CN116650827A - Flow Pulsatility Control System of ECMO Centrifugal Blood Pump Based on RBF Neural Network - Google Patents

Abstract

An ECMO centrifugal blood pump flow pulsatility control system based on RBF neural network, which realizes the pulsatility of centrifugal blood pump blood flow by periodically changing the rotational speed of the centrifugal blood pump, including the main controller module, signal acquisition module, and motor drive module Human-computer interaction module; the signal acquisition module is used to obtain the current inlet pressure and outlet pressure of the centrifugal blood pump and the patient's intra-aortic blood pressure; the controller module calculates the tracking error based on the patient's actual intra-aortic blood pressure and the expected aortic blood pressure , tracking error first-order differential and other parameters, according to the calculated parameters, use the preset RBF neural network adaptive control algorithm to calculate the rotational speed that the centrifugal blood pump should achieve, as the control amount of the centrifugal blood pump-blood circulation coupling system; The invention realizes the control of the flow pulsation of the ECMO blood pump, and further realizes the precise control of the patient's aortic blood pressure waveform.

Description

Flow pulsatility control system of ECMO centrifugal blood pump based on RBF neural network

Technical Field

The present invention belongs to the technical field of adaptive control of complex nonlinear systems, and in particular to an ECMO centrifugal blood pump flow pulsatility control system based on an RBF neural network.

Background Art

ECMO (Extracorporeal Membrane Oxygenation) is called extracorporeal membrane oxygenation in Chinese, commonly known as "ECMO" or "artificial lung". It is a medical emergency device used to provide continuous extracorporeal respiration and circulation to patients with severe heart and lung failure in order to maintain the patient's life. When ECMO is in operation, blood is drawn from the veins, oxygenated through the membrane lung, and carbon dioxide is discharged. The oxygenated blood can be returned to the veins (V-V bypass) or to the arteries (V-A bypass). The essence of ECMO is a modified artificial heart-lung machine. The core parts are the membrane lung and blood pump, which act as artificial lungs and artificial hearts respectively. They can provide long-term cardiopulmonary support for patients with severe heart and lung failure, thus buying precious time for the rescue of critical illnesses.

RBF neural network is a feedforward neural network. The characteristic of feedforward neural network is that there is no feedback link in the network under this structure, and the output of the previous layer directly constitutes the input of the next layer; the input layer of the network is directly composed of source nodes, receiving external input information, and the output layer derives the output signal of the neural network to the outside world; the input and output layers are directly connected to the outside world, so they are called visible layers, and the structure between the input layer and the output layer has no direct connection with the outside world and is called the hidden layer.

The RBF neural network is named because its hidden layer activation function is the radial basis function. It is a feedforward neural network with only one hidden layer. The overall network structure has three layers. The input layer is the interface between the front end of the neural network and the outside world. It is responsible for receiving input signals and importing the signals into the hidden layer. The hidden layer is the core of the network. The neurons in the hidden layer perform nonlinear transformations on the received signals through activation functions, and send the transformed signals to the output layer through linear transformation. The output layer is the interface between the back end of the neural network and the outside world. It is responsible for integrating the network output and exporting it to the outside world.

The characteristic of RBF neural network is that the input layer and the hidden layer are not connected by weights. The input signal is directly mapped to the hidden layer space through the radial basis function, thereby realizing the mapping of the signal from low dimension to high dimension, making the input signal linearly separable in high dimension; this mapping is nonlinear and the mapping relationship is only related to the radial basis function parameters. The input mapping can be adjusted by adjusting the radial basis function parameters. However, the hidden layer and the output layer are connected by weights, that is, the mapping from the hidden layer space to the output space is linear. The mapping from input to output of the entire neural network is nonlinear, while the network output is linear for adjustable parameters. The weight of the network can be directly solved by a set of linear equations, which greatly speeds up the learning speed and avoids the local minimum problem.

