Intraoperative anesthetic drug infusion rate self-adaptive adjustment method based on reinforcement learning

Abstract

This invention relates to the field of infusion rate control technology, specifically to an adaptive adjustment method for intraoperative anesthetic drug infusion rate based on reinforcement learning. The method includes the following steps: monitoring tubing pressure waveforms to extract hysteresis response features; constructing a multidimensional fluid environment state vector by splicing anesthesia depth errors; inputting the vector into a neural network to map and output a composite command containing a base flow rate and resistance torque; establishing a corrected drive signal by superimposing acceleration pulses based on tubing stiffness and compensating for flow leakage; and completing closed-loop verification based on pressure change rate deviation. In this invention, by analyzing the establishment and attenuation features of the pressure waveform to reconstruct the hysteresis response state of the physical environment, tubing compliance is incorporated into the network input space and mapped to generate a composite command containing a base flow rate and resistance torque. This eliminates drug delivery delays caused by tubing expansion and energy absorption, ensuring that motor actions are instantly converted into effective drug injection, thereby improving the real-time response speed and drug delivery control accuracy of anesthesia depth regulation.

Description

Technical Field

[0001] This invention relates to the field of infusion rate control technology, and in particular to an adaptive adjustment method for intraoperative anesthetic drug infusion rate based on reinforcement learning. Background Technology

[0002] The field of infusion rate control technology mainly involves the precise adjustment and monitoring of the flow rate and velocity of liquid drugs entering the body using mechanical drive mechanisms and fluid transmission principles. This technology covers the entire process from the power generated by stepper motors or DC motors to drive the syringe plunger or peristaltic pump wheel assembly, to the delivery of the drug solution through fluid tubing, and includes physical detection of tubing pressure and flow rate deviations. Traditional adaptive adjustment methods for intraoperative anesthetic drug infusion rates address the technical aspects of maintaining the patient's depth of anesthesia and stable vital signs during surgery. These methods utilize population pharmacokinetic and pharmacodynamic models embedded in the infusion device to calculate and estimate blood drug concentrations based on the patient's height, weight, and age parameters, mapping this to the rotation frequency of the infusion pump motor. Alternatively, anesthesiologists can observe the bispectral index (BIS) value on electroencephalogram (EEG) and invasive arterial blood pressure waveforms, and manually set the flow rate value in milliliters per hour on the infusion pump panel according to preset proportional-integral-derivative (PID) control logic or clinical dosing guidelines.

[0003] Current intraoperative anesthetic drug infusion technology relies on fixed pharmacokinetic models or simple feedback logic, ignoring the physical compliance of the infusion tubing and the nonlinear attenuation effect of fluid back pressure fluctuations on drug flow. This results in a significant time lag between motor action and effective drug entry into the body, and fails to detect elastic deformation caused by the softness or hardness of the tubing material. Consequently, the pump flow rate command is buffered and absorbed by tubing expansion, causing an unquantifiable deviation between the actual amount of drug entering the body and the preset target value. This leads to a delay in the response of anesthesia depth control and fluctuations in intraoperative drug delivery accuracy. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide a method for adaptive adjustment of intraoperative anesthetic drug infusion rate based on reinforcement learning, comprising the following steps: To achieve the above objectives, the present invention employs the following technical solution: an adaptive adjustment method for intraoperative anesthetic drug infusion rate based on reinforcement learning, comprising the following steps: S1: Monitor the pressure sensor signal at the end of the anesthetic drug infusion line, record the pressure waveform when the flow rate steps, analyze the rise and fall tail characteristics of the pressure waveform, and obtain the pressure hysteresis response characteristic data of the line. S2: Call the pipeline pressure hysteresis response characteristic data, collect the anesthesia depth and calculate the control error, and splice the pipeline pressure hysteresis response characteristic data, control error and real-time pressure to generate a multi-dimensional fluid environment state mapping vector. S3: Input the multidimensional fluid environment state mapping vector into a deep neural network, map and output the basic flow velocity and resistance torque, separate the flow velocity component that maintains the depth of anesthesia and the torque component that overcomes the back pressure deformation of the pipeline, and obtain a dual-channel flow velocity and torque composite command. S4: For the dual-channel velocity-torque composite command, find the flow loss compensation amount based on the real-time pressure, compare the pipeline pressure hysteresis response characteristic data with the stiffness parameters, and superimpose the acceleration pulse determined by the resistance compensation torque on the speed command to establish a fluid dynamics correction drive pulse signal. S5: By using the fluid dynamics-corrected drive pulse signal, monitor the pressure change rate, compare the real-time pressure change rate with the theoretical pressure change rate, trigger resampling based on the deviation, and generate a closed-loop pressure response verification signal.

[0005] As a further aspect of the present invention, the pipeline pressure hysteresis response characteristic data includes pressure rise time and pressure release tailing time; the multidimensional fluid environment state mapping vector includes mechanical compliance feature vector elements, anesthesia depth deviation value, and real-time pipeline pressure value; the dual-channel velocity-torque composite command includes a basic target velocity command component and a resistance compensation torque command component; the fluid dynamics correction drive pulse signal includes a flow loss compensation step frequency and a dynamic resistance acceleration pulse sequence; and the closed-loop pressure response verification signal includes a pressure change rate deviation metric and a state resampling trigger flag.

[0006] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Monitors the pressure sensor signal at the end of the anesthetic drug infusion line, collects the time-series voltage sequence output by the pressure sensor in real time during the flow rate command change operation of the infusion pump, identifies the time window from the start of the flow rate change to the pressure recovery to steady state, extracts the corresponding signal segment within the time window and performs digital sampling and analog-to-digital conversion processing to generate a step excitation pressure waveform sequence. S102: Based on the step excitation pressure waveform sequence, traverse the data points of the rising and falling edges of the amplitude, calculate the pressure establishment time constant and peak rise slope value of the rising edge stage, extract the pressure attenuation tailing time and amplitude fall rate of the falling edge stage, and aggregate and encode the calculated time and rate values ​​to obtain the waveform establishment and attenuation feature set. S103: Call the waveform establishment and attenuation feature set, calculate the absolute value of the difference between the pressure establishment time constant and the pressure attenuation tail time, calculate the pressure response lag time and waveform asymmetry coefficient under unit flow rate change, map the lag time and asymmetry coefficient into a multi-dimensional vector, and obtain pipeline pressure hysteresis response feature data.

