Electro-hydraulic servo pump control method and system for controlling hydrogen compression based on fuzzy nerve

Abstract

The invention provides an electro-hydraulic servo pump control method and system for controlling hydrogen compression based on fuzzy nerve, and relates to the technical field of hydrogen compression, and the method comprises the steps that the current state of a hydraulic cylinder controlled by an electro-hydraulic servo pump is determined, and an input control variable is obtained; determining an output control variable, establishing a fuzzy control rule, and dividing a basic discourse domain; performing composite defuzzification, and obtaining a corrected weight coefficient through a feedforward neural network; the discrete rule amplitude correction factor and the weight coefficient correction are dynamically optimized and corrected; neural network training is established, a loss function evaluation model is utilized, and hydrogen compression control is achieved through a double-buffering mechanism. Fuzzy PID control is adopted, composite defuzzification is carried out according to the real-time state change of a system in combination with fuzzy logic, a membership function output by the fuzzy PID is divided according to regions, a control signal of a driver is synthesized, and hydrogen compression control is achieved; system output is adjusted according to different working conditions, power loss is reduced, and system control precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrogen compression, and particularly relates to an electro-hydraulic servo pump control method and system for hydrogen compression based on fuzzy neural control. Background Art

[0002] In traditional hydrogen compression valve control systems, there are large throttling losses and overflow losses. Due to the non-linear and time-varying characteristics of flexible loads, frequent speed regulation is required, resulting in increased throttling losses, large power losses, high energy consumption ratios, low efficiency, and serious heating. After long-term operation, the accumulated heat of the system itself leads to poor thermal balance. Usually, multiple hydraulic accessories need to be equipped, with many components and low integration. The hydraulic power components, actuators, control components, and hydraulic accessories are connected by pipelines. The pipeline layout is complex, the floor area occupied by the installation is large, the complex pipelines are not convenient for installation and debugging, and there are potential fault hazards such as pipeline leakage, which is not conducive to troubleshooting and maintenance of system faults. Key components such as valve cores and valve seats in hydraulic valves are prone to wear during long-term operation, resulting in problems such as increased valve leakage and decreased control accuracy, affecting the normal operation of the system. In addition, hydraulic valves have high requirements for the cleanliness of the system oil and are highly sensitive to impurity particles in the oil. Impurity particles brought into the system during long-term operation may block the throttling orifices of hydraulic valves and jam the valve cores, triggering system failures. It is necessary to equip precision filters or replace the oil regularly, resulting in high difficulty in later maintenance and difficult repair. Summary of the Invention

[0003] In order to solve the above-mentioned deficiencies of the prior art, the purpose of the present invention is to provide an electro-hydraulic servo pump control method and system for hydrogen compression based on fuzzy neural control, which can improve the system control accuracy and rapid response ability, shorten the response time of the control system, adopt multiple sensors to perceive the working conditions and environment and flexibly adjust the hydrogen compression control system, simultaneously monitor various key parameters in real time, comprehensively master the working state of the hydraulic system, and improve the stability and reliability of the control system; the integrated cooling structure in the electro-hydraulic servo pump control system makes heat exchange more efficient.

[0004] The present invention provides an electro-hydraulic servo pump control method for hydrogen compression based on fuzzy neural control, which includes the following steps: S1. Calculate the current state pressure of the hydraulic cylinder controlled by the electro-hydraulic servo pump and the hydrogen flexible load force , to obtain the force error and the rate of change of the force error , which are used as input control variables; S2. Determine the output control variable of the fuzzy PID control method for electro-hydraulic servo pump control, and output a control signal through a feedforward neural network based on an optimizer; establish fuzzy control rules, and divide the adjustment amount of the proportional coefficient , the adjustment amount of the integral coefficient , the adjustment amount of the differential coefficient The basic universe of discourse; S3. Use the centroid method and the weighted average method for composite defuzzification. Divide the membership function output by the fuzzy PID according to regions. For continuous regions, calculate the base value of the steady-state component of the control variable output using the centroid method and for discrete single-point regions, calculate the base value of the dynamic component output of the control variable using the weighted average method , where is the base value of the dynamic component output; is the correction factor for the discrete rule amplitude; is the adjustment vector of the PID coefficients; Organize the base value of the dynamic component output into the PID component form; Output the weight coefficient for dynamically adjusting the continuous region based on the neural network through the feedforward neural network based on the optimizer to obtain the corrected weight coefficient as: ; where is the weight coefficient for dynamically adjusting the continuous region based on the neural network; is the corrected weight coefficient; is the weight coefficient correction amount; S4. Dynamically optimize and correct the discrete rule amplitude correction factor in step S3 using a feedforward neural network and the weight coefficient correction amount S5. Establish a neural network training to evaluate the model using a loss function. Achieve asynchronous cooperation between the electro-hydraulic servo pump control system and the neural network optimization layer through a double-buffer mechanism. Use the optimizer to update the parameters in real time, and convert them into PWM signals through the servo driver to drive the servo motor to achieve hydrogen compression control.

[0005] Preferably, step S2 is specifically: S21: Determine that the input control variable and the output control variable of the fuzzy PID control algorithm are proportional fuzzy control variables, and output a control signal through the feedforward neural network based on the optimizer; S22: Establish fuzzy control rules, and the output variable format is ( , , ), representing the corresponding adjustment amounts of the proportional, integral, and differential coefficients; S23: Divide the adjustment amount of the proportional coefficient , the adjustment amount of the integral coefficient and the adjustment amount of the differential coefficient according to the measurement range of the acceleration sensor, the measurement range of the pressure sensor, and the actual hydrogen compression flexible load demand.