Currently, most centrifugal blood pumps in ECMO operate at a constant speed (Li Guanhua, Zhang Yu, Zhao Zongkai, et al. ECMO pulsatile flow generating device: 202121333914.1 [P]. 2021-12-17). Centrifugal blood pumps operating at a constant speed cannot meet the patient's requirements for different cardiac outputs under different activity states and pathological conditions (such as hypertension, hypotension, hypervolemia, hypovolemia, etc.) (PATIBANDLA PK, RAJASEKARAN N S, SHELAR SB, et al. Evaluation of the effect of diminished pulsatility as seen in continuous flow ventricular assist devices on arterial endothelial cell phenotype and function [J]. The Journal of Heart and Lung Transplantation, 2016, 35 (7): 930-932. DOI: 10.1016/j.healun.2016.03.008.). More importantly, the centrifugal blood pump flow operating at a constant speed cannot produce obvious pulsating blood flow stimulation to the blood vessels, which significantly reduces the pulsatility of the blood vessels (SOUCY KG, KOENIG SC, GIRIDHARAN GA, et al. Rotary pumps and diminished pulsatility: do we need a pulse[J]. ASAIO Journal, 2013, 59(4): 355-366. DOI: 10.1097/MAT.0b013e31829f9bb3.). Various complications caused by the reduced vascular pulsatility of constant-speed centrifugal blood pumps have been reported clinically, such as arteriovenous malformations, gastrointestinal bleeding, hemorrhagic stroke, aortic regurgitation and valve fusion (SOUCY KG, KOENIG SC, GIRIDHARAN GA, et al. Defining pulsatility during continuous-flow ventricular assist device support [J]. The Journal of Heart and Lung Transplantation, 2013, 32(6): 581-587. DOI: 10.1016/j.healun.2013.02.010.) and various neurological complications (Xi Shaosong, Zhu Ying, Diao Mengyuan, et al. Analysis of neurological complications in patients receiving venoarterial extracorporeal membrane oxygenation support after cardiac arrest [J]. Chinese Modern Doctor, 2020, 58(28): 34-40.) etc.

Summary of the invention

In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide an ECMO centrifugal blood pump flow pulsatility control system based on RBF neural network, which realizes the control of ECMO blood pump flow pulsatility by approximating the nonlinear terms in the ideal control law through RBF neural network, thereby realizing precise control of the patient's aortic blood pressure waveform.

An ECMO centrifugal blood pump flow pulsatility control system based on RBF neural network realizes the pulsatility of the blood flow of the centrifugal blood pump by periodically changing the rotation speed of the centrifugal blood pump, and includes a main controller module, a signal acquisition module, a motor drive module and a human-computer interaction module; the main controller module uses an adaptive control algorithm based on RBF neural network to control the centrifugal blood pump, and calculates the rotation speed of the centrifugal blood pump according to the patient's actual aortic pressure and the expected aortic pressure.

The adaptive control algorithm based on RBF neural network adopts radial basis function network with adaptive learning function.

The manner of periodically changing the blood pump rotation speed is to adjust according to a preset blood pressure curve or a preset periodic function describing the blood pressure.

The adaptive control algorithm based on the RBF neural network predicts the corresponding centrifugal blood pump speed according to the actual aortic pressure of the patient to maintain the corresponding aortic blood pressure pulsatility.

The ECMO centrifugal blood pump flow pulsatility control system based on the RBF neural network has an automatic adjustment function, and can automatically adjust the rotation speed of the centrifugal blood pump according to the patient's physiological state and needs to provide corresponding blood flow pulsatility.

The method for establishing the adaptive control algorithm based on RBF neural network is as follows:

1) Establish a mathematical model of the human blood circulation system;

An electrical network is used to simulate the dynamic characteristics of the human blood circulation system. The voltage in the electrical network corresponds to the blood in the blood circulation system, the current corresponds to the blood flow in the blood circulation system, the resistance corresponds to the blood flow resistance in the blood circulation system, the capacitance corresponds to the vascular compliance in the blood circulation system, the inductance corresponds to the blood flow inertia in the blood circulation system, and the diode corresponds to the valve in the circulation system. A combined circuit of a controllable voltage source and a controllable current source is used to simulate the contraction of the left and right ventricles. The resulting mathematical model of the blood circulation system is as follows:

The meanings of the variables in the equation are as follows: _x1 : aortic blood pressure; _x2 : blood volume flow into the systemic arterial system; _x3 : systemic circulation inlet blood pressure; _x4 : blood volume flow into the systemic venous system; _x5 : systemic venous system and right atrium inlet blood pressure; _x6 : left ventricular volume; _x7 : pulmonary artery blood pressure; _x8 : blood volume flow into the pulmonary arterial system; _x9 : pulmonary blood pressure; _x10 : blood volume flow into the pulmonary venous system; _x11 : pulmonary vein and left atrium blood pressure; _x12 : left ventricular volume; _R1 : aortic valve resistance; _R2 : aortic and systemic arterial resistance; _R3 : capillary and systemic venous resistance; _R4 : %tricuspid valve resistance; _R5 : %pulmonary valve resistance; _R6 : pulmonary artery resistance; _R7 : other pulmonary resistance; _R8 : mitral valve resistance; _RL : left ventricular internal viscous resistance; _RR : right ventricular internal viscous resistance; _L1 : aorta-systemic artery inertia; _L2 : systemic meridian system inertia; _L3 : pulmonary artery inertia; _L4 : pulmonary vein inertia; _C1 : aortic compliance; _C2 : systemic arterial system compliance; _C3 : systemic venous system compliance; C4: pulmonary artery compliance; _C5 : pulmonary arterial system compliance; _C6 : pulmonary venous system compliance; S _i (t): function that controls the opening and closing of valve S _i , which is related to the _pressure difference across the valve; ΔP _i (t): pressure difference across the valve S _i ;

The equations related to pulmonary circulation in the above equations are omitted; at the same time, the equations related to ventricular contraction in the above equations do not need to be considered, so the above equations are simplified to:

2) Establish a mathematical model of the centrifugal blood pump;

The centrifugal blood pump used in ECMO equipment is described using the following equation:

Where Q represents the blood flow output by the centrifugal blood pump, σ represents the speed of the centrifugal blood pump, P _out represents the outlet pressure of the centrifugal blood pump, P _in represents the inlet pressure of the centrifugal blood pump, β ₀ , β ₁ , and β ₂ are constants obtained by least squares fitting based on the experimental data of the centrifugal blood pump;

When the constants in the centrifugal blood pump mathematical model are determined, the coefficients in the centrifugal blood pump mathematical model are replaced by variables m ₁ , m ₂ , and m ₃ , and the centrifugal blood pump mathematical model becomes:

Centrifugal blood pump-blood circulation coupling system simulation module: The mathematical model of the centrifugal blood pump is combined with the mathematical model of the human blood circulation system to obtain the mathematical model of the centrifugal blood pump-blood circulation coupling system;

The inlet of the centrifugal blood pump is directly connected to the junction of the superior and inferior vena cava of the patient. It is approximately considered that the inlet pressure of the centrifugal blood pump is the inlet blood pressure of the systemic venous system and the right atrium in the human blood circulation system, that is, it is considered that _Pin = _x5 ; in the coupling model, a resistor _Rp is added between the centrifugal blood pump and the circulatory system to represent the resistance of blood circulation in the oxygenator;

For the centrifugal blood pump-circulatory system coupling model, select the new state variables as follows:

z ₁ : aortic blood pressure; z ₂ : current in L1, blood volume flow into the systemic arterial system; z ₃ : voltage on C2, systemic inlet blood pressure; z ₄ : current in L2, blood volume flow into the systemic venous system; z ₅ : voltage on C3, systemic venous system and right atrium blood pressure; z ₆ : centrifugal blood pump flow;

Assuming that the input u is the square of the centrifugal blood pump speed and the output y is the aortic blood pressure, the state equation of the centrifugal blood pump-circulatory system coupling model is as follows:

The output equation of the centrifugal blood pump-circulatory system coupling model is as follows:

y＝z ₁ .

3) Design the control law of the system and the weight update law of the RBF neural network based on Lyapunov stability theory;

The state equation of the centrifugal blood pump-circulatory system coupling model shows that:

Again The derivative is:

make in The above formula can be written as:

The tracking error of the system is:

e＝yy _d ＝z ₁ -z _1d

Where y _d and z _1d are the desired aortic blood pressures;

Define the tracking error function:

Where λ＞0, it is easy to see that when s→0, e→0 and Therefore, designing a control law that makes the error function s asymptotically stable at s = 0 can ensure that the tracking error of the centrifugal blood pump-circulatory system coupling model control is asymptotically stable at e = 0;

Construct the Lyapunov function:

We can get:

make have to:

According to Lyapunov stability theory, it is necessary to make Negative determination, we can order:

The ideal controller can be solved as:

Where η is a constant greater than zero, used to control The distance from zero point; choose to use RBF neural network to approximate the function

