Intelligent analgesia control model establishment method and system

Abstract

The invention relates to the technical field of medical artificial intelligence, and discloses an intelligent analgesia control model establishment method and system, and the system extracts tissue thermal state features through a multi-modal data fusion module, achieves the rolling optimization of a temperature track based on a recurrent neural network predictor and a model prediction controller, and achieves the control of analgesia. And the weight of the predictor is updated online by means of a dynamic calibration module, so that the control model is adaptive to real-time treatment data change, and the long-term precision, stability and individualized analgesic effect of temperature control are improved.

Description

Technical Field

[0001] This invention belongs to the field of medical artificial intelligence technology, specifically relating to a method and system for establishing an intelligent analgesia control model. Background Technology

[0002] In the field of medical device technology, closed-loop automatic control systems, through real-time monitoring and feedback adjustment mechanisms, achieve precise control of treatment parameters, thereby improving the safety and effectiveness of treatment, and have become an important component of modern precision medicine. Among these, non-invasive analgesia devices, as an important type of medical equipment, focus on achieving the desired physiological effects through precise control of energy or substance output.

[0003] Among them, intelligent analgesia control technology aims to precisely regulate the output dosage and rate of media such as cryogens through a closed-loop automatic control mechanism, achieving and maintaining a preset effective treatment temperature range in a specific target area of ​​the human body, thereby achieving non-invasive and controllable analgesia. Existing technologies typically integrate temperature sensing, control algorithms, and actuators to form a closed loop from monitoring to regulation.

[0004] While existing technologies can achieve basic temperature control, they face significant limitations in practical applications: the systems are insufficiently adaptable to individual physiological differences, changes in tissue thermal properties, and environmental disturbances, making it difficult to guarantee the accuracy and stability of temperature control; traditional control models rely on fixed parameters or simple feedback, failing to dynamically learn and optimize based on real-time treatment data, thus affecting the individualization and consistency of analgesic effects; simultaneously, the lack of effective fusion and intelligent analysis of multi-source physiological signals limits the system's accurate perception and early warning of treatment status and risks. Therefore, there are still pressing technical challenges to be solved in achieving safe, precise, and adaptive non-invasive analgesia control. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for establishing an intelligent analgesia control model, in order to solve the problems in the prior art where the temperature control accuracy and stability are insufficient due to individual physiological differences, changes in tissue thermal properties and environmental interference, as well as the inability of traditional control models to perform dynamic learning optimization based on real-time treatment data and the lack of effective fusion and intelligent analysis of multi-source physiological signals.

[0006] To achieve the above objectives, the technical solution adopted in this invention is to construct an intelligent analgesia control system. This system includes a physiological signal acquisition module, a multimodal data fusion processing module, an adaptive predictive control module, and a treatment execution and dynamic calibration module. The physiological signal acquisition module is used to acquire temperature signals, blood perfusion rate signals, and surface electromyographic signals from the target treatment area in real time. The multimodal data fusion processing module is connected to the physiological signal acquisition module and is used to perform time alignment, noise reduction, and feature extraction on the received multi-source heterogeneous physiological signals, and to construct a nonlinear mapping from the original signals to tissue thermal state feature vectors based on a deep belief network.

[0007] The adaptive predictive control module is connected to the multimodal data fusion processing module. It integrates a recurrent neural network predictor with online learning capability and a model predictive controller. The recurrent neural network predictor predicts the tissue temperature change trajectory for multiple time steps in the future based on the current tissue thermal state feature vector and the historical control command sequence. The model predictive controller generates optimized control commands that make the predicted temperature trajectory closest to the preset treatment temperature range by solving a constrained rolling optimization problem.

[0008] The treatment execution and dynamic calibration module is connected to the adaptive predictive control module and the physiological signal acquisition module. It includes a high-precision cryoprotectant flow regulating valve and a microprocessor. On the one hand, this module executes optimized control commands from the model predictive controller to drive the flow regulating valve to precisely control the cryoprotectant output. On the other hand, it continuously compares the error between the predicted temperature trajectory and the actual measured temperature. When the accumulated error exceeds the preset adaptive trigger threshold, it automatically initiates online incremental updates to the weights inside the recurrent neural network predictor, thereby achieving dynamic calibration and self-optimization of the control model.

