Multi-mode intelligent control method and system for electric sickbed

Abstract

The invention relates to the technical field of medical equipment, and discloses a multi-mode intelligent control method and system for an electric sickbed, the system is composed of a sickbed state monitoring layer, an electric sickbed control layer and a control strategy execution layer, and the state of the electric sickbed and physiological parameters of a patient are monitored in real time. A sickbed control model is constructed, a control process is modeled as a multi-objective optimization problem, and a reinforcement learning algorithm is adopted to dynamically adjust a control scheme. The system can make an intelligent decision, automatically adjust a sickbed control strategy according to the real-time state and demand of a patient, and provide personalized and accurate nursing services. And the efficiency evaluation module ensures that the control action is accurate, rapid and energy-saving, and gives consideration to the comfort and safety of the patient. The comfort and functionality of the electric sickbed are remarkably improved, medical resource utilization is optimized, and the treatment effect and nursing quality of a patient are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical equipment, and in particular to a multi-modal intelligent control method and system for an electric hospital bed. Background Art

[0002] In the field of medical care, electric beds are important medical equipment, and their comfort and functionality are crucial to improving the treatment effect and quality of care for patients. At present, electric bed control systems often rely on preset fixed control logic, or require medical staff to manually adjust the bed, such as adjusting the height and tilt angle of the bed. However, this control method has many limitations, such as slow response speed, low control accuracy, and difficulty in dynamic adjustment according to the patient's real-time physiological state and needs, which affects the patient's comfort and treatment effect.

[0003] With the continuous advancement of medical technology and the development of intelligent technology, people have put forward higher requirements for the control system of electric beds. On the one hand, it is hoped that the status of electric beds can be monitored in real time, including the position, angle, motion status of the bed and the physiological parameters of the patient, so as to find problems in time and make adjustments. On the other hand, it is expected that the control system can have intelligent decision-making capabilities and can automatically adjust the control strategy of the bed according to the real-time status and needs of the patient, so as to provide more personalized and accurate nursing services. However, it is difficult for the existing electric bed control system to meet these requirements at the same time. Summary of the invention

[0004] The object of the present invention is to provide a multi-modal intelligent control method and system for an electric hospital bed to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a multi-modal intelligent control system for an electric hospital bed, the system comprising:

[0006] It includes bed status monitoring layer, electric bed control layer and control strategy execution layer;

[0007] The bed status monitoring layer is used to monitor the status of the electric bed and the patient's physiological parameters in real time;

[0008] The electric bed control layer is used to construct a bed control model, which models the control process of the electric bed as a multi-objective optimization problem and dynamically adjusts the control scheme through an algorithm;

[0009] The electric hospital bed control layer further includes:

[0010] The state description module is used to define the current state space of the electric bed and the patient;

[0011] The action definition module is used to define the action space for controlling the electric bed;

[0012] The effectiveness evaluation module is used to define the evaluation function and quantitatively evaluate the control results according to the accuracy, response speed, energy consumption, patient comfort and safety of the control action;

[0013] The model training module uses a reinforcement learning algorithm to select the optimal control action under a given state and optimize the control strategy through continuous iterative learning;

[0014] The control strategy execution layer is used to receive data input from the bed status monitoring module, transfer it to the bed control model, and drive the electric bed and its auxiliary equipment to perform corresponding actions according to the control strategy output by the model.

[0015] Preferably, the patient's physiological parameters include heart rate, respiratory rate, blood pressure and body temperature.

[0016] Preferably, the state space includes the current position, tilt angle, movement speed, acceleration of the electric bed and the patient's physiological parameters.

[0017] Preferably, the action space includes the lifting, tilting, and rotating of the bed surface, as well as the activation and adjustment of massage and turning auxiliary functions.

