WO2020107694A1 - 呼吸机比例阀流量控制方法、装置、计算机设备 - Google Patents

呼吸机比例阀流量控制方法、装置、计算机设备 Download PDF

Info

Publication number
WO2020107694A1
WO2020107694A1 PCT/CN2019/072605 CN2019072605W WO2020107694A1 WO 2020107694 A1 WO2020107694 A1 WO 2020107694A1 CN 2019072605 W CN2019072605 W CN 2019072605W WO 2020107694 A1 WO2020107694 A1 WO 2020107694A1
Authority
WO
WIPO (PCT)
Prior art keywords
ventilator
neural network
network model
operating parameters
proportional valve
Prior art date
Application number
PCT/CN2019/072605
Other languages
English (en)
French (fr)
Inventor
封志纯
吴本清
李秋华
敖伟
罗小锁
陈浪
Original Assignee
深圳市科曼医疗设备有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳市科曼医疗设备有限公司 filed Critical 深圳市科曼医疗设备有限公司
Publication of WO2020107694A1 publication Critical patent/WO2020107694A1/zh

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61M—DEVICES FOR INTRODUCING MEDIA INTO, OR ONTO, THE BODY; DEVICES FOR TRANSDUCING BODY MEDIA OR FOR TAKING MEDIA FROM THE BODY; DEVICES FOR PRODUCING OR ENDING SLEEP OR STUPOR
    • A61M16/00—Devices for influencing the respiratory system of patients by gas treatment, e.g. mouth-to-mouth respiration; Tracheal tubes
    • A61M16/20—Valves specially adapted to medical respiratory devices
    • A61M16/201—Controlled valves
    • A61M16/202—Controlled valves electrically actuated
    • A61M16/203—Proportional
    • A61M16/205—Proportional used for exhalation control
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
    • G16H40/60—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices
    • G16H40/63—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices for local operation
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61M—DEVICES FOR INTRODUCING MEDIA INTO, OR ONTO, THE BODY; DEVICES FOR TRANSDUCING BODY MEDIA OR FOR TAKING MEDIA FROM THE BODY; DEVICES FOR PRODUCING OR ENDING SLEEP OR STUPOR
    • A61M16/00—Devices for influencing the respiratory system of patients by gas treatment, e.g. mouth-to-mouth respiration; Tracheal tubes
    • A61M16/0003—Accessories therefor, e.g. sensors, vibrators, negative pressure
    • A61M2016/003—Accessories therefor, e.g. sensors, vibrators, negative pressure with a flowmeter
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61M—DEVICES FOR INTRODUCING MEDIA INTO, OR ONTO, THE BODY; DEVICES FOR TRANSDUCING BODY MEDIA OR FOR TAKING MEDIA FROM THE BODY; DEVICES FOR PRODUCING OR ENDING SLEEP OR STUPOR
    • A61M2205/00—General characteristics of the apparatus
    • A61M2205/33—Controlling, regulating or measuring
    • A61M2205/3331—Pressure; Flow
    • A61M2205/3334—Measuring or controlling the flow rate

