjusted downward by the rehabilitation therapist according to the user's specific medical history to ensure safety;

[0056] The role of the physiological control theory model is not merely to check whether the heart rate exceeds the threshold, but to dynamically assess the rate and state at which the individual approaches the threshold using the current complete physiological data; the percentage of cardiopulmonary reserve depletion output by this unit is a quantitative representation of the aforementioned degree of approach; this signal is then fed to the coupled risk fusion prediction unit.

[0057] By employing a physiological control theory model and introducing an individualized maximum safe load threshold, this invention solves the problem of the disconnect between physiological monitoring and safety risk in existing technologies. It makes the output physiological load status signal no longer an isolated value, but a decision-making basis with a clear safety boundary meaning that can be used for risk prediction, significantly enhancing the system's sensitivity and accuracy to individual physiological overload risk.

[0058] Example 4:

[0059] The neuromuscular fatigue assessment unit is used to continuously track the downward trend of the median frequency of electromyographic signals using time-frequency analysis methods.

[0060] The dynamic stability assessment unit is used to calculate dynamic physical quantities that characterize the robustness of the user's current attitude in real time, based on biomechanics and dynamic system stability theory, in order to obtain the stability boundary.

[0061] To further clarify the specific implementation of each evaluation unit in the system of Example 1, this example provides a detailed explanation of the working mechanism of the neuromuscular fatigue evaluation unit and the dynamic stability evaluation unit;

[0062] The purpose of the neuromuscular fatigue assessment unit is to quantify muscle fatigue, a key risk factor that leads to a decline in balance control. This unit receives surface electromyography (sEMG) signal streams from the acquisition layer as input. Its core processing logic involves calculating the power spectrum of the sEMG signal within a continuous time window using time-frequency analysis and continuously tracking the decreasing trend of the median frequency (MF). It should be understood that the decreasing trend of MF is a recognized physical characteristic of muscle fatigue. The unit then qualitatively defines this trend as a muscle fatigue index. For example, this index can be set as a normalized index of 0-100%, where the index is mapped to high fatigue when the MF decrease slope reaches a predetermined statistical significance level. The unit generates a fatigue state signal and sends it to the prediction unit.

[0063] The purpose of the dynamic stability assessment unit is to evaluate how far the user is from the point of irreversible instability, i.e., to determine the immediate risk of instability. This unit receives pressure center data and inertial sensor data streams from the acquisition layer as input. Its core processing logic involves calculating a dynamic physical quantity that characterizes the robustness of the user's current posture in real time, based on biomechanical and dynamic system stability theories, to obtain the stability boundary. In this embodiment, this boundary is calculated based on the relationship between the center of mass and the supporting base surface. The system uses IMU data to estimate the position and velocity of the user's center of mass and extrapolates it by a time step to obtain the extrapolated center of mass. Simultaneously, it uses force-sensitive sensor array data to determine the boundary of the supporting base surface. The stability boundary is quantified as the instantaneous distance between the extrapolated center of mass and the nearest supporting base surface boundary. A small or negative stability boundary indicates an extremely high risk of immediate instability. This unit ultimately generates an instantaneous stability signal and sends it to the prediction unit.

[0064] Through the above configuration, the present invention adopts recognized and objective physical quantities in the field to quantify fatigue and stability, replacing the practice of relying on apparent indicators such as completion rate or swaying speed in the prior art; this evaluation method based on the physical and biomechanical layers can penetrate the fog of the user's compensatory posture and obtain more realistic internal state signals, providing reliable and accurate input for subsequent coupling risk prediction.

[0065] Example 5:

[0066] The physiological-equilibrium fusion safety model based on machine learning is a pre-trained model. The pre-trained model is trained on offline training data of near-instability and physiological overload events to learn the coupling feature patterns before the risk occurs.

[0067] The coupled risk fusion prediction unit is used to specifically generate future instability probability signals and physiological overload risk signals; the coupled risk fusion prediction unit is also used to integrate the future instability probability signals and physiological overload risk signals into a comprehensive risk index.

[0068] This embodiment will focus on the core innovative unit of the system in Embodiment 1, namely the coupled risk fusion prediction unit, and its internal physiological-balance fusion safety model and working logic. The purpose of this unit is to predict the catastrophic consequences that will occur in the very short future after the nonlinear coupling amplification of physiological load, muscle fatigue and dynamic stability.