[0018] Preferably, the implementation of the effectiveness evaluation module includes:

[0019] An evaluation function G(s,a) is defined, where s represents the current state space of the electric bed and the patient; a represents the control action taken, which is selected from the control action space; the evaluation function G(s,a) performs weighted summation according to the accuracy P(s,a), response speed T(s,a), energy consumption C(s,a), patient comfort U(s,a) and safety S(s,a) of the control action to obtain a quantitative evaluation value of the effectiveness of the control action; the specific algorithm formula is:

[0020] G(s,a)=k1*P(s,a)-k2*T(s,a)-k3*C(s,a)+k4*U(s,a)+k5*S(s,a)

[0021] Among them, k1, k2, k3, k4, k5 are the weight coefficients of accuracy, response speed, energy consumption, patient comfort and safety, respectively, and k1+k2+k3+k4+k5=1; P(s,a) represents the control accuracy after taking action a in state s, which is obtained by calculating the deviation between the actual position and the target position; T(s,a) represents the response speed, which is obtained by calculating the time from issuing the control command to the completion of the action; C(s,a) represents energy consumption, including electrical energy consumption and mechanical wear; U(s,a) represents patient comfort, which is based on the changes in the patient's physiological parameters and feedback evaluation; S(s,a) represents safety, which is obtained by evaluating the risks generated during the execution of the action.

[0022] Preferably, the model training module uses a deep Q network DQN algorithm in a reinforcement learning algorithm to train the bed control model.

[0023] Preferably, a simulation model of the electric bed and the patient is constructed according to the actual operation of the electric bed and the physiological characteristics of the patient; different states and control actions are simulated by the simulation model, the corresponding immediate rewards are calculated, and a simulated state-action-reward tuple is generated as training data for the bed control model.

[0024] Preferably, the step of the model training module training the bed control model includes:

[0025] Step A: Initialize the deep Q network, which takes the current state space of the electric bed and the patient as input and outputs the Q value estimate for each possible control action; at the same time, initialize the experience replay memory and the target Q network, and the structure of the target Q network is the same as that of the deep Q network;

[0026] Step B: Set the learning rate α, discount factor γ, exploration rate ε, and the size of the experience replay memory N, where α controls the learning speed of network parameters, γ controls the discounted value of future returns, ε controls the balance between exploration and exploitation, and N determines the capacity of the experience replay memory;

[0027] Step C: For each training round:

[0028] Step C01: Initialize the state s of the electric bed and the patient from the training data;

[0029] Step C02: According to the current state s and the ε-greedy strategy, a control action a is selected in combination with the output of the deep Q network, an action is randomly selected with a probability of ε, and the action with the largest current Q value is selected with a probability of 1-ε;

[0030] Step C03: Execute action a, observe the next state s' and the immediate reward r calculated by the performance evaluation unit;

[0031] Step C04: storing the state transition tuple (s, a, r, s') into the experience playback memory;

[0032] Step C05: Randomly extract a small batch of state transition tuples from the experience replay memory to update the parameters of the deep Q network. The update method uses the gradient descent method to minimize the loss function. The loss function is defined as:

[0033] L=(r+γ*max_a'Q'(s',a')-Q(s,a))^2

[0034] Among them, Q(s,a) is the Q value estimate of the deep Q network for the current state s and action a, and Q'(s',a') is the Q value estimate of the target Q network for the next state s' and the action with the largest Q value among all possible actions;

[0035] Step D: At regular intervals, copy the parameters of the deep Q network to the target Q network;

[0036] Step E: Repeat step C until the preset number of training rounds is reached or the output of the deep Q network converges;

[0037] Step F: After training, the deep Q network is used as a multimodal control strategy model. For any given electric bed and patient state s, the action a with the largest Q value is selected as the optimal control action through the output of the deep Q network.

