Electric adjusting system for pressure in anaesthetic mask tube

Abstract

The invention relates to the technical field of medical instruments, and discloses an electric adjusting system for pressure in an anaesthetic mask tube. The system comprises a physiological data acquisition module for acquiring multi-source physiological data to generate a synchronous data set; the multi-modal feature modeling module is used for generating a multi-dimensional joint probability feature matrix by using a Bayesian probability graph model; the dynamic pressure decision module is used for constructing a self-adaptive controller output pressure regulation target value based on deep reinforcement learning; and the multi-target cooperative control module generates a steady-state pressure control strategy through a mixed integer programming model and an improved ant colony algorithm. In addition, the system is further provided with a virtual pressure estimation module for pressure leakage diagnosis and redundant valve switching, and an abnormal interruption protection module for guaranteeing the safety of a patient. By means of the system, precise and intelligent adjustment of the pressure in the anaesthetic mask tube is achieved, and the safety and stability of the anesthesia process are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical devices, in particular to an electric pressure regulating system for an anesthesia mask tube. Background Art

[0002] In modern medical anesthesia, the anesthesia mask is a critical device for maintaining airway patency, delivering anesthetic gases, and ensuring respiratory support. Precisely controlling the pressure within the anesthesia mask tube is crucial for ensuring patient safety and enhancing anesthetic effectiveness. However, existing anesthesia mask pressure regulation technology has numerous shortcomings, making it difficult to meet clinical needs.

[0003] Traditionally, anesthesia mask pressure adjustment relies heavily on manual operation by medical staff, a method that is not only inefficient but also significantly affected by human factors. Medical staff must constantly monitor the patient's physiological status and pressure data, manually adjusting the valve opening to control pressure. In emergency situations, manual adjustment cannot guarantee speed and accuracy, which can easily lead to excessive pressure fluctuations, affecting the patient's respiratory function and anesthesia effectiveness. For example, during surgery, a patient's physiological state may suddenly change, requiring rapid adjustment of the anesthesia mask pressure. However, manual adjustment can be delayed and fail to meet the patient's needs in a timely manner.

[0004] Some anesthesia masks that use simple electronic control systems, although they have achieved a certain degree of automation, have relatively simple functions. These systems can usually only adjust the pressure according to preset fixed parameters, and cannot fully consider the individual differences of patients and the dynamic changes in their physiological status in real time. Different patients have different ages, weights, conditions, and breathing patterns, and their requirements for anesthesia mask pressure also vary. The existing simple control systems cannot flexibly adjust to these differences, which may cause the pressure to be too high or too low, causing discomfort or even harm to the patient. For example, for pediatric patients, their respiratory systems are more fragile and require more precise pressure adjustment. The existing simple control systems are difficult to meet this demand.

[0005] In addition, during anesthesia, the patient's physiological indicators such as respiratory rate and blood oxygen saturation will continue to change. These changes are closely related to the pressure inside the anesthesia mask tube. However, most current regulation systems lack effective integration and analysis of multi-source physiological data, and are unable to establish a dynamic correlation model between pressure and physiological indicators. This makes pressure regulation lack a scientific basis and difficult to achieve precise control. For example, when the patient's respiratory rate increases, the existing regulation system may not be able to adjust the pressure in time to adapt to the patient's breathing needs, thereby affecting the patient's gas exchange and anesthesia effect.

[0006] Existing technologies also have limitations when it comes to pressure monitoring. Pressure monitoring at critical locations, such as the endotracheal tube connection, is inaccurate, making it difficult to detect potential pressure leaks in a timely manner. Pressure leaks not only waste anesthetic gas and increase medical costs, but can also affect the patient's depth of anesthesia and respiratory function, endangering their lives. Furthermore, when system anomalies occur, existing protection mechanisms are inadequate, preventing prompt and effective action to ensure patient safety. Summary of the Invention

[0007] The object of the present invention is to provide an electric pressure regulating system for an anesthesia mask tube to solve the problems raised in the above background technology.

[0008] To achieve the above object, the present invention provides the following technical solution: an electric pressure regulating system for anesthesia mask tube, the system comprising:

[0009] Physiological data acquisition module, used to collect the anesthesia mask internal pressure signal, patient respiratory rate and blood oxygen saturation data in real time, and generate a multi-source physiological synchronous data set;

[0010] A multimodal feature modeling module is used to input the multi-source physiological synchronization data set into a Bayesian probabilistic graphical model, extract the correlation features of respiratory phase, pressure fluctuation and physiological state through node conditional probability distribution and dynamic latent variable inference algorithm, and generate a multidimensional joint probability feature matrix;

[0011] A dynamic pressure decision module is used to construct an adaptive controller based on deep reinforcement learning, perform real-time strategy optimization on the multi-dimensional joint probability feature matrix, and output a pressure regulation target value;

[0012] The multi-objective collaborative control module is used to establish a mixed integer programming model based on the pressure regulation target value, use an improved ant colony algorithm to optimize the opening gradient, response delay and energy consumption of the electric valve, and generate a steady-state pressure control strategy.