Definitions

  • the present application relates to the technical field of ventilator, in particular to a flow control method, device, computer equipment and storage medium of proportional valve of ventilator.
  • Ventilator is a vital medical device that can prevent and treat respiratory failure, reduce complications, save and prolong the life of patients.
  • As an effective method that can artificially replace autonomous ventilation function it has been widely used in various Respiratory failure due to causes, anesthesia and respiratory management during major surgery, respiratory support therapy, and emergency resuscitation occupy a very important position in the field of modern medicine.
  • the proportional valve of the ventilator mostly uses the electromagnetic proportional valve.
  • the electromagnetic proportional valve refers to a proportional valve that uses a proportional electromagnet as an electrical-mechanical conversion element.
  • the proportional electromagnet converts the input current signal into a force and displacement mechanical signal output. Then control parameters such as pressure, flow and direction.
  • External interference factors include pressure difference changes, current interference, sensor noise and other factors.
  • a flow control method for a proportional valve of a ventilator including:
  • the current signal is input to the proportional valve in the ventilator, and the current signal is used to control the flow of the proportional valve.
  • the preset trained neural network model is an Elman (Elman) neural network model.
  • the method further includes:
  • the parameters of the initial Elman neural network model are trained to obtain the trained Elman neural network model.
  • training the initial Elman neural network model includes:
  • the Levenberg-Marquardt algorithm is used to adjust the weight and threshold of the initial Elman neural network model until the optimal weight and threshold are determined.
  • adjusting the weights and thresholds of the initial Elman neural network model according to the Levenberg-Marquardt algorithm until determining the optimal weights and thresholds includes:
  • reading the operating parameters of the ventilator system includes the pressure difference at the current moment, the flow sensor value of the previous frame, and the target flow value of the next frame.
  • the preset Elman neural network model includes 4 input layer neurons, 7 hidden layer neurons and state layer neurons, and 1 output layer neuron.
  • a flow control device for a proportional valve of a ventilator including:
  • Data reading module used to read the operating parameters of the ventilator system
  • the first input module is used to input the operating parameters of the ventilator system into the preset trained neural network model, and the preset trained neural network model is generated based on the training parameters of the historical system of the ventilator;
  • the data acquisition module is used to acquire the current signal output by the preset trained neural network model according to the operating parameters of the ventilator system;
  • the second input module is used to input a current signal to the proportional valve in the ventilator, and the current signal is used to control the flow of the proportional valve.
  • the flow control device of the proportional valve of the ventilator further includes:
  • the training module is used to initialize the connection weights and thresholds of each layer of the initial Elman neural network model; obtain the operating parameters of the historical system of the ventilator; adjust the weights of the initial Elman neural network model using the Levenberg-Marquardt algorithm according to the operating parameters of the historical system of the ventilator Value and threshold until the optimal weight and threshold are determined.
  • a computer device includes a memory and a processor.
  • the memory stores a computer program
  • the processor implements the following steps when the computer program is executed:
  • the current signal is input to the proportional valve in the ventilator, and the current signal is used to control the flow of the proportional valve.
  • the computer program is executed by a processor, the following steps are realized:
  • the current signal is input to the proportional valve in the ventilator, and the current signal is used to control the flow of the proportional valve.
  • the above-mentioned ventilator proportional valve flow control method, device, computer equipment and storage medium read the ventilator system operating parameters, input the ventilator system operating parameters into a preset neural network model generated based on the training of the ventilator historical system operating parameters, and obtain Let the trained neural network model output the current signal according to the operating parameters of the ventilator system, and input the current signal to the proportional valve in the ventilator to control the flow of the proportional valve.
  • FIG. 1 is a prediction block diagram of flow control of a proportional valve of a ventilator in an embodiment
  • FIG. 2 is a flowchart of a flow control method of a proportional valve of a ventilator in an embodiment
  • FIG. 3 is a flowchart of a proportional valve flow control method of a ventilator in an embodiment
  • FIG. 4 is a schematic structural diagram of an Elman neural network in an embodiment
  • FIG. 5 is a schematic diagram of the three-dimensional curved surface fitting effect of the pressure difference-flow-current DA of the low-pressure valve obtained by using the pressure difference-flow two-dimensional table query;
  • FIG. 6 is a schematic diagram of the three-dimensional surface fitting effect of the pressure difference-flow-current DA of the low pressure valve fitted by the Elman neural network model;
  • FIG. 7 is a schematic structural diagram of a flow control device for a proportional valve of a ventilator in an embodiment
  • FIG. 8 is a schematic diagram of the internal structure of a computer device in an embodiment.
  • FIG. 1 is a prediction block diagram of the flow control of the proportional valve of the ventilator.
  • the ventilator includes a differential pressure sensor and a flow sensor, a proportional valve, and the proportional valve in the ventilator is Low-pressure electromagnetic proportional valve, the working principle of the electromagnetic proportional valve is controlled by a proportional solenoid, so that the output pressure or flow rate is proportional to the input current, so the method of changing the input electrical signal can be used to continuously control the pressure and flow rate.
  • the low-pressure electromagnetic proportional valve receives the current signal output by the neural network model to continuously control the flow of the ventilator.
  • Differential pressure sensor 10 is used to measure the difference between two pressures.
  • Flow sensor 20 is used to measure the air flow into the ventilator engine at every moment.
  • the flow sensor 20 mainly outputs the previous frame flow sensor value and the current frame sensor value, and the flow command 30 is the target flow value of the next frame.
  • Variables are input to a trained multi-input dynamic regression neural network, that is, the Elman neural network model, dynamically predicts the current signal required to generate the target flow value of the next frame, and inputs the predicted current signal into the ventilator proportional valve. The flow of the proportional valve of the ventilator is controlled.
  • a method for controlling the flow rate of a proportional valve of a ventilator is provided.
  • the method is applied to a microprocessor of a ventilator as an example for illustration, and includes the following steps:
  • Step S200 Read the operating parameters of the ventilator system.
  • the operating parameters of the ventilator system are the data generated by the ventilator during normal operation, which can reflect the operating state of the ventilator.
  • the microprocessor of the ventilator needs to collect the operating parameters of the ventilator system, read the breath from the memory controller
  • the operation data collected by each component of the machine in real time is sufficient.
  • the ventilator operating parameters may include parameters such as exhalation pressure, inspiratory pressure, pressure difference sensor value, flow sensor value, and inspiratory flow rate.
  • the ventilator system operating parameters read by the ventilator processor may include parameters such as current pressure difference, current frame flow sensor value, previous frame flow sensor value, next frame target flow value, current signal, etc.
  • the differential pressure is collected by the differential pressure sensor in real time.
  • the current frame flow sensor value and the previous frame flow sensor value are collected by the flow sensor.
  • the next frame target flow value is a preset value.
  • step S400 the operating parameters of the ventilator system are input into a preset trained neural network model, and the preset trained neural network model is generated based on training of operating parameters of the historical system of the ventilator.
  • the microprocessor inputs the read current pressure difference, current frame flow sensor value, previous frame flow sensor value, and next frame target flow value into a preset neural network model, the preset neural network model is based on the ventilator history system Running parameter training is generated.
  • the preset trained neural network model is a trained neural network model, which is used to output a reasonable current signal to control the flow, eliminate the noise, airflow speed, Factors such as pressure difference changes.
  • the operating parameters of the historical system of the ventilator include the pressure difference value, the flow value at each moment, and the current signal.
  • Step S600 Obtain a current signal output by the preset trained neural network model according to the operating parameters of the ventilator system.
  • the current signal is actually a current analog signal after digital-to-analog conversion of the DA (Digital Analog) converter, that is, the current DA value.
  • the analog signal is distributed in all corners of nature, such as temperature changes, and the digital signal is Artificially abstracted signals that are discontinuous in magnitude.
  • Electrical analog signals mainly refer to electrical signals with continuous amplitude and phase. This signal can be subjected to various calculations by analog circuits, such as amplification, addition and multiplication.
  • the current DA value is used to continuously control the flow of the proportional valve of the ventilator.
  • step S800 a current signal is input to the proportional valve in the ventilator, and the current signal is used to control the flow of the proportional valve.
  • the microprocessor inputs the current signal output from the preset neural network, that is, the current DA value, to the proportional valve of the ventilator.
  • the proportional valve receives the current DA value, and uses the current DA value to continuously control the flow rate of the ventilator.
  • the proportional valve is usually composed of a spool and a control coil. The coil passes current and controls the magnitude of the current to change the position of the spool, that is, to control the opening of the proportional valve, thereby controlling the flow rate.
  • the preset trained neural network model is an Elman neural network model.
  • the method further includes: S100, establishing an initial Elman neural network model; S120, according to Levenberg-Marquardt algorithm and the operating parameters of the ventilator history system, the parameters of the initial Elman neural network model are trained, and the trained Elman neural network model is obtained.
  • the adaptive adjustment ability of this application selects the Elman neural network model trained based on Levenberg-Marquardt algorithm to control the output of proportional valve flow.
  • the selected Elman neural network is a typical local regression network, which belongs to the feedback neural network. It is very similar to the forward neural network and has stronger computing power. Its outstanding advantage is that it has a strong optimization calculation and Associative memory function.
  • the basic Elman neural network is composed of input layer, hidden layer, state layer and output layer. Compared with the BP (back propagation) network, the Elman neural network has an additional state layer to form local feedback.
  • the transfer function of the state layer is a linear function, but there is an additional delay unit, so The state layer can memorize the past state, and at the next moment together with the input of the network as the input of the hidden layer, so that the network has a dynamic memory function.
  • the initial Elman neural network model is a neural network model based on the Elman neural network.
  • the construction process includes: collecting the operating parameters of the ventilator historical system as a sample set, normalizing the sample set, and training method of the neural network model It is the Levenberg-Marquardt algorithm.
  • the Levenberg-Marquardt algorithm is the most widely used nonlinear least squares algorithm.
  • the Chinese is the Levenberg-Marquardt method. It is an algorithm that uses the gradient to find the maximum (small) value.
  • the Levenberg-Marquardt algorithm is one of the optimization algorithms. Optimization is to find the parameter vector that minimizes the function value. Its application fields are very wide, such as: economics, management optimization, network analysis, optimal design, mechanical or electronic design, etc. It can be understood that the Levenberg-Marquardt algorithm can also be Gradient descent, Newton’s method, Quasi-Newton method, and other training algorithms, which are not limited herein.
  • the output of the hidden layer can be automatically connected to the input of the hidden layer.
  • the sensitivity of the neural network to historical data Significantly strengthened, and there is a feedback network inside to improve the network's ability to process dynamic information and better achieve dynamic modeling; the initial Elman neural network model is trained using the Levenberg-Marquardt algorithm to make the error function converge faster and more accurate higher.
  • the parameters for training the initial Elman neural network model include: S122, initializing the connection weights of each layer of the initial Elman neural network model and Thresholds of each layer; S124, obtaining the operating parameters of the historical system of the ventilator; S126, adjusting the weights and thresholds of the initial Elman neural network model according to the operating parameters of the historical system of the ventilator, using the Levenberg-Marquardt algorithm, until the optimal weights and thresholds are determined .
  • the transfer function of the neuron has a large difference between [0,1], when the input is greater than 1, the transfer function value does not change much, and its derivative or slope is relatively small, which is not conducive to the implementation of the backpropagation algorithm.
  • the propagation algorithm needs to use the gradient information of the transfer function of each neuron.
  • the initial weight of the initial Elman neural network model is normalized by input, and the initial weight and threshold are normalized to between 0 and 1.
  • the initial weight and The threshold can be normalized to between -1 and 1.
  • the neural network includes 4 input neurons, the input of the first input neuron is constant at 1, and the weight of each hidden layer neuron connected to the neuron to the neuron is equivalent to the bias; 7 hidden layers Neuron, where the input of the first hidden layer neuron is always 1, which is equivalent to the hidden layer to the output layer bias, no memory function, no local feedback, the remaining 6 hidden layers have local feedback, corresponding to 6 states Layer neurons, where the state layer is also called the receiving layer.