[0038] Preferably, before the model training begins, the training data is further preprocessed, including:

[0039] Perform one-hot encoding conversion on discrete features;

[0040] The Z-score standardization method is used to standardize continuous features. Specifically, each numerical data is subtracted from its mean and divided by its standard deviation so that the processed data conforms to the standard normal distribution. The standardization formula is: Z = (X-μ) / σ, where X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

[0041] Preferably, a multi-modal intelligent control method for an electric hospital bed comprises:

[0042] S1: Real-time monitoring of the status of the electric bed and the patient's physiological parameters as data input for the bed status monitoring layer;

[0043] S2: Construct a bed control model and model the control process of the electric bed as a multi-objective optimization problem. The specific steps include:

[0044] S201: define the current state space of the electric bed and the patient;

[0045] S202: defining the action space for controlling the electric hospital bed;

[0046] S203: defining an evaluation function to quantitatively evaluate the control results according to the accuracy, response speed, energy consumption, patient comfort and safety of the control action;

[0047] S204: Using a reinforcement learning algorithm, select the optimal control action under a given state, and optimize the control strategy through continuous iterative learning;

[0048] S3: Receive data input from the bed status monitoring layer, pass it to the bed control model, and drive the electric bed and its auxiliary equipment to perform corresponding actions according to the control strategy output by the model.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] The electric bed control layer has built a bed control model based on multi-objective optimization, combined with a reinforcement learning algorithm, which can automatically and dynamically adjust the control strategy according to the patient's real-time status and needs. This intelligent decision-making mechanism not only improves the accuracy and efficiency of control, but also provides patients with more personalized and accurate nursing services to meet the specific needs of different patients, thereby significantly improving the patient's comfort and treatment effect.

[0051] Through intelligent control, the system reduces the manual operation of medical staff and the risk of human error, thereby improving the safety and reliability of nursing. At the same time, the system can continuously monitor the patient's physiological state, timely detect and warn of potential health risks, provide strong decision-making support for medical staff, and further ensure the safety of patients. The introduction of the performance evaluation module enables the system to comprehensively consider multiple factors such as the accuracy, response speed, and energy consumption of the control action during the control process, and select the optimal control solution through quantitative evaluation. This not only optimizes the utilization of resources and reduces energy consumption, but also conforms to the development trend of green medicine and helps reduce the operating costs of medical institutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a structural diagram of a multi-modal intelligent control system for an electric hospital bed according to the present invention;

[0053] Figure 2 It is a schematic diagram of the design of the evaluation function in the performance evaluation module;

[0054] Figure 3 Training flow chart for the bed control model. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] See also Figure 1-3 The present invention provides a technical solution: a multi-modal intelligent control system for an electric hospital bed, the system comprising:

[0057] Bed status monitoring layer: The bed status monitoring layer is mainly responsible for real-time monitoring of the status of the electric bed and the patient's physiological parameters. In order to achieve this function, the system integrates a variety of sensors, including but not limited to position sensors, angle sensors, acceleration sensors, and medical devices for monitoring patient physiological parameters, such as heart rate monitors, respiratory rate monitors, blood pressure monitors, and thermometers. These sensors collect the current position, tilt angle, movement speed, acceleration of the electric bed, and the patient's heart rate, respiratory rate, blood pressure, body temperature and other physiological parameters in real time, and transmit these data to the control strategy execution layer via wired or wireless means.

[0058] Electric bed control layer: The electric bed control layer is the core part of the system. It builds a bed control model, which models the control process of the electric bed as a multi-objective optimization problem. The electric bed control layer is further subdivided into the following modules:

[0059] State description module: This module receives data from the bed state monitoring layer and defines the current state space of the electric bed and the patient. The state space includes but is not limited to the current position, tilt angle, movement speed, acceleration of the electric bed, and the patient's heart rate, respiratory rate, blood pressure, body temperature and other physiological parameters.

[0060] Action definition module: This module defines the action space for controlling the electric hospital bed, including the lifting, tilting, and rotating of the bed surface, as well as the activation and adjustment of massage and turning auxiliary functions. These actions constitute the optional control strategy set of the system.

[0061] Effectiveness evaluation module: This module defines an evaluation function to quantitatively evaluate the control results based on multiple dimensions such as control action accuracy, response speed, energy consumption, patient comfort and safety. The evaluation function is an important basis for the system optimization control strategy.