[0013] Preferably, noise suppression and time alignment processing are performed on the multi-source physiological synchronization data set based on Kalman filtering and dynamic time warping algorithm, specifically including:

[0014] Separate the high-frequency noise component and the low-frequency trend component in the pressure signal, and use the Kalman gain matrix to recursively filter the high-frequency component;

[0015] Dynamic time warping algorithm is used to align the respiratory rate sequence and pressure waveform to eliminate the time offset of physiological signals;

[0016] The blood oxygen saturation data are locally smoothed by sliding window interpolation method to generate a multi-source dataset with consistent time domain.

[0017] Preferably, generating a multi-dimensional joint probability feature matrix includes:

[0018] Define respiratory phase nodes, pressure state nodes, and physiological indicator nodes as Bayesian network vertices and construct a conditional probability transition table;

[0019] The variational inference algorithm is used to calculate the posterior distribution of latent variables and iteratively update the dependency weights between nodes;

[0020] The joint probability density function is generated through the Gibbs sampling strategy, and the output is a feature matrix containing nonlinear interaction relationships.

[0021] Preferably, the construction of an adaptive controller based on deep reinforcement learning includes:

[0022] The designed state space is a triplet of current pressure deviation, respiratory cycle phase angle, and historical adjustment actions;

[0023] The reward function is defined as a linear combination of pressure stability weight, valve action smoothness and energy consumption penalty;

[0024] A dual-depth Q network architecture is used to train the strategy network and value network in parallel to output the optimal valve control instructions.

[0025] Preferably, the step of constructing the mixed integer programming model includes:

[0026] The objective function is set as the Pareto trade-off of minimizing the pressure fluctuation variance, minimizing the valve life loss coefficient, and maximizing the energy efficiency;

[0027] The added constraints are the maximum allowable pressure deviation threshold, valve mechanical stroke limit, and breathing cycle synchronization tolerance interval;

[0028] The pheromone volatilization factor is introduced to adaptively adjust the exploration ability of the ant colony algorithm and screen the non-inferior solution set to generate the control strategy.

[0029] Preferably, the multi-source physiological synchronization dataset includes:

[0030] The real-time pressure waveform collected by the integrated piezoelectric sensor, the tidal volume curve recorded by the infrared respiratory sensor, and the physiological parameters monitored by the photoelectric blood oxygen sensor;

[0031] A tensor decomposition algorithm is used to complete the missing data with low-rank approximation and construct a complete spatiotemporal data tensor;

[0032] Adaptive quantile normalization is used to eliminate dimension differences and distribution offsets between sensors.

[0033] Preferably, the system further comprises:

[0034] A virtual pressure estimation module based on a generative adversarial network was constructed to simulate and reconstruct the turbulent pressure distribution at the endotracheal tube connection.

[0035] The reconstruction result is compared with the measured pressure data to generate a residual vector, which triggers the pressure leak diagnosis and redundant valve switching logic.

[0036] Preferably, the system further comprises:

[0037] Design an abnormal interrupt protection module based on fuzzy logic. When a sudden change in respiratory rate or an excess of blood oxygen saturation is detected, the membership function is used to calculate the risk level.

[0038] The preset safety pressure curve is switched according to the risk level, and the buzzer alarm signal and valve emergency locking command are triggered.

[0039] Preferably, the present invention also includes an electronic device, which includes a processor, a memory, and a computer program stored in the memory and runnable on the processor, and when the processor executes the computer program, the functions of each module in the above-mentioned electric pressure adjustment system for the anesthesia mask tube are realized.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] The electric pressure regulation system for anesthesia mask tubes presented in this invention demonstrates significant benefits in multiple areas, significantly improving the accuracy, safety, and intelligence of pressure regulation during anesthesia. Regarding precise data acquisition and processing, the physiological data acquisition module integrates multiple sensors to collect real-time data on the anesthesia mask's internal pressure signal, the patient's respiratory rate, and blood oxygen saturation. Missing data is supplemented using a tensor decomposition algorithm, adaptive quantile normalization eliminates dimensional differences, and Kalman filtering and dynamic time warping algorithms are used for noise suppression and time alignment. This ensures that the acquired multi-source physiological synchronized dataset is accurate, complete, and time-domain consistent, providing a solid foundation for subsequent precise analysis and decision-making. For example, Kalman filtering effectively suppresses high-frequency noise components in the pressure signal, preventing noise interference with pressure monitoring and regulation, enabling the system to more accurately perceive the patient's true pressure needs. The dynamic time warping algorithm aligns the respiratory rate sequence with the pressure waveform, eliminating time offsets in the physiological signals and ensuring data relevance and synchronization, allowing the system to accurately regulate pressure based on the patient's real-time respiratory status.