  • Each state layer neuron memorizes the state of the hidden layer neuron at the time before the frame, and connects to each hidden layer neuron through weight feedback connection.
  • the specific process includes: Let the error index function E(w k ) be: In the formula, Y i is the current expected output vector, Y i ′ is the actual current output vector, p is the number of samples, w is the vector composed of the weights and thresholds of the initial Elman neural network model, and e i (w) is the error;
  • ⁇ w [J T (W)J(w)+ ⁇ I] -1 J T (W)e(w), where I is the unit matrix and ⁇ is user-defined J(w) is the Jacobian matrix, or Jacobian matrix, which defines the Jacobian matrix to be composed of the partial derivatives of the error terms to the parameters; the specific steps are as follows:
  • the process of training the Elman neural network model using the Levenberg-Marquardt algorithm is actually a process of finding the optimal parameters, that is, learning and adjusting the weights and thresholds of the network model, so that the network model realizes the given input/output mapping Relationship, complete the identification of the system.
  • the preset training error allowable value is 0.001.
  • the weights and thresholds of the Elman neural network model are continuously adjusted until the Elman neural network when the error index function value is determined to be less than the training error
  • the weights and thresholds of the model that is, when the value of the error indicator function is less than 0.001, it is regarded as the end of training, and the new weights and thresholds determined by the trained neural network model can predict the new output value.
  • the allowable value of the training error may be 0.01, 0.0099, and other allowable values of error, which are not limited herein.
  • the Levenberg-Marquardt algorithm is used to update the training data in real time to obtain the optimal weights and thresholds suitable for the current moment. The error function converges faster and the prediction accuracy is higher.
  • the parameters for training the initial Elman neural network model include: calculating the loss value, gradient, and approximate Hessian matrix, and then determining the attenuation parameters and attenuation coefficients .
  • the Levenberg-Marquardt algorithm is mainly aimed at the loss function of the square sum error class.
  • the Levenberg-Marquardt algorithm is also known as the least square method of attenuation. It is for the loss function to be in the form of a square sum error. It does not require accurate calculation of the Hessian matrix, and a gradient vector and a Jacobian matrix are required.
  • n is the number of parameters of the neural network
  • the size of the Jacobian matrix is m ⁇ n.
  • the attenuation factor ⁇ is set to 0, it is equivalent to Newton's method. If ⁇ is set very large, this is equivalent to a gradient descent method with a small learning rate. The initial value of the parameter ⁇ is very large, so the first few updates are along the gradient descent direction. If iterative update fails at a certain step, ⁇ is expanded. Otherwise, ⁇ decreases as the loss value decreases, Levenberg-Marquardt is close to Newton's method, this process can accelerate the speed of convergence.
  • the operating parameters of the ventilator system include the pressure difference at the current moment, the flow sensor value of the previous frame, and the target flow value of the next frame.
  • the current differential pressure is collected by the differential pressure sensor
  • the flow sensor value of the previous frame is collected by the flow sensor
  • the target flow value of the next frame is a preset value, which is stored in the memory of the ventilator in advance.
  • the current flow control parameters and methods are not only related to the target flow in the next frame, but will also be affected by the current valve opening and the output flow of the previous and subsequent frames and the target flow change span. The control parameters inevitably required for the flow span and the small flow span are different.
  • this application extracts the current pressure difference, the previous frame flow sensor value and the next frame target flow value from the original data as the main component analysis, and analyzes the read current time pressure difference, the previous frame flow sensor value and the next frame target flow value Perform standardization processing, construct a new feature vector, and input it into the preset Elman neural network model, and dynamically fit and predict the current signal required to generate the target flow value of the next frame, with a view to the Elman neural network model according to the current time
  • the pressure difference, the flow sensor value of the previous frame and the target flow value of the next frame output a reasonable current signal.
  • Figure 5 is the three-dimensional surface fitting effect of low pressure valve pressure difference-flow-current DA obtained by querying the differential pressure-flow two-dimensional table
  • Figure 6 is the low pressure fitting using the Elman neural network model.
  • Valve pressure difference-flow-current DA three-dimensional surface fitting effect we can see that by inputting the current pressure difference, the previous frame flow sensor value and the next frame target flow value into the trained Elman neural network The model makes the method of dynamically predicting the current signal required to generate the target flow value of the next frame.
  • the control surface of the proportional valve is smoother, which means that the fitting accuracy is higher, the output current signal is more reasonable, and the proportional valve can be In the process of flow control, it can better adapt to the influence of external factors such as pressure difference change, flow span, noise interference and so on.
  • the preset Elman neural network model includes 4 input layer neurons, 7 hidden layer neurons and state layer neurons, and 1 output layer neuron.
  • the structural size of the initial Elman neural network model used is 4-7-1, that is, 4 input layer neurons, 7 hidden layer neurons, and 1 output layer neuron
  • the number of state layer neurons is equal to the number of hidden layer neurons, that is, 7 state layer neurons.
  • the first hidden layer neuron has no memory function and no local feedback, and the remaining 6 hidden layers have local feedback, corresponding to 6 state layer neurons.
  • Each state layer neuron memorizes the state of the hidden layer neuron at the time before the frame, and connects to each hidden layer neuron through weight feedback connection.
  • the neural network structure uses only one hidden layer by default. If multiple hidden layers are used, the number of neurons in each hidden layer is the same. It is understandable that the Elman neural network structure size can also be other The structure is not limited here. In this embodiment, 7 hidden layer neurons are used, and the classification effect is better.
  • the above-mentioned ventilator proportional valve flow control method, device, computer equipment and storage medium read the ventilator system operating parameters, input the ventilator system operating parameters into a preset neural network model generated based on the training of the ventilator historical system operating parameters, and obtain Let the trained neural network model output the current signal according to the operating parameters of the ventilator system, and input the current signal to the proportional valve in the ventilator to control the flow of the proportional valve.
  • steps in the flowcharts of FIGS. 2-3 are displayed in order according to the arrows, the steps are not necessarily executed in the order indicated by the arrows. Unless clearly stated in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in FIGS. 2-3 may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but may be executed at different times. These sub-steps or stages The execution order of is not necessarily sequential, but may be executed in turn or alternately with at least a part of other steps or sub-steps or stages of other steps.
  • a proportional valve flow control device for a ventilator including: a data reading module 710, a first input module 720, a data acquisition module 730, and a second input module 740, wherein :
  • the data reading module 710 is used to read the operating parameters of the ventilator system.
  • the first input module 720 is configured to input the operating parameters of the ventilator system into a preset trained neural network model, and the preset trained neural network model is generated based on training of operating parameters of the historical system of the ventilator.
  • the data obtaining module 730 is used to obtain the current signal output by the preset trained neural network model according to the operating parameters of the ventilator system.
  • the second input module 740 is used to input a current signal to the proportional valve in the ventilator, and the current signal is used to control the flow of the proportional valve.
  • the ventilator proportional valve flow control device further includes: a training module 750 for initializing the connection weights and thresholds of each layer of the initial Elman neural network model; obtaining historical system operating parameters of the ventilator; according to the ventilator Historical system operating parameters, using the Levenberg-Marquardt algorithm to adjust the weights and thresholds of the initial Elman neural network model until the optimal weights and thresholds are determined.
  • the training module 750 is also used to establish an initial Elman neural network model; according to the Levenberg-Marquardt algorithm and ventilator historical system operating parameters, the parameters of the initial Elman neural network model are trained to obtain the trained Elman neural network model .
  • the data reading module 710 is further used to read the pressure difference at the current moment, the flow sensor value of the previous frame and the target flow value of the next frame.
  • Each module in the above-mentioned ventilator proportional valve flow control device may be implemented in whole or in part by software, hardware, or a combination thereof.
  • the above modules may be embedded in the hardware or independent of the processor in the computer device, or may be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
  • a computer device is provided.
  • the computer device may be a terminal, and its internal structure may be as shown in FIG. 8.
  • the computer equipment includes a processor, a memory, a network interface, a display screen, and an input device connected through a system bus.
  • the processor of the computer device is used to provide computing and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium stores an operating system and computer programs.
  • the internal memory provides an environment for the operating system and computer programs in the non-volatile storage medium.
  • the network interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, a flow control method of the proportional valve of the ventilator is realized.
  • the display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen
  • the input device of the computer device may be a touch layer covered on the display screen, or may be a button, a trackball or a touchpad provided on the computer device housing , Can also be an external keyboard, touchpad or mouse.
  • FIG. 8 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer equipment to which the solution of the present application is applied.
  • the specific computer equipment may It includes more or fewer components than shown in the figure, or some components are combined, or have a different component arrangement.
  • a computer device which includes a memory and a processor.
  • a computer program is stored in the memory.
  • the processor executes the computer program, the following steps are realized: reading the operating parameters of the ventilator system; operating the ventilator system
  • the parameter input is preset to the trained neural network model.
  • the preset trained neural network model is generated based on the training parameters of the ventilator historical system; obtain the current signal output from the preset trained neural network model according to the ventilator system operating parameters;
  • the current signal is input to the proportional valve in the ventilator, and the current signal is used to control the flow of the proportional valve.
  • the processor also implements the following steps when executing the computer program: establishing an initial Elman neural network model; training the parameters of the initial Elman neural network model according to the preset Levenberg-Marquardt algorithm and the operating parameters of the ventilator historical system to obtain training After the Elman neural network model.
  • the processor also implements the following steps when executing the computer program: Initializing the initial connection weights and thresholds of each layer of the initial Elman neural network model; obtaining the historical system operating parameters of the ventilator; using the historical system operating parameters of the ventilator, using The Levenberg-Marquardt algorithm adjusts the weights and thresholds of the initial Elman neural network model until the optimal weights and thresholds are determined.
  • the processor also implements the following steps when executing the computer program: reading the pressure difference at the current time, the value of the flow sensor in the previous frame, and the target flow value in the next frame.
  • a computer-readable storage medium on which a computer program is stored.
  • the following steps are realized: reading the ventilator system operating parameters; inputting the ventilator system operating parameters into a pre- Set the trained neural network model, the preset trained neural network model is generated based on the training parameters of the ventilator historical system; obtain the current signal output by the preset trained neural network model according to the ventilator system operating parameters; input the current signal To the proportional valve in the ventilator, the current signal is used to control the flow of the proportional valve.
  • the following steps are also implemented: the initial Elman neural network model is established; the parameters of the initial Elman neural network model are trained according to the preset Levenberg-Marquardt algorithm and ventilator historical system operating parameters Elman neural network model after training.
  • the following steps are also implemented: initializing the connection weights and thresholds of each layer of the initial Elman neural network model; obtaining the historical system operating parameters of the ventilator; according to the historical system operating parameters of the ventilator, The Levenberg-Marquardt algorithm is used to adjust the weights and thresholds of the initial Elman neural network model until the optimal weights and thresholds are determined.
  • the computer program executes the computer program's the weight increment ⁇ w.
  • Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory can include random access memory (RAM) or external cache memory.
  • RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
  • SRAM static RAM
  • DRAM dynamic RAM
  • SDRAM synchronous DRAM
  • DDRSDRAM double data rate SDRAM
  • ESDRAM enhanced SDRAM
  • SLDRAM synchronous chain (Synchlink) DRAM
  • SLDRAM synchronous chain (Synchlink) DRAM
  • Rambus direct RAM
  • DRAM direct memory bus dynamic RAM
  • RDRAM memory bus dynamic RAM