[0007] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Based on the pipeline pressure hysteresis response characteristic data, monitor the patient's real-time anesthesia depth characterization parameters, retrieve the preset anesthesia depth target benchmark value, perform the difference operation between the real-time parameters and the target benchmark value to obtain the deviation amplitude, perform discretization differentiation processing on the deviation amplitude to obtain the deviation change rate, perform linear weighted summation on the deviation amplitude and the deviation change rate according to the preset weight coefficient, quantify the degree of deviation between the real-time control state and the target state, and generate anesthesia depth control error scalar; S202: Call the scalar of the anesthesia depth control error, synchronously collect the real-time fluid pressure value output by the pressure sensor at the end of the infusion line, take the moment of flow rate change as the time reference point, perform time-series synchronous calibration of hysteresis characteristics, control error and real-time pressure, and cascade and splice the calibrated multiple data according to the feature dimension order to construct a multi-source heterogeneous state combination sequence. S203: Based on the multi-source heterogeneous state combination sequence, calculate the arithmetic mean and standard deviation of multi-dimensional values ​​within a sliding window, perform a centering translation operation on the sequence elements using the arithmetic mean, and scale the translated data in combination with the standard deviation, uniformly mapping the data with different physical dimensions to the target numerical range, avoiding the order-of-magnitude differences between feature dimensions, and generating a multi-dimensional fluid environment state mapping vector.

[0008] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Based on the multidimensional fluid environment state mapping vector, call the preset network weight parameter matrix, perform multi-level weighted summation operation on the vector elements, perform dimensionality reduction mapping on the operation values ​​in the feature space, convert the multidimensional environment state features into numerical expressions of fluid delivery control dimensions, analyze the response values ​​of the output layer nodes, determine the real-time required liquid propulsion speed and the magnitude of the torque against pipeline resistance, and generate the predicted values ​​of basic flow velocity and resistance torque. S302: Call the predicted values ​​of the basic flow rate and the resistance torque, extract the flow rate numerical channel that characterizes the drug metabolism demand, calculate the steady-state flow rate benchmark required to maintain the real-time anesthesia depth in combination with pharmacokinetic parameters, and simultaneously separate the torque numerical channel that characterizes the mechanical properties of the pipeline. Calculate the additional compensation torque required for the elastic deformation caused by the pipeline back pressure, complete the decoupling process of the physiological demand control quantity and the physical property compensation quantity, and obtain the components of the maintenance flow rate and deformation compensation torque. S303: Based on the maintained flow rate and deformation compensation torque components, a parallel encoding method is used to digitally encapsulate the steady-state flow rate reference and the additional compensation torque. A timing synchronization check bit is added to the instruction frame header, and the flow rate control signal is mapped to the main injection channel, while the torque compensation signal is mapped to the auxiliary boosting channel to generate a dual-channel flow rate and torque composite instruction.

[0009] As a further aspect of the present invention, the process of calculating the steady-state flow rate benchmark required to maintain the real-time anesthesia depth by combining pharmacokinetic parameters is specifically as follows: calling the preset drug central compartment distribution volume parameter and drug clearance rate constant, parsing the value in the flow rate numerical channel characterizing the drug metabolism demand into the target effect room concentration value, and constructing a plasma concentration decay curve per unit time based on the drug clearance rate constant. Based on the drug central compartment distribution volume parameter, the drug replenishment rate required to maintain the target effect compartment concentration is inverted. The drug replenishment rate is divided by the standard density of the drug solution, and the resulting volume value is quantified into a flow rate unit of milliliters per hour to generate a steady-state flow rate benchmark.

[0010] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Call the dual-channel flow velocity torque composite command, separate the flow velocity control channel data, synchronously collect the real-time fluid pressure value at the end of the injection pipeline, retrieve the preset correspondence matrix between pressure amplitude and flow loss, locate the flow compensation coefficient of the interval to which the real-time pressure value belongs, perform linear superposition operation on the flow compensation coefficient and the flow velocity control channel data, and generate a pressure-related flow compensation correction value. S402: Call the pressure-related flow compensation correction value, load the preset pipeline material stiffness reference parameter, calculate the difference between the hysteresis response characteristics and the stiffness reference and quantify the degree of pipeline deformation, perform gain adjustment operation based on the difference and the resistance torque component, convert the adjusted torque value into the corresponding instantaneous high-frequency pulse sequence, determine the pulse duration and trigger phase, and generate stiffness-adaptive resistance acceleration pulse. S403: Based on the stiffness-adaptive resistance acceleration pulse, the flow compensation correction value is converted into the basic drive frequency of the stepper motor. The resistance acceleration pulse is embedded at the beginning of each cycle of the basic drive frequency. The duty cycle and frequency of the drive waveform are jointly modulated to synthesize an electrical signal sequence with dynamic fluid resistance compensation characteristics, and a fluid dynamics correction drive pulse signal is generated.

[0011] As a further aspect of the present invention, the process of embedding an anti-resistance acceleration pulse at the start point of each cycle of the basic drive frequency and jointly modulating the duty cycle and frequency of the drive waveform is specifically as follows: converting the flow compensation correction value into a basic drive signal in the form of a square wave according to the step angle of the stepper motor, and capturing the trigger time of each rising edge of the basic drive signal in real time. Within the time window following the rising edge trigger moment, a stiffness-adapted resistance acceleration pulse is superimposed onto the high-level front segment of the basic drive signal. While keeping the period length of the basic drive signal constant, the duration of the high level and the steepness of the waveform front are changed to synthesize a hydrodynamically corrected drive pulse signal.

[0012] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Based on the fluid dynamics correction drive pulse signal, drive the fluid delivery execution unit to act, synchronously activate the pipeline end pressure acquisition circuit, acquire the real-time discrete pressure value sequence according to the preset period, perform sliding window differentiation operation on the real-time discrete pressure value sequence, calculate the pressure value increment per unit time, extract the physical quantity reflecting the real-time fluid drive response agility, and generate the real-time pipeline pressure change rate. S502: Call the real-time pipeline pressure change rate, analyze the duty cycle parameter of the drive pulse, establish a fluid transmission model and calculate the expected pressure rise slope under the theoretical interference-free state, perform a differential comparison operation between the real-time change rate and the expected rise slope, calculate the numerical deviation between the two, quantify the dynamic following error of the real-time output to the drive command, and generate a pressure following response deviation value. S503: Based on the pressure following response deviation value, retrieve the preset dynamic response allowable error benchmark value, compare the pressure following response deviation value with the dynamic response allowable error benchmark value, and trigger the data resampling mechanism when the dynamic response allowable error benchmark value is exceeded. Optimize the sampling frequency and identify the waveform segment during the deviation period. Encapsulate the waveform segment and deviation verification bit into a data frame to generate a closed-loop pressure response verification signal.