The RBF neural network is divided into input layer, hidden layer and output layer, where the activation function of the hidden layer neurons is expressed as:

Among them, x is the input vector of the neuron, c _j is the center position of the jth neuron activation function, and b _j is the width of the jth neuron activation function;

When using RBF neural network to approximate function f(x), there must be an ideal weight vector W ^* such that:

f(x)＝W ^*Th (x)+∈

Where W ^* is the ideal weight vector of the neural network, ∈ is the approximation error of the neural network, |∈|＜∈ _N , ∈ _N is any real number greater than zero; use RBF neural network to approximate the function Let the actual output of the neural network be:

is the actual weight vector of the RBF neural network, and the approximation error of the neural network can be expressed as:

The weight approximation error Design the adaptive law so that The value of Asymptotically stable at ; design a Lyapunov function that describes both the tracking error and the approximation error:

Where γ＞0, we get:

First, ensure that the tracking error is asymptotically stable to 0, and design the control law as:

When η＞|∈| _max , s∈-sηsgn(s) is negative definite, and the adaptive law is adopted:

have to:

When η＞|∈| _max Negative definite, the whole control system is asymptotically stable when both the tracking error and the network approximation error approach zero;

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

The present invention achieves the pulsatility of the blood flow of the blood pump by periodically changing the rotation speed of the blood pump, thereby improving the pulsatility of the aortic blood pressure of patients using ECMO; at the same time, the present invention adopts a control algorithm including a radial basis function neural network with adaptive learning function, so that the present invention can automatically adjust the rotation speed of the blood pump according to the patient's physiological state and needs to provide appropriate blood flow pulsatility.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a simplified diagram of the system structure of the present invention.

FIG. 2 is a schematic diagram of the RBF neural network structure used in an embodiment of the present invention.

FIG3 is a schematic diagram of a model of a human blood circulation network constructed according to an embodiment of the present invention.

DETAILED DESCRIPTION

The present invention is described in detail below with reference to the accompanying drawings and embodiments.

As shown in FIG1 , an ECMO centrifugal blood pump flow pulsatility control system based on an RBF neural network realizes the pulsatility of the blood flow of the centrifugal blood pump by periodically changing the rotation speed of the centrifugal blood pump, thereby improving the aortic blood pressure pulsatility of the patient, and includes: a main controller module, a signal acquisition module, a motor drive module and a human-computer interaction module;

The main controller module uses an adaptive control algorithm based on RBF neural network to control the centrifugal blood pump and calculates the rotation speed of the centrifugal blood pump according to the patient's actual aortic pressure and the expected aortic pressure. The adaptive control algorithm based on RBF neural network uses a radial basis function network with adaptive learning function to predict the appropriate rotation speed of the centrifugal blood pump according to the patient's actual aortic pressure to maintain appropriate aortic blood pressure pulsatility.

The system has an automatic regulating function and can automatically adjust the rotation speed of the centrifugal blood pump according to the patient's physiological state and needs to provide appropriate blood flow pulsatility.

The main controller module uses ZYNQ series chips, the specific model is XC7Z020CLG400-2;

The signal acquisition module is used to collect the patient's aortic pressure, centrifugal blood pump flow and other related parameters in real time by installing pressure sensors inside the ECMO arterial cannula and at the inlet of the ECMO centrifugal blood pump, and installing flow sensors on the outlet pipe of the centrifugal blood pump;

The motor drive module uses a three-phase half-bridge drive circuit to drive a brushless DC motor connected to a centrifugal blood pump;

The human-computer interaction module includes a touch screen and an alarm system. The touch screen is used to display the important operating parameters and operating status of the system and alarm information, and is also used to input control instructions; the alarm system is used to monitor abnormal centrifugal blood pump speed and abnormal patient aortic blood pressure, and issue corresponding alarms.

The core of this system is an adaptive control algorithm based on RBF neural network. The actual aortic pressure and the expected aortic pressure of the patient are used as inputs, and the speed that the centrifugal blood pump should currently reach is calculated by RBF neural network. The schematic diagram of the RBF neural network structure in the adaptive control algorithm based on RBF neural network is shown in Figure 2. The characteristic of RBF neural network is that the input layer and the hidden layer are not connected by weights. The input signal is directly mapped to the hidden layer space through the radial basis function, thereby realizing the mapping of the signal from low dimension to high dimension, so that the input signal can be linearly separated in high dimension. This mapping is nonlinear and the mapping relationship is only related to the radial basis function parameters. The input mapping can be adjusted by adjusting the radial basis function parameters. However, the hidden layer and the output layer are connected by weights, that is, the mapping from the hidden layer space to the output space is linear. The mapping from input to output of the entire neural network is nonlinear, while the network output is linear with respect to the adjustable parameters. The weight of the network can be directly solved by the linear equation group, thereby greatly accelerating the learning speed and avoiding the local minimum problem.