[0042] The multimodal feature modeling module inputs multi-source physiological synchronous datasets into a Bayesian probabilistic graphical model. It constructs a conditional probability transition table by defining respiratory phase, pressure state, and physiological indicator nodes. It uses a variational inference algorithm to calculate the posterior distribution of latent variables and update node dependency weights. It then utilizes a Gibbs sampling strategy to generate a multidimensional joint probability feature matrix. This process can deeply explore the complex nonlinear interactions between respiratory phase, pressure fluctuations, and physiological state, providing rich and accurate feature information for pressure decision-making. Compared with traditional methods, it is no longer limited to simple linear relationship analysis, but rather comprehensively captures the potential connections between data. For example, when analyzing the relationship between pressure fluctuations and blood oxygen saturation at different respiratory phases of patients, it can more accurately grasp its inherent laws, thereby providing a more targeted basis for pressure regulation.

[0043] The dynamic pressure decision module constructs an adaptive controller based on deep reinforcement learning, designs a reasonable state space and reward function, and uses a dual deep Q-network architecture to parallel train the policy network and value network. In this way, the system can perform real-time policy optimization based on real-time pressure deviation, respiratory cycle phase angle, and historical adjustment actions, outputting a more accurate pressure regulation target value. The reward function comprehensively considers the pressure stability weight, valve movement smoothness, and energy consumption penalty terms, allowing the system to not only ensure pressure stability during pressure regulation, but also optimize valve movement and reduce energy consumption. For example, during surgery, the patient's respiratory status may change at any time. The module can quickly adjust the pressure regulation strategy based on real-time data to ensure that the pressure in the anesthesia mask tube is always within the appropriate range, while reducing unnecessary and frequent valve movements, extending valve life, and reducing energy consumption.

[0044] The multi-objective collaborative control module establishes a mixed integer programming model based on the target pressure regulation value, sets appropriate objective functions and constraints, and uses an improved ant colony algorithm to optimize the electric valve's opening gradient, response delay, and energy consumption to generate a steady-state pressure control strategy. This module achieves collaborative optimization of multiple objectives, minimizing the variance of pressure fluctuations while reducing valve life loss coefficients and improving energy efficiency. For example, by optimizing the valve opening gradient, pressure regulation is smoother, preventing the adverse effects of sudden pressure changes on patients; by properly controlling response delay, the system can respond promptly to pressure changes; and by optimizing energy consumption, medical costs are reduced.

[0045] The system also incorporates a virtual pressure estimation module based on a generative adversarial network and an emergency interruption protection module based on fuzzy logic. The virtual pressure estimation module simulates and reconstructs the turbulent pressure distribution at the endotracheal tube connection. By comparing the reconstructed results with measured pressure data, it generates a residual vector, which implements pressure leak diagnosis and redundant valve switching logic, significantly improving system reliability and safety. When a pressure leak is detected, it promptly switches to a redundant valve, ensuring the continuity of the anesthesia process and preventing pressure leaks from compromising the patient's anesthesia or endangering their life. When the emergency interruption protection module detects a sudden change in respiratory rate or an excessive oxygen saturation, it uses a membership function to calculate the risk level. Based on the risk level, it switches to a preset safety pressure curve, triggering a buzzer alarm and an emergency valve lock command, effectively ensuring patient safety during anesthesia. For example, if a patient experiences an emergency situation such as a sudden increase in respiratory rate or a sharp drop in oxygen saturation, the module can quickly respond and take appropriate measures to prevent further danger. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a working principle diagram of an electric pressure regulating system for anesthesia mask tube according to the present invention;

[0047] Figure 2 Schematic diagram for building an adaptive controller based on deep reinforcement learning;

[0048] Figure 3 Diagram of the steps for processing multi-source physiological synchronized datasets;

[0049] Figure 4 Flowchart of virtual pressure estimation and related logic. DETAILED DESCRIPTION

[0050] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0051] See also Figures 1-4 The present invention provides an electric pressure adjustment system for anesthesia mask tubes, which aims to achieve precise and intelligent adjustment of the pressure inside the anesthesia mask tubes to meet the physiological needs of patients during anesthesia. The overall implementation scheme is as follows:

[0052] The physiological data acquisition module is responsible for real-time acquisition of the anesthesia mask internal pressure signal, the patient's respiratory rate, and blood oxygen saturation data, generating a multi-source synchronized physiological data set. By integrating multiple sensors, such as a piezoelectric sensor for real-time pressure waveform acquisition, an infrared respiratory sensor for recording tidal volume curves, and a photoelectric blood oxygen sensor for monitoring physiological parameters, comprehensive and accurate physiological information is obtained.