Abstract

一种呼吸机比例阀（50）流量控制方法、装置、计算机设备，读取呼吸机系统运行参数（S200），将呼吸机系统运行参数输入基于呼吸机历史系统运行参数训练生成的预设神经网络模型（S400），获取预设已训练的神经网络模型根据呼吸机系统运行参数输出的电流信号（S600），将电流信号输入至呼吸机中比例阀（50），来控制比例阀的流量（S800）。通过将当前呼吸机系统运行参数输入至基于呼吸机历史系统运行参数训练生成的预设神经网络模型中，能够得到一个合理的电流信号，根据这个合理的电流信号能够有效地控制呼吸机比例阀（50）的流量，使比例阀（50）在流量控制过程中，能够适应外部干扰因素变化带来的影响。

Description

呼吸机比例阀流量控制方法、装置、计算机设备 技术领域
本申请涉及呼吸机技术领域，特别是涉及一种呼吸机比例阀流量控制方法、装置、计算机设备和存储介质。
背景技术
呼吸机是一种能够起到预防和治疗呼吸衰竭，减少并发症，挽救及延长病人生命的至关重要的医疗设备，作为一项能人工替代自主通气功能的有效手段，已普遍用于各种原因所致的呼吸衰竭、大手术期间的麻醉呼吸管理、呼吸支持治疗和急救复苏中，在现代医学领域内占有十分重要的位置。
日常生活中，呼吸机比例阀多采用电磁比例阀，电磁比例阀是指采用比例电磁铁作为电气一机械转换元件的比例阀，比例电磁铁将输入的电流信号转换成力、位移机械信号输出.进而控制压力、流量及方向等参数。
目前，现有的比例阀的流量控制方法都是离线生成表格，在流量控制过程中很难适应外部干扰因素变化带来的影响，外部干扰因素包括压差变化、电流干扰、传感器噪声等因素。
发明内容
基于此，有必要针对比例阀很难适应难适应外部干扰因素变化带来的影响的问题，提供一种具有一定自适应调整能力的呼吸机比例阀流量控制方法、装置、计算机设备和存储介质。
一种呼吸机比例阀流量控制方法，包括：
读取呼吸机系统运行参数；
将呼吸机系统运行参数输入预设已训练的神经网络模型，预设已训练的神经网络模型基于呼吸机历史系统运行参数训练生成；
获取预设已训练的神经网络模型根据呼吸机系统运行参数输出的电流信号；
将电流信号输入至呼吸机中比例阀，电流信号用于控制比例阀的流量。
在其中一个实施例中，预设已训练的神经网络模型为Elman(艾尔曼)神经网络模型，将呼吸机系统运行参数输入预设已训练的神经网络模型之前还包括：
建立初始Elman神经网络模型；
根据预设Levenberg-Marquardt算法和呼吸机历史系统运行参数，训练初始Elman神经网络模型的参数，得到训练后的Elman神经网络模型。
在其中一个实施例中，根据Levenberg-Marquardt(列文伯格-马夸尔特)算法和呼吸机历史系统运行参数，训练初始Elman神经网络模型包括：
初始化初始Elman神经网络模型各层连接权值和各层阈值；
获取呼吸机历史系统运行参数；
根据呼吸机历史系统运行参数，利用Levenberg-Marquardt算法调整初始Elman神经网络模型的权值和阈值，直至确定最优的权值和阈值。
在其中一个实施例中，根据Levenberg-Marquardt算法调整初始Elman神经网络模型的权值和阈值，直至确定最佳的权值和阈值包括：
给出训练误差允许值ε，常数μ 0和β(0＜β＜1)，初始化权值和阈值组成的向量，令k＝0,μ＝μ 0；
计算初始Elman神经网络模型的输出值及误差指标函数E(w k)；
计算Jacobian矩阵J(w k)；
计算权值增量Δw；
若E(w k)＜ε,则结束训练；
若E(w k)≥ε，以w k+1＝w k+Δw为新的权值和阈值向量，计算误差指标函数E(w k)，若E(w k+1)＜E(w k)，则令k＝k+1，μ＝μβ，返回计算初始Elman神经网络模型的输出值及误差指标函数E(w k)，否则令μ＝μ/β，返回计算权值增量Δw的步骤。
在其中一个实施例中，读取呼吸机系统运行参数包括当前时刻压差、前帧流量传感器值以及下帧目标流量值。
在其中一个实施例中，预设Elman神经网络模型包括4个输入层神经元、7个隐含层神经元和状态层神经元以及1个输出层神经元。
一种呼吸机比例阀流量控制装置，包括：
数据读取模块，用于读取呼吸机系统运行参数；
第一输入模块，用于将呼吸机系统运行参数输入预设已训练的神经网络模型，预设已训练的神经网络模型基于呼吸机历史系统运行参数训练生成；
数据获取模块，用于获取预设已训练的神经网络模型根据呼吸机系统运行参数输出的电流信号；
第二输入模块，用于将电流信号输入至呼吸机中比例阀，电流信号用于控制比例阀的流量。
在其中一个实施例中，呼吸机比例阀流量控制装置还包括：
训练模块，用于初始化初始Elman神经网络模型各层连接权值和各层阈值；获取呼吸机历史系统运行参数；根据呼吸机历史系统运行参数，利用Levenberg-Marquardt算法调整初始Elman神经网络模型的权值和阈值，直至确定最优的权值和阈值。
一种计算机设备，包括存储器和处理器，存储器存储有计算机程序，处理器执行计算机程序时实现以下步骤：
读取呼吸机系统运行参数；
将呼吸机系统运行参数输入预设已训练的神经网络模型，预设已训练的神经网络模型基于呼吸机历史系统运行参数训练生成；
获取预设已训练的神经网络模型根据呼吸机系统运行参数输出的电流信号；
将电流信号输入至呼吸机中比例阀，电流信号用于控制比例阀的流量。
一种计算机可读存储介质，其上存储有计算机程序，计算机程序被处理器执行时实现以下步骤：
读取呼吸机系统运行参数；
将呼吸机系统运行参数输入预设已训练的神经网络模型，预设已训练的神经网络模型基于呼吸机历史系统运行参数训练生成；
获取预设已训练的神经网络模型根据呼吸机系统运行参数输出的电流信号；
将电流信号输入至呼吸机中比例阀，电流信号用于控制比例阀的流量。
上述呼吸机比例阀流量控制方法、装置、计算机设备和存储介质，读取呼吸机系统运行参数，将呼吸机系统运行参数输入基于呼吸机历史系统运行参数训练生成的预设神经网络模型，获取预设已训练的神经网络模型根据呼吸机系统运行参数输出的电流信号，将电流信号输入至呼吸机中比例阀，来控制比例阀的流量。通过将当前呼吸机系统运行参数输入至基于呼吸机历史系统运行参数训练生成的预设神经网络模型中，能够得到一个合理的电流信号，根据这个合理的电流信号能够有效地控制呼吸机比例阀的流量，使比 例阀在流量控制过程中，能够适应外部干扰因素变化带来的影响。
附图说明
图1为一个实施例中呼吸机比例阀流量控制的预测框图；
图2为一个实施例中呼吸机比例阀流量控制方法的流程图；
图3为一个实施例中呼吸机比例阀流量控制方法的流程图；
图4为一个实施例中Elman神经网络的结构示意图；
图5为采用压差-流量二维表查询获得的低压阀压差-流量-电流DA三维曲面拟合效果示意图；
图6为采用Elman神经网络模型拟合的低压阀压差-流量-电流DA三维曲面拟合效果示意图；
图7为一个实施例中呼吸机比例阀流量控制装置的结构示意图；
图8为一个实施例中计算机设备的内部结构示意图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
本申请提供的呼吸机比例阀流量控制方法，图1为呼吸机比例阀流量控制的预测框图，如图1所示，呼吸机包括压差传感器和流量传感器、比例阀，呼吸机中比例阀为低压电磁比例阀，电磁比例阀的工作原理是采用比例电磁铁控制,使输出的压力或流量与输入的电流成正比，所以可用改变输入电信号的方法对压力、流量进行连续控制，本申请中低压电磁比例阀接收神经网络模型输出的电流信号对呼吸机流量进行连续控制。压差传感器10用来测量两个压力之间差值的传感器，通常用于测量某一设备或部件前后两端的压差，流量传感器20是用于测定每一时刻吸入呼吸机发动机的空气流量，本申请中流量传感器20主要输出前帧流量传感器值、当前帧传感器值，流量指令30即下帧目标流量值，本方法是将当前压差、前帧流量传感器值、下帧目标流量值等多个变量输入到一个经过训练的多输入动态回归神经网络即Elman神经网络模型对产生下帧目标流量值需要的电流信号进行动态拟合预测，并将预测的电流信号输入到呼吸机比例阀中对呼吸机比例阀的流量进行控制。
其中，在一个实施例中，如图2所示，提供了一种呼吸机比例阀流量控制方法，以该方法应用于呼吸机微处理器为例进行说明，包括以下步骤：
步骤S200，读取呼吸机系统运行参数。
其中，呼吸机系统运行参数是呼吸机在正常工作时生成的数据，其能够反映出呼吸机的运行状态，当呼吸机微处理器需要采集呼吸机系统运行参数时，从内存控制器读取呼吸机各个部件实时采集的运行数据即可。具体来说，呼吸机运行参数可以包括呼气压、吸气压、压差传感器值、流量传感器值以及吸气流率等参数。在实际应用中，呼吸机处理器读取的呼吸机系统运行参数可以包括当前压差、当前帧流量传感器值、前帧流量传感器值、下帧目标流量值、当前电流信号等参数，其中，当前压差由压差传感器实时采集，当前帧流量传感器值、前帧流量传感器值由流量传感器采集，下帧目标流量值为预先设定的值，这些参数用于输出理想的电流信号。
步骤S400，将呼吸机系统运行参数输入预设已训练的神经网络模型，预设已训练的神经网络模型基于呼吸机历史系统运行参数训练生成。
微处理器将读取到的当前压差、当前帧流量传感器值、前帧流量传感器值、下帧目标流量值输入到预设的神经网络模型中，预设的神经网络模型基于呼吸机历史系统运行参数训练生成，该预设已训练的神经网络模型是经过训练的神经网络模型，用于输出合理的控制流量的电流信号，消除呼吸机中比例阀在控制流量过程中受噪声、气流速度、压差变化等因素的干扰。呼吸机历史系统运行参数包括压差值、各个时刻的流量值以及电流信号。
步骤S600，获取预设已训练的神经网络模型根据呼吸机系统运行参数输出的电流信号。
其中，该电流信号实际为经过DA(Digital Analog，数字模拟)转换器数模转换后的电流模拟信号，即电流DA值，模拟信号分布于自然界的各个角落，如气温的变化，而数字信号是人为的抽象出来的在幅 度取值上不连续的信号。电学上的模拟信号主要是指幅度和相位都连续的电信号，此信号可以被模拟电路进行各种运算，如放大，相加乘等。电流DA值用于连续控制呼吸机比例阀的流量。
步骤S800，将电流信号输入至呼吸机中比例阀，电流信号用于控制比例阀的流量。
微处理器将预设神经网络输出的电流信号也就是电流DA值输入至呼吸机比例阀，比例阀接收该电流DA值，利用该电流DA值连续控制呼吸机流量。具体的，比例阀通常由阀芯和控制线圈构成，线圈通过电流，控制电流的大小来改变阀芯的位置，即控制比例阀的开度，从而控制流量。