The method for establishing the adaptive control algorithm based on RBF neural network is as follows:

1) Use an electrical network to simulate the human blood circulation network. The voltage in the electrical network corresponds to the blood in the blood circulation system, the current corresponds to the blood flow in the blood circulation system, the resistance corresponds to the blood flow resistance in the blood circulation system, the capacitance corresponds to the vascular compliance in the blood circulation system, the inductance corresponds to the blood flow inertia in the blood circulation system, and the diode corresponds to the valve in the circulation system. Use a combination circuit of a controllable voltage source and a controllable current source to simulate the contraction of the left and right ventricles, as shown in Figure 3. The resulting mathematical model of the blood circulation system is as follows:

The meanings of the variables in the equation are as follows: _x1 : aortic blood pressure; _x2 : blood volume flow into the systemic arterial system; _x3 : systemic circulation inlet blood pressure; _x4 : blood volume flow into the systemic venous system; _x5 : systemic venous system and right atrium inlet blood pressure; _x6 : left ventricular volume; _x7 : pulmonary artery blood pressure; _x8 : blood volume flow into the pulmonary arterial system; _x9 : pulmonary blood pressure; _x10 : blood volume flow into the pulmonary venous system; _x11 : pulmonary vein and left atrium blood pressure; _x12 : left ventricular volume; _R1 : aortic valve resistance; _R2 : aortic and systemic arterial resistance; _R3 : capillary and systemic venous resistance; _R4 : %tricuspid valve resistance; _R5 : %pulmonary valve resistance; _R6 : pulmonary artery resistance; _R7 : other pulmonary resistance; _R8 : mitral valve resistance; _RL : left ventricular internal viscous resistance; _RR : right ventricular internal viscous resistance; _L1 : aorta-systemic artery inertia; _L2 : systemic meridian system inertia; _L3 : pulmonary artery inertia; _L4 : pulmonary vein inertia; _C1 : aortic compliance; _C2 : systemic arterial system compliance; _C3 : systemic venous system compliance; C4: pulmonary artery compliance; _C5 : pulmonary arterial system compliance; _C6 : pulmonary venous system compliance; S _i (t): function that controls the opening and closing of valve S _i , which is related to the _pressure difference across the valve; ΔP _i (t): pressure difference across the valve S _i ;

Considering the application scenario of ECMO, when the patient adapts to ECMO, blood no longer flows through the pulmonary circulation system, so the equations related to pulmonary circulation in the above equations can be omitted; at the same time, when VA-ECMO is used, ECMO directly pumps blood from the junction of the superior and inferior vena cava to the aorta, and the contraction and relaxation of the heart have almost no direct impact on the dynamic characteristics of the circulatory system. Therefore, there is no need to consider the equations related to ventricular contraction in the above equations, so the above equations can be simplified to:

2) Establish a mathematical model of the centrifugal blood pump;

The centrifugal blood pump used in ECMO equipment is described using the following equation:

Wherein, Q represents the blood flow output by the centrifugal blood pump, σ represents the rotation speed of the centrifugal blood pump, P _out represents the outlet pressure of the centrifugal blood pump, P _in represents the inlet pressure of the centrifugal blood pump, β ₀ , β ₁ , and β ₂ are constants, which can be obtained by least squares fitting based on the experimental data of the centrifugal blood pump;

When the constants in the centrifugal blood pump mathematical model are determined, the coefficients in the centrifugal blood pump mathematical model are replaced by variables m ₁ , m ₂ , and m ₃ , and the centrifugal blood pump mathematical model becomes:

Centrifugal blood pump-blood circulation coupling system simulation module: The mathematical model of the centrifugal blood pump is combined with the mathematical model of the human blood circulation system to obtain the mathematical model of the centrifugal blood pump-blood circulation coupling system;

