Elastic socks capable of preventing lower limb venous thrombosis

Abstract

The invention belongs to the technical field of intelligent medical treatment, and provides a stretch sock for preventing lower limb venous thrombosis, comprising: a wrapping body having a pressure layer located on the outer side, a contact layer located on the inner side, and a functional layer located between the pressure layer and the contact layer; the physiological index sensor is arranged on the functional layer and used for collecting physiological index data of the user; the stimulating electrode is arranged on the contact layer, is in contact with the skin of the user and is used for stimulating muscles of the user; the controller is connected with the physiological index sensor and the stimulating electrode and used for receiving the physiological index data and generating a control signal of the stimulating electrode based on the physiological index data and the user baseline data so as to control the stimulating electrode to execute the stimulating parameters corresponding to the control signal. According to the stretch sock for preventing the lower limb venous thrombosis, personalized stimulation can be carried out based on the personalized physiological indexes of the current wearing user, and the prevention effect of the lower limb venous thrombosis is improved.

Description

Technical Field

[0001] This application relates to the field of intelligent medical technology, and in particular to an elastic stocking for preventing deep vein thrombosis in the lower extremities. Background Technology

[0002] Venous thrombosis is a highly dangerous clinical condition. It often develops insidiously, but once a thrombus breaks off, it can cause life-threatening conditions such as pulmonary embolism. For high-risk patients, such as those undergoing surgery, those with cancer, pregnant women, or those who are bedridden for extended periods, physical prevention using lower limb compression devices is necessary.

[0003] Currently, the conventional lower limb compression device is an intermittent air pump. This pump applies pressure to the lower limbs intermittently to promote venous return and reduce thrombus formation. However, in practice, healthcare professionals typically place the device on the user's lower limb and apply pressure intermittently using fixed parameters (such as fixed pressure and fixed interval inflation duration). This can lead to significant differences in user experience and effectiveness among different users, making it difficult to achieve personalized prevention of lower limb deep vein thrombosis.

[0004] Therefore, how to implement personalized venous thrombosis prevention for different users has become an urgent technical problem to be solved. Summary of the Invention

[0005] This application provides an elastic stocking for preventing lower extremity venous thrombosis, in order to solve the following technical problem in the prior art: how to provide personalized venous thrombosis prevention for different users.

[0006] According to a first aspect, this application provides an elastic stocking for preventing lower extremity deep vein thrombosis, comprising: a wrapping body having a pressure layer on the outer side and a contact layer on the inner side, and a functional layer located between the pressure layer and the contact layer; a physiological index sensor disposed in the functional layer for collecting physiological index data of a user; a stimulation electrode disposed in the contact layer and in contact with the user's skin for stimulating the user's muscles; and a controller connected to the physiological index sensor and the stimulation electrode respectively, for receiving the physiological index data and generating a control signal for the stimulation electrode based on the physiological index data and user baseline data, so as to control the stimulation electrode to execute stimulation parameters corresponding to the control signal.

[0007] In one embodiment, the physiological indicator sensor includes:

[0008] At least one of the following: lower limb physiological characteristic sensor, lower limb hemodynamic sensor, lower limb muscle function sensor, and behavioral state sensor, wherein...

[0009] The lower limb physiological feature sensor is used to collect the user's lower limb physiological feature data;

[0010] The lower limb hemodynamic sensor is used to collect the user's lower limb hemodynamic data;

[0011] The lower limb muscle function sensor is used to collect the user's lower limb muscle function data;

[0012] The controller is used to compare at least one of the lower limb physiological characteristic data, lower limb hemodynamic data, and lower limb muscle function data with the user baseline data to obtain physiological indicator change characteristics; and to generate the control signal based on the physiological indicator change characteristics.

[0013] In one embodiment, the controller deploys a reinforcement learning model based on the TinyML framework;

[0014] When the stimulation electrode executes the stimulation parameters, the controller optimizes the stimulation parameters for subsequent rounds using the reinforcement learning model based on the stimulation parameters of the current round and the physiological index change characteristics of the current round.

[0015] In one embodiment, the reinforcement learning model includes an Actor network and a Critic network;

[0016] The controller optimizes the stimulation parameters for subsequent rounds using the reinforcement learning model based on the stimulation parameters and physiological change characteristics of the current round, including:

[0017] The controller constructs a multi-dimensional input vector based on the physiological index change characteristics of the current round;

[0018] The multidimensional input vector is input into the Actor network to obtain the initial adjustment parameters;

[0019] The controller generates an adjustment control signal according to the initial adjustment parameters, so that the stimulation electrode executes the adjustment stimulation parameters corresponding to the adjustment control signal.

[0020] In one embodiment, the controller acquires the physiological index change characteristics of the stimulation electrode in the next round after the adjustment of stimulation parameters, calculates the parameter adjustment reward value based on the physiological index change characteristics in the next round, and updates the reinforcement learning model based on the adjustment reward value.

[0021] In one embodiment, the adjusted reward value includes an immediate reward and a cumulative reward;

[0022] The step of updating the reinforcement learning model based on the adjusted reward value includes:

[0023] The adjusted stimulus parameters, the physiological index change characteristics of the current round, the physiological index change characteristics of the next round, and the parameter adjustment reward value are combined into empirical data.

[0024] Obtain the timestamp and user tag when the adjustment of the stimulus parameters is performed, wherein the user tag includes user symptom and user behavior status data;

[0025] The timestamp and the user tag are bound to the experience data to obtain experience data with execution scenario tags;

[0026] The experience data with execution scenario labels is cached until a set quantity is reached, and the cumulative reward value is calculated based on the cached experience data.

[0027] The reinforcement learning model is updated based on the accumulated reward value.

[0028] In one embodiment, adjusting the reward value based on the physiological indicator change characteristics of the next round includes:

[0029] The parameter adjustment reward value is calculated based on the following formula:

[0030] R = w1·R1 + W2·R2 + ... + w n ·R n ;

[0031] Wherein, R is the parameter adjustment reward value, R1, R2...R n These represent the characteristics of physiological indicator changes, w1, w2…w n The physiological indicators are characterized by changes in R1, R2...R n The corresponding weighting coefficients are determined based on the user tags, and different user tags have different degrees of importance to the physiological indicator change characteristics.