[0062] Model training module: This module uses a reinforcement learning algorithm to select the optimal control action under a given state. Through continuous iterative learning, the system continuously tries and adjusts the control strategy to maximize the value of the evaluation function, thereby optimizing the control strategy.

[0063] Control strategy execution layer: The control strategy execution layer is responsible for receiving the data transmitted by the bed status monitoring layer and passing it to the bed control model in the electric bed control layer. According to the optimal control strategy output by the model, the control strategy execution layer drives the electric bed and its auxiliary equipment to perform corresponding actions, such as adjusting the height and tilt angle of the bed, starting massage or turning auxiliary functions, etc. In order to achieve this function, the control strategy execution layer is closely connected with the drive system of the electric bed, and drives the motor, air pump and other actuators through control signals to achieve precise adjustment of the bed.

[0064] The present invention will be further described below in conjunction with Examples 1 to 3:

[0065] Embodiment 1:

[0066] The implementation of the effectiveness evaluation module of the present invention involves defining a comprehensive evaluation function G(s, a), which is used to quantitatively evaluate the effectiveness of taking a control action a under a specific state s. The construction of the evaluation function G(s, a) and the implementation of its components are described in detail below.

[0067] Definition of evaluation function G(s,a): The evaluation function G(s,a) is a weighted sum function that comprehensively considers the five dimensions of control action accuracy P(s,a), response speed T(s,a), energy consumption C(s,a), patient comfort U(s,a) and safety S(s,a). The specific form of the function is:

[0068] G(s,a)=k1*P(s,a)-k2*T(s,a)-k3*C(s,a)+k4*U(s,a)+k5*S(s,a)

[0069] Among them, k1, k2, k3, k4, and k5 are weight coefficients of the above five dimensions respectively, and satisfy k1+k2+k3+k4+k5=1 to ensure the comprehensiveness and balance of the evaluation function.

[0070] Implementation of each component:

[0071] Precision P(s,a): Precision P(s,a) indicates the degree of deviation between the actual position of the electric bed and the target position after taking action a in state s. In order to achieve quantitative evaluation of precision, the system uses a position sensor to measure the actual position of the electric bed in real time and compare it with the target position. The smaller the deviation, the higher the precision and the larger the value of P(s,a).

[0072] Response speed T(s,a): The response speed T(s,a) reflects the time from issuing a control command to the electric bed completing the corresponding action. The system records the time point when the control command is issued and the time point when the action is completed, and calculates the time difference as a measure of the response speed. The faster the response speed, the smaller the value of T(s,a).

[0073] Energy consumption C(s,a): Energy consumption C(s,a) includes the power consumption and mechanical wear of the electric bed during the execution of the action. The system comprehensively evaluates the energy consumption level by monitoring the power consumption of the electric bed and the wear of the motor and transmission mechanism. The lower the energy consumption, the smaller the value of C(s,a).

[0074] Patient comfort U(s,a): Patient comfort U(s,a) is evaluated based on the changes and feedback of the patient's physiological parameters. The system monitors the patient's heart rate, respiratory rate, blood pressure and other physiological parameters in real time, and combines the patient's feedback information (such as pain, discomfort, etc.) to comprehensively judge the patient's comfort during the use of the electric bed. The higher the comfort, the greater the value of U(s,a).

[0075] Safety S(s,a): Safety S(s,a) is obtained by evaluating the risks generated during the execution of the action. The system considers the safety risks such as collision, pinching, and slipping that may occur during the execution of the action of the electric bed, as well as the stability and reliability of the electric bed. The higher the safety, the greater the value of S(s,a).

[0076] Taking the lifting action of an electric bed as an example, assume that the current state s is that the electric bed is in the lowest position, the patient is lying flat on the bed, and physiological parameters such as heart rate and respiratory rate are normal. The system considers taking action a to raise the electric bed to a certain height. When evaluating the effectiveness of this action, the system will:

[0077] The deviation between the actual lifting height and the target height is measured by the position sensor, and the accuracy P(s,a) is calculated;

[0078] Record the time from issuing the lifting command to the bed reaching the target height, and calculate the response speed T(s,a);

[0079] Monitor the power consumption and mechanical wear during the lifting process and calculate the energy consumption C(s,a);

[0080] Monitor the patient's heart rate, respiratory rate and other physiological parameters in real time, and evaluate the comfort level U(s,a) based on the patient's feedback;

[0081] Assess possible safety risks during the lifting process, such as collision, slipping, etc., and calculate the safety S(s,a).