[0013] As a further aspect of the present invention, the process of establishing a fluid transport model and calculating the expected pressure rise slope under theoretically undisturbed conditions is as follows: based on the duty cycle parameter of the drive pulse, the pre-stored motor speed characteristic curve is retrieved to obtain the corresponding piston propulsion speed; the theoretical instantaneous flow rate is calculated by combining the piston cross-sectional area of ​​the syringe; the derivative value of the pressure change with time under conditions without external disturbance is solved; and the derivative value is set as the expected pressure rise slope.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, the hysteresis response state of the physical environment is reconstructed by analyzing the establishment and attenuation characteristics of the pressure waveform. The pipeline compliance is incorporated into the network input space and mapped to generate a composite command containing the basic flow velocity and the resistance torque. The resistance torque component is used to actively generate an acceleration pulse to counteract the elastic deformation of the pipeline to achieve overdrive compensation. The flow loss caused by back pressure leakage is corrected based on the fluid dynamics model. The drug delivery delay caused by pipeline expansion and energy absorption is eliminated, ensuring that the motor action is instantly converted into effective drug injection, thereby improving the real-time response speed and drug delivery control accuracy of anesthesia depth regulation. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the accompanying drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] Please see Figure 1 This invention provides a reinforcement learning-based adaptive adjustment method for intraoperative anesthetic drug infusion rate, comprising the following steps: S1: Monitor the pressure sensor signal at the end of the anesthetic drug infusion line, record the pressure waveform when the infusion pump starts or the flow rate setting changes stepwise, analyze the rise and build-up characteristics of the pressure waveform rise phase, identify the attenuation tail characteristics of the pressure natural attenuation phase after the motor stops, and obtain the pressure hysteresis response characteristic data of the line. S2: Call the pipeline pressure hysteresis response characteristic data, collect the patient's real-time anesthesia depth monitoring data, calculate the control error between the real-time depth and the target depth, splice the pipeline pressure hysteresis response characteristic data, control error and pipeline real-time pressure data, construct a numerical sequence and perform standardization processing, and generate a multi-dimensional fluid environment state mapping vector. S3: Input the multidimensional fluid environment state mapping vector into the deep neural network model, perform the multiplication operation between the input sequence and the network weight matrix, map and output the basic target flow velocity and the resistance compensation torque, separate the flow velocity component that maintains the depth of anesthesia and the torque component that overcomes the back pressure deformation of the pipeline, and obtain the dual-channel flow velocity and torque composite command. S4: For the basic target flow velocity in the dual-channel flow velocity and torque composite command, retrieve the preset pressure-flow-leakage fluid dynamics model, find the flow loss compensation amount based on the real-time back pressure, compare the pipeline pressure hysteresis response characteristic data with the standard stiffness parameters, and superimpose the acceleration pulse determined by the resistance compensation torque on the speed command of the basic target flow velocity conversion to establish the fluid dynamics correction drive pulse signal. S5: Convert the fluid dynamics correction drive pulse signal into the stepper motor drive frequency, execute the intraoperative anesthetic drug injection action, monitor the pressure change rate inside the infusion line in real time, compare the real-time pressure change rate with the theoretical pressure change rate based on the real-time action prediction, trigger the state resampling and strategy update process according to the deviation amplitude, and generate a closed-loop pressure response verification signal. The pipeline pressure hysteresis response characteristic data includes the pressure build-up time value and the pressure release tailing time value; the multidimensional fluid environment state mapping vector includes mechanical compliance feature vector elements, anesthesia depth deviation value, and real-time pipeline pressure value; the dual-channel velocity-torque composite command includes the basic target velocity command component and the resistance compensation torque command component; the fluid dynamics correction drive pulse signal includes the flow loss compensation step frequency and the dynamic resistance acceleration pulse sequence; and the closed-loop pressure response verification signal includes the pressure change rate deviation metric value and the state resampling trigger flag.

[0020] Please see Figure 2 The specific steps of S1 are as follows: S101: Monitors the pressure sensor signal at the end of the anesthetic drug infusion line, collects the time-series voltage sequence output by the pressure sensor in real time during the flow rate command change operation of the infusion pump, identifies the time window from the start of the flow rate change to the pressure recovery to steady state, extracts the corresponding signal segment within the time window and performs digital sampling and analog-to-digital conversion processing to generate a step excitation pressure waveform sequence. Initiate continuous monitoring of the microelectromechanical (MEMS) pressure sensor at the end of the fluid delivery loop, with the sensor sampling frequency set to [value missing]. Hz (i.e., sampling per second) (This is repeated several times) to ensure that microsecond-level pressure fluctuation details can be captured. The monitoring process continuously reads the analog voltage signal output by the sensor, which is linearly related to the hydrostatic pressure of the fluid in the pipeline. When a flow rate change command is received (e.g., changing the flow rate from...), the sensor will detect the change. The mL / h step adjustment was made to... When the flow rate is mL / h, that moment is marked as time zero. This step locks from Start, until the sensor feedback voltage fluctuation amplitude is continuous The number of sampling points is lower than The time period up to mV (i.e., the time until steady state is determined) is identified as the effective time window. Within this time window, a segment of length [mV] is extracted. The original voltage signal segment of each data point is called. A high-precision analog-to-digital converter performs a digital conversion on this analog voltage segment, mapping the analog level to... to The digital values ​​between these points are encoded. During this process, digital filtering is performed, and a five-point moving average algorithm is used to remove high-frequency electromagnetic noise interference. For example, if the digital values ​​of five consecutive sampling points are respectively... , , , , The filtered output value at that point is then calculated as the arithmetic mean of these five values: After full sequence processing, a set of discretized digital sequences reflecting the step change of pressure over time is generated, namely the step excitation pressure waveform sequence.