如图3所示，在其中一个实施例中，预设已训练的神经网络模型为Elman神经网络模型，读取呼吸机系统运行参数之前还包括：S100,建立初始Elman神经网络模型；S120,根据Levenberg-Marquardt算法和呼吸机历史系统运行参数，训练初始Elman神经网络模型的参数，得到训练后的Elman神经网络模型。
现有比例阀在流量控制过程中，容易受到压差变化、气流速度、流量跨度、噪声干扰等因素影响，为了让比例阀的流量控制算法和模块能适应外部因素影响，要求控制算法必须具有一定的自适应调整能力，故本申请选用基于Levenberg-Marquardt算法训练得到的Elman神经网络模型，控制比例阀流量的输出。本实施例中，选用的Elman神经网络是一种典型的局部回归网络，属于反馈神经网络，与前向神经网络非常相似，具有更强的计算能力，其突出优点是具有很强的优化计算和联想记忆功能。基本的Elman神经网络由输入层、隐含层、状态层和输出层组成。Elman神经网络在结构上与BP(back propagation，反向传播)网络相比，多了一个状态层，用于构成局部反馈，其状态层的传输函数为线性函数，但多了一个延迟单元，所以状态层可以记忆过去的状态，并且在下一时刻与网络的输入一起作为隐含层的输入，使网络具有动态记忆功能。该初始Elman神经网络模型则是基于Elman神经网络构建的神经网络模型，构建过程包括：采集呼吸机历史系统运行参数作为样本集，对该样本集进行归一化处理，该神经网络模型的训练方法为Levenberg-Marquardt算法，Levenberg-Marquardt算法是使用最广泛的非线性最小二乘算法，中文为列文伯格-马夸尔特法。它是利用梯度求最大(小)值的算法，形象的说，属于“爬山”法的一种。它同时具有梯度法和牛顿法的优点。当λ很小时，步长等于牛顿法步长，当λ很大时，步长约等于梯度下降法的步长。Levenberg-Marquardt算法是最优化算法中的一种。最优化是寻找使得函数值最小的参数向量。它的应用领域非常广泛，如：经济学、管理优化、网络分析、最优设计、机械或电子设计等等。可以理解的是，Levenberg-Marquardt算法还可以是梯度下降法(Gradient descent)、牛顿算法(Newton’s method)、柯西-牛顿法(Quasi-Newton method)以及其他训练算法，在此不做限制。基于Elamn神经网络的特殊结构，通过状态层也就是状态层的延迟与存储，隐含层地输出可以自动联接到隐含层的输入，基于这种自联方式，神经网络对历史数据的敏感性显著加强，并且其内部设有反馈网络，提高网络本身处理动态信息的能力，更好地实现动态建模；使用Levenberg-Marquardt算法训练初始Elman神经网络模型是为了使误差函数收敛速度更快，精度更高。
如图3所示，在其中一个实施例中，根据Levenberg-Marquardt算法和呼吸机历史系统运行参数，训练初始Elman神经网络模型的参数包括：S122,初始化初始Elman神经网络模型各层连接权值和各层阈值；S124,获取呼吸机历史系统运行参数；S126,根据呼吸机历史系统运行参数，利用Levenberg-Marquardt算法调整初始Elman神经网络模型的权值和阈值，直至确定最优的权值和阈值。
其中，我们将微处理器采集到的呼吸机历史系统运行参数作为原始数据，并提取原始数据中的当前时刻压差、前帧流量传感器值、当前帧流量传感器值和电流DA值作为主成分分析，并对其进行标准化处理，构建成新的特征向量，并将得到的特征向量进行归一化处理，得到新的训练样本集。具体的，在各种压差范围条件下逐步改变电流DA值，采集记录压差传感器值、流量传感器值，得到当前时刻k的压差DPress、前帧流量传感器值PreFlow、当前帧流量传感器值和电流DA值训练样本值[DPress(k),PreFlow(k),DFlow(k)；DA(k)]，其中DA(k)为第k个训练样本的电流目标值，k＝150*4096＝614400(压差范围1-150hPa，电流DA范围0-4095)。由于神经元的传输函数在[0,1]之间区别比较大，当输入大于1以后，传输函数值变化不大，其导数或斜率比较小，不利于反向传播算法的执行，而反向传播算法需要用到各个神经元传输函数的梯度信息，当神经元的输入太大时(比如大于1)，相应的该点自变量梯度值就过小，导致无法顺利实现权值和阈值的调整，故本实施例中，初始Elman神经网络模型的初始权值采用输入归一化，将初始权值和阈值归一化到0到1之间，当然，在其他实施例中，初始权值和阈值可归一化到-1到1之间。该神经网络包括4个输入神经元，第1个输入神经元输入恒定为1，与该神经元连接的各个隐含层神经元 到该神经元的权值相当于偏置；7个隐含层神经元，其中第1个隐含层神经元输入恒定为1，相当于隐含层到输出层偏置，没有记忆功能，无局部反馈，其余6个隐含层有局部反馈，对应6个状态层神经元，其中，状态层亦称作承接层。每个状态层神经元记忆前帧时刻对应隐含层神经元的状态，并通过权值反馈连接输入到各个隐含层神经元。将训练样本值DPress(k),PreFlow(k),DFlow(k)，DA(k)输入Elman神经网络模型中，计算Elman神经网络模型输出的电流实际输出向量以及误差指标函数，通过Levenberg-Marquardt算法调整初始Elman神经网络模型的权值和阈值，直至确定该误差指标函数的值满足训练误差值时的最优的权值和阈值，得到训练后的Elman神经网络模型。
在其中一个实施例中，根据Levenberg-Marquardt算法调整初始Elman神经网络模型的权值和阈值，直至确定最佳的权值和阈值包括：给出训练误差允许值ε，常数μ 0和β(0＜β＜1)，初始化权值和阈值组成的向量，令k＝0,μ＝μ 0；计算初始Elman神经网络模型的输出值及误差指标函数E(w k)；计算Jacobian矩阵J(w k)；计算权值增量Δw；若E(w k)＜ε,则结束训练；以w k+1＝w k+Δw为新的权值和阈值向量，计算误差指标函数E(w k)，若E(w k+1)＜E(w k)，则令k＝k+1，μ＝μβ，返回计算初始Elman神经网络模型的输出值及误差指标函数E(w k)，否则令μ＝μ/β，返回计算权值增量Δw的步骤。
具体过程包括：设误差指标函数E(w k)为：
Figure PCTCN2019072605-appb-000001
式中，Y i为电流期望输出向量，Y i′为电流实际输出向量，p为样本数目，w为初始Elman神经网络模型的权值和阈值所组成的向量，e i(w)为误差；
设w k+1表示第k次迭代的权值和阈值所组成的向量，新的权值和阈值所组成的向量w k+1为w k+1＝w k+Δw，Δw为权值增量，权值增量Δw计算公式为：Δw＝[J T(W)J(w)+μI] -1J T(W)e(w)，式中，I为单位矩阵，μ为用户定义的学习率，J(w)为Jacobian矩阵即雅各布矩阵，定义雅各布矩阵由误差项对参数的偏导数组成；具体步骤如下：
1)给出训练误差允许值ε，常数μ 0和β(0＜β＜1)，初始化权值和阈值组成的向量，令k＝0,μ＝μ 0；
2)计算初始Elman神经网络模型的输出值及误差指标函数E(w k)；
3)计算Jacobian矩阵J(w k)；
4)计算权值增量Δw；
5)若E(w k)＜ε,则返回步骤7)；
6)以w k+1＝w k+Δw为新的权值和阈值向量，计算误差指标函数E(w k)，若E(w k+1)＜E(w k)， 则令k＝k+1，μ＝μβ，返回步骤2)，否则令μ＝μ/β，返回步骤4)；
7)算法结束。
采用Levenberg-Marquardt算法训练Elman神经网络模型的过程实际是一个寻找最优参数的过程，也就是对该网络模型的权值和阈值进行学习和调整，使该网络模型实现给定的输入/输出映射关系，完成对系统的辨识。本实施例中，预设的训练误差允许值为0.001，通过上述Levenberg-Marquardt算法，不断调整该Elman神经网络模型的权值和阈值，直到确定误差指标函数值小于训练误差值时的Elman神经网络模型的权值和阈值，也就是当该误差指标函数的的值小于0.001时，即视为训练结束，通过训练后的神经网络模型确定的新的权值和阈值能够对新的输出值进行预测。在其他实施例中，训练误差允许值可以为0.01、0.0099以及其他误差允许值，在此不做限制。本实施中，使用Levenberg-Marquardt算法实时更新训练数据，得到适合当前时刻的最优权值和阈值，误差函数收敛速度更快，预测精度更高。
在其中一个实施例中，根据预设Levenberg-Marquardt算法和呼吸机历史系统运行参数，训练初始Elman神经网络模型的参数包括：计算损失值、梯度和近似海森矩阵，然后确定衰减参数和衰减系数。
由于Levenberg-Marquardt算法主要针对平方和误差类的损失函数。Levenberg-Marquardt算法又称为衰减的最小平方法，其是针对损失函数是平方和误差的形式，不需要准确计算海森矩阵，需要用到梯度向量和雅各布矩阵。
具体的，定义损失函数的雅各布矩阵由误差项对参数的偏导数组成，J i,jf(w)＝de i/dw j,(i＝1,…,m&j＝1,…,n)，其中，m是训练集中的样本个数，n是神经网络的参数个数，雅各布矩阵的规模是m·n。定义损失函数的梯度向量为
Figure PCTCN2019072605-appb-000002
其中e是所有误差项组成的向量。
用上述雅各布矩阵和梯度向量的表达式来估计计算海森矩阵，Hf≈2J T(J+λI)，其中λ是衰减因子，以确保海森矩阵是正的，I是单位矩阵。此算法的参数更新公式如下：
Figure PCTCN2019072605-appb-000003
若衰减因子λ设为0，相当于是牛顿法。若λ设置的非常大，这就相当于是学习率很小的梯度下降法。参数λ的初始值非常大，因此前几步更新是沿着梯度下降方向的。如果某一步迭代更新失败，则λ扩大一些。否则，λ随着损失值的减小而减小，Levenberg-Marquardt接近牛顿法，这个过程可以加快收敛的速度。
在其中一个实施例中，呼吸机系统运行参数包括当前时刻压差、前帧流量传感器值以及下帧目标流量值。