[0082] Finally, according to the weight coefficients k1, k2, k3, k4, and k5, the evaluation values ​​of the above five dimensions are weighted and summed to obtain the quantitative evaluation value G(s,a) of the effectiveness of action a in state s. By comparing the G(s,a) values ​​corresponding to different actions, the system can select the optimal control action to realize the intelligent control of the electric bed.

[0083] Embodiment 2:

[0084] The model training module of the present invention uses the deep Q network (DQN) algorithm in the reinforcement learning algorithm to train the bed control model. The specific implementation method of the model training module to train the bed control model includes:

[0085] ① Initialization phase:

[0086] Deep Q network initialization: Build a deep neural network that takes the current state space of the electric bed and the patient as input, and the output layer is the Q value estimate for each possible control action. The network structure can be designed according to specific task requirements, such as using a multi-layer perceptron (MLP) or a convolutional neural network (CNN).

[0087] Initialize the experience replay memory: Set up a fixed-size memory to store state transition tuples (s, a, r, s'), where s is the current state, a is the action taken, r is the immediate reward, and s' is the next state after the action is taken. The memory size N is determined according to the task complexity and memory resources.

[0088] Target Q network initialization: Build a network with the same structure as the deep Q network to calculate the target Q value. In the early stage of training, the parameters of the target Q network are the same as those of the deep Q network.

[0089] Parameter settings: Set the learning rate α, discount factor γ, exploration rate ε, and the size of the experience replay memory N. The learning rate α controls the learning speed of the network parameters, the discount factor γ determines the discounted value of future returns, the exploration rate ε is used to balance exploration and utilization, and the memory size N determines the capacity of the experience replay memory.

[0090] ②Training process:

[0091] State initialization: For each training round, the state s of the electric bed and the patient are randomly initialized from the training data.

[0092] Action selection: According to the current state s and the ε-greedy strategy, a control action a is selected in combination with the output of the deep Q network. Specifically, an action is randomly selected with a probability of ε, and the action with the largest current Q value is selected with a probability of 1-ε. As training progresses, ε gradually decreases, allowing the model to make more use of known information.

[0093] Execute action and observe: Execute the selected action a, observe the next state s' of the electric bed and the patient, and calculate the immediate reward r according to the performance evaluation unit. The immediate reward r is obtained by comprehensive evaluation of multiple dimensions such as the accuracy, response speed, energy consumption, patient comfort and safety of the control action.

[0094] Storing experience: Store the state transition tuple (s, a, r, s') in the experience replay memory. When the memory is full, the new tuple will replace the oldest tuple.

[0095] Network update: Randomly extract a small batch of state transition tuples from the experience replay memory to update the parameters of the deep Q network. The update method uses the gradient descent method to minimize the loss function, which is defined as:

[0096] L=(r+γ*max_a'Q'(s',a')-Q(s,a))^2

[0097] Among them, Q(s,a) is the Q value estimated by the deep Q network for the current state s and action a, and Q'(s',a') is the Q value estimated by the target Q network for the next state s' and the action with the largest Q value among all possible actions. By minimizing the loss function, the deep Q network can estimate the Q value more accurately.

[0098] Target network update: At regular intervals (e.g., every C rounds), the parameters of the deep Q network are copied to the target Q network to maintain the stability of the target Q network.

[0099] Repeat training: Repeat the above steps until the preset number of training rounds is reached or the output of the deep Q network converges. The convergence criterion can be that the value of the loss function is lower than a certain threshold, or the fluctuation range of the Q value is less than a certain threshold.