[0021] S102: Based on the step excitation pressure waveform sequence, traverse the data points of the rising and falling edges of the amplitude, calculate the pressure establishment time constant and peak rise slope value of the rising edge stage, extract the pressure attenuation tail time and amplitude fall rate of the falling edge stage, and aggregate and encode the calculated time and rate values ​​to obtain the waveform establishment and attenuation feature set. The rate of change of values ​​in the sequence is scanned point by point to locate the rising and falling edges of the waveform. For the rising edge phase, the pressure value is identified from the steady-state reference value. Climb to The time taken is defined as the pressure build-up time constant. Simultaneously, the difference between adjacent data points within this interval is calculated, and the maximum positive difference is divided by the sampling interval to obtain the peak rise slope. During the falling edge phase (i.e., when the flow rate stops or decreases), the pressure value is measured to decay from the peak value to the reference value. The duration of this interval is defined as the pressure decay tail time. The amplitude reduction per unit time during this stage is calculated as the amplitude fall rate. The four physical quantities, namely the pressure build-up time constant, the peak rise slope, the pressure decay tail time, and the amplitude fall rate, are numerically aggregated. The aggregation process adopts a binary encoding concatenation method, and the first and last parts are spliced ​​together in sequence to form a feature encoding string, thus obtaining the waveform build-up and decay feature set. See Table 1, which shows the feature data extraction results of pipelines of different materials under the flow velocity step test. Table 1: Experimental Data for Pipeline Pressure Waveform Feature Extraction Pipe type number Pressure build-up time constant (ms) Peak rise slope (kPa / s) Pressure decay tail time (ms) Amplitude reduction rate (kPa / s) PVC-Soft-01 450 12.5 620 8.2 PE-Hard-02 120 35.8 150 28.4 PU-Comp-03 280 18.2 340 14.6 As shown in Table 1, by performing the above waveform feature extraction on different pipelines, the physical differences in pressure response between flexible hoses and rigid pipes can be quantitatively distinguished. The data shows that the pressure decay tailing time of PVC hoses ( ms) is significantly longer than that of PE rigid pipe ( (ms), verifying that the hose material has a more significant elastic hysteresis effect.

[0022] S103: Call the waveform establishment and decay feature set, calculate the absolute value of the difference between the pressure establishment time constant and the pressure decay tail time, calculate the pressure response lag time and waveform asymmetry coefficient under unit flow rate change, map the lag time and asymmetry coefficient into a multi-dimensional vector, and obtain pipeline pressure hysteresis response feature data. The pressure build-up time constant and pressure decay tail time are analyzed, and the difference is calculated by subtracting the pressure build-up time constant from the pressure decay tail time and taking the absolute value of the result. For example, referring to the data of PVC-Soft-01 in Table 1, if the decay time is... ms, creation time is If ms, then the absolute value of the difference is ms, to obtain the current flow rate step magnitude (e.g. (mL / h), divide the absolute value of the aforementioned difference by the flow rate step amplitude to calculate the pressure response hysteresis time per unit flow rate change. In the example above, the calculation process is as follows: ms / (mL / h), and simultaneously, calculate the ratio of the settling time constant to the decay tail time, using this ratio as the waveform asymmetry coefficient. The calculation process is as follows: (Retain three decimal places) Map the lag time value and the asymmetry coefficient value to a two-dimensional vector space. The first dimension is the lag time and the second dimension is the asymmetry, thereby constructing pipeline pressure hysteresis response characteristic data that can characterize the viscoelastic deformation characteristics of the pipeline under the current operating conditions.

[0023] Please see Figure 3 The specific steps of S2 are as follows: S201: Based on the hysteresis response characteristic data of the pipeline pressure, monitor the real-time anesthesia depth characterization parameters of the patient, retrieve the preset target benchmark value of anesthesia depth, perform the difference operation between the real-time parameters and the target benchmark value to obtain the deviation amplitude, perform discretization differentiation processing on the deviation amplitude to obtain the deviation change rate, perform linear weighted summation on the deviation amplitude and the deviation change rate according to the preset weight coefficient, quantify the degree of deviation between the real-time control state and the target state, and generate anesthesia depth control error scalar; Connect to the real-time data interface of the patient's bispectral index (BIS) monitor, according to The current anesthesia depth characterization parameters are read once per second, and simultaneously, a preset anesthesia depth target baseline value (e.g., set to) is retrieved from memory. The subtraction operation is performed, subtracting the target baseline value from the real-time BIS value to obtain the deviation amplitude. The deviation amplitudes of the three most recent samples are then discretized and differentiated, i.e., the current deviation is subtracted from the previous deviation, and then divided by the sampling time interval to obtain the deviation change rate. A preset weighting coefficient library is then called, where the weighting coefficient for the deviation amplitude is set to... The weighting coefficient for the rate of change of deviation is set to The coefficients were determined through regression analysis of large-sample clinical data, and a linear weighted summation operation was performed: the deviation magnitude was multiplied by... Multiply the rate of change of the deviation by And add the products of the two items together. For example, if the current deviation magnitude is... The rate of change of the deviation is The calculation process is as follows: The calculation result is the quantified scalar of the anesthesia depth control error, which comprehensively reflects the distance of the current state from the target and the speed of the deviation trend.

[0024] S202: Call the scalar of anesthesia depth control error, synchronously collect the real-time fluid pressure value output by the pressure sensor at the end of the infusion line, take the moment of flow rate change as the time reference point, perform time-series synchronous calibration of hysteresis characteristics, control error and real-time pressure, and cascade and splice the calibrated multiple data according to the feature dimension order to construct a multi-source heterogeneous state combination sequence. The anesthesia depth control error scalar and tubing pressure hysteresis response characteristic data are retrieved simultaneously. The real-time fluid pressure value output by the pressure sensor at the end of the infusion line is used. Due to differences in the acquisition frequency and transmission delay of different data, this step performs a time synchronization calibration operation: a linear interpolation algorithm is used to convert low-frequency anesthesia depth data ( Hz) and hysteresis characteristic data upsampled to high-frequency pressure data ( On a time axis consistent with Hz, ensure that each millisecond time slice corresponds to the above three types of data. After calibration, concatenate and stitch together data points at the same time point in the order of "hysteresis feature vector - control error scalar - real-time pressure value". For example, the hysteresis feature vector is... The control error scalar is Real-time pressure is kPa, then the elements of the original sequence after splicing are In this way, a multi-source heterogeneous state combination sequence is constructed.