[0032] In one embodiment, the Actor network is an ontology policy generation model under a federated learning framework. The Actor network has a shared layer and a personalized layer. The controller is also connected to a cloud server. The cloud server also deploys a global policy generation model. After the Actor network updates the reward value based on the parameters, it uploads the shared layer parameters to the cloud server for parameter aggregation and accepts the global model parameters issued by the cloud server to update the shared layer parameters.

[0033] In one embodiment, the physiological indicator sensor further includes:

[0034] A behavior state sensor is used to collect the user's behavior state, which includes sitting / lying down, standing, and walking states.

[0035] When the user wears the elastic stockings, the controller establishes first baseline data corresponding to the sitting / lying state, second baseline data corresponding to the standing state, and third baseline data corresponding to the walking state based on the pressure data and pulse wave data in the sitting / lying state, standing state, and walking state, respectively.

[0036] When determining the thrombosis warning result, the user's behavioral status is determined based on the real-time behavioral status information collected by the behavioral status sensor;

[0037] The baseline data is selected based on the user's current behavior state.

[0038] In one embodiment, the lower limb physiological characteristic sensor includes: a plurality of pressure sensors arranged in a ring in a first predetermined area of ​​the wrapping body, for collecting real-time pressure data of the first predetermined area of ​​the lower limb;

[0039] The lower limb hemodynamic sensor includes a pulse wave sensor, which is located in a second designated area of ​​the elastic wrapping body and is used to collect the user's pulse wave data.

[0040] In one embodiment, when the physiological indicator changes beyond a set amount, the controller outputs a stop control signal to stop the stimulation electrode from working.

[0041] According to a second aspect, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the control method of the controller as described in the first aspect above.

[0042] According to a third aspect, this application provides a controller including one or more processors, one or more memories, and one or more computer program instructions, which, when executed by the processor, implement the control method of the controller as described in the first aspect above.

[0043] The present invention has at least the following beneficial effects:

[0044] This invention proposes an elastic stocking for preventing lower extremity deep vein thrombosis, comprising: a wrapping body having an outer pressure layer and an inner contact layer, and a functional layer between the pressure layer and the contact layer; a physiological index sensor disposed in the functional layer for collecting the user's physiological index data; a stimulation electrode disposed in the contact layer, in contact with the user's skin, for stimulating the user's muscles; and a controller connected to both the physiological index sensor and the stimulation electrode for receiving the physiological index data and generating a control signal for the stimulation electrode based on the physiological index data and the user's baseline data, thereby controlling the stimulation electrode to execute the stimulation parameters corresponding to the control signal. By collecting initial physiological index data when the user first wears the elastic stocking to establish a user baseline, and continuously collecting physiological index data during the wearing process, the collected real-time physiological index data is compared with the user baseline. Based on the comparison result, a corresponding control signal for the stimulation electrode is generated, controlling the stimulation electrode to execute the corresponding stimulation parameters. Therefore, the elastic stocking of this application can provide personalized stimulation based on the individual physiological index of the current wearer, improving the prevention effect of lower extremity deep vein thrombosis.

[0045] Furthermore, by using behavioral state sensors to establish baselines for different behavioral states of the user, when the user wears the elastic stockings, the controller establishes first baseline data for the sitting / lying state, second baseline data for the standing state, and third baseline data for the walking state based on pressure data and pulse wave data from the sitting / lying, standing, and walking states, respectively. When determining the thrombosis warning result, the user's behavioral state is determined based on the real-time behavioral state information collected by the behavioral state sensors; the corresponding baseline data is selected as the user's baseline data based on the user's current behavioral state. When comparing the user's physiological indicators with the user's baseline data, the user's current behavioral state can be determined first through the behavioral state data collected by the behavioral state sensors, and then the baseline data corresponding to the behavioral state can be called. Then, the user's current physiological indicators are compared with the called baseline data, which can greatly improve the matching degree of stimulation parameters and enhance the effectiveness of lower limb thrombosis prevention for users in different states. This prevents inaccurate warnings caused by changes in baseline data when the user is standing, walking, or sitting / lying for extended periods.

[0046] Furthermore, the controller is equipped with a reinforcement learning model based on the TinyML framework. When the stimulation electrodes execute the stimulation parameters, the controller optimizes the stimulation parameters for subsequent rounds based on the current round's stimulation parameters and the physiological indicator changes. Feedback / closed-loop reinforcement learning, using physiological indicator data after the user receives electrode stimulation, enables the stimulation parameters to adapt to the user's personalized physiological needs. Through closed-loop feedback between stimulation parameters and stimulation effects (i.e., physiological indicator data collected after the stimulation parameters are executed), the stimulation parameters are continuously optimized based on the user's personalized physiological needs and performance. Compared to traditional methods, such as elastic stockings with fixed pressure and application time, which cannot adapt to the physiological differences of different groups and may lead to insufficient or excessive intervention, the stimulation parameters are matched to individual physiological characteristics through continuous interaction with the user during electrostimulation therapy using stimulation electrodes. In the prevention of venous thrombosis, this significantly improves the adaptation rate of stimulation parameters and enhances the preventive effect, especially for special populations such as pregnant women and cancer patients. Attached Figure Description

[0047] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0048] Figure 1 A schematic diagram of the modular structure of elastic stockings for preventing lower extremity deep vein thrombosis provided for some embodiments of this application;

[0049] Figure 2 A schematic diagram of the stacked structure of the package body provided for some embodiments of this application;

[0050] Figure 3 Schematic diagram of the structure of elastic stockings for preventing lower extremity deep vein thrombosis provided for some embodiments of this application;

[0051] Figure 4 Schematic diagram of the structure of elastic stockings for preventing lower extremity venous thrombosis provided for other embodiments of this application;

[0052] Figure 5 A schematic diagram of a controller in elastic stockings for preventing deep vein thrombosis of the lower extremities, provided for some embodiments of this application. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0054] Based on this, this application proposes an elastic stocking for preventing lower extremity venous thrombosis, such as... Figure 1-4 As shown, it includes: a wrapping body 10 having a pressure layer 11 on the outside and a contact layer 13 on the inside, and a functional layer 12 located between the pressure layer 11 and the contact layer 13; a physiological index sensor 20 disposed on the functional layer 12 for collecting the user's physiological index data; a stimulation electrode 30 disposed on the contact layer 13, in contact with the user's skin, for stimulating the user's muscles; and a controller 40 connected to the physiological index sensor 20 and the stimulation electrode 30 respectively, for receiving the physiological index data and generating a control signal for the stimulation electrode 30 based on the physiological index data and the user's baseline data, so as to control the stimulation electrode 30 to execute the stimulation parameters corresponding to the control signal.