[0006] Preferably, the control signal output by the feedforward neural network based on the optimizer in step S21 is: ; where is the proportional fuzzy control variable; is the integral fuzzy control variable; is the differential fuzzy control variable; is the original proportional coefficient; is the original integral coefficient; is the original differential coefficient; is the steady-state component base value of the proportional coefficient adjustment amount; is the steady-state component base value of the integral coefficient adjustment amount; is the steady-state component base value of the differential coefficient adjustment amount; is the dynamic component base value of the proportional coefficient adjustment amount; is the dynamic component base value of the integral coefficient adjustment amount; is the dynamic component base value of the differential coefficient adjustment amount.

[0007] Preferably, step S3 is specifically: S31: Calculate the steady-state component base value of the control variable output using the centroid method in the continuous region ; S32: Calculate the dynamic component output base value of the control variable for suppressing sudden disturbances using the weighted average method at discrete single points, i.e., extreme points ; Arrange the dynamic component output base value in the form of PID components; S33: Dynamically adjust the weight coefficient in the continuous region through the neural network in the control signal output by the feedforward neural network based on the optimizer , to obtain the corrected weight coefficient .

[0008] Preferably, the steady-state component base value in step S31 is specifically: ; where is the steady-state component base value of the control variable output; is the steady-state component of the proportional coefficient adjustment amount; is the steady-state component of the integral coefficient adjustment amount; is the steady-state component of the differential coefficient adjustment amount; is the triangular membership function of the PID controller output by the proportional link in the continuous region [a, b], is the triangular membership function of the PID controller output by the integral link in the continuous region [a, b], is the triangular membership function of the output of the differential link of the PID controller within the continuous region [a, b]; [a, b] is the range of the action interval of the continuous membership function, is the lower limit of the action interval, is the upper limit of the action interval, which is divided by the fuzzy control rule in step S22.

[0009] Preferably, step S32 arranges the dynamic component output base value into the form of PID components, specifically: ; Among them, is the proportional coefficient adjustment amount; is the integral coefficient adjustment amount; is the differential coefficient adjustment amount; is the discrete rule amplitude correction factor.

[0010] Preferably, step S4 is specifically: S41: Set the input layer of the feedforward neural network to 4 nodes to obtain the four-dimensional input vector of the feedforward neural network ; S42: Determine the hidden layer of the feedforward neural network through Bayesian optimization. Based on minimizing the error of the validation set, use the ReLU activation function to filter the fluctuating part of the signal variance noise; S43: Set the output layer of the feedforward neural network to 2 nodes to obtain the output vector of the feedforward neural network .

[0011] Preferably, in step S5, the neural network training uses the loss function to evaluate the model, specifically: ; Among them, is the loss function of the neural network training; is the control error at time t; is the weight coefficient correction amount; is the regularization coefficient.

[0012] On the other hand, the present invention provides an electro-hydraulic servo pump control system based on fuzzy neural control for hydrogen compression, which includes: a power output module, a filter and oil replenishment module, a safety overflow module, a hot oil cooling module, a pressure and flow monitoring module, and an actuator and monitoring module; The power output module includes a servo motor and a bi-directional fixed displacement hydraulic pump. The servo motor is connected to the bi-directional fixed displacement hydraulic pump through a coupling and serves as a power source to provide the required pressure and flow rate. The filter and oil replenishment module includes a first check valve and a second check valve, which are used to compensate for external leakage and filter the oil. The safety overflow module includes a first safety overflow valve and a second safety overflow valve. When the pressure in the system exceeds 21 MPa during normal operation, the overflow valve opens to the working position, returning the excess oil to the accumulator and stabilizing the pressure within a safe range. The hot oil cooling module includes a low-pressure side hot oil shuttle valve, a high-pressure cooler, a back-pressure overflow valve, and a pressure measuring joint, which monitor the system overflow pressure and oil temperature, and take a part of the system flow rate to cool the motor. The pressure and flow rate monitoring module includes a flow meter, a first pressure relay, and a second pressure relay, which ensure the safe operation of the system by real-time detecting the pressure and flow rate of the system, and transmit the pressure and flow rate data to the controller after acquisition. The actuator and monitoring module includes a hydraulic cylinder, an acceleration sensor, and a displacement sensor to achieve hydrogen compression.

[0013] Preferably, the first oil inlet of the low-pressure side hot oil shuttle valve of the hot oil cooling module is connected to the high-pressure pipeline, the second oil inlet of the low-pressure side hot oil shuttle valve is connected to the low-pressure pipeline through the cooler, the oil inlet of the back-pressure overflow valve is connected to the oil port of the low-pressure side hot oil shuttle valve, the oil outlet of the back-pressure overflow valve is connected to the motor cooling oil passage, the pressure measuring joint is connected to the oil outlet of the back-pressure overflow valve. The low-pressure side hot oil shuttle valve extracts the oil in the low-pressure pipeline, passes through the motor cooling oil passage to cool the motor and reduce the motor temperature. At the same time, the cooler cools the system oil, controls the system temperature. After the heat balance of the motor, the hot oil merges with the leakage oil of the bi-directional fixed displacement hydraulic pump through the leakage confluence passage and enters the filter through the filter passage.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention adopts fuzzy PID control to adapt to flexible loads. The fuzzy PID control combines fuzzy logic based on the real-time state changes of the system, uses the centroid method and the weighted average method for composite defuzzification, divides the membership function output by the fuzzy PID according to regions, synthesizes the control signal of the driver, realizes hydrogen compression control, provides conditions for coping with complex working conditions changes, adjusts the system output according to different working conditions, reduces power loss, improves the system control accuracy, improves the rapid response ability of the system, and shortens the control system response time.