[0025] S203: Based on the multi-source heterogeneous state combination sequence, calculate the arithmetic mean and standard deviation of multi-dimensional values ​​within a sliding window, use the arithmetic mean to perform a centering translation operation on the sequence elements, and combine the standard deviation to scale the translated data, uniformly map the data with different physical dimensions to the target numerical range, avoid the order of magnitude difference between feature dimensions, and generate a multi-dimensional fluid environment state mapping vector. Set a length of A sliding window is used to calculate the arithmetic mean and standard deviation of each feature dimension (i.e., lag time, asymmetry, control error, and real-time pressure) within the window. For example, for the real-time pressure dimension, the arithmetic mean and standard deviation within the window are calculated. Average of pressure values kPa, standard deviation kPa, perform Z-score standardization: subtract the average value from the current real-time pressure value in the sequence, and then divide the difference by the standard deviation. Assume the current pressure value is... kPa, the calculation process is as follows Perform this operation on the values ​​in all dimensions of the sequence, transforming the original physical quantities (time unit ms, pressure unit kPa, dimensionless coefficients, etc.) with vastly different orders of magnitude (from... arrive The data is uniformly mapped to the mean. variance is Within the standard normal distribution range, it effectively avoids the misleading influence of the order-of-magnitude difference between feature dimensions on the subsequent algorithm weights, and generates a multidimensional fluid environment state mapping vector.

[0026] Please see Figure 4 The specific steps of S3 are as follows: S301: Based on the multidimensional fluid environment state mapping vector, the preset network weight parameter matrix is ​​called to perform multi-level weighted summation operation on the vector elements, and the dimensionality reduction mapping of the operation values ​​in the feature space is performed to convert the multidimensional environment state features into numerical expressions of the fluid delivery control dimension. The response values ​​of the output layer nodes are analyzed to determine the real-time required liquid propulsion speed and the magnitude of the torque against pipeline resistance, and to generate the predicted values ​​of basic flow velocity and resistance torque. The multidimensional fluid environment state mapping vector is used as input and fed into a pre-built deep neural network model. This network model contains an input layer, three hidden layers, and an output layer. The number of nodes in the input layer corresponds to the vector dimension (e.g., ...). (Dimension), the first hidden layer contains The second hidden layer contains 10 neurons. The third hidden layer contains 1 neuron. The output layer contains neurons, and the output layer contains There are nodes, and the layers are fully connected. During the operation, each neuron in the first hidden layer receives the weighted sum of the output of the previous layer and the corresponding weight parameter matrix, adds a bias term, and performs a nonlinear transformation through the ReLU (Rectified Linear Unit) activation function. Specifically, for the first hidden layer... There are *n* neurons, whose input is the sum of the products of each element of the input vector and its corresponding weight. Assume the input vector is a standardized... The corresponding weight is , bias is The weighted sum calculation process is as follows: After processing with the ReLU function (i.e., taking a positive value), the output of this neuron is: This process propagates layer by layer forward until it reaches the output layer. The two nodes in the output layer represent the "powder propulsion speed" and the "torque against pipeline resistance," respectively. The response values ​​of these two nodes are directly analyzed. For example, the value of output node 1 is... (Representing the predicted base velocity), the value of node 2 is... (Representing the predicted resistance torque), thereby generating the predicted base flow velocity and resistance torque.

[0027] S302: Call the predicted values ​​of basic flow velocity and resistance torque, extract the flow velocity numerical channel that characterizes the drug metabolism demand, calculate the steady-state flow velocity benchmark required to maintain the real-time anesthesia depth in combination with pharmacokinetic parameters, and simultaneously separate the torque numerical channel that characterizes the mechanical properties of the pipeline. Calculate the additional compensation torque required for the elastic deformation caused by pipeline back pressure, complete the decoupling process of physiological demand control quantity and physical property compensation quantity, and obtain the components of the maintenance flow velocity and deformation compensation torque. Extract flow velocity numerical channel data ( The system calls upon pre-stored propofol pharmacokinetic parameters (such as central compartment distribution volume and clearance rate) and uses the drug metabolism rate equation to calculate the steady-state flow rate baseline required to maintain the current target depth of anesthesia. It assumes that, according to the equation, the metabolic rate of propofol... The actual physical steady-state flow rate corresponding to the unit drug prediction value is: mL / h, simultaneously, separate torque numerical channel data ( To address the elastic deformation caused by pipeline back pressure, the required additional compensation torque is calculated. Using a fluid volume compensation model based on the pipeline wall's elastic modulus, the predicted torque is converted into the additional torque compensation required by the motor. For example, the model calculates that to overcome the current back pressure, an additional torque needs to be added on top of the base torque. The compensation in Nm (Newton-meters), through the above separation calculation, decoupled the physiological demand control quantity (related only to drug metabolism) from the physical characteristic compensation quantity (related only to pipeline mechanical characteristics), and obtained the maintenance flow rate ( mL / h) and deformation compensation torque component ( Nm).

[0028] S303: Based on the components of maintaining flow velocity and deformation compensation torque, a parallel coding method is used to digitally encapsulate the steady-state flow velocity reference and the additional compensation torque. A timing synchronization check bit is added to the instruction frame header, and the flow velocity control signal is mapped to the main infusion channel, and the torque compensation signal is mapped to the auxiliary boosting channel to generate a dual-channel flow velocity and torque composite instruction. Define a The instruction frame structure is a bit-based structure, with the frame header preceding the instruction frame. Bits are written into a specific binary sequence (such as...) () is used as a timing synchronization check bit to indicate the start time of the instruction, followed by... The bit is used to encapsulate the steady-state flow rate reference value, mL / h is converted to fixed-point number encoding, and then... The bit is used to encapsulate the additional compensation torque value, Nm is converted to fixed-point encoding, and finally The bit is a cyclic redundancy check code. This step maps the encapsulated flow rate control signal logic to the main injection channel that controls the main pump injection speed, and maps the torque compensation signal logic to the auxiliary booster channel that adjusts the motor output torque. In this way, a dual-channel flow rate and torque composite command is generated.