[0053] The multimodal feature modeling module inputs multi-source physiological synchronized datasets into a Bayesian probabilistic graphical model. Within this model, respiratory phase nodes, pressure state nodes, and physiological indicator nodes are defined as Bayesian network vertices, and a conditional probability transition table is constructed. Next, a variational inference algorithm is used to calculate the posterior distribution of latent variables, iteratively updating the dependency weights between nodes. Finally, a Gibbs sampling strategy is used to generate a multidimensional joint probability feature matrix containing correlated features of respiratory phase, pressure fluctuation, and physiological state.

[0054] The dynamic pressure decision module constructs an adaptive controller based on deep reinforcement learning. The state space is designed as a triplet of current pressure deviation, respiratory cycle phase angle, and historical regulation actions. The reward function is defined as a linear combination of pressure stability weights, valve motion smoothness, and energy consumption penalty terms. A dual deep Q-network architecture is used to train the policy network and value network in parallel, enabling real-time policy optimization of the multidimensional joint probability feature matrix to output the target pressure regulation value.

[0055] The multi-objective collaborative control module establishes a mixed integer programming model based on the pressure regulation target. The objective function is set as a Pareto trade-off between minimizing the pressure fluctuation variance, minimizing the valve life loss coefficient, and maximizing energy efficiency. Constraints such as the maximum allowable pressure deviation threshold, valve mechanical travel limit, and respiratory cycle synchronization tolerance are also added. An improved ant colony algorithm is used to optimize the electric valve's opening gradient, response delay, and energy consumption, generating a steady-state pressure control strategy to effectively regulate the pressure within the anesthesia mask tube.

[0056] The implementation of the present invention will be further described below with reference to Examples 1 to 6.

[0057] Example 1:

[0058] In practical applications, in order to ensure the quality of the collected multi-source physiological synchronous data sets, a series of preprocessing operations are required on the data.

[0059] During the data acquisition phase, the integrated piezoelectric sensor, infrared respiratory sensor, and photoelectric blood oxygen sensor are used to collect real-time pressure waveforms, tidal volume curves, and physiological parameters. Since the data collected by the sensors may contain missing values, the tensor decomposition algorithm is used to perform low-rank approximation to complete the missing data. Assume that the collected data tensor is X∈R I×J×K, where I represents the time dimension, J represents the sensor type dimension, and K represents the sample dimension. Through the tensor decomposition algorithm, X is decomposed into a core tensor G and multiple factor matrices U i (i=1,2,…), that is, X≈G×1U1×2U2×3U3. In this way, the missing data is completed and a complete spatiotemporal data tensor is constructed.

[0060] In order to eliminate the dimension difference and distribution offset between sensors, the adaptive quantile normalization method is used. For the data sequence x=[x1,x2,…,x n ], calculate its quantile q(x), and then normalize the data according to the adaptive rule to make the data of different sensors comparable.

[0061] In the data processing stage, noise suppression and time alignment are performed on multi-source physiological synchronization data sets based on Kalman filtering and dynamic time warping algorithms.

[0062] For the pressure signal, it is separated into high-frequency noise components and low-frequency trend components. Assuming that the pressure signal is p(t), through a specific filtering algorithm, it can be expressed as p(t) = p high (t)+p low (t), where p high (t) is the high-frequency noise component, p low (t) is the low-frequency trend component. The Kalman gain matrix K is used to calculate the high-frequency component p high (t) performs recursive filtering. The Kalman gain matrix K is calculated based on the system's state transition matrix A, observation matrix H, process noise covariance matrix Q, and observation noise covariance matrix R. In each recursive step, the Kalman gain matrix K is updated based on the current estimated state and observation value, thereby effectively suppressing high-frequency noise components.

[0063] The dynamic time warping algorithm is used to align the respiratory rate sequence and the pressure waveform. The respiratory rate sequence r(t) and the pressure waveform p(t) may have time offsets due to physiological differences or acquisition equipment. The dynamic time warping algorithm aligns the two sequences in time by finding an optimal time warping path, eliminating the time offset of the physiological signals. Specifically, the distance matrix D(i,j) between the two sequences is calculated, where i and j represent the time points in the two sequences, respectively. Using a dynamic programming algorithm, an optimal path is found from the starting point (1,1) to the end point (m,n) (m and n are the lengths of the two sequences, respectively), so that the sum of the distances on the path is minimized, thereby achieving alignment of the respiratory rate sequence and the pressure waveform.