Considering that the inlet of the centrifugal blood pump can be directly connected to the connection between the superior and inferior vena cava of the patient, it can be approximately considered that the inlet pressure of the centrifugal blood pump is the blood pressure of the systemic venous system and the right atrium in the human blood circulation system, that is, it is considered that _Pin = _x5 ; for the outlet of the centrifugal blood pump, considering that there is an oxygenator between the outlet of the centrifugal blood pump and the patient's aorta, a resistor _Rp is added between the centrifugal blood pump and the circulatory system in the coupling model to represent the resistance of blood circulation in the oxygenator; for the centrifugal blood pump-circulatory system coupling model, the new state variables are selected as follows:

z ₁ : aortic blood pressure; z ₂ : current in L1, blood volume flow into the systemic arterial system; z ₃ : voltage on C2, systemic inlet blood pressure; z ₄ : current in L2, blood volume flow into the systemic venous system; z ₅ : voltage on C3, systemic venous system and right atrium blood pressure; z ₆ : centrifugal blood pump flow;

Assuming that the input u is the square of the centrifugal blood pump speed and the output y is the aortic blood pressure, the state equation of the centrifugal blood pump-circulatory system coupling model is as follows:

The output equation of the centrifugal blood pump-circulatory system coupling model is as follows:

y＝z ₁

3) Design the control law of the system and the weight update law of the RBF neural network based on Lyapunov stability theory;

The state equation of the centrifugal blood pump-circulatory system coupling model shows that:

Again The derivative is:

make in The above formula can be written as:

The tracking error of the system is:

e＝yy _d ＝z ₁ -z _1d

Where y _d and z _1d are the desired aortic blood pressures;

Define the tracking error function:

Where λ＞0, it is easy to see that when s→0, e→0 and Therefore, designing a control law that makes the error function s asymptotically stable at s = 0 can ensure that the tracking error of the centrifugal blood pump-circulatory system coupling model control is asymptotically stable at e = 0;

Construct the Lyapunov function:

We can get:

make have to:

According to Lyapunov stability theory, it is necessary to make Negative determination, we can order:

The ideal controller can be solved as:

Where η is a constant greater than zero, used to control The distance from zero; in the above formula, the function The nonlinearity is high and it is difficult to find its expression, so we choose to use RBF neural network to approximate the function

In the RBF neural network used in the present invention, the input layer has 5 neurons, the hidden layer has 21 neurons, and the output layer has 1 neuron, wherein the general expression of the activation function of the hidden layer neuron is:

Among them, x is the input vector of the neuron, c _j is the center position of the jth neuron activation function, and b _j is the width of the jth neuron activation function;

Theoretically, RBF neural network can approximate a nonlinear function with arbitrary precision. Therefore, when using RBF neural network to approximate function f(x), there must be an ideal weight vector W ^* such that:

f(x)＝W ^*Th (x)+∈

Where W ^* is the ideal weight vector of the neural network, ∈ is the approximation error of the neural network, |∈|＜∈ _N , ∈ _N is any real number greater than zero; therefore, the RBF neural network can be used to approximate the function Let the actual output of the neural network be:

is the actual weight vector of the RBF neural network, and the approximation error of the neural network can be expressed as:

The weight approximation error Design the adaptive law so that The value of In order to make the control globally stable, a Lyapunov function that describes both the tracking error and the approximation error is designed:

Where γ＞0, we can get:

First, ensure that the tracking error is asymptotically stable to 0, and design the control law as:

When η＞|∈| _max , s∈-sηsgn(s) is negative definite, and the adaptive law is adopted:

We can get:

It can be seen that when η＞|∈| _max Negative definite, the whole control system is asymptotically stable when both the tracking error and the network approximation error approach zero;

The operation process of the adaptive control algorithm based on RBF neural network in the main controller module is as follows:

1) Initialization; setting the initial operating parameters based on the RBF neural network, the operating parameters include: the center position matrix c _ij of the radial basis function in the RBF neural network, the width vector b _j of the radial basis function, the weight vector W of the RBF neural network, the constant λ in the error function, and the adaptive gain matrix T;

2) Run the adaptive control algorithm based on RBF neural network; the preset RBF neural network will calculate the control quantity of the current centrifugal blood pump-blood circulation coupling system according to the current weights and network input parameters. At the same time, the preset RBF neural network weight update law will correct the weight vector of the RBF neural network in real time according to the current tracking error to minimize the tracking error.