[0029] Please see Figure 5 The specific steps of S4 are as follows: S401: Call the dual-channel flow velocity torque composite command, separate the flow velocity control channel data, synchronously collect the real-time fluid pressure value at the end of the injection pipeline, retrieve the preset correspondence matrix between pressure amplitude and flow loss, locate the flow compensation coefficient of the interval to which the real-time pressure value belongs, perform linear superposition operation on the flow compensation coefficient and the flow velocity control channel data, and generate the pressure-related flow compensation correction value. Parse the flow rate control channel data ( (mL / h), while simultaneously, the fluid pressure value at the end of the infusion line is collected in real time (e.g., mL / h). (kPa), retrieve the pre-set pressure amplitude and flow loss correspondence matrix in the memory. This matrix is ​​constructed by measuring the difference between the actual liquid output and the theoretical liquid output in the pipeline under different back pressures using a precision balance. See Table 2, which shows the correspondence between some pressure and flow compensation coefficients. Table 2: Correspondence between Pressure and Flow Compensation Coefficients Real-time pressure range (kPa) Flow compensation coefficient (dimensionless) The flow loss rate (%) was measured in the experiment. Based on real-time pressure values kPa, locate in Table 2 Extract the corresponding flow compensation coefficient from the interval. Perform linear superposition operation: This involves analyzing the flow rate control channel data... Multiply by the flow compensation coefficient The calculation process is as follows: The calculation result is the pressure-related flow compensation correction value ( The flow rate (mL / h) is essentially an additional flow rate added on top of the commanded flow rate to offset the actual loss of liquid output caused by pipeline expansion under high pressure.

[0030] S402: Call the pressure-related flow compensation correction value, load the preset pipeline material stiffness reference parameters, calculate the difference between the hysteresis response characteristics and the stiffness reference and quantify the degree of pipeline deformation, perform gain adjustment operation based on the difference and the resistance torque component, convert the adjusted torque value into the corresponding instantaneous high-frequency pulse sequence, determine the pulse duration and trigger phase, and generate stiffness-adaptive resistance acceleration pulse. Load the preset pipeline material stiffness reference parameters (e.g., standard stiffness modulus) for (MPa), read the hysteresis response characteristic data obtained in step S103 above, calculate the difference between it and the stiffness benchmark, and assume that the equivalent stiffness after hysteresis characteristic conversion is MPa, then the difference MPa, this difference quantifies the degree of softening or deformation of the pipeline relative to the standard state. Based on this difference, gain adjustment calculation is performed on the anti-resistance torque component, and the gain adjustment formula is set as follows: in, To adjust the torque, The original resisting torque ( Nm), Gain coefficient (set to) Substitute the numerical values ​​into the calculation: Nm, the adjusted torque value Nm is converted into an instantaneous high-frequency pulse sequence, and the duration of the pulse is determined as follows: ms, the trigger phase is set before the stepper motor drive cycle. This generates stiffness-adaptive resistance-resistant acceleration pulses.

[0031] S403: Based on stiffness-adaptive resistance acceleration pulse, the flow compensation correction value is converted into the basic drive frequency of the stepper motor. Resistance acceleration pulse is embedded at the beginning of each cycle of the basic drive frequency. The duty cycle and frequency of the drive waveform are jointly modulated to synthesize an electrical signal sequence with dynamic fluid resistance compensation characteristics and generate a fluid dynamics correction drive pulse signal. Adjust the traffic compensation value ( Convert mL / h) to the basic drive frequency of the stepper motor, assuming the conversion coefficients corresponding to the motor step angle and transmission ratio are as follows: If the fundamental driving frequency is Hz / (mL / h), then it is... At the beginning of each sinusoidal cycle of the fundamental driving frequency (i.e., at phase 0), the aforementioned impedance acceleration pulse is physically embedded. This is equivalent to superimposing a short high-energy pulse onto the originally smooth driving waveform. This step jointly modulates the duty cycle and frequency of the driving waveform: during the pulse embedding period, the duty cycle is instantaneously increased to Hz. To output high torque; during non-pulse periods, maintain the duty cycle corresponding to the fundamental frequency (e.g., This synthesis method generates fluid dynamics-corrected drive pulse signals.

[0032] Please see Figure 6 The specific steps of S5 are as follows: S501: Based on the fluid dynamics correction drive pulse signal, drive the fluid delivery execution unit to act, synchronously activate the pipeline end pressure acquisition circuit, acquire the real-time discrete pressure value sequence according to the preset period, perform sliding window differentiation operation on the real-time discrete pressure value sequence, calculate the pressure value increment per unit time, extract the physical quantity reflecting the real-time fluid drive response agility, and generate the real-time pipeline pressure change rate. The drive motor rotates and pushes the syringe piston. Simultaneously, the pressure acquisition circuit at the end of the tubing is activated, according to... The system acquires a real-time discrete pressure numerical sequence at a preset period of ms, performs a sliding window differentiation operation on the sequence, and takes the time span of the 5 most recent pressure points. ms), calculate the current pressure value and The difference between the pressure values ​​before the time interval (ms) is then divided by the time interval. seconds, for example, the current pressure is kPa, ms before kPa, the difference is kPa, the calculation process is as follows kPa / s, this value is the real-time pipeline pressure change rate.

[0033] S502: Call the real-time pipeline pressure change rate, analyze the duty cycle parameter of the drive pulse, establish a fluid transmission model and calculate the expected pressure rise slope under the theoretical interference-free state, perform differential comparison operation between the real-time change rate and the expected rise slope, calculate the numerical deviation between the two, quantify the dynamic following error of the real-time output to the drive command, and generate the pressure following response deviation value. Call the real-time pipeline pressure change rate ( (kPa / s), and analyze the duty cycle parameters of the current driving pulse (e.g., average duty cycle is kPa / s). A fluid transport model based on Poiseuille's law is established, inputting the current flow velocity command and pipeline flow resistance constant. The expected pressure rise slope under theoretically undisturbed conditions is calculated. Assuming the model calculates that... Driven by a flow rate of mL / h, the theoretical pressure rise slope should be: kPa / s, the rate of change in real time ( ) and the expected upward slope ( Perform a difference comparison operation to calculate the numerical deviation: Converting it to a percentage gives approximately This deviation quantifies the dynamic following error of the real-time output to the drive command, generating a pressure following response deviation value.