[0009] Furthermore, the specific operation flow of the multimodal data fusion processing module is as follows. First, the module performs unified timestamp synchronization on the temperature signal, blood perfusion rate signal, and electromyography signal input from the physiological signal acquisition module to ensure that all signal data are on the same time base. Next, a joint denoising algorithm based on wavelet transform is used to denoise the synchronized multi-channel signals to suppress power frequency interference and physiological motion artifacts.

[0010] Then, time-domain features, frequency-domain features, and nonlinear dynamic features are extracted from each denoised signal to form a high-dimensional original feature set. Finally, the high-dimensional original feature set is input into a pre-trained offline deep belief network. This network, through its multi-layer nonlinear transformation, compresses and maps the high-dimensional original feature set into a low-dimensional, dense tissue thermal state feature vector. This feature vector comprehensively represents the thermal dynamic characteristics and physiological response state of the target tissue at the current moment.

[0011] Furthermore, the offline training process of the deep belief network specifically includes the following steps: A large amount of historical physiological signal data and corresponding tissue temperature data covering different individuals and treatment sites are collected as a training sample set. A deep belief network structure containing 3 visible layers and 5 hidden layers is constructed. The contrastive divergence algorithm is used for unsupervised layer-by-layer pre-training of the network to initialize the network weights. After pre-training, the network is subjected to supervised fine-tuning training using labeled tissue temperature data to minimize the reconstruction error between the network's output tissue thermal state feature vector and the actual tissue temperature data. Finally, the trained network parameters are embedded into the multimodal data fusion processing module.

[0012] Furthermore, the recurrent neural network predictor in the adaptive predictive control module is constructed as follows. This predictor employs a long short-term memory (LSTM) network structure. Its input layer receives the current tissue thermal state feature vector output by the multimodal data fusion processing module, as well as the control command sequence for the past five time steps. The network's hidden layer contains 128 LSM units to capture long-range temporal dependencies in tissue thermodynamics.

[0013] The output layer predicts tissue temperature values ​​for the next three time steps. The mathematical description of the rolling optimization problem of the model predictive controller is as follows: in each control cycle, find a set of future control command sequences that minimizes the variance between the predicted temperature output given by the recurrent neural network predictor and the preset treatment temperature range reference trajectory, while satisfying the physical constraints of the rate of change of control commands and the absolute value of the cryocoolant flow rate.

[0014] Furthermore, the execution mechanism of online incremental updates in the treatment execution and dynamic calibration module is as follows: The microprocessor continuously calculates the root mean square error (RMSE) between the temperature predictions and actual measurements for the next three time steps using a recurrent neural network predictor. This module presets an adaptive trigger threshold of 0.5 degrees Celsius. When the calculated RMS error exceeds 0.5 degrees Celsius for 10 consecutive control cycles, the microprocessor automatically triggers the online learning program.

[0015] The online learning program uses the actual control commands and measured temperature data from the most recent 50 time steps to form an incremental training dataset. It employs a backpropagation algorithm over time to slightly adjust the weights of the recurrent neural network predictor, with a learning rate set to 0.001. The weight update process is completed within one control cycle, ensuring that the real-time performance of the control loop is not affected.

[0016] Furthermore, the intelligent analgesia control system also integrates a safety monitoring and intervention module. This module monitors the tissue thermal state feature vector output by the multimodal data fusion processing module in real time and calculates the mutation index of its feature values. The safety monitoring and intervention module has a pre-stored feature template library for normal treatment and abnormal states. When the calculated mutation index exceeds the safety threshold of 2.0 for three consecutive cycles, and the feature vector matches the pre-stored abnormal state template with a 90% degree of accuracy, the module sends a highest-priority overlay command to the adaptive predictive control module, forcing the model predictive controller to output a preset safety control mode, that is, reducing the refrigerant flow rate to the minimum level required to maintain the temperature, and simultaneously sending an alarm signal to the external monitoring terminal.

[0017] Furthermore, the control cycle of the system of this invention is fixed at 100 milliseconds. The sampling frequency of the physiological signal acquisition module is 1000 Hz. The control accuracy of the high-precision cryotherapy flow regulating valve in the treatment execution and dynamic calibration module reaches 0.1% of the full range. The coordinated design of these timing and accuracy parameters jointly ensures the real-time performance, accuracy, and stability of the system control.