[0100] ③Model application:

[0101] After training, the deep Q network is used as a multimodal control strategy model. For any given electric bed and patient state s, the action a with the largest Q value is selected as the optimal control action through the output of the deep Q network. In this way, the model can automatically select the most appropriate control action according to the current state, realizing intelligent control of the electric bed.

[0102] During the training process, the state space may include the current height of the electric bed, the patient's weight, the patient's physiological parameters (such as heart rate, blood pressure, etc.), etc.; the action space may include control actions such as raising, lowering, and maintaining. Through the training of the DQN algorithm, the model can learn what actions to take in different states to maximize the patient's comfort and minimize energy consumption. In actual application, you only need to input the current state of the electric bed and the patient into the model, and the model can output the optimal control action to achieve automatic lifting and lowering control of the electric bed.

[0103] Embodiment 3:

[0104] In order to build a simulation model of an electric bed and a patient and generate data for training the bed control model, the specific implementation steps are described in detail below.

[0105] Electric bed simulation model construction: According to the actual mechanical structure and operating parameters of the electric bed, such as motor performance, transmission mechanism, bed surface size, etc., the physical model of the electric bed is established. The simulation code of the electric bed is written to simulate the movement state of the electric bed under different control instructions, including lifting, tilting, translation and other actions.

[0106] Patient model construction: Collect the patient's physiological characteristics data, such as weight, height, age, gender, underlying diseases, etc. According to the patient's physiological characteristics, build the patient's physiological model to simulate the patient's physiological reactions in different bed conditions, such as heart rate changes, blood pressure fluctuations, comfort feelings, etc.

[0107] State-action-reward mechanism design: Define the state space of the electric bed and the patient, including the position, speed, acceleration of the electric bed and the patient's physiological parameters.

[0108] Design the control action space, including the control instructions of the electric bed's lifting, tilting, and translation.

[0109] According to the operation of the electric bed and the patient's physiological response, an immediate reward function is designed to calculate the immediate reward corresponding to each state-action pair. The reward function can comprehensively consider multiple dimensions such as the accuracy of the control action, response speed, energy consumption, patient comfort and safety.

[0110] Simulation data generation: Use the constructed simulation model to simulate different states and control actions and calculate the corresponding immediate rewards. Generate simulated state-action-reward tuples as training data for the bed control model.

[0111] Before model training begins, the generated training data is preprocessed to improve the training effect and convergence speed of the model.

[0112] Discrete feature processing: Perform one-hot encoding conversion on discrete features in the training data (such as the patient's gender, underlying disease type, etc.).

[0113] One-hot encoding converts each discrete feature into a binary vector. Each position of the vector corresponds to an eigenvalue. When the eigenvalue appears, the corresponding position is 1, otherwise it is 0.

[0114] Continuous feature processing: The continuous features in the training data (such as the patient's weight, heart rate, blood pressure, etc.) are standardized using the Z-score standardization method.

[0115] The formula for standardization is: Z = (X-μ) / σ, where X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

[0116] Through standardization, the data of continuous features is converted into a form that conforms to the standard normal distribution, which helps the model better learn the inherent laws of the data.

[0117] The specific implementation is as follows:

[0118] Construct a physical model of the electric bed and simulate its movement during the lifting process.

[0119] Collect the patient's physiological characteristics data such as weight, height, and underlying diseases, establish the patient's physiological model, and simulate the patient's physiological reactions during the lifting process.

[0120] The design state space includes the height, speed, heart rate, blood pressure, etc. of the electric bed; the design action space includes lifting and lowering control instructions (such as rising, falling, stopping, etc.).

[0121] Design an immediate reward function based on the lifting and lowering of the electric bed and the patient's physiological response. For example, when the electric bed is lifted and lowered smoothly and the patient's heart rate and blood pressure remain within the normal range, a positive reward is given; when the electric bed is lifted and lowered too quickly or the patient's physiological parameters are abnormal, a negative reward is given.

[0122] A simulation model is used to generate a large number of state-action-reward tuples as training data.