Claims (10)

1. An ECMO centrifugal blood pump flow pulsatility control system based on RBF neural network, characterized in that the pulsatility of the blood flow of the centrifugal blood pump is achieved by periodically changing the rotation speed of the centrifugal blood pump, comprising a main controller module, a signal acquisition module, a motor drive module and a human-computer interaction module; the main controller module uses an adaptive control algorithm based on RBF neural network to control the centrifugal blood pump, and calculates the rotation speed of the centrifugal blood pump according to the patient's actual aortic pressure and the expected aortic pressure. 2. The system according to claim 1 is characterized in that the adaptive control algorithm based on RBF neural network adopts a radial basis function network with adaptive learning function. 3. The system according to claim 1 is characterized in that the manner of periodically changing the blood pump speed is adjusted according to a preset blood pressure curve or a preset periodic function describing blood pressure. 4. The system according to claim 1 is characterized in that the adaptive control algorithm based on the RBF neural network predicts the corresponding centrifugal blood pump speed according to the patient's actual aortic pressure to maintain the corresponding aortic blood pressure pulsatility. 5. The system according to claim 1 is characterized in that it has an automatic adjustment function and can automatically adjust the rotation speed of the centrifugal blood pump according to the patient's physiological state and needs to provide corresponding blood flow pulsatility. 6. The system according to claim 1 is characterized in that: the main controller module adopts ZYNQ series chip, model number is XC7Z020CLG400-2. 7. The system according to claim 1 is characterized in that: the data acquisition module is used to collect the patient's aortic pressure, centrifugal blood pump flow and other related parameters in real time by installing pressure sensors inside the ECMO arterial cannula and at the inlet of the ECMO centrifugal blood pump and installing a flow sensor on the outlet pipe of the centrifugal blood pump. 8. The system according to claim 1, characterized in that: the motor drive module uses a three-phase half-bridge drive circuit to drive a brushless DC motor connected to the centrifugal blood pump. 9. The system according to claim 1 is characterized in that: the human-computer interaction module includes a touch screen and an alarm system, the touch screen is used to display the operating parameters and operating status of the system and alarm information, and is also used to input control instructions; the alarm system is used to monitor abnormal blood pump speed and abnormal patient aortic blood pressure, and issue corresponding alarms. 10. The system according to claim 3, characterized in that the method for establishing the adaptive control algorithm based on RBF neural network is as follows: 1) Use an electrical network to simulate the human blood circulation network. The voltage in the electrical network corresponds to the blood in the blood circulation system, the current corresponds to the blood flow in the blood circulation system, the resistance corresponds to the blood flow resistance in the blood circulation system, the capacitance corresponds to the vascular compliance in the blood circulation system, the inductance corresponds to the blood flow inertia in the blood circulation system, and the diode corresponds to the valve in the circulation system. Use a combination circuit of a controllable voltage source and a controllable current source to simulate the contraction of the left and right ventricles. The resulting mathematical model of the blood circulation system is as follows: The meanings of the variables in the equation are as follows: _x1 : aortic blood pressure; _x2 : blood volume flow into the systemic arterial system; _x3 : systemic circulation inlet blood pressure; _x4 : blood volume flow into the systemic venous system; _x5 : systemic venous system and right atrium inlet blood pressure; _x6 : left ventricular volume; _x7 : pulmonary artery blood pressure; _x8 : blood volume flow into the pulmonary arterial system; _x9 : pulmonary blood pressure; _x10 : blood volume flow into the pulmonary venous system; _x11 : pulmonary vein and left atrium blood pressure; _x12 : left ventricular volume; _R1 : aortic valve resistance; _R2 : aortic and systemic arterial resistance; _R3 : capillary and systemic venous resistance; _R4 : %tricuspid valve resistance; _R5 : %pulmonary valve resistance; _R6 : pulmonary artery resistance; _R7 : other pulmonary resistance; _R8 : mitral valve resistance; _RL : left ventricular internal viscous resistance; _RR : right ventricular internal viscous resistance; _L1 : aorta-systemic artery inertia; _L2 : systemic meridian system inertia; _L3 : pulmonary artery inertia; _L4 : pulmonary vein inertia; _C1 : aortic compliance; _C2 : systemic arterial system compliance; _C3 : systemic venous system compliance; C4: pulmonary artery compliance; _C5 : pulmonary arterial system compliance; _C6 : pulmonary venous system compliance; S _i (t): function that controls the opening and closing of valve S _i , which is related to the _pressure difference across the valve; ΔP _i (t): pressure difference across the valve S _i ; The equations related to pulmonary circulation in the above equations are omitted; and the equations related to ventricular contraction in the above equations do not need to be considered. The above equations are simplified to: 2) Establish a mathematical model of the centrifugal blood pump; The centrifugal blood pump used in ECMO equipment is described using the following equation: Where Q represents the blood flow output by the centrifugal blood pump, σ represents the speed of the centrifugal blood pump, P _out represents the outlet pressure of the centrifugal blood pump, P _in represents the inlet pressure of the centrifugal blood pump, β ₀ , β ₁ , and β ₂ are constants obtained by least squares fitting based on the experimental data of the centrifugal blood pump; When the constants in the centrifugal blood pump mathematical model are determined, the coefficients in the centrifugal blood pump mathematical model are replaced by variables m ₁ , m ₂ , and m ₃ , and the centrifugal blood pump mathematical model becomes: The inlet of the centrifugal blood pump is directly connected to the junction of the superior and inferior vena cava of the patient. It is considered that the inlet pressure of the centrifugal blood pump is the inlet blood pressure of the systemic venous system and the right atrium in the human blood circulation system, that is, it is considered that _Pin = _x5 ; in the coupling model, a resistor _Rp is added between the centrifugal blood pump and the circulatory system to represent the resistance of blood circulation in the oxygenator; For the centrifugal blood pump-circulatory system coupling model, select the new state variables as follows: z ₁ : aortic blood pressure; z ₂ : current in L1, blood volume flow into the systemic arterial system; z ₃ : voltage on C2, systemic inlet blood pressure; z ₄ : current in L2, blood volume flow into the systemic venous system; z ₅ : voltage on C3, systemic venous system and right atrium blood pressure; z ₆ : centrifugal blood pump flow; Assuming that the input u is the square of the centrifugal blood pump speed and the output y is the aortic blood pressure, the state equation of the centrifugal blood pump-circulatory system coupling model is as follows: The output equation of the centrifugal blood pump-circulatory system coupling model is as follows: y＝z ₁ 3) Design the control law of the system and the weight update law of the RBF neural network based on Lyapunov stability theory; From the state equation of the centrifugal blood pump-circulatory system coupling model, we know: Again The derivative is: make in The above formula is written as: The tracking error of the system is: e＝yy _d ＝z ₁ -z _1d Where y _d and z _1d are the desired aortic blood pressures; Define the tracking error function: Where λ＞0, when s→0, e→0 and Therefore, a control law is designed to make the error function s asymptotically stable at s = 0, that is, to ensure that the tracking error of the centrifugal blood pump-circulatory system coupling model control is asymptotically stable at e = 0; Construct the Lyapunov function: have to: make have to: According to Lyapunov stability theory, it is necessary to make Negative determination, order: The ideal controller is solved as: Where η is a constant greater than zero, used to control The distance from zero point; choose to use RBF neural network to approximate the function The RBF neural network is divided into input layer, hidden layer and output layer, where the activation function of the hidden layer neurons is expressed as: Among them, x is the input vector of the neuron, c _j is the center position of the jth neuron activation function, and b _j is the width of the jth neuron activation function; When using RBF neural network to approximate function f(x), there must be an ideal weight vector W ^* such that: f(x)＝W ^*Th (x)+∈ Where W ^* is the ideal weight vector of the neural network, ∈ is the approximation error of the neural network, |∈|＜∈ _N , ∈ _N is any real number greater than zero; use RBF neural network to approximate the function Let the actual output of the neural network be: is the actual weight vector of the RBF neural network, and the approximation error of the neural network is expressed as: The weight approximation error Design the adaptive law so that The value of Asymptotically stable at ; design a Lyapunov function that describes both the tracking error and the approximation error: Where γ＞0, we get: First, ensure that the tracking error is asymptotically stable to 0, and design the control law as: When η＞|∈| _max , s∈-sηsgn(s) is negative definite, and the adaptive law is adopted: have to: When η＞|∈| _max The whole control system is asymptotically stable when both the tracking error and the network approximation error approach zero.