[0055] In this embodiment, the physiological indicator sensor 20 includes a lower limb physiological characteristic sensor, such as a pressure sensor 21, a skin temperature sensor, or other sensors that directly collect lower limb physiological characteristics. For example, the lower limb physiological characteristic sensor is described using the pressure sensor 21 as an example: the pressure sensor 21 can be an array of pressure sensors 21 arranged in a ring, positioned in a ring within the first designated area of ​​the wrapping body 10, for collecting real-time pressure data of the first designated area of ​​the lower limb. The pressure sensor 21 can be a piezoresistive flexible pressure sensor array, with a thermoplastic polyurethane (TPU) substrate and a 0.5mm thick medical-grade silicone protective layer, combining flexibility and biocompatibility. A hydrophobic nano-coating (such as polytetrafluoroethylene) is added within the medical-grade silicone protective layer to improve resistance to sweat corrosion. The first designated area can be found in [reference needed]. Figure 3 and Figure 4 The area shown can be, for example, the gastrocnemius muscle region, the ankle joint, or the mid-thigh. The ring-shaped pressure sensors 21 can collect data on the varying pressure applied to the wrapping body 10 by different lower limb circumferences, thus enabling real-time determination of the lower limb circumference's real-time value and changes. A sharp increase in lower limb circumference may indicate venous blood stasis, an early sign of thrombosis.

[0056] Skin temperature sensors can use infrared temperature sensors to collect skin temperature in the ankle, patella, gastrocnemius muscle area, or thigh area. Skin temperature sensors can detect abnormal skin temperature caused by impaired blood circulation at the site of lower limb thrombosis.

[0057] The physiological index sensor 20 may also include a lower limb hemodynamic sensor. This sensor employs a pulse sensor, such as a piezoelectric pulse sensor, a piezoresistive pulse sensor, or a photoelectric pulse sensor. In this embodiment, a photoelectric pulse sensor may be used, for example, a multi-wavelength PPG sensor embedded in an elastic layer, located above the space between the first and second metatarsal bones on the dorsum of the foot, in contact with the skin, using a medical-grade polyurethane film with a light transmittance ≥90%. To prevent ambient light interference, a ring-shaped light-shielding gasket is placed around the sensor to block ambient light interference. Pulse wave data is collected by the pulse sensor. This pulse wave data is the result of the combined effects of cardiac ejection, arterial elasticity, and peripheral vascular resistance, and its periodic variation is directly related to lower limb hemodynamic indicators. An increased coefficient of variation in the pulse wave waveform may indicate fluctuations in cardiac output or abnormal peripheral vascular resistance, which is positively correlated with the risk of thrombosis.

[0058] Furthermore, pulse wave data can characterize changes in vascular compliance by measuring the variation in pulse wave propagation time, reflecting fluctuations in vascular elasticity and thus indicating the risk of thrombosis. Therefore, in this embodiment, pulse wave data is used to characterize lower limb blood flow data.

[0059] The physiological indicator sensor 20 may also include a lower limb muscle function sensor, which may employ an electromyography (EMG) sensor to acquire the energy characteristics of EMG signals from areas such as the gastrocnemius muscle. For example, the root mean square of the EMG signal may be used as lower limb muscle function data. For instance, weakened EMG activity may indicate decreased muscle pump function, and since venous return depends on muscle contraction, this may increase the risk of thrombosis.

[0060] In some alternative embodiments, the physiological index sensor 20 may also include a bioimpedance sensor for acquiring bioimpedance and biocapacitance. The ratio of the dynamic changes in bioimpedance and biocapacitance characterizes the changes in tissue fluid volume and blood distribution. For example, during venous thrombosis, impedance decreases (blood stasis) while capacitance increases (tissue edema). This ratio can quantify the coupling state of blood flow and edema and can reflect changes in vascular elasticity and tissue fluid distribution.

[0061] In some optional embodiments, the physiological indicator sensor 20 may further include a behavior state sensor, which collects the current user's behavior state, such as whether the user is sitting / lying down, standing, or walking, and the duration of each behavior state. In this embodiment, the behavior state sensor may be, for example, a gyroscope or accelerometer used to collect posture sensor data, and by collecting the acceleration components of the x, y, and z axes during lower limb movement, it is used to monitor data such as behavior state, intensity, and duration.

[0062] After obtaining the physiological index data, the controller 40 generates a control signal for the stimulation electrode 30 based on the physiological index data and the user baseline data. Specifically, the controller 40 compares at least one of the lower limb physiological characteristic data, lower limb hemodynamic data, and lower limb muscle function data with the user baseline data to obtain physiological index change characteristics; and generates the control signal based on the physiological index change characteristics. This controls the stimulation electrode 30 to execute the stimulation parameters corresponding to the control signal.

[0063] When a user first wears the compression stockings, the user's current baseline data is collected. For example, the lower limb circumference data is recorded by the pressure sensor 21, and the pulse wave data is recorded by the pulse wave sensor 22 to construct the user's baseline data. The user baseline data may also include muscle function data collected by the lower limb muscle function sensor at the time of first wear.

[0064] After obtaining baseline data, the collected physiological index data is compared with the user's baseline data to determine the degree of deviation. Based on this deviation, corresponding stimulation parameters are determined, and a control signal for the stimulation electrode 30 is generated. This control signal is then output to the stimulation electrode 30 to control it to execute the stimulation parameters. By collecting initial physiological index data when the user first wears the elastic stockings to construct a user baseline, and continuously collecting physiological index data during the wearing process, the real-time physiological index data is compared with the user baseline. Based on the comparison results, a corresponding control signal for the stimulation electrode 30 is generated, controlling the stimulation electrode 30 to execute the corresponding stimulation parameters. Therefore, the elastic stockings in this application can provide personalized stimulation based on the current user's individual physiological indicators, improving the prevention effect of lower extremity venous thrombosis.