[0034] S503: Based on the pressure follow-response deviation value, retrieve the preset dynamic response allowable error benchmark value, compare the pressure follow-response deviation value with the dynamic response allowable error benchmark value, and trigger the data resampling mechanism when the dynamic response allowable error benchmark value is exceeded. Optimize the sampling frequency and identify the waveform segment during the deviation period. Encapsulate the waveform segment and deviation verification bit into a data frame to generate a closed-loop pressure response verification signal. Based on the pressure follow-response deviation value ( This step retrieves the dynamic response allowable error baseline value, which is set to... , the deviation value Compared with the benchmark value Numerical comparison was performed because If the current response error is determined to exceed the allowable range, a data resampling mechanism is triggered, automatically adjusting the sampling frequency of the pressure sensor from... Hz temporarily increased to Hz, and identify the specific time period in which the deviation occurred (e.g., in the past). (ms), extract waveform segments within this time period, and add specific offset check bits (such as binary) to the end of the data packet. The waveform data and the check bit are encapsulated into a data frame to generate a closed-loop pressure response check signal. This signal will be fed back to the upper controller to correct the neural network weights in S3 or the compensation coefficients in S4 in the next control cycle.

[0035] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the described technical solutions.

Claims

1. A method for adaptive adjustment of an infusion rate of an intraoperative anesthetic drug based on reinforcement learning, characterized in that, Includes the following steps: S1: Monitor the pressure sensor signal at the end of the anesthetic drug infusion line, record the pressure waveform when the flow rate steps, analyze the rise and fall tail characteristics of the pressure waveform, and obtain the pressure hysteresis response characteristic data of the line. S2: Call the pipeline pressure hysteresis response characteristic data, collect the anesthesia depth and calculate the control error, and splice the pipeline pressure hysteresis response characteristic data, control error and real-time pressure to generate a multi-dimensional fluid environment state mapping vector. S3: Input the multidimensional fluid environment state mapping vector into a deep neural network, map and output the basic flow velocity and resistance torque, separate the flow velocity component that maintains the depth of anesthesia and the torque component that overcomes the back pressure deformation of the pipeline, and obtain a dual-channel flow velocity and torque composite command. S4: For the dual-channel velocity-torque composite command, find the flow loss compensation amount based on the real-time pressure, compare the pipeline pressure hysteresis response characteristic data with the stiffness parameters, and superimpose the acceleration pulse determined by the resistance compensation torque on the speed command to establish a fluid dynamics correction drive pulse signal. S5: By using the fluid dynamics-corrected drive pulse signal, monitor the pressure change rate, compare the real-time pressure change rate with the theoretical pressure change rate, trigger resampling based on the deviation, and generate a closed-loop pressure response verification signal.

2. The method for adaptive adjustment of intraoperative anesthetic drug infusion rate based on reinforcement learning according to claim 1, characterized in that, The pipeline pressure hysteresis response characteristic data includes pressure rise time and pressure release tail time. The multidimensional fluid environment state mapping vector includes mechanical compliance feature vector elements, anesthesia depth deviation, and real-time pipeline pressure. The dual-channel velocity-torque composite command includes a basic target velocity command component and a resistance compensation torque command component. The fluid dynamics correction drive pulse signal includes a flow loss compensation step frequency and a dynamic resistance acceleration pulse sequence. The closed-loop pressure response verification signal includes a pressure change rate deviation metric and a state resampling trigger flag.

3. The method for adaptive adjustment of intraoperative anesthetic drug infusion rate based on reinforcement learning according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Monitors the pressure sensor signal at the end of the anesthetic drug infusion line, collects the time-series voltage sequence output by the pressure sensor in real time during the flow rate command change operation of the infusion pump, identifies the time window from the start of the flow rate change to the pressure recovery to steady state, extracts the corresponding signal segment within the time window and performs digital sampling and analog-to-digital conversion processing to generate a step excitation pressure waveform sequence. S102: Based on the step excitation pressure waveform sequence, traverse the data points of the rising and falling edges of the amplitude, calculate the pressure establishment time constant and peak rise slope value of the rising edge stage, extract the pressure attenuation tailing time and amplitude fall rate of the falling edge stage, and aggregate and encode the calculated time and rate values ​​to obtain the waveform establishment and attenuation feature set. S103: Call the waveform establishment and attenuation feature set, calculate the absolute value of the difference between the pressure establishment time constant and the pressure attenuation tail time, calculate the pressure response lag time and waveform asymmetry coefficient under unit flow rate change, map the lag time and asymmetry coefficient into a multi-dimensional vector, and obtain pipeline pressure hysteresis response feature data.

4. The method for adaptive adjustment of intraoperative anesthetic drug infusion rate based on reinforcement learning according to claim 3, characterized in that, The specific steps of S2 are as follows: S201: Based on the pipeline pressure hysteresis response characteristic data, monitor the patient's real-time anesthesia depth characterization parameters, retrieve the preset anesthesia depth target benchmark value, perform the difference operation between the real-time parameters and the target benchmark value to obtain the deviation amplitude, perform discretization differentiation processing on the deviation amplitude to obtain the deviation change rate, perform linear weighted summation on the deviation amplitude and the deviation change rate according to the preset weight coefficient, quantify the degree of deviation between the real-time control state and the target state, and generate anesthesia depth control error scalar; S202: Call the scalar of the anesthesia depth control error, synchronously collect the real-time fluid pressure value output by the pressure sensor at the end of the infusion line, take the moment of flow rate change as the time reference point, perform time-series synchronous calibration of hysteresis characteristics, control error and real-time pressure, and cascade and splice the calibrated multiple data according to the feature dimension order to construct a multi-source heterogeneous state combination sequence. S203: Based on the multi-source heterogeneous state combination sequence, calculate the arithmetic mean and standard deviation of multi-dimensional values ​​within a sliding window, perform a centering translation operation on the sequence elements using the arithmetic mean, and scale the translated data in combination with the standard deviation, uniformly mapping the data with different physical dimensions to the target numerical range, avoiding the order-of-magnitude differences between feature dimensions, and generating a multi-dimensional fluid environment state mapping vector.