[0015] (2) The present invention adopts a feedforward neural network to dynamically optimize the fuzzy PID control parameters; realizes the asynchronous cooperation between the fuzzy PID control system and the neural network optimization layer through a double-buffer mechanism, and uses an optimizer to update the fuzzy PID control parameters in real time to realize the real-time control of hydrogen compression.

[0016] (3) The present invention adopts a multi-sensor sensing working condition and environmental adjustment system with a wide range of parameter coverage. It can simultaneously monitor various key parameters involved in different parts and different operating stages of the hydraulic system in real time, grasp the working status of the hydraulic system, and avoid the occurrence of monitoring blind spots. Multi-sensors accurately monitor various working conditions such as startup, normal operation, and overload, so that abnormal parameters in the system under various conditions can be collected, thereby improving the stability and reliability of the system.

[0017] (4) The present invention adopts an integrated cooling structure of a low-pressure side hot oil shuttle valve and a high-pressure cooler and a valve block to perform heat balance on the hydraulic system, eliminating the motor oil cooling pipeline and equipment, optimizing the heat exchange process, allowing the motor to obtain cooling medium more quickly, reducing unnecessary heat accumulation and transfer paths in the system, making the heat exchange process more efficient and quickly removing heat. The integrated structure integrates the relevant cooling parameters into the system monitoring, allowing for more intuitive monitoring of the cooling status, saving installation procedures, and reducing floor space. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of the electro-hydraulic servo pump control method of hydrogen compression based on fuzzy neural control of the present invention; Figure 2 It is a functional block diagram of the electro-hydraulic servo pump control based on fuzzy neural control of hydrogen compression of the present invention; Figure 3 The flowchart of the electro-hydraulic servo pump control of hydrogen compression based on fuzzy neural control of the present invention is as follows; Figure 4 This is a control flow chart of the electro-hydraulic servo pump control based on fuzzy neural control of hydrogen compression of the present invention; Figure 5 This is an experimental curve diagram of simulation according to a specific embodiment of the present invention.

[0019] Main reference numerals: 1. Servo motor; 2. Coupling; 3. Bidirectional fixed-displacement hydraulic pump; 4. First one-way valve; 5. Second one-way valve; 6. First safety relief valve; 7. Second safety relief valve; 8. Filter; 9. Pressure measuring joint; 10. Valve block; 11. Make-up oil pressure relay; 12. Temperature sensor; 13. Back pressure relief valve; 14. High-pressure cooler; 15. Low-pressure side hot oil shuttle valve; 16. Flowmeter; 17. First pressure relay; 18. Second pressure relay; 19. Hydraulic cylinder; 20. Acceleration sensor; 21. Displacement sensor; 22. Accumulator; 23. High-pressure pipeline; 24. Low-pressure pipeline; 25. First oil inlet of the hydraulic cylinder; 26. Second oil inlet of the hydraulic cylinder; A. Power output module; B. Filter and make-up oil module; C. Safety relief module; D. Hot oil cooling module; E. Pressure and flow monitoring module; F. Actuator and monitoring module; L1. Filter pore channel; L2. Motor cooling oil pore channel; L3. Leakage confluence pore channel; L4. Make-up oil overflow pore channel. Detailed implementation manners

[0020] Hereinafter, the implementation manners of the present invention will be described with reference to the drawings.

[0021] The present invention provides an electro-hydraulic servo pump control method based on fuzzy neural control for hydrogen compression. As Figure 1 shown, calculate the current state pressure of the hydraulic cylinder of the electro-hydraulic servo pump control and the hydrogen flexible load force to obtain the input control variables; determine the output control variables of the electro-hydraulic servo pump control fuzzy PID control, divide the basic domain, and establish fuzzy control rules; use the centroid method and the weighted average method for composite defuzzification, calculate the base values of the output steady-state component and the output dynamic component of the control variables, and synthesize the control signal of the driver; use a feedforward neural network to optimize the parameters, realize asynchronous cooperative control through a double-buffer mechanism, update the synthesized control signal of the driver, and realize hydrogen compression control; specifically include the following steps: Step S1: Calculate the current state pressure of the hydraulic cylinder of the electro-hydraulic servo pump control and the hydrogen flexible load force to obtain the force error and the force error change rate as the system input control variables.

[0022] Step S11: Calculate the current state force applied by the electro-hydraulic servo pump hydraulic cylinder as: ; wherein, is the current state pressure of the hydraulic cylinder; is the real-time pressure of the hydraulic system; is the piston diameter of the hydraulic cylinder; is the rod diameter of the hydraulic cylinder; is the constant of pi.

[0023] Step S12: Calculate the hydrogen flexible load force as: ; Among them, is the flexible load force of hydrogen; is the initial pressure of hydrogen; is the initial volume of hydrogen; is the displacement of the hydraulic cylinder rod.

[0024] Step S13: Use the current state pressure of the hydraulic cylinder and the flexible load force of hydrogen to calculate the force error and the rate of change of the force error , the force error and the rate of change of the force error are the input control variables.