[0065] In one embodiment, a behavior state sensor can be used to establish baselines corresponding to different behavior states for the user, preventing inaccurate warnings caused by changes in baseline data when the user is standing, walking, sitting / lying down, or in other states. In this embodiment, behavior states include sitting / lying down, standing, and walking. When the user wears the elastic stockings, the controller 40 establishes first baseline data for the sitting / lying down state, second baseline data for the standing state, and third baseline data for the walking state based on the pressure data and pulse wave data in the sitting / lying down, standing, and walking states, respectively. When determining the thrombosis warning result, the user's behavior state is determined based on the real-time behavior state information collected by the behavior state sensor; the corresponding baseline data is selected as the user's baseline data based on the user's current behavior state.

[0066] When comparing a user's physiological indicators with baseline data, the user's current behavioral state can be determined first by using behavioral state data collected by behavioral state sensors. Then, the baseline data corresponding to the behavioral state can be retrieved. Finally, the user's current physiological indicators can be compared with the retrieved baseline data. This can greatly improve the matching degree of stimulation parameters and enhance the effectiveness of lower limb thrombosis prevention for users in different states.

[0067] In one embodiment, energy management can also be adaptively managed based on the user's behavioral state. For example, the sampling frequency of the physiological indicator sensor 20 can vary depending on the user's behavioral state. For instance, when the user is active, such as walking or jogging, venous blood flow is faster due to muscle activity, resulting in a lower probability of thrombosis. Therefore, the sampling frequency can be reduced to save power. When the user is at rest, such as during prolonged bed rest, sitting, or sleep, venous blood flow slows down. Therefore, the sampling frequency needs to be increased to achieve real-time acquisition of dynamic changes in physiological indicator data, thus enabling immediate prevention of lower extremity venous thrombosis.

[0068] To further enhance the prevention of lower extremity deep vein thrombosis and enable more personalized physiotherapy for different users, in one embodiment, the controller 40 is equipped with a reinforcement learning model based on the TinyML framework. When the stimulation electrode 30 executes the stimulation parameters, the controller 40 optimizes the stimulation parameters for subsequent rounds based on the stimulation parameters of the current round and the physiological index changes of the current round using the reinforcement learning model. Feedback / closed-loop reinforcement learning based on the user's physiological index data after receiving electrode stimulation enables the stimulation parameters to adapt to the user's personalized physiological needs. Through closed-loop feedback between stimulation parameters and stimulation effects (i.e., physiological index data collected after the stimulation parameters are executed), the stimulation parameters suitable for the current user are continuously optimized based on the user's personalized physiological needs and performance. Compared to traditional solutions, such as elastic stockings with fixed pressure and application time, which cannot adapt to the physiological differences of different populations and may lead to insufficient or excessive intervention. During electrical stimulation therapy via the stimulation electrode 30, reinforcement learning is used to match the stimulation parameters with individual physiological characteristics through continuous interaction with the user (e.g., if the patient initially responds poorly to high-frequency electrotherapy after surgery, the model gradually reduces the frequency to 20Hz and extends the duration). This significantly improves the fit of stimulation parameters in the prevention of venous thrombosis, enhancing the preventative effect, especially for special populations such as pregnant women and cancer patients.

[0069] In one embodiment, the reinforcement learning model includes an Actor network and a Critic network, both of which are pre-trained network models. For example, they can be pre-trained using electrotherapy / massage intervention cases covering different populations such as post-operative patients, pregnant women, and cancer patients, and already possess basic stimulation parameter adjustment capabilities in the initial state. The Actor network is responsible for generating specific stimulation parameters, while the Critic network is responsible for evaluating the effectiveness of the stimulation parameters.

[0070] The controller 40 constructs a multidimensional input vector based on the physiological index change characteristics and stimulation parameters of the current round. The physiological index change characteristics may include the lower limb circumference change rate, arterial pulsation coefficient of variation, skin temperature difference, plantar pressure center of gravity offset distance, root mean square of electromyographic signal, venous compliance index, and triaxial acceleration; the stimulation parameters may include the current stimulation current frequency, stimulation intensity, and stimulation duration.

[0071] For example, the indicators in the physiological indicator change characteristics can be:

[0072] The rate of change in lower limb circumference can be calculated or characterized from the pressure data measured by pressure sensor 21, representing the rate of change of real-time lower limb circumference data relative to baseline lower limb circumference data: Rc =(R r -R b ) / R b ×100%, where R c R represents the rate of change in lower limb circumference. r For real-time lower limb circumference data, R b Baseline lower limb circumference data.

[0073] The coefficient of variation of arterial pulsation can be measured by the degree of variation of real-time pulse wave data collected by pulse wave sensor 22 compared with baseline pulse wave data in baseline data.

[0074] The skin temperature difference can be calculated by comparing the real-time lower limb skin temperature data collected by a skin temperature sensor with the baseline skin temperature in the baseline data, ΔT = T. r -T b Where ΔT is the skin temperature difference, T r For real-time skin temperature, T b Baseline skin temperature.

[0075] The offset distance of the plantar pressure center of gravity can be acquired by the ring-arranged pressure sensors 21, which can be used to collect the offset of the pressure distribution center relative to the pressure distribution center in the baseline data when the lower limb is bearing weight.

[0076]

[0077] in, The coordinates of the real-time pressure center are (x0, y0), and the coordinates of the baseline pressure center are (x0, y0).

[0078] The root mean square of the electromyographic (EMG) signal can be calculated from the EMG signal acquired by the EMG sensor using the following formula:

[0079]

[0080] Where, x i denoted as the electromyographic signal sample value, and N as the sampling window length.

[0081] The venous compliance index can be calculated as the ratio of the rate of change of bioimpedance acquired by a bioimpedance sensor to the rate of change of biological impedance.

[0082]

[0083] Where ΔZ is bioimpedance, ΔC is biocapacitance, and Z r For real-time bioimpedance, Z b Baseline bioimpedance; C r For real-time biocapacitance, C b Baseline biocapacitance.

[0084] Three-axis acceleration can be used to determine the real-time behavior of a user by collecting the values ​​of three-axis acceleration from an attitude sensor.