[0123] The discrete features in the training data (such as the patient's underlying disease type) are converted to one-hot encoding; the continuous features (such as the patient's weight, heart rate, blood pressure, etc.) are normalized by Z-score.

[0124] The present invention also includes a multi-modal intelligent control method for an electric hospital bed, the method comprising:

[0125] S1: Real-time monitoring of the status of the electric bed and the patient's physiological parameters as data input for the bed status monitoring layer;

[0126] S2: Construct a bed control model and model the control process of the electric bed as a multi-objective optimization problem. The specific steps include:

[0127] S201: define the current state space of the electric bed and the patient;

[0128] S202: defining the action space for controlling the electric hospital bed;

[0129] S203: defining an evaluation function to quantitatively evaluate the control results according to the accuracy, response speed, energy consumption, patient comfort and safety of the control action;

[0130] S204: Using a reinforcement learning algorithm, select the optimal control action under a given state, and optimize the control strategy through continuous iterative learning;

[0131] S3: Receive data input from the bed status monitoring layer, pass it to the bed control model, and drive the electric bed and its auxiliary equipment to perform corresponding actions according to the control strategy output by the model.

[0132] The implementation method of the method is the same as the above embodiment and will not be described in detail in the specification.

[0133] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0134] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-modal intelligent control system for an electric hospital bed, characterized in that: It includes bed status monitoring layer, electric bed control layer and control strategy execution layer; The bed status monitoring layer is used to monitor the status of the electric bed and the patient's physiological parameters in real time; The electric bed control layer is used to construct a bed control model, which models the control process of the electric bed as a multi-objective optimization problem and dynamically adjusts the control scheme through an algorithm; The electric hospital bed control layer further includes: The state description module is used to define the current state space of the electric bed and the patient; The action definition module is used to define the action space for controlling the electric bed; The effectiveness evaluation module is used to define the evaluation function and quantitatively evaluate the control results according to the accuracy, response speed, energy consumption, patient comfort and safety of the control action; The model training module uses a reinforcement learning algorithm to select the optimal control action under a given state and optimize the control strategy through continuous iterative learning; The control strategy execution layer is used to receive data input from the bed status monitoring module, transfer it to the bed control model, and drive the electric bed and its auxiliary equipment to perform corresponding actions according to the control strategy output by the model.

2. The multi-modal intelligent control system of an electric hospital bed according to claim 1, characterized in that: The patient's physiological parameters include heart rate, respiratory rate, blood pressure and body temperature.

3. The multi-modal intelligent control system of an electric hospital bed according to claim 1, characterized in that: The state space includes the current position, tilt angle, movement speed, acceleration of the electric bed and the physiological parameters of the patient.

4. The multi-modal intelligent control system of an electric hospital bed according to claim 1, characterized in that: The action space includes the lifting, tilting, and rotating of the bed surface, as well as the activation and adjustment of massage and turning auxiliary functions.

5. The multi-modal intelligent control system of an electric hospital bed according to claim 1, characterized in that: The implementation of the effectiveness evaluation module includes: An evaluation function G(s,a) is defined, where s represents the current state space of the electric bed and the patient; a represents the control action taken, which is selected from the control action space; the evaluation function G(s,a) performs weighted summation according to the accuracy P(s,a), response speed T(s,a), energy consumption C(s,a), patient comfort U(s,a) and safety S(s,a) of the control action to obtain a quantitative evaluation value of the effectiveness of the control action; the specific algorithm formula is: G(s,a)=k1*P(s,a)-k2*T(s,a)-k3*C(s,a)+k4*U(s,a)+k5*S(s,a) Among them, k1, k2, k3, k4, k5 are the weight coefficients of accuracy, response speed, energy consumption, patient comfort and safety, respectively, and k1+k2+k3+k4+k5=1; P(s,a) represents the control accuracy after taking action a in state s, which is obtained by calculating the deviation between the actual position and the target position; T(s,a) represents the response speed, which is obtained by calculating the time from issuing the control command to the completion of the action; C(s,a) represents energy consumption, including electrical energy consumption and mechanical wear; U(s,a) represents patient comfort, which is based on the changes in the patient's physiological parameters and feedback evaluation; S(s,a) represents safety, which is obtained by evaluating the risks generated during the execution of the action.