[0025] Step S2: Determine the output control variables for the fuzzy PID control method used in the electro-hydraulic servo pump control, and establish fuzzy control rules.

[0026] Step S21: Determine that the input control variables of the fuzzy PID control algorithm are the force error and the rate of change of the force error and the output control variables are the proportional fuzzy control variable , the integral fuzzy control variable and the differential fuzzy control variable . The universes of discourse of the fuzzy PID control algorithm all adopt a variety of fuzzy control variables. In the embodiments of the present invention, 7-level fuzzy control quantities are used, specifically: the first-level fuzzy control variable FK, the second-level fuzzy control variable FG, the third-level fuzzy control variable PS, the fourth-level fuzzy control variable ZW, the fifth-level fuzzy control variable NP, the sixth-level fuzzy control variable NM, and the seventh-level fuzzy control variable NK. The output of the fuzzy PID control method is the corresponding adjustment amounts , , of the proportional, integral, and differential coefficients; finally, the output control signal based on the optimizer's feedforward neural network is: ; Among them, is the proportional fuzzy control variable; is the integral fuzzy control variable; is the differential fuzzy control variable; is the original proportional coefficient; is the original integral coefficient; is the original differential coefficient; is the steady-state component base value of the proportional coefficient adjustment amount; is the steady-state component base value of the integral coefficient adjustment amount; is the steady-state component base value of the differential coefficient adjustment amount; the steady-state components of the above PID coefficients are obtained in step S31; is the dynamic component base value of the proportional coefficient adjustment amount; is the dynamic component base value of the integral coefficient adjustment amount; is the dynamic component base value of the differential coefficient adjustment amount; the dynamic components of the above PID coefficients are obtained in step S32.

[0027] In the embodiment of the present invention, a neural network-based dynamic adjustment of the continuous region weight coefficient is added to the above formula, and the weight coefficient is updated in real time dynamically using the neural network.

[0028] According to the force error and the force error change rate , the basic domain of the input control variable is divided as: ; wherein, is the force error; is the force error change rate.

[0029] To match the non-linearity and time-variability characteristics of the electro-hydraulic servo pump control system for compressed hydrogen, the accurate measured value is converted into the "degree of membership" of fuzzy logic. After determining the basic domain, the membership function globally used is selected as the triangular membership function. The basic domain membership function is obtained by selecting the triangular membership function according to the requirements, specifically: ; wherein, is the output result of the basic domain membership function; is the input of the basic domain membership function; is the left boundary point of the membership function, the point where the membership degree starts to be 0; is the vertex value of the membership function, the point where the membership degree is the highest at 1; is the right boundary point of the membership function, the point where the membership degree ends at 0.

[0030] Step S22 establishes the fuzzy control rules as shown in Table 1 below. The output variable format in the table is ( , , ), representing the corresponding adjustment amounts of the proportional, integral, and differential coefficients.

[0031] Table 1 Fuzzy control rule table Step S23: According to the measurement range of the acceleration sensor, the measurement range of the pressure sensor, and the actual hydrogen compression flexible load demand, divide the adjustment amounts of the proportional, integral, and differential coefficients, that is, the proportional coefficient adjustment amount , the integral coefficient adjustment amount , and the differential coefficient adjustment amount The basic domain is: ; wherein, is the proportional coefficient adjustment amount; is the integral coefficient adjustment amount; is the differential coefficient adjustment amount.

[0032] The triangular membership function of the proportional / integral / differential coefficient adjustment amount in the continuous region [a, b] is obtained by using step S21. For example: ; Then the triangular membership function of the proportional / integral / differential coefficient adjustment amount in the continuous region [a, b] is: ; such as Figure 3 is the working process block diagram of the electro-hydraulic servo pump control based on fuzzy neural control for hydrogen compression of the present invention. The sensor detects external information in real time, and controls the servo motor through the fuzzy PID controller combined with the neural network trained by the optimizer, and finally realizes the real-time feedback control of the hydraulic cylinder.

[0033] Step S3: The composite defuzzification is carried out by using the centroid method and the weighted average method. The membership function output by the fuzzy PID is divided by region, and the steady-state component base value of the control variable output is calculated by the centroid method in the continuous region , and the dynamic component output base value of the control variable is calculated by the weighted average method in the discrete single-point region , and finally the control signal of the driver is synthesized to realize the hydrogen compression control, specifically: Step S31: Calculate the steady-state component base value of the control variable output by the centroid method in the continuous region as: ; wherein, is the steady-state component base value of the control variable output, which represents the smooth control amount of the steady-state regulation in the continuous region and is composed of the steady-state components of the adjustment amounts of each coefficient of the PID; is the steady-state component of the proportional coefficient adjustment amount; is the steady-state component of the integral coefficient adjustment amount; is the steady-state component of the differential coefficient adjustment amount; is the triangular membership function of the PID controller output by the proportional link in the continuous region [a, b], is the triangular membership function of the PID controller output by the integral link in the continuous region [a, b], is the triangular membership function of the PID controller output by the differential link in the continuous region [a, b], which is determined by step S21; [a, b] is the range of the action interval of the continuous membership function, is the lower limit of the action interval, It is the upper line of the action interval, which is divided by the fuzzy control rule in step S22.