[0085] The physiological indicators and stimulation parameters of the current round are used to construct a multidimensional input vector, which is then input into the Actor network to obtain initial adjustment parameters. After receiving the current multidimensional input vector, the Actor network outputs a set of initial adjustment parameters, including stimulation current frequency adjustment (Δf, range -20Hz to +20Hz), stimulation intensity (Δi, range -1 level to +1 level), and stimulation duration (t, range 5-30 minutes).

[0086] To avoid the model relying solely on historical experience and ignoring potential better strategies, thus falling into local optima, a time-varying, stable noise, such as Ornstein-Uhlenbeck noise, is added to the stimulus parameters in the initial rounds. The noise intensity dynamically decays as training progresses: the standard deviation of the noise can be 0.2 in the initial stage, and after 500 cumulative interactions, the standard deviation drops to 0.05, focusing on effective parameters for interaction validation.

[0087] The generated stimulation parameters are compared with safety thresholds to ensure that all stimulation parameters are within the clinically safe range (e.g., stimulation current frequency is limited to 1-100Hz, stimulation intensity is limited to 0-10 levels, and stimulation duration is limited to 30 minutes). If they exceed the range, they are automatically cropped to the critical value.

[0088] The controller 40 generates an adjustment control signal according to the initial adjustment parameters, so that the stimulation electrode 30 executes the adjustment stimulation parameters corresponding to the adjustment control signal. During execution, the sensor continuously monitors physiological indicators to ensure stable action execution.

[0089] After the action is completed, physiological indicators are collected again to generate the multidimensional input vector for the next round.

[0090] The controller 40 acquires the physiological index change characteristics of the stimulation electrode 30 in the next round after the adjustment of stimulation parameters, calculates the parameter adjustment reward value based on the physiological index change characteristics in the next round, and updates the reinforcement learning model based on the adjustment reward value.

[0091] The parameter adjustment reward value is calculated based on the following formula:

[0092] R = w1·R1 + W2·R2 + ... + w n ·R ；

[0093] Wherein, R is the parameter adjustment reward value, R1, R2...R n These represent the characteristics of physiological indicator changes, w1, w2…w nThe physiological indicators are characterized by changes in R1, R2...R n The corresponding weighting coefficients are determined based on the user tags, and different user tags have different degrees of importance to the physiological indicator change characteristics.

[0094] For example, the physiological parameters of change include the coefficient of variation of arterial pulsatility, the rate of change of lower limb circumference, skin temperature difference, root mean square of electromyography signal, venous compliance index, and the offset distance of plantar pressure, to illustrate the reward value:

[0095] R = w1·ΔCV + w2·ΔRMS m +w3·ΔCOP d +w4·ΔT+w5·ΔZ / ΔC+w6·R c ;

[0096] Where w1, w2…w6 are the weighting coefficients corresponding to the physiological index changes, ΔCV is the coefficient of variation of arterial pulsation, and ΔRMS is… m ΔCOP is the difference between the root mean square (RMS) of the electromyography (EMG) signal and the baseline RMS of the EMG signal. d ΔT is the distance of the center pressure offset of the foot, ΔZ / ΔC is the skin temperature difference, and R is the venous compliance index. c This represents the rate of change in lower limb circumference.

[0097] For example, if the pulse wave conduction time variation coefficient decreases compared to before the action, indicating improved venous blood flow, a positive reward is given; the greater the decrease, the higher the reward value. If the root mean square of the electromyographic signal increases compared to before the action, i.e., ΔRMS... m If the temperature difference decreases compared to before the procedure, muscle activity increases, resulting in a positive reward. The greater the increase, the higher the reward value. Comfort reward: If the skin temperature difference decreases compared to before the procedure, the local skin temperature tends to normalize, improving comfort and resulting in a positive reward. The greater the skin temperature difference, the lower the reward value. The reward value can also be a negative reward.

[0098] After obtaining the reward value, the adjusted stimulus parameters, the physiological indicator change characteristics of the current round, the physiological indicator change characteristics of the next round, and the parameter adjustment reward value are merged into empirical data. The timestamp and user tag of the time the adjusted stimulus parameters were executed are obtained, where the user tag includes user symptom and user behavior status data. The timestamp and user tag are bound to the empirical data to obtain empirical data with execution scenario tags. The empirical data with execution scenario tags is cached until a set quantity is reached, and the cumulative reward value is calculated based on the cached empirical data. The reinforcement learning model is updated based on the cumulative reward value. Each piece of empirical data is accompanied by a timestamp and user tag (e.g., tags such as day n post-surgery, after m hours of sitting, etc., scenario tags including timestamp and user tags, such as week x of pregnancy), to facilitate subsequent targeted learning of strategies for different scenarios by the model.

[0099] When the accumulated empirical data in the buffer reaches, for example, 10, 20, 50, or 100 data points, the following update measures are initiated: multiple empirical data points are randomly extracted from the buffer to avoid the influence of temporal correlation on training performance; the Critic network calculates the average value (i.e., cumulative reward value) of the stimulus parameters based on the extracted empirical data and optimizes the network parameters through gradient descent to make the evaluation more accurate; the Actor network adjusts the parameters through the policy gradient algorithm based on the evaluation results of the Critic network to generate better stimulus parameters; the Actor network and the Critic network each maintain a target network, and after each preset number of model updates, the target network parameters are softly updated at a ratio of 99.9% old parameters + 0.1% new parameters to avoid excessive parameter fluctuations.

[0100] In one embodiment, the Critic network evaluates the effectiveness of stimulus parameters by combining a cumulative reward value formed from multiple immediate rewards and the predicted future reward value corresponding to the current stimulus parameter to determine the long-term effectiveness of the stimulus parameter. The immediate reward, as the basis for the cumulative reward, is combined with the predicted future reward value for updating both the Critic and Actor networks.

[0101] Specifically, the Critic network first calculates the reward value R using immediate physiological data collected after adjusting the stimulus parameters, serving as short-term effect feedback. The Critic network then uses this immediate physiological data to predict the maximum reward value achievable with the adjusted stimulus parameters in the current round, which is then used as the future reward value. In this embodiment, when predicting the future reward value, the current parameters of the Critic network can be fixed first. Based on the changes in physiological indicators collected after execution, the network can then calculate the long-term cumulative reward sum that can be obtained by executing the optimal stimulus parameters.