[0069] To achieve this objective, the machine learning-based physiological-equilibrium fusion safety model built into this unit is configured as a pre-trained model. The construction of this pre-trained model is described below: it is not based on a simple IF-THEN rule, but rather built by training on offline training data of near-instability and physiological overload events. As a specific, non-limiting implementation, the machine learning-based model can employ various models known to those skilled in the art, such as, but not limited to, support vector machines, random forests, or a recurrent neural network or long short-term memory network capable of handling time-series relationships. The model's input feature vector specifically includes, but is not limited to: the current physiological load state signal, fatigue state signal, and instantaneous stability signal; preferably, it also includes statistical characteristics of these three signals over a short time window, such as mean, variance, and slope of change, to enable the model to capture dynamic trends. It should be understood that offline training data can be pre-acquired through controlled experiments achievable by those skilled in the art or through validated biomechanical simulation models.

[0070] Through this training, the model learns coupled characteristic patterns prior to risk occurrence. To further clarify, coupled characteristic patterns refer to, for example, the model learning a combination of: Feature A: physiological load state signal showing 70%+; Feature B: fatigue state signal showing 60%+; Feature C: instantaneous stability signal rapidly decreasing by 0.1 meters in the past second. The predicted probability of instability in the next second is far greater than the sum of risks assessed independently by these three signals. This reflects the nonlinear amplification effect of physiological fatigue on the decline in balance control ability, which is something that existing technologies cannot capture.

[0071] During system operation, the coupled risk fusion prediction unit receives the output signals of the aforementioned three units as real-time input and uses this trained model for forward-looking calculations; the output of this unit is no longer the lagging current state, but rather two specific, forward-looking prediction signals:

[0072] Signals of future instability probability;

[0073] Physiological overload risk signals;

[0074] This unit is also configured to integrate the future instability probability signal and the physiological overload risk signal into a comprehensive risk index; this integration logic is not a simple weighted summation, but is configured as a risk decision function; as a specific implementation, this risk decision function is configured to take the maximum value of the two input signals to achieve a veto logic; specifically, the function performs the following calculations:

[0075] Comprehensive Risk Index = Max (Signal of future instability probability, signal of physiological overload risk)

[0076] Among them, the future instability probability signal and the physiological overload risk signal are both normalized; this configuration ensures that the risk in any dimension will dominate the comprehensive risk index, forcibly raising it to the critical risk level, thereby prioritizing the cardiovascular safety of users in a clear and achievable way.

[0077] This comprehensive risk index is the final quantitative indicator that characterizes the overall security boundary of the user. It is output as the sole control signal to the adaptive training control unit.

[0078] By employing this offline-trained machine learning model to fuse multimodal state signals and explicitly outputting two forward-looking signals—the probability of future instability and the risk of physiological overload—this invention achieves a qualitative leap from state monitoring to risk prediction. This design can accurately capture the instantaneous collapse risk generated by physiological-equilibrium coupling, solving the fundamental defect of existing technologies that cannot predict risks, and is the core technological foundation for realizing subsequent adaptive safety control.

[0079] Example 6:

[0080] The adaptive training control unit is used for:

[0081] The comprehensive risk index is compared with the preset safety training threshold and the preset critical risk threshold.

[0082] When the overall risk index is lower than the safe training threshold, an instruction to increase the training difficulty is generated.

[0083] When the comprehensive risk index approaches or exceeds the critical risk threshold, an instruction to immediately reduce the training difficulty or an active safety intervention instruction is generated.

[0084] In a further implementation of the system of Example 1, the specific decision-making logic of the adaptive training control unit was defined; the ultimate goal of this unit is to use the risk predicted by the previous unit to achieve personalized extreme safety training, that is, to dynamically push the user to the optimal neuroplasticity region where the user is about to become unstable but will not become unstable.

[0085] This unit receives the comprehensive risk index from the coupled risk fusion prediction unit as its core control input; its internal decision-making logic is as follows:

[0086] This unit continuously compares the comprehensive risk index with two preset thresholds; it should be understood that the setting of these thresholds is the core of the present invention to achieve safe training.

[0087] The safety training threshold can be set to a low overall risk index; this value indicates that the user is in an absolutely safe zone, the training load is insufficient, and the system should proactively increase the challenge.

[0088] The critical risk threshold is a key safety boundary that can be set as a high comprehensive risk index; this value corresponds to a boundary that a person skilled in the art can accept based on safety considerations as the highest predicted probability of instability or physiological overload.

[0089] These thresholds are not fixed, but can be dynamically optimized and personalized based on the user's historical training data;

[0090] This unit performs conditional judgments based on the comparison results and generates immediate decision instructions:

[0091] Condition 1: When the overall risk index is lower than the safe training threshold, this unit determines that the user's current training load is too low; at this time, it generates a training difficulty increase instruction to improve the training effect.