[0034] Step S32: At the discrete single point, i.e., the extreme value point, the weighted average method is used to calculate the control variable for suppressing the dynamic component of the sudden disturbance and output the base value. It is: ; Among them, is the base value of the dynamic component output; is the discrete rule amplitude correction factor, with an initial value of 1, which is updated by the output of the feedforward neural network before step S43; is the adjustment vector of the PID coefficients, which is composed of the adjustment amounts of each PID coefficient.

[0035] Arrange the base value of the dynamic component output in the form of PID components: ; In the embodiment of the present invention, for the output of the dynamic component and the steady-state component, through different defuzzification methods, the calculation of different state components of the same PID coefficient signal is realized.

[0036] Step S33: In step S21, finally, the weight coefficient of the continuous region in the control signal is dynamically adjusted based on the output of the feedforward neural network based on the Adam optimizer. , and the corrected weight coefficient and the weight coefficient correction amount are: ; Among them, is the weight coefficient of the continuous region dynamically adjusted based on the neural network; α new is the corrected weight coefficient; is the weight coefficient correction amount.

[0037] The innovation in the embodiment of the present invention lies in updating the correction coefficient introduced in the control signal by using the feedforward neural network based on the optimizer.

[0038] Step S4: Dynamically optimize and correct the discrete rule amplitude correction factor and the weight coefficient correction amount in step S3 by using the feedforward neural network; realize the asynchronous cooperation between the electro-hydraulic servo pump control system and the neural network optimization layer through the double-buffer mechanism, and use the optimizer to update the parameters in real time to realize the real-time control of hydrogen compression. In a preferred embodiment, the feedforward neural network is a three-layer feedforward neural network.

[0039] Step S41: If the input layer of the feedforward neural network has 4 nodes, the four-dimensional input vector of the feedforward neural network is: ; Among them, is the four-dimensional input vector of the feedforward neural network; is the absolute value of the force error; is the absolute value of the error change rate; is the difference degree between the original continuous and discrete outputs, which is calculated from the base value u of the original output steady-state component A and the base value u of the original dynamic component output B Calculate; is the variance of the historical control signal, which is used to evaluate the stability of the system control signal; is the th historical control signal value; is the mean value of the historical control signal; is the total number of historical control signals.

[0040] Step S42: Determine that the hidden layer of the feedforward neural network has 8 nodes through Bayesian optimization. On the premise of ensuring real-time performance, minimize the validation set error. The hidden layer has 8 nodes, and the ReLU activation function is used to filter the fluctuating part of the signal variance noise. Specifically: ; Among them, is the output result of the hidden layer of the feedforward neural network, which is an 8-dimensional vector; is the activation function; is the hidden layer weight matrix. The input layer has 4 nodes and the hidden layer has 8 nodes. The dimension is 4×8; is the four-dimensional input vector of the feedforward neural network determined by S41; is the hidden layer bias vector, and the dimension is 8. The following explains the hidden layer weight matrix and the hidden layer bias vector form.

[0041] Hidden layer weight matrix The specific form is: ; Among them, is the connection strength from the input layer node to the hidden layer node, which is jointly determined by initializing and generating according to the Xavier condition and Adam optimization. In actual implementation, the initial value of the matrix is automatically generated by the algorithm and the parameters are updated in real time by step S43.

[0042] For example, the initial value that meets the Xavier condition is: ; Hidden layer bias vector The specific form is: ; Among them, is initialized with a small random value, that is, sampled from a normal distribution with a mean of 0 and a standard deviation of σ = 0.01. In actual implementation, the initialization value is automatically generated by the algorithm and the parameters are updated in real time by step S5.

[0043] For example: ; Step S43: The output layer of the feedforward neural network has 2 nodes, and the output vector of the feedforward neural network is set as: ; Among them, is the output vector of the feedforward neural network; is the weight matrix from the hidden layer to the output layer, is the first weight matrix from the hidden layer to the output layer, is the second weight matrix from the hidden layer to the output layer. Both adopt He initialization, with the number of neurons being 8, the standard deviation being 0.5, the uniform distribution range being (-0.866, +0.866), and the dimension being 8×2. It is automatically generated by the algorithm and updated in real time by step S5; is the bias vector of the output layer, is the first bias vector of the output layer, is the second bias vector of the output layer, initialized as a two-dimensional vector of zero, and updated in real time by step S5; is the weight coefficient correction amount; is the discrete rule amplitude correction factor.

[0044] The weight matrix from the hidden layer to the output layer The specific structure is: ; Among them, is the element of the weight matrix from the hidden layer to the output layer.

[0045] is the threshold for controlling the correction amount, for the weight coefficient correction amount and the discrete rule amplitude correction factor Adopt the Sigmoid activation function for linear scaling and translation, specifically: ; Among them, is the weight coefficient correction amount, after being scaled by the Sigmoid function ϵ[-0.1, +0.1]; is the discrete rule amplitude correction factor, after being scaled by the Sigmoid function ϵ[0.8, 1.2], u A原 is the base value of the original output steady-state component; u B原is the original dynamic component output base value; the weight coefficient correction amount, is the original parameter that updates the discrete rule amplitude correction factor in steps S32 and S33 respectively and .

[0046] Step S5: The operation period of the fuzzy PID layer of the hydrogen compression control system is 5 ms, and the operation period of the neural network optimization layer is 50 ms. That is, for every ten cycles of the fuzzy PID layer operation, the neural network parameters are updated once. Asynchronous coordination needs to be achieved through a double-buffer mechanism. The neural network training uses a loss function to evaluate the difference between the model prediction value and the target value: ; Among them, is the loss function for neural network training; is the control error at time t; is the weight coefficient correction amount, and the sum of its squares regularization term is used to suppress the excessive influence of high noise on the weights; is the regularization coefficient, which is used to balance error optimization and weight stability, = 0.01.