[0102] In this embodiment, when storing experience data, the adjustment parameter reward value in each piece of experience data can be the sum of the immediate reward and the future reward value. In this embodiment, the weights of the immediate reward value and the future reward value in the summation are different for different user tags. The weight of the immediate reward value is positively correlated with the severity of the illness in the user tag. For example, for the postoperative acute phase, where rapid improvement of blood flow is needed, the weight of the immediate reward value is higher; for patients with chronic diseases, such as paralysis, bedridden individuals, or those who require long-term positional muscle function, or those who sit / stand for long periods, the weight of the future reward value should be higher. This approach can balance short-term improvement with long-term prevention and safety for different types of users when preventing lower extremity deep vein thrombosis.

[0103] In one embodiment, the occurrence of thrombosis is closely related to patient age, type of surgery, lifestyle habits, physiological characteristics, and current behavioral status. However, the patient population of a single medical institution is often limited. To expand the data scale, integrate diverse data, increase sample size, adapt to localization needs, and make thrombosis prevention strategies more accurate and personalized, this embodiment employs a federated learning architecture to deploy the stimulus strategy generation model. Specifically, a local stimulus strategy generation model is deployed within the controller, while a global stimulus strategy generation model is deployed on a cloud server. In one embodiment, the local stimulus strategy generation model and the global stimulus strategy generation model have the same structure (i.e., the overall model architecture is consistent) to facilitate parameter aggregation. To save local storage resources, the local stimulus strategy generation model has a smaller model parameter scale, fewer network layers, and fewer neurons compared to the global stimulus strategy generation model.

[0104] In this embodiment, the Actor Network is deployed as a local stimulus strategy generation model within the local controller. The Actor Network has a shared layer and a personalized layer. The Actor Network updates parameters based on cumulative reward values, i.e., updating the shared layer parameters and personalized layer parameters. The shared layer parameters and cumulative reward values ​​are then encrypted and uploaded to the cloud server. Upon receiving the shared layer parameters uploaded by a preset number of controllers, the cloud server performs weighted aggregation of the shared layer parameters from multiple different controllers using a federated averaging algorithm to generate global shared layer parameters. This process extracts common knowledge across devices, such as the general dependency between postoperative duration and stimulus strategies, and the dependency between lower limb circumference change rate and stimulus parameters. In this embodiment, the weighting coefficients can be determined based on the cumulative reward values, where the cumulative reward values ​​are positively correlated with the weighting coefficients. The cloud server distributes the aggregated global shared layer parameters to each controller. The local Actor Network replaces its own shared layer parameters with the global shared layer parameters, retaining the personalized layer parameters unchanged, forming a fusion model of global common knowledge and local personalized strategies, which is then used for local intervention decision-making.

[0105] In another embodiment, the cloud generates group-specific sub-models (such as postoperative patient sub-models, pregnant women sub-models, etc.) based on the global model, and adapts them to specific group characteristics by adjusting the shared layer parameters. The local Actor network can load the shared layer parameters of the corresponding group's sub-models and combine them with its own personalized layer parameters to further improve the adaptability of the strategy to the local user's group.

[0106] In this embodiment, either a Long Short-Term Memory (LSTM) network or a Gate Recurrent Unit (GRU) network can be used as the local Actor network. This embodiment will use the LSTM network as an example.

[0107] The local stimulus strategy generation model comprises an input layer, a temporal feature enhancement module (LSTM), a shared layer, an attention mechanism module, a personalization layer, and an output layer. The input layer constructs a temporal feature sequence and extracts the temporal feature sequence from a multidimensional input vector. For example, the multidimensional input vector may include six dimensions: the coefficient of variation of arterial pulsation, the difference between the root mean square of the electromyography (EMG) signal and the root mean square of the baseline EMG signal, the offset distance of the plantar pressure center, the skin temperature difference, the venous compliance index, and the rate of change of lower limb circumference.

[0108] The input feature matrix is:

[0109]

[0110] The study used 12 time steps (t=12), each time step being 5 minutes, covering 60 minutes of data to reflect the slow evolution characteristics of the pre-thrombotic phase (6-12 hours). ΔCV represents the coefficient of variation of arterial pulsatility, and ΔRMS... m ΔCOP is the difference between the root mean square (RMS) of the electromyography (EMG) signal and the baseline RMS of the EMG signal. d ΔT is the distance of the center pressure offset of the foot, ΔZ / ΔC is the skin temperature difference, and R is the venous compliance index. c This represents the rate of change in lower limb circumference.

[0111] The multidimensional input vector is normalized based on the user's baseline data, for example, in R. c (t)∈[-1,1], where 0 represents the healthy baseline and 1 represents the thrombosis risk threshold.

[0112] The temporal feature enhancement module includes a 1-layer LSTM network with 20 hidden units, employing the ReLU activation function. It captures the dynamic trends of lower limb physiological indicators by processing the normalized input feature matrix, including the temporal dependencies of parameters such as arterial pulsatility variation coefficient, the difference between the root mean square of electromyography (EMG) signal and the root mean square of baseline EMG signal, the plantar center pressure offset distance, skin temperature difference, venous compliance index, and lower limb circumference change rate.

[0113] The temporal feature enhancement module can remember R through the forgetting gate. c The long-term trends of ΔZ / ΔC are used as long-term features, and the short-term fluctuations of ΔCV and ΔT are captured by the input gate as short-term features. The cell state is fused with long-term and short-term features to encode the evolution trajectory of thrombosis risk, and the output hidden state contains the fusion result of temporal features of 12 time steps.

[0114] In this embodiment, the shared portion of the temporal feature enhancement module may include forget gate weights and input gate weights to extract general temporal patterns across devices, such as the correlation between the rate of change in lower limb circumference and stimulation parameters, and the temporal dependence of postoperative duration on massage intensity. The personalized portion may include output gate weights and cell state update weights to adapt to individual temporal deviations and capture personalized features of patients in both long-term and short-term features.

[0115] The attention mechanism module can be structured as a single fully connected network, dynamically weighting the importance of different features, such as R in postoperative patients, by dynamically weighting the hidden states at 12 time steps. c The weighting is higher than that of temperature difference, and the weighting of temperature difference is even higher in pregnant women. The formula for calculating the weighting is as follows:

[0116]

[0117] Where, α i For the feature attention weights, ω i The attention parameters (obtained through federated aggregation); x i For input features.