[0064] For the blood oxygen saturation data, local smoothing is performed using the sliding window interpolation method. Assume that the blood oxygen saturation data sequence is o(t), set a sliding window w, and the window size is N. In each window, the interpolation algorithm is used to smooth the data. For example, for the data in the window [o i ,o i+1 ,…,o i+N-1 ], linear interpolation or spline interpolation methods are used to calculate the smoothed intermediate values and generate a multi-source data set that is consistent in the time domain, providing a high-quality data foundation for subsequent data analysis and processing.

[0065] Example 2:

[0066] Generating a multi-dimensional joint probability feature matrix in the multimodal feature modeling module is a key step in achieving intelligent decision-making of the system.

[0067] In the Bayesian probabilistic graphical model, respiratory phase nodes, pressure state nodes, and physiological indicator nodes are defined as Bayesian network vertices. Respiratory phase nodes can represent different stages of the patient's breathing process, such as the start, middle, and end of the inspiration and expiration phases. Pressure state nodes reflect the changes in pressure within the anesthesia mask, such as high pressure, low pressure, and stable pressure. Physiological indicator nodes cover the states corresponding to physiological parameters such as the patient's respiratory rate and blood oxygen saturation.

[0068] A conditional probability transition table is constructed to describe the probabilistic dependencies between nodes. Assume that the respiratory phase node has m states, the pressure state node has n states, and the physiological indicator node has k states. For each respiratory phase state i (i = 1, …, m), pressure state j (j = 1, …, n), and physiological indicator state l (l = 1, …, k), a conditional probability P(j|i, l) is defined, representing the probability of the pressure state being j given the known respiratory phase and physiological indicator state. Through statistical analysis of extensive historical data, the values of these conditional probabilities are determined, and a complete conditional probability transition table is constructed.

[0069] A variational inference algorithm is used to calculate the posterior distribution of the latent variable and iteratively update the dependency weights between nodes. The variational inference algorithm introduces a variational distribution q(z) to approximate the true posterior distribution P(z|x) of the latent variable z, where x is the observed data. The parameters of the variational distribution are continuously optimized by minimizing the KL divergence (KL(q(z)||P(z|x)) between the variational distribution q(z) and the true posterior distribution P(z|x). In each iteration, the dependency weights between nodes are updated based on the current variational distribution and the observed data, allowing the model to better reflect the underlying relationships in the data.

[0070] The joint probability density function is generated by the Gibbs sampling strategy. Gibbs sampling is a sampling method based on Markov Chain Monte Carlo (MCMC). Starting from the initial state, each node is sampled in turn, and the sampling probability of the node is calculated based on the current state of other nodes and the conditional probability transition table. For example, for the respiratory phase node i, when other nodes are fixed, according to the conditional probability P(i|j1,…,j s )(j1,…,j s The system samples the states of other nodes. This process generates a series of samples whose distribution approximates the joint probability density function. Ultimately, these samples generate a feature matrix containing nonlinear interactions. This matrix comprehensively reflects the complex relationships between respiratory phase, pressure fluctuations, and physiological state, providing rich feature information for subsequent dynamic pressure decision-making.

[0071] Example 3:

[0072] The dynamic pressure decision module achieves optimized output of the pressure regulation target value by constructing an adaptive controller based on deep reinforcement learning.

[0073] The design state space is a triplet of current pressure deviation, respiratory cycle phase angle, and historical adjustment action. The current pressure deviation Δp is defined as the current actual pressure p actual With the preset target pressure p target The difference, that is, Δp=p actual -p target The respiratory cycle phase angle θ describes the patient's current position in the respiratory cycle. Its value range is typically [0, 2π] and can be calculated by monitoring respiratory rate and time. The historical adjustment action records previous adjustments to the electric valve, such as the change in valve opening and adjustment time.

[0074] The reward function is defined as a linear combination of the pressure stability weight, valve action smoothness, and energy consumption penalty. Let the reward function be R, the pressure stability weight be w1, the valve action smoothness weight be w2, and the energy consumption penalty weight be w3. The pressure stability weight w1 reflects the importance attached to pressure stability. The larger w1 is, the more the system tends to maintain pressure stability. The valve action smoothness weight w2 is used to measure the smoothness of valve action and avoid frequent and drastic valve adjustments. The energy consumption penalty weight w3 is used to punish high-energy consumption adjustment actions, prompting the system to reduce energy consumption while adjusting pressure. The specific reward function can be expressed as R = w1×S-w2×|aa prev |-w3×E, where S is the pressure stability index, for example, it can be the inverse of the pressure deviation; a is the current valve adjustment action, a previs the last valve adjustment action; E is the energy consumption index, for example, it can be the power consumption during the electric valve adjustment process.