5. The method for adaptive adjustment of intraoperative anesthetic drug infusion rate based on reinforcement learning according to claim 4, characterized in that, The specific steps for S3 are as follows: S301: Based on the multidimensional fluid environment state mapping vector, call the preset network weight parameter matrix, perform multi-level weighted summation operation on the vector elements, perform dimensionality reduction mapping on the operation values ​​in the feature space, convert the multidimensional environment state features into numerical expressions of fluid delivery control dimensions, analyze the response values ​​of the output layer nodes, determine the real-time required liquid propulsion speed and the magnitude of the torque against pipeline resistance, and generate the predicted values ​​of basic flow velocity and resistance torque. S302: Call the predicted values ​​of the basic flow rate and the resistance torque, extract the flow rate numerical channel that characterizes the drug metabolism demand, calculate the steady-state flow rate benchmark required to maintain the real-time anesthesia depth in combination with pharmacokinetic parameters, and simultaneously separate the torque numerical channel that characterizes the mechanical properties of the pipeline. Calculate the additional compensation torque required for the elastic deformation caused by the pipeline back pressure, complete the decoupling process of the physiological demand control quantity and the physical property compensation quantity, and obtain the components of the maintenance flow rate and deformation compensation torque. S303: Based on the maintained flow rate and deformation compensation torque components, a parallel encoding method is used to digitally encapsulate the steady-state flow rate reference and the additional compensation torque. A timing synchronization check bit is added to the instruction frame header, and the flow rate control signal is mapped to the main injection channel, while the torque compensation signal is mapped to the auxiliary boosting channel to generate a dual-channel flow rate and torque composite instruction.

6. The method for adaptive adjustment of intraoperative anesthetic drug infusion rate based on reinforcement learning according to claim 5, characterized in that, The process of calculating the steady-state flow rate baseline required to maintain the depth of real-time anesthesia by combining pharmacokinetic parameters is as follows: calling the preset drug central compartment distribution volume parameter and drug clearance constant, parsing the value in the flow rate numerical channel that characterizes the drug metabolism requirement into the target effect room concentration value, and constructing a plasma concentration decay curve per unit time based on the drug clearance constant. Based on the drug central compartment distribution volume parameter, the drug replenishment rate required to maintain the target effect compartment concentration is inverted. The drug replenishment rate is divided by the standard density of the drug solution, and the resulting volume value is quantified into a flow rate unit of milliliters per hour to generate a steady-state flow rate benchmark.

7. The method for adaptive adjustment of intraoperative anesthetic drug infusion rate based on reinforcement learning according to claim 5, characterized in that, The specific steps of S4 are as follows: S401: Call the dual-channel flow velocity torque composite command, separate the flow velocity control channel data, synchronously collect the real-time fluid pressure value at the end of the injection pipeline, retrieve the preset correspondence matrix between pressure amplitude and flow loss, locate the flow compensation coefficient of the interval to which the real-time pressure value belongs, perform linear superposition operation on the flow compensation coefficient and the flow velocity control channel data, and generate a pressure-related flow compensation correction value. S402: Call the pressure-related flow compensation correction value, load the preset pipeline material stiffness reference parameter, calculate the difference between the hysteresis response characteristics and the stiffness reference and quantify the degree of pipeline deformation, perform gain adjustment operation based on the difference and the resistance torque component, convert the adjusted torque value into the corresponding instantaneous high-frequency pulse sequence, determine the pulse duration and trigger phase, and generate stiffness-adaptive resistance acceleration pulse. S403: Based on the stiffness-adaptive resistance acceleration pulse, the flow compensation correction value is converted into the basic drive frequency of the stepper motor. The resistance acceleration pulse is embedded at the beginning of each cycle of the basic drive frequency. The duty cycle and frequency of the drive waveform are jointly modulated to synthesize an electrical signal sequence with dynamic fluid resistance compensation characteristics, and a fluid dynamics correction drive pulse signal is generated.

8. The method for adaptive adjustment of intraoperative anesthetic drug infusion rate based on reinforcement learning according to claim 7, characterized in that, The process of embedding an anti-resistance acceleration pulse at the start of each cycle of the basic drive frequency and jointly modulating the duty cycle and frequency of the drive waveform is as follows: the flow compensation correction value is converted into a basic drive signal in the form of a square wave according to the step angle of the stepper motor, and the trigger time of each rising edge of the basic drive signal is captured in real time. Within the time window following the rising edge trigger moment, a stiffness-adapted resistance acceleration pulse is superimposed onto the high-level front segment of the basic drive signal. While keeping the period length of the basic drive signal constant, the duration of the high level and the steepness of the waveform front are changed to synthesize a hydrodynamically corrected drive pulse signal.

9. The method for adaptive adjustment of intraoperative anesthetic drug infusion rate based on reinforcement learning according to claim 7, characterized in that, The specific steps of S5 are as follows: S501: Based on the fluid dynamics correction drive pulse signal, drive the fluid delivery execution unit to act, synchronously activate the pipeline end pressure acquisition circuit, acquire the real-time discrete pressure value sequence according to the preset period, perform sliding window differentiation operation on the real-time discrete pressure value sequence, calculate the pressure value increment per unit time, extract the physical quantity reflecting the real-time fluid drive response agility, and generate the real-time pipeline pressure change rate. S502: Call the real-time pipeline pressure change rate, analyze the duty cycle parameter of the drive pulse, establish a fluid transmission model and calculate the expected pressure rise slope under the theoretical interference-free state, perform a differential comparison operation between the real-time change rate and the expected rise slope, calculate the numerical deviation between the two, quantify the dynamic following error of the real-time output to the drive command, and generate a pressure following response deviation value. S503: Based on the pressure following response deviation value, retrieve the preset dynamic response allowable error benchmark value, compare the pressure following response deviation value with the dynamic response allowable error benchmark value, and trigger the data resampling mechanism when the dynamic response allowable error benchmark value is exceeded. Optimize the sampling frequency and identify the waveform segment during the deviation period. Encapsulate the waveform segment and deviation verification bit into a data frame to generate a closed-loop pressure response verification signal.

10. The method for adaptive adjustment of intraoperative anesthetic drug infusion rate based on reinforcement learning according to claim 9, characterized in that, The process of establishing a fluid transport model and calculating the expected pressure rise slope under theoretically undisturbed conditions is as follows: based on the duty cycle parameter of the drive pulse, the pre-stored motor speed characteristic curve is retrieved to obtain the corresponding piston propulsion speed. The theoretical instantaneous flow rate is calculated by combining the piston cross-sectional area of ​​the syringe. The derivative value of the pressure change with time under the condition of no external disturbance is solved, and the derivative value is set as the expected pressure rise slope.

Images