[0118] The shared layer can be a two-layer fully connected network used to extract general features from the shared parts, such as forget gate weights, input gate weights, and attention parameters ω. i .

[0119] The personalization layer can use a fully connected network to output individual adaptation coefficients, such as the stimulus intensity in the stimulus parameters, to adapt to the differences in patients' sensitivity to the stimulus parameters.

[0120] The output layer generates continuous action parameters for the stimulation electrodes, such as stimulation current frequency adjustment (Δf, range -20Hz to +20Hz), stimulation intensity (Δi, range -1 level to +1 level), stimulation duration (t, range 5-30 minutes), etc.

[0121] This technical solution employs a Long Short-Term Memory (LSTM) network model architecture as the local Actor network. It captures and accurately models the dynamic process of thrombosis formation through long-term temporal dependencies, achieves a balance between global knowledge sharing and local personalization through a federated hierarchical strategy, and optimizes deployment for edge devices (within the controller of compression stockings) through lightweight optimization. Its core value lies in leveraging the modeling capabilities of LSTM for delayed rewards (such as the long-term effects of thrombosis prevention) while overcoming data silos through federated learning. Ultimately, it significantly improves the accuracy and temporal coherence of lower extremity venous thrombosis prevention strategies while ensuring privacy and security.

[0122] In one embodiment, the elastic stockings for preventing lower extremity deep vein thrombosis can also be other forms of wearable devices. The physiological indicator sensor 20 is illustrated using the pressure sensor 21 and the multi-wavelength PPG sensor as examples:

[0123] The pressure sensor 21 employs a piezoresistive flexible pressure sensor array. The substrate is thermoplastic polyurethane, covered with a medical-grade silicone protective layer, combining flexibility and biocompatibility. A hydrophobic nano-coating is added within the medical-grade silicone protective layer to enhance resistance to sweat corrosion. The sensor is distributed in a ring array across the calf muscle region, ankle joint, and mid-thigh area. Each distribution area is connected to the controller 40 via a flexible circuit with a serpentine wiring design. The circuit line width is ≤0.1mm to avoid affecting the elasticity of the compression stockings. Upon initial wear, the controller 40 records initial lower limb circumference data. This data can be from any one or more states, such as static standing, walking, sitting / lying, etc., to establish baseline data for the corresponding behavioral state. A sliding window filter is used to eliminate noise from postural changes; the baseline data is compared in real-time. If a leg circumference change of ≥5% is detected three consecutive times, an edema warning is triggered for baseline data calibration.

[0124] The pulse wave sensor 22 employs a multi-wavelength PPG sensor, embedded above the space between the first and second metatarsal bones on the dorsum of the foot in an elastic stocking. A medical-grade polyurethane film with a light transmittance of ≥90% is used at the skin contact point. A ring-shaped silicone gasket is placed around the sensor to block ambient light interference.

[0125] like Figure 4As shown, the stimulation electrode 30 is a self-adhesive conductive fiber electrode, such as a silver fiber blend, symmetrically arranged at the movement points of the gastrocnemius muscle. The power supply control circuit of the stimulation electrode 30 can adopt a miniaturized PCB integrated H-bridge drive circuit, supporting bidirectional current output, and a rechargeable lithium polymer battery to achieve constant current output through a Buck-Boost circuit, with a battery life of ≥48 hours.

[0126] In one embodiment, the pressure layer of the elastic stockings may be made of an 80% nylon + 20% spandex blend with a gradient pressure design, for example, the pressure from the ankle to below the knee decreases from 30 mmHg to 20 mmHg, and a diamond-shaped breathable mesh is formed using 3D knitting technology.

[0127] The physiological indicator sensor 20 and circuitry are seamlessly embedded in the functional layer of the elastic stockings using hot melt adhesive. The cables employ a sandwich structure, for example, a thermoplastic polyurethane substrate layer, a sensor layer, and a hydrophobic coating layer, to enhance durability. Furthermore, redundant wiring space is reserved in the functional layer to accommodate leg circumferences of 25-50cm.

[0128] The contact layer of the compression stockings uses antibacterial materials, such as antibacterial silver fibers, to inhibit the growth of Staphylococcus aureus through ion exchange. The cuff structure of the compression stockings uses elastic silicone anti-slip strips and adjustable Velcro, and the Velcro backing uses breathable mesh fabric to avoid pressure in the popliteal fossa.

[0129] Figure 5 This is a structural block diagram of an optional controller according to an embodiment of this application, such as... Figure 5 As shown, it includes a processor 501, a communication interface 502, a memory 503, and a communication bus 504. The processor 501, communication interface 502, and memory 503 communicate with each other via the communication bus 504.

[0130] Memory 503 is used to store computer programs;

[0131] When the processor 501 executes the computer program stored in the memory 503, it implements any one of the control methods executed by the controller in the above embodiments.

[0132] Optionally, in this embodiment, the communication bus described above can be...

[0133] PCI (Peripheral Component Interconnect) bus, or

[0134] EISA (Extended Industry Standard Architecture) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0135] The communication interface is used for communication between the aforementioned controller and other devices.

[0136] The memory may include RAM or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0137] The processor mentioned above can be a general-purpose processor, and may include, but is not limited to:

[0138] CPU (Central Processing Unit), NP (Network Processor), etc.; it can also be DSP (Digital Signal Processor), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0139] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0140] Those skilled in the art will understand that Figure 5 The structure shown is for illustrative purposes only. The terminal device can be a smartphone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (MID), a PAD, or other terminal devices. Figure 5 This does not limit the structure of the aforementioned electronic device. For example, the terminal device may also include components that are more... Figure 5 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 5 The different configurations shown.

[0141] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, ROM, RAM, disk or optical disk, etc.

[0142] According to another aspect of the embodiments of this application, a storage medium is also provided. Optionally, in this embodiment, the storage medium can be used to execute program code of the control method executed by the controller in the above embodiments.

[0143] Optionally, in this embodiment, the storage medium may be located on at least one of the network devices in the network shown in the above embodiment.

[0144] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated in this embodiment.