6. The multi-modal intelligent control system of an electric hospital bed according to claim 1, characterized in that: The model training module uses the deep Q network DQN algorithm in the reinforcement learning algorithm to train the bed control model.

7. The multi-modal intelligent control system of an electric hospital bed according to claim 6, characterized in that: According to the actual operation of the electric bed and the physiological characteristics of the patient, a simulation model of the electric bed and the patient is constructed. Different states and control actions are simulated through the simulation model, and the corresponding immediate rewards are calculated to generate simulated state-action-reward tuples as training data for the bed control model.

8. The multi-modal intelligent control system of an electric hospital bed according to claim 7, characterized in that: The steps of the model training module training the bed control model include: Step A: Initialize the deep Q network, which takes the current state space of the electric bed and the patient as input and outputs the Q value estimate for each possible control action; at the same time, initialize the experience replay memory and the target Q network, and the structure of the target Q network is the same as that of the deep Q network; Step B: Set the learning rate α, discount factor γ, exploration rate ε, and the size of the experience replay memory N, where α controls the learning speed of network parameters, γ controls the discounted value of future returns, ε controls the balance between exploration and exploitation, and N determines the capacity of the experience replay memory; Step C: For each training round: Step C01: Initialize the state s of the electric bed and the patient from the training data; Step C02: According to the current state s and the ε-greedy strategy, a control action a is selected in combination with the output of the deep Q network, an action is randomly selected with a probability of ε, and the action with the largest current Q value is selected with a probability of 1-ε; Step C03: Execute action a, observe the next state s' and the immediate reward r calculated by the performance evaluation unit; Step C04: storing the state transition tuple (s, a, r, s') into the experience playback memory; Step C05: Randomly extract a small batch of state transition tuples from the experience replay memory to update the parameters of the deep Q network. The update method uses the gradient descent method to minimize the loss function. The loss function is defined as: L=(r+γ*max_a'Q'(s',a')-Q(s,a))^2 Among them, Q(s,a) is the Q value estimate of the deep Q network for the current state s and action a, and Q'(s',a') is the Q value estimate of the target Q network for the next state s' and the action with the largest Q value among all possible actions; Step D: At regular intervals, copy the parameters of the deep Q network to the target Q network; Step E: Repeat step C until the preset number of training rounds is reached or the output of the deep Q network converges; Step F: After training, the deep Q network is used as a multimodal control strategy model. For any given electric bed and patient state s, the action a with the largest Q value is selected as the optimal control action through the output of the deep Q network.

9. The multi-modal intelligent control system of an electric hospital bed according to claim 8, characterized in that: Before model training begins, the training data is further preprocessed, including: Perform one-hot encoding conversion on discrete features; The Z-score standardization method is used to standardize continuous features. Specifically, each numerical data is subtracted from its mean and divided by its standard deviation so that the processed data conforms to the standard normal distribution. The standardization formula is: Z = (X-μ) / σ, where X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

10. A multi-modal intelligent control method for an electric hospital bed, characterized in that: The method comprises: S1: Real-time monitoring of the status of the electric bed and the patient's physiological parameters as data input for the bed status monitoring layer; S2: Construct a bed control model and model the control process of the electric bed as a multi-objective optimization problem. The specific steps include: S201: define the current state space of the electric bed and the patient; S202: defining the action space for controlling the electric hospital bed; S203: defining an evaluation function to quantitatively evaluate the control results according to the accuracy, response speed, energy consumption, patient comfort and safety of the control action; S204: Using a reinforcement learning algorithm, select the optimal control action under a given state, and optimize the control strategy through continuous iterative learning; S3: Receive data input from the bed status monitoring layer, pass it to the bed control model, and drive the electric bed and its auxiliary equipment to perform corresponding actions according to the control strategy output by the model.

Images