其中，当前时刻压差由压差传感器采集，前帧流量传感器值由流量传感器采集，下帧目标流量值为预先设定的值，其预先存储于呼吸机的存储器中。比例阀、特别是低压阀在流量控制过程中，当前流量控制参数和方法不仅与下帧目标流量有关，而且也将受到当前阀开度和前后帧输出流量和目标流量变化跨度的影响，大的流量跨度和小的流量跨度势必要求的控制参数不一样。故本申请从原始数据中提取当前时刻压差、前帧流量传感器值以及下帧目标流量值作为主成分分析，并对读取的当前时刻压差、前帧流量传感器值以及下帧目标流量值进行标准化处理，构建成新的特征向量，并将其输入到预设的Elman神经网络模型中，对产生下帧目标流量值需要的电流信号进行动态拟合预测，以期Elman神经网络模型根据当前时刻压差、前帧流量传感器值以及下帧目标流量值输出合理的电流信号。如图5、图6所示，图5为采用压差-流量二维表查询获得的低压阀压差-流量-电流DA三维曲面拟合效果，图6为采用Elman神经网络模型拟合的低压阀压差-流量-电流DA三维曲面拟合效果，通过图5和图6，我们能够看出通过将当前时刻压差、前帧流量传感器值以及下帧目标流量值输入训练后的Elman神经网络模型使对产生下帧目标流量值需要的电流信号进行动态预测的方法，比例阀的控制曲面更加光滑，也意味着拟合精度更高，输出的电流信号更 为合理，更能使比例阀在流量控制过程中更能适应压差变化、流量跨度、噪声干扰等外界因素的带来的影响。
在其中一个实施例中，预设Elman神经网络模型包括4个输入层神经元、7个隐含层神经元和状态层神经元以及1个输出层神经元。
如图4所示，本实施例中，采用的初始Elman神经网络模型的结构尺寸是4-7-1，即4个输入层神经元、7个隐含层神经元以及1个输出层神经元，状态层神经元数目与隐含层神经元数目相等，即7个状态层神经元。其中，7个隐含层神经元，第1个隐含层神经元没有记忆功能，无局部反馈，其余6个隐含层有局部反馈，对应6个状态层神经元。每个状态层神经元记忆前帧时刻对应隐含层神经元的状态，并通过权值反馈连接输入到各个隐含层神经元。在实际应用中，神经网络结构默认只用一个隐含层，如果用多个隐含层，则每个隐含层的神经元数目一样，可以理解的是，Elman神经网络结构尺寸还可以是其他结构，在此不做限制，本实施例采用7个隐含层神经元，其分类效果更好。
上述呼吸机比例阀流量控制方法、装置、计算机设备和存储介质，读取呼吸机系统运行参数，将呼吸机系统运行参数输入基于呼吸机历史系统运行参数训练生成的预设神经网络模型，获取预设已训练的神经网络模型根据呼吸机系统运行参数输出的电流信号，将电流信号输入至呼吸机中比例阀，来控制比例阀的流量。通过将当前呼吸机系统运行参数输入至基于呼吸机历史系统运行参数训练生成的预设神经网络模型中，能够得到一个合理的电流信号，根据这个合理的电流信号能够有效地控制呼吸机比例阀的流量，使比例阀在流量控制过程中，能够适应外部干扰因素变化带来的影响。
应该理解的是，虽然图2-3的流程图中的各个步骤按照箭头的指示依次显示，但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，这些步骤可以以其它的顺序执行。而且，图2-3中的至少一部分步骤可以包括多个子步骤或者多个阶段，这些子步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，这些子步骤或者阶段的执行顺序也不必然是依次进行，而是可以与其它步骤或者其它步骤的子步骤或者阶段的至少一部分轮流或者交替地执行。
在一个实施例中，如图7所示，提供了一种呼吸机比例阀流量控制装置，包括：数据读取模块710、第一输入模块720、数据获取模块730和第二输入模块740，其中：
数据读取模块710，用于读取呼吸机系统运行参数。
第一输入模块720，用于将呼吸机系统运行参数输入预设已训练的神经网络模型，预设已训练的神经网络模型基于呼吸机历史系统运行参数训练生成。
数据获取模块730，用于获取预设已训练的神经网络模型根据呼吸机系统运行参数输出的电流信号。
第二输入模块740，用于将电流信号输入至呼吸机中比例阀，电流信号用于控制比例阀的流量。
在其中一个实施例中，呼吸机比例阀流量控制装置还包括：训练模块750，用于初始化初始Elman神经网络模型各层连接权值和各层阈值；获取呼吸机历史系统运行参数；根据呼吸机历史系统运行参数，利用Levenberg-Marquardt算法调整初始Elman神经网络模型的权值和阈值，直至确定最优的权值和阈值。
在其中一个实施例中，训练模块750还用于建立初始Elman神经网络模型；根据Levenberg-Marquardt算法和呼吸机历史系统运行参数，训练初始Elman神经网络模型的参数，得到训练后的Elman神经网络模型。
在其中一个实施例中，训练模块750还用于给出训练误差允许值ε，常数μ 0和β(0＜β＜1)，初始化权值和阈值组成的向量，令k＝0,μ＝μ 0；计算初始Elman神经网络模型的输出值及误差指标函数E(w k)；计算Jacobian矩阵J(w k)；计算权值增量Δw；若E(w k)＜ε,则结束训练；以w k+1＝w k+Δw 为新的权值和阈值向量，计算误差指标函数E(w k)，若E(w k+1)＜E(w k)，则令k＝k+1，μ＝μβ，返回计算初始Elman神经网络模型的输出值及误差指标函数E(w k)，否则令μ＝μ/β，返回计算权值增量Δw的步骤。
在其中一个实施例中，数据读取模块710还用于读取当前时刻压差、前帧流量传感器值以及下帧目标流量值。
关于呼吸机比例阀流量控制装置的具体限定可以参见上文中对于呼吸机比例阀流量控制方法的限定，在此不再赘述。上述呼吸机比例阀流量控制装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。
在一个实施例中，提供了一种计算机设备，该计算机设备可以是终端，其内部结构图可以如图8所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口、显示屏和输入装置。其中，该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统和计算机程序。该内存储器为非易失性存储介质中的操作系统和计算机程序的运行提供环境。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机程序被处理器执行时以实现一种呼吸机比例阀流量控制方法。该计算机设备的显示屏可以是液晶显示屏或者电子墨水显示屏，该计算机设备的输入装置可以是显示屏上覆盖的触摸层，也可以是计算机设备外壳上设置的按键、轨迹球或触控板，还可以是外接的键盘、触控板或鼠标等。
本领域技术人员可以理解，图8中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
在一个实施例中，提供了一种计算机设备，包括存储器和处理器，存储器中存储有计算机程序，该处理器执行计算机程序时实现以下步骤：读取呼吸机系统运行参数；将呼吸机系统运行参数输入预设已训练的神经网络模型，预设已训练的神经网络模型基于呼吸机历史系统运行参数训练生成；获取预设已训练的神经网络模型根据呼吸机系统运行参数输出的电流信号；将电流信号输入至呼吸机中比例阀，电流信号用于控制比例阀的流量。
在一个实施例中，处理器执行计算机程序时还实现以下步骤：建立初始Elman神经网络模型；根据预设Levenberg-Marquardt算法和呼吸机历史系统运行参数，训练初始Elman神经网络模型的参数，得到训练后的Elman神经网络模型。
在一个实施例中，处理器执行计算机程序时还实现以下步骤：初始化初始Elman神经网络模型各层连接权值和各层阈值；获取呼吸机历史系统运行参数；根据呼吸机历史系统运行参数，利用Levenberg-Marquardt算法调整初始Elman神经网络模型的权值和阈值，直至确定最优的权值和阈值。
在一个实施例中，处理器执行计算机程序时还实现以下步骤：给出训练误差允许值ε，常数μ 0和β(0＜β＜1)，初始化权值和阈值组成的向量，令k＝0,μ＝μ 0；计算初始Elman神经网络模型的输出值及误差指标函数E(w k)；计算Jacobian矩阵J(w k)；计算权值增量Δw；若E(w k)＜ε,则结束训练；以w k+1＝w k+Δw为新的权值和阈值向量，计算误差指标函数E(w k)，若E(w k+1)＜E(w k)，则令k＝k+1，μ＝μβ，返回计算初始Elman神经网络模型的输出值及误差指标函数E(w k)，否则令μ＝μ/β，返回计算权值增量Δw的步骤。
在一个实施例中，处理器执行计算机程序时还实现以下步骤：读取当前时刻压差、前帧流量传感器值 以及下帧目标流量值。
在一个实施例中，提供了一种计算机可读存储介质，其上存储有计算机程序，计算机程序被处理器执行时实现以下步骤：读取呼吸机系统运行参数；将呼吸机系统运行参数输入预设已训练的神经网络模型，预设已训练的神经网络模型基于呼吸机历史系统运行参数训练生成；获取预设已训练的神经网络模型根据呼吸机系统运行参数输出的电流信号；将电流信号输入至呼吸机中比例阀，电流信号用于控制比例阀的流量。
在一个实施例中，计算机程序被处理器执行时还实现以下步骤：建立初始Elman神经网络模型；根据预设Levenberg-Marquardt算法和呼吸机历史系统运行参数，训练初始Elman神经网络模型的参数，得到训练后的Elman神经网络模型。
在一个实施例中，计算机程序被处理器执行时还实现以下步骤：初始化初始Elman神经网络模型各层连接权值和各层阈值；获取呼吸机历史系统运行参数；根据呼吸机历史系统运行参数，利用Levenberg-Marquardt算法调整初始Elman神经网络模型的权值和阈值，直至确定最优的权值和阈值。
在一个实施例中，计算机程序被处理器执行时还实现以下步骤：给出训练误差允许值ε，常数μ 0和β(0＜β＜1)，初始化权值和阈值组成的向量，令k＝0,μ＝μ 0；计算初始Elman神经网络模型的输出值及误差指标函数E(w k)；计算Jacobian矩阵J(w k)；计算权值增量Δw；若E(w k)＜ε,则结束训练；以w k+1＝w k+Δw为新的权值和阈值向量，计算误差指标函数E(w k)，若E(w k+1)＜E(w k)，则令k＝k+1，μ＝μβ，返回计算初始Elman神经网络模型的输出值及误差指标函数E(w k)，否则令μ＝μ/β，返回计算权值增量Δw的步骤。在一个实施例中，计算机程序被处理器执行时还实现以下步骤：读取当前时刻压差、前帧流量传感器值以及下帧目标流量值。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，的计算机程序可存储于一非易失性计算机可读取存储介质中，该计算机程序在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
以上实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上实施例仅表达了本申请的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对发明专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。