[0145] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, ROMs, RAMs, portable hard drives, magnetic disks, or optical disks.

[0146] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0147] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0148] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0149] In the several embodiments provided in this application, it should be understood that the disclosed client can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between units or modules, and may be electrical or other forms.

[0150] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the solution provided in this embodiment, depending on actual needs.

[0151] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0152] For any parts not mentioned in this application, existing technologies may be used or referenced.

[0153] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0154] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. An elastic stocking for preventing lower extremity deep vein thrombosis, characterized in that, include: The package body has a pressure layer on the outside and a contact layer on the inside, as well as a functional layer between the pressure layer and the contact layer; A physiological indicator sensor, installed in the functional layer, is used to collect the user's physiological indicator data; Stimulation electrodes, disposed in the contact layer, come into contact with the user's skin and are used to stimulate the user's muscles; The controller is connected to the physiological index sensor and the stimulation electrode, respectively, and is used to receive the physiological index data and generate a control signal for the stimulation electrode based on the physiological index data and user baseline data, so as to control the stimulation electrode to execute the stimulation parameters corresponding to the control signal.

2. The elastic stockings for preventing lower extremity venous thrombosis as described in claim 1, characterized in that, The physiological indicator sensor includes: At least one of the following: a lower limb physiological characteristic sensor, a lower limb hemodynamic sensor, and a lower limb muscle function sensor, wherein... The lower limb physiological feature sensor is used to collect the user's lower limb physiological feature data; The lower limb hemodynamic sensor is used to collect the user's lower limb hemodynamic data; The lower limb muscle function sensor is used to collect the user's lower limb muscle function data; The controller is used to compare at least one of the lower limb physiological characteristic data, lower limb hemodynamic data, and lower limb muscle function data with the user baseline data to obtain physiological indicator change characteristics; and to generate the control signal based on the physiological indicator change characteristics.

3. The elastic stockings for preventing lower extremity venous thrombosis as described in claim 1, characterized in that, The controller is equipped with a reinforcement learning model based on the TinyML framework. When the stimulation electrode executes the stimulation parameters, the controller optimizes the stimulation parameters for subsequent rounds using the reinforcement learning model based on the stimulation parameters of the current round and the physiological index change characteristics of the current round.

4. The elastic stockings for preventing lower extremity venous thrombosis as described in claim 3, characterized in that, The reinforcement learning model includes an Actor network and a Critic network; The controller optimizes the stimulation parameters for subsequent rounds using the reinforcement learning model based on the stimulation parameters and physiological change characteristics of the current round, including: The controller constructs a multi-dimensional input vector based on the physiological index change characteristics of the current round; The multidimensional input vector is input into the Actor network to obtain the initial adjustment parameters; The controller generates an adjustment control signal according to the initial adjustment parameters, so that the stimulation electrode executes the adjustment stimulation parameters corresponding to the adjustment control signal; The controller acquires the physiological index change characteristics of the stimulation electrode in the next round after the adjustment of stimulation parameters. The Critic network calculates the parameter adjustment reward value based on the physiological index change characteristics in the next round. The controller uses the parameter adjustment reward value to update the reinforcement learning model.

5. The elastic stockings for preventing lower extremity venous thrombosis as described in claim 4, characterized in that, The controller updates the reinforcement learning model based on the adjusted reward value, including: The adjusted stimulus parameters, the physiological index change characteristics of the current round, the physiological index change characteristics of the next round, and the parameter adjustment reward value are combined into empirical data. Obtain the timestamp and user tag when the adjustment of the stimulus parameters is performed, wherein the user tag includes user symptom and user behavior status data; The timestamp and the user tag are bound to the experience data to obtain experience data with execution scenario tags; The experience data with execution scenario labels is cached until a set quantity is reached, and the cumulative reward value is calculated based on the cached experience data. The reinforcement learning model is updated based on the accumulated reward value.

6. The elastic stockings for preventing lower extremity venous thrombosis as described in claim 5, characterized in that, The parameter adjustment reward value based on the physiological indicator change characteristics of the next round includes: The parameter adjustment reward value is calculated based on the following formula: R=w1·R1+w2·R2+...+w n ·R n ; Where R is the parameter adjustment reward value, R1, R2...R n These represent the characteristics of physiological indicator changes, w1, w2…w n The physiological indicators are characterized by changes in R1, R2...R n The corresponding weighting coefficients are determined based on the user tags, and different user tags have different degrees of importance to the physiological indicator change characteristics.

7. The elastic stockings for preventing lower extremity deep vein thrombosis as described in any one of claims 4-6, characterized in that, The Actor network is an ontology policy generation model under a federated learning framework. The Actor network has a shared layer and a personalized layer. The controller is also connected to a cloud server. The cloud server also deploys a global policy generation model. After the Actor network updates the reward value based on the parameters, it uploads the shared layer parameters to the cloud server for parameter aggregation and receives global model parameters from the cloud server to update the shared layer parameters.

8. The elastic stockings for preventing lower extremity venous thrombosis as described in claim 2, characterized in that, The lower limb physiological characteristic sensor includes: multiple pressure sensors arranged in a ring in the first designated area of ​​the wrapping body, used to collect real-time pressure data of the first designated area of ​​the lower limb; The lower limb hemodynamic sensor includes a pulse wave sensor, which is located in a second designated area of ​​the elastic wrapping body and is used to collect the user's pulse wave data.

9. The elastic stockings for preventing lower extremity venous thrombosis as described in claim 8, characterized in that, The physiological indicator sensor also includes: A behavior state sensor is used to collect the user's behavior state, which includes: sitting / lying down, standing, and walking. When the user wears the elastic stockings, the controller establishes first baseline data corresponding to the sitting / lying state, second baseline data corresponding to the standing state, and third baseline data corresponding to the walking state based on the pressure data and pulse wave data in the sitting / lying state, standing state, and walking state, respectively. When determining the thrombosis warning result, the user's behavioral status is determined based on the real-time behavioral status information collected by the behavioral status sensor; The baseline data is selected based on the user's current behavior state.

10. The elastic stockings for preventing lower extremity venous thrombosis as described in claim 2, characterized in that, When the changes in the physiological indicators exceed the set amount, the controller outputs a stop control signal to stop the stimulation electrode from working.

Images