[0047] Backpropagation process: Calculate the gradient through the chain rule to prevent the gradient from being too high when |Δe| fluctuates violently, limit the gradient upper limit, and improve the online learning efficiency. The gradient calculation result of the chain rule is: ; Among them, is the gradient from the regularization term; is the control error backpropagation gradient; We get: ; ; Among them, is the initial weight coefficient correction amount; is the initial discrete rule amplitude correction factor; Use the optimizer to update the neural network parameters, that is, update the weight matrices W1, W2 ( , ) and the bias vectors B1, B2 ( , ). Combine the momentum method and adaptive learning to dynamically adjust the learning rate of each neural network parameter to adapt to the gradient change. The parameter update formula is: ; Among them, is the optimizer learning rate; is the optimized neural network weight matrix; is the optimized neural network bias vector; is the neural network weight matrix before optimization; is the neural network bias vector before optimization. Substitute the updated weight matrix and bias vector into step S4.

[0048] According to the optimized amplitude correction factor and the weight coefficient correction amount recalculate the control signal of the composite driver in step S3 , which is converted into a PWM signal by the servo driver to drive the servo motor, realizing hydrogen compression control.

[0049] Such as Figure 4 is the control flowchart of the electro-hydraulic servo pump control based on fuzzy neural control for hydrogen compression in the present invention. Input the working load of the system, the flow rate and pressure of the hydraulic cylinder, and detect the real-time rotation angle, speed and torque of the servo motor to realize servo motor control.

[0050] Such as Figure 5 is the simulation experiment curve graph of the embodiment of the present invention. The step input curve and the sine input curve obtained by simulating different input signals. The response results of the attached figure curves show that the method has good tracking effect.

[0051] The present invention provides an electro-hydraulic servo pump control system based on fuzzy neural control for hydrogen compression, such as Figure 2 is the functional block schematic diagram of the electro-hydraulic servo pump control based on fuzzy neural control for hydrogen compression in the present invention, which includes: a power output module A, a filter and oil replenishment module B, a safety overflow module C, a hot oil cooling module D, a pressure and flow monitoring module E, and an actuator and monitoring module F.

[0052] The power output module A includes: a servo motor 1 and a bidirectional fixed-displacement hydraulic pump 3. The servo motor 1 is connected to the bidirectional fixed-displacement hydraulic pump 3 through a coupling 2, serving as the power source of the system to provide the required pressure and flow for the system. The motor oil cooling inlet is connected to the motor cooling oil passage L2, and the motor oil cooling outlet is connected to the leakage confluence passage L3.

[0053] The filter and oil replenishment module B includes: a first check valve 4, a second check valve 5. The inlets of the first check valve 4 and the second check valve 5 are connected to the accumulator 22, and the outlets are connected to the accumulator 22. The inlet of the filter 8 is connected to the leakage filtration passage L1, and the outlet of the filter 8 is connected to the accumulator 22. The oil replenishment pressure relay 11 and the temperature sensor 12 are connected to the overflow and oil replenishment passage L4, and the accumulator 22 is connected to the oil replenishment overflow passage L4; it plays two roles, one is to compensate for the external leakage of the system, and the other is to filter the system oil.

[0054] The safety overflow module C includes: a first safety overflow valve 6 and a second safety overflow valve 7. The oil inlet of the first safety overflow valve 6 is connected to the high-pressure pipeline 23, and the oil inlet of the second safety overflow valve 7 is connected to the low-pressure pipeline 24. Their oil outlets are connected to the oil replenishing and overflow passage L4. When an abnormal high pressure, that is, the pressure is greater than 21 MPa, occurs during the normal operation of the system, the overflow valve opens and is in the working position, overflowing the excess oil back to the accumulator to stabilize the pressure within the safe range.

[0055] The hot oil cooling module D includes: a low-pressure side hot oil shuttle valve 15, a high-pressure cooler 14, a back-pressure overflow valve 13, and a pressure measuring joint 9. The first oil inlet A of the low-pressure side hot oil shuttle valve 15 is connected to the high-pressure pipeline 23, and the second oil inlet B of the low-pressure side hot oil shuttle valve 15 is connected to the low-pressure pipeline 24 through the cooler 14. The oil inlet of the back-pressure overflow valve 13 is connected to the T oil port of the low-pressure side hot oil shuttle valve 15, the oil outlet of the back-pressure overflow valve 13 is connected to the motor cooling oil passage L2, and the pressure measuring joint 9 is connected to the oil outlet of the back-pressure overflow valve 13. The low-pressure side hot oil shuttle valve 15 takes out the oil in the low-pressure pipeline, passes through the motor cooling oil passage L2 to cool the motor and reduce the motor temperature. At the same time, the cooler 14 continuously cools the system oil to control the system temperature. After the hot oil is thermally balanced for the motor, it is combined with the leakage oil of the bidirectional fixed-displacement hydraulic pump 3 through the leakage confluence passage L3 and enters the filter 8 through the filter passage L1; monitor the system overflow pressure and oil temperature, and take a part of the system flow to cool the motor.