[0018] Compared with the prior art, the advantages and positive effects of the present invention are as follows: 1. This invention achieves a fundamental shift in analgesia control models from static fixed parameters to dynamic self-calibration by constructing an adaptive predictive control module that integrates an online learning recurrent neural network predictor. The system can continuously optimize its internal predictive model based on real-time treatment data, effectively adapting to individual physiological differences and the time-varying characteristics of tissue thermal properties during treatment. This significantly improves the long-term accuracy and stability of temperature control, ensuring the individualization and consistency of analgesic effects.

[0019] 2. The multimodal data fusion processing module designed in this invention employs a deep belief network to extract and fuse deep features from multi-source heterogeneous physiological signals such as temperature, blood perfusion rate, and electromyography, generating feature vectors that comprehensively reflect tissue thermal dynamics and physiological states. This deep learning-based intelligent analysis method overcomes the shortcomings of traditional systems, such as strong dependence on single signals and susceptibility to interference, enhancing the perception of complex treatment states and potential risks, and providing a richer and more reliable information foundation for advanced control decisions.

[0020] 3. This invention achieves closed-loop autonomy in the control model optimization process through an error monitoring and automatic triggering mechanism based on preset thresholds in the treatment execution and dynamic calibration module. The system can autonomously determine the degree of model mismatch and perform incremental learning without external intervention, ensuring the system's adaptive maintenance capability and robustness during long-term operation. Simultaneously, the integrated high-priority safety monitoring and intervention module provides proactive safety protection for the treatment process through feature mutation analysis and template matching, further enhancing the clinical applicability and reliability of the entire system. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall technical architecture of the intelligent analgesia control system proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework of the multimodal data fusion processing module in this invention; Figure 3 This is a logical flow diagram of the adaptive predictive control module in this invention; Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow of the treatment execution and dynamic calibration module in this invention; Figure 5 This is a schematic diagram of the principle framework of the safety monitoring and intervention module in this invention. Detailed Implementation

[0022] Example 1: This example details an intelligent analgesia control system that achieves precise tissue temperature control and model self-optimization through multi-module collaboration. Please refer to the appendix. Figure 1 The system's overall architecture includes a physiological signal acquisition module, a multimodal data fusion and processing module, an adaptive predictive control module, a treatment execution and dynamic calibration module, and a safety monitoring and intervention module. These modules are connected via a high-speed data bus, forming a closed-loop control circuit. The system control cycle is fixed at 100 milliseconds, and the physiological signal sampling frequency is 1000 Hz, ensuring real-time performance and data synchronization.

[0023] The physiological signal acquisition module is responsible for acquiring multi-source physiological signals from the target treatment area in real time. This module integrates a temperature sensor, a laser Doppler flowmeter, and surface electromyography electrodes. The temperature sensor uses a thermocouple structure with a measurement accuracy of ±0.1 degrees Celsius. It is directly attached to the skin surface or implanted in subcutaneous tissue to acquire tissue temperature signals.

[0024] The laser Doppler flowmeter calculates local blood perfusion rate by emitting a 780 nm laser beam and analyzing the frequency shift of backscattered light, with output units of milliliters per minute per 100 grams of tissue. The surface electromyography (EMG) electrodes, made of silver-silver chloride, acquire EMG signals in differential amplification mode, achieving a common-mode rejection ratio greater than 100 dB to effectively suppress environmental electromagnetic interference. All sensor signals are digitized by a 24-bit analog-to-digital converter and timestamped using a synchronous clock circuit to ensure a consistent data time base.

[0025] The multimodal data fusion processing module receives the digitized signals transmitted from the physiological signal acquisition module. Please refer to the appendix. Figure 2 The module first performs a time alignment operation, aligning the temperature signal, blood perfusion rate signal, and electromyography signal to the same time axis based on the hardware timestamp, with a maximum allowable time delay deviation of less than 1 millisecond. The aligned signals then enter the joint noise reduction unit, which uses the discrete wavelet transform algorithm and selects the Db4 wavelet basis function to perform a 5-level decomposition on each signal.