[0075] The dual deep Q network architecture is used to train the policy network and value network in parallel. The dual deep Q network (DDQN) reduces the overestimation problem existing in the deep Q network (DQN) by decoupling action selection and action evaluation. The policy network π(s; θ π ) Output an action a according to the current state s, where θ π are the parameters of the policy network. The value network Q(s,a;θ Q ) evaluates the value of performing action a in state s, where θ Q is the parameter of the value network. During the training process, an action a is first selected according to the policy network, and then the value network is used to evaluate the value of the action. By minimizing the loss function L(θ Q ) to update the parameters of the value network. The loss function is typically calculated based on the temporal difference error. Simultaneously, the parameters of the value network are periodically copied to the target value network for target value calculation, improving training stability. Through continuous training, the policy network gradually learns the optimal valve control strategy and outputs the optimal valve control instructions to effectively regulate the pressure within the anesthesia mask tube.

[0076] Example 4:

[0077] The multi-objective collaborative control module generates a steady-state pressure control strategy by constructing a mixed integer programming model and using an improved ant colony algorithm to optimize the opening gradient, response delay and energy consumption of the electric valve.

[0078] The steps to construct the mixed integer programming model are as follows:

[0079] The objective function is set as the Pareto trade-off of minimizing the pressure fluctuation variance, minimizing the valve life loss coefficient, and maximizing the energy efficiency. Assume that the pressure fluctuation variance is The valve life loss coefficient is c and the energy efficiency is η. The objective function can be expressed as ω1, ω2, and ω3 are weight coefficients used to balance the importance of different objectives. By adjusting these weight coefficients, you can optimize different objectives based on actual needs. For example, if pressure stability is a high requirement, you can increase the value of w1 appropriately.

[0080] The added constraints are the maximum allowable pressure deviation threshold, valve mechanical stroke limit and breathing cycle synchronization tolerance range. Maximum allowable pressure deviation threshold Δp max The range of pressure fluctuation is limited to ensure that the pressure inside the anesthesia mask does not exceed the safe range. The valve mechanical stroke limit stipulates the maximum and minimum opening of the electric valve. Assume that the maximum opening of the valve is amax , the minimum opening is a min , then the valve opening a must satisfy a min ≤a≤a max The respiratory cycle synchronization tolerance interval is used to ensure the synchronization of valve adjustment and patient respiratory cycle. Assuming the respiratory cycle is T, the synchronization tolerance interval is [T min ,T max ] During the adjustment process, the valve action time must be within this tolerance range to avoid adverse effects on the patient's breathing.

[0081] A pheromone volatility factor is introduced to adaptively adjust the ant colony algorithm's exploration capability, screening a set of non-inferior solutions to generate a control strategy. The ant colony algorithm (ACA) is an optimization algorithm that simulates the foraging behavior of ants. Ants release pheromones along a path, guiding other ants to find the optimal path. In this improved ACA, a pheromone volatility factor, ρ, is introduced. As the number of iterations increases, the pheromone gradually evaporates, preventing premature convergence. Furthermore, the value of the pheromone volatility factor is adaptively adjusted based on the characteristics of the problem. During the algorithm's execution, ants select paths based on pheromone concentration and heuristic information, updating the pheromone concentration after each iteration. Through multiple iterations, a set of non-inferior solutions is identified—a set of solutions that meet multiple objectives. From this set of non-inferior solutions, a suitable solution is selected as the steady-state pressure control strategy to optimize the electric valve opening gradient, response delay, and energy consumption, ensuring that the pressure inside the anesthesia mask tube remains stable within an appropriate range.

[0082] Example 5:

[0083] This system introduces a virtual pressure estimation module based on a generative adversarial network, whose purpose is to simulate and reconstruct the turbulent pressure distribution at the endotracheal tube connection, thereby realizing the functions of pressure leak diagnosis and redundant valve switching, thereby improving the stability and reliability of the system.

[0084] The Generative Adversarial Network (GAN) is mainly composed of a generator and a discriminator. The generator receives a random noise vector as input. In practical application scenarios, the random noise vector can be a set of data randomly extracted from a specific distribution (such as a normal distribution), represented by the symbol z. The generator processes the input random noise vector through its own complex neural network structure and finally outputs the simulated turbulent pressure distribution data at the endotracheal tube connection, which is recorded as The role of the discriminator is to judge the authenticity of the input data. Its input includes the real pressure data p and the simulated data generated by the generator. The neural network inside the discriminator extracts and analyzes the features of the input data, and then outputs a judgment result indicating whether the input data is real data or simulated data.