Claims (10)

  1. 一种呼吸机比例阀流量控制方法，其特征在于，所述方法包括：
    读取呼吸机系统运行参数；
    将所述呼吸机系统运行参数输入预设已训练的神经网络模型，所述预设已训练的神经网络模型基于所述呼吸机历史系统运行参数训练生成；
    获取所述预设已训练的神经网络模型根据所述呼吸机系统运行参数输出的电流信号；
    将所述电流信号输入至所述呼吸机中比例阀，所述电流信号用于控制所述比例阀的流量。
  2. 根据权利要求1所述的呼吸机比例阀流量控制方法，其特征在于，所述预设已训练的神经网络模型为Elman神经网络模型，所述将所述呼吸机系统运行参数输入预设已训练的神经网络模型之前还包括：
    建立初始Elman神经网络模型；
    根据预设Levenberg-Marquardt算法和所述呼吸机历史系统运行参数，训练所述初始Elman神经网络模型的参数，得到训练后的Elman神经网络模型。
  3. 根据权利要求2所述的呼吸机比例阀流量控制方法，其特征在于，所述根据预设Levenberg-Marquardt算法和所述呼吸机历史系统运行参数，训练所述初始Elman神经网络模型的参数包括：
    初始化所述初始Elman神经网络模型各层连接权值和各层阈值；
    获取所述呼吸机历史系统运行参数；
    根据所述呼吸机历史系统运行参数，利用Levenberg-Marquardt算法调整所述初始Elman神经网络模型的权值和阈值，直至确定最优的权值和阈值。
  4. 根据权利要求3所述的呼吸机比例阀流量控制方法，其特征在于，所述根据所述Levenberg-Marquardt算法调整所述初始Elman神经网络模型的权值 和阈值，直至确定最佳的权值和阈值包括：
    给出训练误差允许值ε，常数μ 0和β(0＜β＜1)，初始化权值和阈值组成的向量，令k＝0,μ＝μ 0；
    计算所述初始Elman神经网络模型的输出值及误差指标函数E(w k)；
    计算Jacobian矩阵J(w k)；
    计算权值增量Δw；
    若E(w k)＜ε,则结束训练；
    若E(w k)≥ε，以w k+1＝w k+Δw为新的权值和阈值向量，计算误差指标函数E(w k)，若E(w k+1)＜E(w k)，则令k＝k+1，μ＝μβ，返回所述计算所述初始Elman神经网络模型的输出值及所述误差指标函数E(w k)，否则令μ＝μ/β，返回所述计算权值增量Δw的步骤。
  5. 根据权利要求1所述的呼吸机比例阀流量控制方法，其特征在于，所述呼吸机系统运行参数包括当前时刻压差、前帧流量传感器值以及下帧目标流量值。
  6. 根据权利要求2至5中任一项所述的呼吸机比例阀流量控制方法，其特征在于，所述预设Elman神经网络模型包括4个输入层神经元、7个隐含层神经元和状态层神经元以及1个输出层神经元。
  7. 一种呼吸机比例阀流量控制装置，其特征在于，所述装置包括：
    数据读取模块，用于读取呼吸机系统运行参数；
    第一输入模块，用于将所述呼吸机系统运行参数输入预设已训练的神经网络模型，所述预设已训练的神经网络模型基于所述呼吸机历史系统运行参数训练生成；
    数据获取模块，用于获取所述预设已训练的神经网络模型根据所述呼吸机系统运行参数输出的电流信号；
    第二输入模块，用于将所述电流信号输入至所述呼吸机中比例阀，所述电流信号用于控制所述比例阀的流量。
  8. 根据权利要求7所述的呼吸机比例阀流量控制装置，其特征在于，所述装置还包括：
    训练模块，用于初始化所述初始Elman神经网络模型各层连接权值和各层阈值；获取所述呼吸机历史系统运行参数；根据所述呼吸机历史系统运行参数，利用Levenberg-Marquardt算法调整所述初始Elman神经网络模型的权值和阈值，直至确定最优的权值和阈值。
  9. 一种计算机设备，包括存储器和处理器，所述存储器存储有计算机程序，其特征在于，所述处理器执行所述计算机程序时实现权利要求1至6中任一项所述方法的步骤。
  10. 一种计算机可读存储介质，其上存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现权利要求1至6中任一项所述的方法的步骤。
PCT/CN2019/072605 2018-11-30 2019-01-22 呼吸机比例阀流量控制方法、装置、计算机设备 WO2020107694A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811459736.XA CN109498952B (zh) 2018-11-30 2018-11-30 呼吸机比例阀流量控制方法、装置、计算机设备
CN201811459736.X 2018-11-30