[0056] The pressure and flow monitoring module E includes: a flowmeter 16, a first pressure relay 17, and a second pressure relay 18. The first pressure relay 17 and the second pressure relay 18 are respectively connected to the high-pressure pipeline 23 and the low-pressure pipeline 24, and the flowmeter 16 is connected in series on the high-pressure pipeline 23; by real-time detecting the system pressure and the system flow of the system, ensure the safe operation of the system and collect the pressure and flow data and transmit them to the controller.

[0057] The actuator and monitoring module F includes a hydraulic cylinder 19, an acceleration sensor 20, and a displacement sensor 21. The first oil inlet 25 of the hydraulic cylinder is connected to the high-pressure pipeline of the valve block 10, the second oil inlet 26 of the hydraulic cylinder is connected to the low-pressure pipeline of the valve block, the acceleration sensor 20 is connected to the end of the cylinder rod of the hydraulic cylinder 19, the position sensor 21 is connected to the other end of the cylinder rod of the hydraulic cylinder 19, and the cylinder rod of the hydraulic cylinder 19 pushes the hydrogen compression device.

[0058] The beneficial effects of the present invention are as follows: the present invention extracts the force error and the force error change rate in the hydrogen compression process as input control variables, adopts fuzzy PID control, adaptive flexible load, and adopts the center of gravity method and weighted average method for composite defuzzification, divides the membership function of the fuzzy PID output by region, synthesizes the control signal of the driver, realizes hydrogen compression control, provides conditions for coping with complex working condition changes, adjusts the system output according to different working conditions, reduces power loss, improves the system control accuracy, improves the rapid response capability, and shortens the system response time; adopts a feedforward neural network to dynamically optimize the fuzzy PID control parameters; realizes asynchronous coordination of the fuzzy PID control system and the neural network optimization layer through a double buffer mechanism, uses an optimizer to update the fuzzy PID control parameters in real time, and realizes real-time control of hydrogen compression; adopts multiple sensors and an environmental adjustment system to perceive the working conditions, with a wide range of parameter coverage, and simultaneously monitors various key parameters in the hydraulic system in real time to ensure that abnormal parameters in the system are collected in any state, thereby improving the stability and reliability of the system. The integrated cooling structure of the low-pressure side hot oil shuttle valve and the high-pressure cooler and valve block is used to perform heat balance on the hydraulic system, eliminating the motor oil cooling pipeline and equipment, optimizing the heat exchange process, allowing the motor to obtain cooling medium more quickly, making the heat exchange process more efficient and taking away heat quickly.

[0059] The embodiments described above are only descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. An electro-hydraulic servo pump control method for hydrogen compression based on fuzzy neural control, characterized in that, It includes: S1. Calculate the current state pressure of the electro-hydraulic servo pump-controlled hydraulic cylinder and the hydrogen flexible load force , and obtain the force error and the force error change rate , which are used as input control variables; S2. Determine the output control variable of the fuzzy PID control method for electro-hydraulic servo pump control, and output the control signal through the feedforward neural network based on the optimizer; Establish fuzzy control rules and divide the adjustment ranges of the proportionality coefficient , the integral coefficient , and the derivative coefficient in their basic domains; S3. The centroid method and the weighted average method are used for composite defuzzification. The membership function output by the fuzzy PID is divided by region. The steady-state component base value of the control variable output is calculated by the centroid method in the continuous region , and the dynamic component output base value of the control variable is calculated by the weighted average method in the discrete single-point region , where is the dynamic component output base value; is the discrete rule amplitude correction factor; is the adjustment vector of the PID coefficient; The dynamic component output base value is sorted into the PID component form; The weight coefficient of the continuous region is dynamically adjusted based on the output of the feedforward neural network based on the optimizer , and the corrected weight coefficient is:[[]]END]] ； Among them, dynamically adjusts the weight coefficient of the continuous region based on the neural network; is the corrected weight coefficient; is the weight coefficient correction amount; S4. Dynamically optimize and correct the discrete rule amplitude correction factor in step S3 using a feedforward neural network and the weight coefficient correction amount ; S5. Establish a neural network training loss function evaluation model, realize the asynchronous cooperation between the electro-hydraulic servo pump control system and the neural network optimization layer through the double-buffer mechanism, use the optimizer to update the parameters in real time, and convert them into PWM signals through the servo driver to drive the servo motor to achieve hydrogen compression control.

2. The electro-hydraulic servo pump control method based on fuzzy neural control for hydrogen compression according to claim 1, wherein: Step S2 is specifically as follows: S21. Determine that the input control variable and output control variable of the fuzzy PID control algorithm are proportional fuzzy control variables, and output the control signal through the feedforward neural network based on the optimizer; S22. Establish fuzzy control rules, and the output variable format is ( , , ), representing the corresponding adjustment amounts of the proportional, integral, and differential coefficients; S23. Divide the adjustment amount of the proportionality coefficient, the adjustment amount of the integral coefficient, and the adjustment amount of the differential coefficient according to the measurement range of the acceleration sensor, the measurement range of the pressure sensor, and the actual hydrogen compression flexible load requirement.

3. The electro-hydraulic servo pump control method based on fuzzy neural control for hydrogen compression according to claim 1, characterized in that: The control signal output by the feedforward neural network based on the optimizer in step S21 is: ； wherein, is the proportional fuzzy control variable; is the integral fuzzy control variable; is the derivative fuzzy control variable; is the original proportional coefficient; is the original integral coefficient; is the original derivative coefficient; is the steady-state component base value of the proportional coefficient adjustment amount; is the steady-state component base value of the integral coefficient adjustment amount; is the steady-state component base value of the derivative coefficient adjustment amount; is the dynamic component base value of the proportional coefficient adjustment amount; is the dynamic component base value of the integral coefficient adjustment amount; is the dynamic component base value of the derivative coefficient adjustment amount.