[0085] When training a generative adversarial network, the generator and discriminator engage in adversarial training. The generator's goal is to continuously adjust the parameters of its neural network so that the generated simulated data increasingly resembles the actual pressure distribution, thereby deceiving the discriminator into interpreting its output as real data. The discriminator, in turn, strives to improve its ability to distinguish between real and simulated data. In this process, both parties continuously optimize their parameters, gradually reaching a state of equilibrium. As training continues, the generator is able to produce simulated data that closely resembles the actual pressure distribution.

[0086] After completing the simulation reconstruction, the reconstruction result is compared with the measured pressure data to generate a residual vector. Assume that the measured pressure data is p real , the simulated reconstructed pressure data is p sim , then the residual vector r is calculated as r = p real -p sim In order to determine whether there is a pressure leak, a preset threshold ∈ needs to be set. When a component of the residual vector exceeds this preset threshold ∈, the system will determine that there may be a pressure leak problem.

[0087] Once a pressure leak is detected, the system quickly triggers pressure leak diagnosis and redundant valve switching logic. First, the system conducts an in-depth analysis of the residual vector, examining its characteristics (such as its magnitude and changing trend) as well as its temporal and spatial distribution to determine the location and extent of the leak. For example, if at a certain moment, the residual vector component in a specific area increases sharply and remains high for a period of time, it can be preliminarily determined that a pressure leak exists in that area. After determining the leak, the system rapidly switches to the backup valve based on the pre-set redundant valve configuration. During the switching process, the system ensures that the backup valve's opening and adjustment strategy match those of the primary valve, thereby maintaining stable pressure within the anesthesia mask tube and ensuring the patient's normal anesthesia. The system also records detailed information about the pressure leak, including the time of occurrence, leak location, and related pressure data, for subsequent troubleshooting and system maintenance.

[0088] Example 6:

[0089] The system is specially designed with an abnormal interruption protection module based on fuzzy logic, which plays a key role in ensuring the safety of the patient's anesthesia process.

[0090] When the system detects a sudden change in respiratory rate or an over-limit of blood oxygen saturation, it uses the membership function to calculate the risk level. Assuming that the respiratory rate is represented by f, the preset normal respiratory rate range is [f min ,f max ]; blood oxygen saturation The normal blood oxygen saturation range is [S min ,S max ]. For the cases of respiratory rate mutation and blood oxygen saturation exceeding the limit, the membership function μ is defined respectively. f (f) and Taking the membership function of respiratory frequency mutation as an example, the Gaussian membership function can be selected, and its expression is where f center Represents the center value of normal respiratory rate, which can be taken from the normal respiratory rate range [f min ,f max ], that is σ is the standard deviation, which determines the shape and change speed of the membership function and can be adjusted according to actual conditions. When the respiratory rate f exceeds the normal range, (ff center ) 2 The value of will increase, resulting in μ f The value of (f) also increases, which means that the possibility of sudden changes in respiratory rate increases. Similarly, the membership function for blood oxygen saturation exceeding the limit can also be defined in a similar function form to accurately reflect the degree to which blood oxygen saturation deviates from the normal range.

[0091] Based on the calculated membership function values for respiratory rate mutation and blood oxygen saturation exceeding the limit, the system calculates the risk level through fuzzy inference rules. Generally, the risk level is divided into three levels: low risk, medium risk, and high risk. For example, if the membership function values for respiratory rate mutation and blood oxygen saturation exceeding the limit are both low, it means that although the patient's physiological state has undergone certain changes, it is still within a relatively safe range. In this case, the risk level is judged to be low risk; if one of the membership function values is high and the other is at a medium level, it indicates that the patient's physiological state has a certain risk, and the risk level is medium risk; when both membership function values are high, it means that the patient's physiological state faces a greater risk, and the risk level is high risk.

[0092] The system will take appropriate measures for different risk levels. When the risk level is low, the system will issue a prompt message to remind medical staff to pay attention to changes in the patient's physiological state so that potential problems can be detected in time. If the risk level is medium, the system will immediately switch to the preset safety pressure curve. This safety pressure curve is pre-set based on a large amount of clinical data and professional knowledge. While ensuring patient safety, the pressure in the anesthesia mask tube is appropriately adjusted to maintain the patient's physiological stability. At the same time, the system will trigger a buzzer alarm signal to attract the attention of medical staff, enabling them to take further measures in a timely manner. When the risk level reaches high, in addition to switching to the safety pressure curve and triggering the buzzer alarm signal, the system will immediately trigger the valve emergency lock command. This command will quickly close the relevant valves to prevent the abnormal pressure from causing further harm to the patient, thereby effectively ensuring the patient's safety during anesthesia.