[0026] Detail coefficients are processed with soft thresholding to suppress high-frequency noise, while approximation coefficients retain low-frequency physiological components. The denoised signals are then fed into the feature extraction engine: the time-domain feature extraction unit calculates the signal mean, variance, peak factor, and zero-crossing rate; the frequency-domain feature extraction unit calculates the power spectrum centroid and band energy ratio using Fast Fourier Transform; and the nonlinear dynamics feature extraction unit calculates the maximum Lyapunov exponent and approximate entropy using phase space reconstruction. All features constitute a high-dimensional original feature set of dimension 256.

[0027] The high-dimensional original feature set is input into a deep belief network for feature compression. This network structure consists of 3 visible layers and 5 hidden layers, with visible layer nodes of 256, 128, and 64 respectively, and hidden layer nodes of 128, 64, 32, 16, and 8 respectively. The network is constructed using stacked Restricted Boltzmann Machines. During the pre-training phase, the contrastive divergence algorithm is used for 1000 iterations with a learning rate of 0.01.

[0028] The fine-tuning phase employs the backpropagation algorithm with mean squared error as the loss function, and uses the Adam optimizer for 500 iterations. After training, the network maps its 256-dimensional features into an 8-dimensional tissue thermal state feature vector, which comprehensively represents the tissue thermal conductivity, blood flow heat dissipation efficiency, and muscle activity heat generation characteristics.

[0029] The adaptive predictive control module receives the tissue thermal state characteristic vector and historical control commands. Please refer to the appendix. Figure 3The recurrent neural network predictor employs a long short-term memory (LSTM) network structure. The input layer contains 8 feature nodes and 5 control instruction nodes. The hidden layer deploys 128 LSM units, with the forget gate bias initialized to 1.0 to enhance long-term memory. The output layer generates temperature predictions for the next three time steps through a fully connected layer. The model predictive controller solves a rolling optimization problem in each control cycle.

[0030] in, To predict temperature in k steps, For reference trajectory, To control the increment, The weighting coefficient is set to 0.1. Optimization constraints include controlling the command change rate to not exceed 10% of full scale per second, and limiting the refrigerant flow rate to 0 to 100 ml / min. The solver uses the effective set algorithm, with computation time controlled within 20 milliseconds.

[0031] The treatment execution and dynamic calibration module includes a high-precision refrigerant flow control valve and a microprocessor. The control valve employs a piezoelectric ceramic drive structure, achieving a resolution of 0.1% across the full range with a response time of less than 50 milliseconds. The microprocessor executes the optimized instructions output by the model predictive controller while continuously calculating the root mean square error (RMSE) between the predicted and actual temperatures. When the RMSSE exceeds 0.5 degrees Celsius for 10 consecutive control cycles, an incremental learning program is triggered. The program extracts the most recent 50 sets of control instructions and temperature data from the circular buffer, executes a backpropagation algorithm over time with a learning rate of 0.001, and updates only the weights of the recurrent neural network predictor output layer; the update process is completed within a single control cycle.

[0032] The safety monitoring and intervention module analyzes the tissue thermal state characteristic vector in real time. Please refer to the appendix. Figure 5 The feature mutation index calculation unit uses a sliding window to calculate the Mahalanobis distance of the feature vectors, with a window length of 10 periods. When the mutation index exceeds 2.0 for three consecutive periods and the cosine similarity with the abnormal state template is greater than 0.9, the safety logic controller sends a coverage command with priority 9 to the adaptive predictive control module. This command forces the model predictive controller to switch to safety mode, limiting the refrigerant flow to 20% of the baseline flow, and simultaneously sending an alarm message containing a timestamp, anomaly code, and feature vector to the monitoring terminal via the RS485 interface.

[0033] The system hardware platform adopts a distributed architecture. The main controller uses a dual-core ARM Cortex-A53 processor and runs a real-time Linux system. The signal acquisition unit is equipped with an isolated power supply module to ensure the electrical safety of patients. The communication bus adopts the CAN2.0B protocol, and the data transmission rate is set to 500 kilobits per second. All modules have passed electromagnetic compatibility testing and meet the Class II protection standard for medical equipment.