4. The electro-hydraulic servo pump control method for hydrogen compression based on fuzzy neural control according to claim 1, characterized in that: Step S3 is specifically as follows: S31. Calculate the base value of the steady-state component of the control variable output using the centroid method in the continuous region ; S32. At discrete single points, use the weighted average method to calculate the control variable for suppressing the dynamic component of the sudden disturbance and output the base value. Organize the base value of the dynamic component output into the form of PID components. S33. Dynamically adjust the weight coefficients of the continuous region in the output control signal of the feedforward neural network based on the optimizer, i.e., adjust the weight coefficients of the continuous region in the output control signal of the feedforward neural network based on the neural network, to obtain the corrected weight coefficients , and obtain the corrected weight coefficients .

5. The electro-hydraulic servo pump control method based on fuzzy neural control for hydrogen compression according to claim 4, characterized in that: Steady-state component base value in step S31 , specifically: ； Among them, is the base value of the steady-state component of the control variable output; is the steady-state component of the proportional coefficient adjustment amount; is the steady-state component of the integral coefficient adjustment amount; is the steady-state component of the differential coefficient adjustment amount; is the triangular membership function of the proportional link output PID controller in the continuous region [a, b], is the triangular membership function of the integral link output PID controller in the continuous region [a, b], is the triangular membership function of the differential link output PID controller in the continuous region [a, b]; [a, b] is the range of the action interval of the continuous membership function, is the lower limit of the action interval, is the upper limit of the action interval, which is divided by the fuzzy control rule in step S22.

6. The electro-hydraulic servo pump control method for hydrogen compression based on fuzzy neural control according to claim 4, characterized in that: In step S32, organize the dynamic component output base value into the form of PID components, specifically: ； Among them, is the adjustment amount of the proportionality coefficient; is the adjustment amount of the integral coefficient; is the adjustment amount of the differential coefficient; is the correction factor of the discrete rule amplitude.

7. The electro-hydraulic servo pump control method for hydrogen compression based on fuzzy neural control according to claim 1, characterized in that: Step S4 is specifically as follows: S41. Set the input layer of the feedforward neural network to have 4 nodes to obtain a four-dimensional input vector of the feedforward neural network ; S42. Determine the hidden layer of the feedforward neural network through Bayesian optimization, and filter the fluctuating part of the signal variance noise by using the ReLU activation function based on minimizing the validation set error; S43. Set the output layer of the feedforward neural network to 2 nodes to obtain the output vector of the feedforward neural network .

8. The electro-hydraulic servo pump control method based on fuzzy neural control for hydrogen compression according to claim 1, characterized in that: In step S5, the neural network training uses the loss function to evaluate the model, specifically: ； Among them, is the loss function for neural network training; is the control error at time t; is the weight coefficient correction amount; is the regularization coefficient.

9. An electro-hydraulic servo pump control system for the electro-hydraulic servo pump control method based on fuzzy neural control of hydrogen compression according to any one of claims 1 to 8, characterized in that, It includes: Power output module, filter and oil replenishment module, safety overflow module, hot oil cooling module, pressure and flow monitoring module, and actuator and monitoring module; The power output module includes a servo motor and a bidirectional fixed-displacement hydraulic pump. The servo motor is connected to the bidirectional fixed-displacement hydraulic pump through a coupling to provide the required pressure and flow as the power source; the filter and oil replenishment module includes a first one-way valve and a second one-way valve for compensating for external leakage and filtering the oil; the safety overflow module includes a first safety overflow valve and a second safety overflow valve. When the normal operating pressure of the system is greater than the pressure threshold, the overflow valve opens and is in the working position, overflowing the excess oil back to the accumulator to stabilize the pressure within a safe range; the hot oil cooling module includes a low-pressure side hot oil shuttle valve, a high-pressure cooler, a back-pressure overflow valve, and a pressure measuring joint to monitor the system overflow pressure and oil temperature, and take part of the system flow to cool the motor; The pressure and flow monitoring module includes a flow meter, a first pressure relay, and a second pressure relay. By detecting the pressure and flow of the system in real time, it ensures the safe operation of the system and transmits the pressure and flow data to the controller after acquisition; the actuator and monitoring module includes a hydraulic cylinder, an acceleration sensor, and a displacement sensor to achieve hydrogen compression.

10. The electro-hydraulic servo pump control system according to claim 9, characterized in that: The first oil inlet of the low-pressure side hot oil shuttle valve of the hot oil cooling module is connected to the high-pressure pipeline, the second oil inlet of the low-pressure side hot oil shuttle valve is connected to the low-pressure pipeline through the cooler, the oil inlet of the back-pressure overflow valve is connected to the oil port of the low-pressure side hot oil shuttle valve, the oil outlet of the back-pressure overflow valve is connected to the motor cooling oil passage, the pressure measuring joint is connected to the oil outlet of the back-pressure overflow valve. The low-pressure side hot oil shuttle valve takes out the oil in the low-pressure pipeline, cools the motor through the motor cooling oil passage, and at the same time the cooler cools the system oil, controls the system temperature, and after thermal balance of the motor, the hot oil merges with the leakage oil of the bidirectional fixed-displacement hydraulic pump through the leakage confluence passage and enters the filter through the filter passage.

Images