[0093] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0094] While 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 these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An electric pressure regulating system for anesthesia mask tube, characterized in that: The system comprises: Physiological data acquisition module, used to collect the internal pressure signal of the anesthesia mask, the patient's respiratory rate and blood oxygen saturation data in real time, and generate a multi-source physiological synchronous data set; A multimodal feature modeling module is used to input the multi-source physiological synchronization data set into a Bayesian probabilistic graphical model, extract the correlation features of respiratory phase, pressure fluctuation and physiological state through node conditional probability distribution and dynamic latent variable inference algorithm, and generate a multidimensional joint probability feature matrix; A dynamic pressure decision module is used to construct an adaptive controller based on deep reinforcement learning, perform real-time strategy optimization on the multi-dimensional joint probability feature matrix, and output a pressure regulation target value; The multi-objective collaborative control module is used to establish a mixed integer programming model based on the pressure regulation target value, use an improved ant colony algorithm to optimize the opening gradient, response delay and energy consumption of the electric valve, and generate a steady-state pressure control strategy.

2. The electric pressure regulating system for the anesthesia mask tube according to claim 1, characterized in that: The multi-source physiological synchronization data set is subjected to noise suppression and time series alignment processing based on Kalman filtering and dynamic time warping algorithms, specifically including: Separate the high-frequency noise component and the low-frequency trend component in the pressure signal, and use the Kalman gain matrix to recursively filter the high-frequency component; Dynamic time warping algorithm is used to align the respiratory rate sequence and pressure waveform to eliminate the time offset of physiological signals; The blood oxygen saturation data are locally smoothed by sliding window interpolation method to generate a multi-source dataset with consistent time domain.

3. The electric pressure regulating system for the anesthesia mask tube according to claim 1, characterized in that: The generating of a multi-dimensional joint probability feature matrix includes: Define respiratory phase nodes, pressure state nodes, and physiological indicator nodes as Bayesian network vertices and construct a conditional probability transition table; The variational inference algorithm is used to calculate the posterior distribution of latent variables and iteratively update the dependency weights between nodes; The joint probability density function is generated through the Gibbs sampling strategy, and the output is a feature matrix containing nonlinear interaction relationships.

4. The electric pressure regulating system for anesthesia mask tube according to claim 1, characterized in that: The method comprises: constructing an adaptive controller based on deep reinforcement learning; The designed state space is a triplet of current pressure deviation, respiratory cycle phase angle, and historical adjustment actions; The reward function is defined as a linear combination of pressure stability weight, valve action smoothness and energy consumption penalty; A dual-depth Q network architecture is used to train the strategy network and value network in parallel to output the optimal valve control instructions.

5. The electric pressure regulating system for the anesthesia mask tube according to claim 1, characterized in that: The steps of constructing the mixed integer programming model include: The objective function is set as the Pareto trade-off of minimizing the pressure fluctuation variance, minimizing the valve life loss coefficient and maximizing the energy efficiency; The added constraints are the maximum allowable pressure deviation threshold, valve mechanical stroke limit, and breathing cycle synchronization tolerance interval; The pheromone volatilization factor is introduced to adaptively adjust the exploration ability of the ant colony algorithm and screen the non-inferior solution set to generate the control strategy.

6. The electric pressure regulating system for anesthesia mask tube according to claim 1, characterized in that: The multi-source physiological synchronization dataset includes: The real-time pressure waveform collected by the integrated piezoelectric sensor, the tidal volume curve recorded by the infrared respiratory sensor, and the physiological parameters monitored by the photoelectric blood oxygen sensor; A tensor decomposition algorithm is used to complete the missing data with low-rank approximation to construct a complete spatiotemporal data tensor; Adaptive quantile normalization is used to eliminate dimension differences and distribution offsets between sensors.

7. The electric pressure regulating system for anesthesia mask tube according to claim 1, characterized in that: The system further comprises: A virtual pressure estimation module based on a generative adversarial network was constructed to simulate and reconstruct the turbulent pressure distribution at the endotracheal tube connection. The reconstruction result is compared with the measured pressure data to generate a residual vector, which triggers the pressure leak diagnosis and redundant valve switching logic.

8. The electric pressure regulating system for anesthesia mask tube according to claim 1, characterized in that: The system further comprises: Design an abnormal interrupt protection module based on fuzzy logic. When a sudden change in respiratory rate or an excess of blood oxygen saturation is detected, the membership function is used to calculate the risk level. The preset safety pressure curve is switched according to the risk level, and the buzzer alarm signal and valve emergency locking command are triggered.

9. An electronic device, characterized in that: The system comprises a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the functions of the various modules of the electric pressure regulating system for anesthesia mask tube according to any one of claims 1 to 8 when executing the computer program.

Images