[0034] In practice, the system performs a self-check upon startup: the temperature sensor is calibrated using a freezing point reference, the blood perfusion instrument undergoes zero-point calibration, and the electromyography electrode impedance is less than 5 kΩ. At the start of treatment, the operator sets the target temperature range via the touchscreen, with a default range of 4 to 8 degrees Celsius. The system initialization phase uses PID control mode, automatically switching to adaptive predictive control mode once sufficient data is collected. During treatment, the temperature curve, control command curve, and safety status indicator lights are displayed in real time.

[0035] The system possesses adaptive parameter capabilities for different tissue characteristics. For areas rich in blood vessels, the weighting coefficient for blood perfusion rate is automatically increased to 0.7; for areas with well-developed muscles, the weighting of electromyographic signal frequency characteristics is increased to 0.6. These parameter adjustments are achieved by querying a preset tissue characteristic database, which contains typical parameter combinations for over 100 tissue types.

[0036] The fault handling mechanism employs a multi-layered strategy. When a sensor signal is lost for more than three cycles, the system automatically switches to a backup sensor; if communication interruption lasts for 10 cycles, the control command retains the last valid value and triggers an audible and visual alarm; in the event of a power failure, the backup battery can maintain system operation for 30 minutes. All abnormal events are recorded in non-volatile memory, supporting later analysis and tracing.

[0037] Example 2: This example provides an intelligent analgesia control scheme for highly dynamic blood flow changes. Based on Example 1, the multimodal data fusion processing module is optimized for scenarios with sudden changes in blood perfusion rate. When the rate of change in blood perfusion rate exceeds 15% per second for two consecutive cycles, the feature extraction engine activates a fast response mode: the time-domain feature calculation window is shortened from 100 milliseconds to 50 milliseconds, frequency domain analysis focuses on the 0.5 to 3 Hz frequency band, and nonlinear features retain only approximate entropy calculations. The deep belief network is switched to a lightweight version, with the number of hidden layer nodes compressed to 64, 32, and 16, and the forward inference time reduced from 15 milliseconds to 8 milliseconds.

[0038] The adaptive predictive control module adds a blood flow change feedforward channel. A blood flow differential feature node is added to the input layer of the recurrent neural network predictor, and a blood flow disturbance compensation term is added to the objective function of the model predictive controller.

[0039] in, For feedforward control quantities based on changes in blood flow, Set to 0.05. The treatment execution module is equipped with an upgraded dual-valve structure. The main valve is responsible for baseline flow control, while the auxiliary valve is dedicated to rapidly compensating for blood flow fluctuations, reducing the response time to 20 milliseconds.

[0040] The safety monitoring module enhances the vasospasm detection function. When the blood perfusion rate decreases by more than 60% within 5 seconds and the electromyography signal amplitude increases by 200%, a risk of vasospasm is identified. The cryotherapy flow rate is immediately reduced to a safe threshold, and a local rewarming procedure is initiated. The rewarming rate is controlled to within 2 degrees Celsius per minute to avoid tissue thermal shock. The system simultaneously records multimodal data for 120 seconds before and after the event for subsequent pathological analysis.

[0041] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0042] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent analgesia control system, characterized in that, include: The physiological signal acquisition module is used to acquire temperature signals, blood perfusion rate signals, and surface electromyography signals from the target treatment area in real time. The multimodal data fusion processing module is connected to the physiological signal acquisition module. It is used to perform time alignment, noise reduction and feature extraction on the received multi-source heterogeneous physiological signals, and to construct a nonlinear mapping from the original signal to the tissue thermal state feature vector based on a deep belief network. The adaptive predictive control module is connected to the multimodal data fusion processing module. It integrates a recurrent neural network predictor with online learning capability and a model predictive controller. The recurrent neural network predictor predicts the tissue temperature change trajectory for multiple time steps in the future based on the current tissue thermal state feature vector and the historical control command sequence. The model predictive controller generates optimized control commands that make the predicted temperature trajectory closest to the preset treatment temperature range by solving a constrained rolling optimization problem. The treatment execution and dynamic calibration module, connected to the adaptive predictive control module and the physiological signal acquisition module, includes a high-precision cryoprotectant flow regulating valve and a microprocessor. This module executes optimized control commands from the model predictive controller to drive the flow regulating valve to precisely control the cryoprotectant output. At the same time, it continuously compares the error between the predicted temperature trajectory and the actual measured temperature. When the accumulated error exceeds the preset adaptive trigger threshold, it automatically initiates online incremental updates to the weights inside the recurrent neural network predictor.