[0092] Condition 2: When the comprehensive risk index approaches or exceeds the critical risk threshold, the unit determines that the user is about to become unstable or experience physiological overload, and immediate intervention is necessary; at this time, it generates instructions to immediately reduce the training difficulty or to proactively intervene in safety matters.

[0093] These instructions are sent to the execution mechanism of the training platform, thereby realizing a dynamic coupling closed loop of predictive risk and training load;

[0094] By setting safe training thresholds and critical risk thresholds, and executing explicit decision-making logic to increase or decrease / intervene difficulty based on these thresholds, this invention truly achieves reverse control of training difficulty. This mechanism ensures that training always takes place within the individual's safety boundaries, both by increasing instructions to challenge the user's limits to improve rehabilitation efficiency and by intervening instructions to proactively avoid danger before it occurs. This forms a millisecond-level adaptive negative feedback safety loop, pioneering a new form of biological adaptive training driven by the user's real-time physiological-balance fusion risk.

[0095] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A balance training system for the elderly integrating cardiopulmonary monitoring and fall prevention functions, characterized in that, include: A multimodal data acquisition layer is used to simultaneously acquire the user's physiological data, biomechanical data, and electromyographic signals; The physiological load assessment unit is used to calculate the percentage of cardiopulmonary reserve depletion based on the physiological data and obtain a physiological load status signal. A neuromuscular fatigue assessment unit is used to track the decreasing trend of its median frequency based on electromyographic signals. The neuromuscular fatigue assessment unit is also used to qualitatively transform the downward trend of the median frequency into a muscle fatigue index to obtain a fatigue state signal. The dynamic stability assessment unit is used to calculate the user's stability boundary based on the biomechanical data and obtain an instantaneous stability signal. The coupled risk fusion prediction unit is used to receive the physiological load state signal, the fatigue state signal and the instantaneous stability signal, and calculate and generate a forward-looking comprehensive risk index through a preset machine learning-based physiological-balance fusion safety model. An adaptive training control unit is used to dynamically adjust the training difficulty based on the comprehensive risk index and generate immediate decision instructions.

2. The balance training system for the elderly integrating cardiopulmonary monitoring and fall prevention functions according to claim 1, characterized in that, The physiological data includes heart rate, heart rate variability, and blood oxygen saturation obtained through continuous ECG and SpO2 sensors; the biomechanical data includes pressure center data obtained through a force-sensitive sensor array, and posture and angular velocity data obtained through an inertial sensor; the electromyographic signals are obtained through a surface electromyography sensor.

3. The balance training system for the elderly integrating cardiopulmonary monitoring and fall prevention functions according to claim 1, characterized in that, The physiological load assessment unit is used to process data based on a physiological control theory model to calculate how close the user's current cardiopulmonary load is to a preset maximum safe load threshold, so as to obtain the percentage of cardiopulmonary reserve depletion.

4. The balance training system for the elderly integrating cardiopulmonary monitoring and fall prevention functions according to claim 1, characterized in that, The neuromuscular fatigue assessment unit is used to continuously track the decreasing trend of the median frequency of the electromyographic signal using a time-frequency analysis method.

5. A balance training system for the elderly integrating cardiopulmonary monitoring and fall prevention functions according to claim 1, characterized in that, The dynamic stability assessment unit is used to calculate, in real time, dynamic physical quantities that characterize the robustness of the user's current posture based on biomechanics and dynamic system stability theory, so as to obtain the stability boundary.

6. The balance training system for the elderly integrating cardiopulmonary monitoring and fall prevention functions according to claim 1, characterized in that, The machine learning-based physiological-equilibrium fusion safety model is a pre-trained model; the pre-trained model is trained on offline training data of near-instability and physiological overload events to learn coupling feature patterns before the risk occurs.

7. A balance training system for the elderly integrating cardiopulmonary monitoring and fall prevention functions according to claim 1, characterized in that, The coupled risk fusion prediction unit is used to specifically generate future instability probability signals and physiological overload risk signals; the coupled risk fusion prediction unit is also used to integrate the future instability probability signals and the physiological overload risk signals into the comprehensive risk index.

8. A balance training system for the elderly integrating cardiopulmonary monitoring and fall prevention functions according to claim 1, characterized in that, The adaptive training control unit is used for: The comprehensive risk index is compared with the preset safety training threshold and the preset critical risk threshold. When the comprehensive risk index is lower than the safety training threshold, a training difficulty increase instruction is generated; When the comprehensive risk index approaches or exceeds the critical risk threshold, an instruction to immediately reduce the training difficulty or an active safety intervention instruction is generated.

Images