Publications (1)

Publication Number Publication Date
WO2020107694A1 true WO2020107694A1 (zh) 2020-06-04

Family

ID=65749835

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/072605 WO2020107694A1 (zh) 2018-11-30 2019-01-22 呼吸机比例阀流量控制方法、装置、计算机设备

Country Status (2)

Country Link
CN (1) CN109498952B (zh)
WO (1) WO2020107694A1 (zh)

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110051454A (zh) * 2019-05-28 2019-07-26 深圳市瑞沃德生命科技有限公司 呼吸机的通气控制方法
CN112128450B (zh) * 2019-06-24 2022-04-22 中国科学院理化技术研究所 一种风阀调节方法及风阀装置
CN111324154B (zh) * 2020-01-20 2022-06-24 深圳市科曼医疗设备有限公司 混合腔压力控制方法、呼吸机设备和计算机可读存储介质
DE102020203848A1 (de) * 2020-03-25 2021-09-30 Siemens Healthcare Gmbh Verfahren und Vorrichtung zur Ansteuerung eines medizinischen Geräts
CN112190800A (zh) * 2020-09-27 2021-01-08 湖南明康中锦医疗科技发展有限公司 一种呼吸支持设备的压力控制系统及方法
CN112190801A (zh) * 2020-09-27 2021-01-08 湖南明康中锦医疗科技发展有限公司 一种呼吸支持设备的氧气流量控制系统及方法
CN112791294A (zh) * 2020-12-31 2021-05-14 河北谊安奥美医疗设备有限公司 一种压电比例阀气路控制系统
CN113521460B (zh) * 2021-05-20 2024-02-23 深圳先进技术研究院 机械通气人机异步检测方法、装置及计算机可读存储介质
CN113835449B (zh) * 2021-11-29 2022-03-18 常州高凯电子有限公司 一种基于压力波动快速调节流量控制器阀门的控制方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5953713A (en) * 1995-10-26 1999-09-14 Board Of Regents, The University Of Texas System Method and apparatus for treatment of sleep disorder breathing employing artificial neural network
CN1468077A (zh) * 2000-10-02 2004-01-14 ����ҽѧ�ɷ����޹�˾ 使用神经网络的cpap的方法和装置
CN101610808A (zh) * 2007-02-12 2009-12-23 Ric投资有限责任公司 具有自动舒适度特征修改的压力支持系统和方法
CN107802993A (zh) * 2017-11-30 2018-03-16 北京航空航天大学 一种基于呼吸肌电信号反馈的肺康复训练系统
CN108187206A (zh) * 2018-01-02 2018-06-22 佳木斯大学 一种双腔支气管导管及控制方法
CN108619597A (zh) * 2017-03-17 2018-10-09 深圳市大雅医疗技术有限公司 一种呼吸机流道的压力补偿方法

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5598508A (en) * 1991-10-18 1997-01-28 Goldman; Julian M. Real-time waveform analysis using artificial neural networks
CN101337101B (zh) * 2007-07-06 2011-04-20 深圳迈瑞生物医疗电子股份有限公司 麻醉机和呼吸机的通气系统及压力监控方法
CN202263271U (zh) * 2011-09-30 2012-06-06 南京普澳医疗设备有限公司 一种呼吸压力模糊控制式呼吸机
CN105125215B (zh) * 2015-10-08 2018-05-11 湖南明康中锦医疗科技发展有限公司 基于神经网络的呼吸机状态分析方法和装置
CN108498930A (zh) * 2018-03-31 2018-09-07 湖南明康中锦医疗科技发展有限公司 呼吸辅助设备比例阀的流量控制方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5953713A (en) * 1995-10-26 1999-09-14 Board Of Regents, The University Of Texas System Method and apparatus for treatment of sleep disorder breathing employing artificial neural network
CN1468077A (zh) * 2000-10-02 2004-01-14 ����ҽѧ�ɷ����޹�˾ 使用神经网络的cpap的方法和装置
CN101610808A (zh) * 2007-02-12 2009-12-23 Ric投资有限责任公司 具有自动舒适度特征修改的压力支持系统和方法
CN108619597A (zh) * 2017-03-17 2018-10-09 深圳市大雅医疗技术有限公司 一种呼吸机流道的压力补偿方法
CN107802993A (zh) * 2017-11-30 2018-03-16 北京航空航天大学 一种基于呼吸肌电信号反馈的肺康复训练系统
CN108187206A (zh) * 2018-01-02 2018-06-22 佳木斯大学 一种双腔支气管导管及控制方法

Also Published As

Publication number Publication date
CN109498952A (zh) 2019-03-22
CN109498952B (zh) 2020-08-28

Similar Documents

Publication Publication Date Title
WO2020107694A1 (zh) 呼吸机比例阀流量控制方法、装置、计算机设备
Pan et al. Composite learning from adaptive backstepping neural network control
Ahn et al. Online tuning fuzzy PID controller using robust extended Kalman filter
Yeung et al. Sensitivity analysis for neural networks
CN108132599B (zh) 一种基于迭代反馈整定的ude控制系统设计方法
JP2019525367A (ja) 強化学習に基づくインスリンの評価
Franco et al. Energy shaping control with integral action for soft continuum manipulators
Davanipour et al. A solution for enhancement of transient performance in nonlinear adaptive control: Optimal adaptive reset based on barrier Lyapunov function
Balakrishnan et al. A deep analysis on optimization techniques for appropriate PID tuning to incline efficient artificial pancreas
Dora et al. A sequential learning algorithm for a spiking neural classifier
CN106033189B (zh) 飞行机器人位姿神经网络预测控制器
Kashihara et al. Adaptive predictive control of arterial blood pressure based on a neural network during acute hypotension
Hajizadeh et al. Ensuring stability and fidelity of recursively identified control-relevant models
Radhimeenakshi et al. Prediction of heart disease using neural network with back propagation
CN114185276A (zh) 一种基于多维泰勒网的非线性严格系统输出反馈控制方法
Wang et al. Active fault-tolerant control for the dual-valve hydraulic system with unknown dead-zone
Allen et al. Neural network approach to energy confinement scaling in tokamaks
Balasaheb et al. Arterial blood pressure regulation using a novel fractional order based sliding mode elephant herd controller
CN117555376B (zh) 基于大数据的居住建筑环境节能调节方法及系统
Constantinescu et al. Geometrical form recognition using ‘one-step-secant’algorithm in case of neural network
Kashihara Nonlinear system identification based on convolutional neural networks for multiple drug interactions
Nirmala et al. Unconstrained and constrained control of drug infusion system for patients under critical care using linear model predictive control strategy
Chang et al. A Smart Healthcare Design of Carryable Drip Frame with Fuzzy-PID Controllers
Pomprapa et al. Policy iteration algorithm for the control of oxygenation
Dai et al. A body simulator with delayed health state transition

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19889422

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19889422

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 19889422

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 16/03/2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19889422

Country of ref document: EP

Kind code of ref document: A1