2. The intelligent analgesia control system according to claim 1, characterized in that, The specific operation process of the multimodal data fusion processing module includes: synchronizing the temperature signal, blood perfusion rate signal, and electromyography signal input from the physiological signal acquisition module with unified timestamps; denoising the synchronized multi-channel signals using a joint denoising algorithm based on wavelet transform; extracting time-domain features, frequency-domain features, and nonlinear dynamic features from each denoised signal to form a high-dimensional original feature set; inputting the high-dimensional original feature set into a pre-trained offline deep belief network, which, through its multi-layer nonlinear transformation, compresses and maps the high-dimensional original feature set into a low-dimensional tissue thermal state feature vector.

3. The intelligent analgesia control system according to claim 2, characterized in that, The offline training process of the deep belief network includes: collecting a large amount of historical physiological signal data and corresponding tissue temperature data covering different individuals and different treatment sites as training sample sets; constructing a deep belief network structure containing 3 visible layers and 5 hidden layers; using the contrastive divergence algorithm to perform unsupervised layer-by-layer pre-training of the network to initialize the network weights; after the pre-training is completed, using labeled tissue temperature data to perform supervised fine-tuning training of the network to minimize the reconstruction error between the tissue thermal state feature vector output by the network and the real tissue temperature data.

4. The intelligent analgesia control system according to claim 1, characterized in that, The recurrent neural network predictor adopts a long short-term memory network structure. Its input layer receives the current tissue thermal state feature vector output by the multimodal data fusion processing module, as well as the control instruction sequence of the past 5 time steps. The network hidden layer contains 128 long short-term memory units to capture the long-range time dependence in tissue thermodynamics. The output layer predicts the tissue temperature value for the next 3 time steps.

5. The intelligent analgesia control system according to claim 4, characterized in that, The mathematical description of the rolling optimization problem of the model predictive controller is as follows: In each control cycle, find a set of future control command sequences such that the variance between the predicted temperature output given by the recurrent neural network predictor and the preset treatment temperature range reference trajectory is minimized, while satisfying the physical constraints of the rate of change of control commands and the absolute value of the refrigerant flow rate.

6. The intelligent analgesia control system according to claim 1, characterized in that, The online incremental update execution mechanism in the treatment execution and dynamic calibration module includes: the microprocessor continuously calculates the root mean square error between the predicted temperature values ​​and the actual measured values ​​for the next three time steps by the recurrent neural network predictor; a preset adaptive trigger threshold of 0.5 degrees Celsius is set; when the root mean square error calculated within 10 consecutive control cycles exceeds 0.5 degrees Celsius, the microprocessor automatically triggers the online learning program; the online learning program uses the actual control commands and measured temperature data of the most recent 50 time steps to form an incremental training dataset, and uses a backpropagation algorithm over time to slightly adjust the weights of the recurrent neural network predictor, with the learning rate set to 0.

001.

7. The intelligent analgesia control system according to claim 6, characterized in that, The weight update process is completed within one control cycle, ensuring that the real-time performance of the control loop is not affected.

8. The intelligent analgesia control system according to claim 1, characterized in that, It also includes a safety monitoring and intervention module, which monitors the tissue thermal state feature vector output by the multimodal data fusion processing module in real time and calculates the mutation index of its feature values. The safety monitoring and intervention module has a pre-stored feature template library of normal treatment and abnormal state. When the calculated mutation index exceeds the safety threshold of 2.0 for three consecutive cycles and the feature vector matches the pre-stored abnormal state template by 90%, the module will send a highest priority coverage instruction to the adaptive predictive control module, forcing the model predictive controller to output a preset safety control mode.

9. The intelligent analgesia control system according to claim 8, characterized in that, The safety control mode includes reducing the refrigerant flow rate to the minimum level required to maintain the temperature, while simultaneously sending an alarm signal to an external monitoring terminal.

10. A method for establishing an intelligent analgesia control model, characterized in that, Analgesia control is achieved using the intelligent analgesia control system described in any one of claims 1-9.

Images