阅读说明：本技术 基于模糊神经网络pid控制器的阀位串级控制方法 (Fuzzy neural network PID controller-based valve position cascade control method ) 是由 朱敏 赵聪聪 于 2020-06-24 设计创作，主要内容包括：本发明涉及一种基于模糊神经网络PID控制器的阀位串级控制方法,属于自动控制领域。该方法建立了包括调节阀阀位控制回路和比例阀阀后压力控制回路的阀位串级控制模型,调节阀阀位控制回路为主环,以调节阀阀位为主环控制对象；比例阀阀后压力控制回路为副环,以比例阀阀后压力为副环控制对象；阀位串级控制模型以调节阀阀位为控制目标,调节阀阀位控制回路采用模糊神经网络PID算法。本发明解决了阀位控制过程复杂多变,难以建立精确数学模型,致使传统PID控制控制效果不佳以及单回路控制难以消除外界扰动的难题,能够对阀位控制过程进行动态实时控制,提升了控制过程的快速性、准确性和鲁棒性,利于调节阀稳定连续地工作。(The invention relates to a valve position cascade control method based on a fuzzy neural network PID controller, and belongs to the field of automatic control. The method establishes a valve position cascade control model comprising a regulating valve position control loop and a proportional valve back pressure control loop, wherein the regulating valve position control loop is a main loop and a regulating valve position is a main loop control object; the pressure control loop behind the proportional valve is an auxiliary ring, and the pressure behind the proportional valve is taken as an auxiliary ring control object; the valve position cascade control model takes the valve position of the regulating valve as a control target, and a fuzzy neural network PID algorithm is adopted in a valve position control loop of the regulating valve. The invention solves the problems that the traditional PID control effect is poor and the single-loop control is difficult to eliminate external disturbance due to the fact that the valve position control process is complex and changeable and an accurate mathematical model is difficult to establish, can carry out dynamic real-time control on the valve position control process, improves the rapidity, the accuracy and the robustness of the control process, and is beneficial to stable and continuous work of the regulating valve.)

1. A valve position cascade control method based on a fuzzy neural network PID controller is characterized in that a valve position control system related to the valve position cascade control method comprises a first comparator, a valve position controller, a second comparator, a pressure controller, a torque motor, a proportional valve, a pneumatic actuator and a regulating valve which are sequentially arranged, wherein the valve position controller comprises a fuzzy neural network and the PID controller, and the fuzzy neural network is in one-way electric connection with the PID controller; the first comparator is respectively in one-way electric connection with the fuzzy neural network and the PID controller, the PID controller is in one-way electric connection with the second comparator, the second comparator is in one-way electric connection with the pressure controller, the pressure controller is in one-way electric connection with the torque motor, the torque motor is in one-way pneumatic connection with the proportional valve, the proportional valve is in one-way pneumatic connection with the pneumatic actuator, and the pneumatic actuator is in one-way pneumatic connection with the regulating valve to regulate the valve position of the regulating valve; the position transmitter is arranged on the regulating valve, detects the actual valve position of the regulating valve in real time and transmits data to the first comparator through electric connection; a pressure transmitter is arranged between the proportional valve and the pneumatic actuator, and the pressure transmitter detects the pressure behind the proportional valve in real time and transmits data to the second comparator through electric connection;

the valve position cascade control method comprises the following specific steps:

step 1, setting the current time as n, and giving a valve position set value y at the current time_dnThe position transmitter detects and obtains the actual valve position y of the regulating valve at the current moment_nAnd the pressure I is transmitted to a first comparator, and the pressure transmitter detects the pressure I behind the proportional valve at the current moment_pnAnd transmitted to the second comparator;

step 2, the first comparator adjusts the actual valve position y of the valve at the current moment_nAnd the set value y of the valve position at the current moment_dnComparing and calculating to obtain the valve position deviation e at the current moment_nAnd the rate of change Δ e of the valve position deviation at the present time_nAnd the valve position at the current moment is deviated by e_nAnd the rate of change Δ e of the valve position deviation at the present time_nTransmitting to fuzzy neural network, and calculating the valve position deviation e at the current moment_nTransmitting to PID controller;

valve position deviation e at the present time_nAnd the rate of change Δ e of the valve position deviation at the present time_nIs calculated as follows:

Δe_n＝e_n-e_n-1

wherein e is_n-1Indicating the valve position deviation at the previous moment;

step 3, the fuzzy neural network receives the valve position deviation e at the current moment sent by the first comparator_nAnd the rate of change Δ e of the valve position deviation at the present time_nAfter the signal is sent, the valve position at the current moment is determinedDeviation e_nAnd the rate of change Δ e of the valve position deviation at the present time_nAs network input, real-time online learning is carried out through a trained fuzzy neural network model, and the following network output signals are obtained: coefficient of proportionality K_PIntegral coefficient K_IAnd a differential coefficient K_DThe fuzzy neural network transmits the network output signal to the PID controller;

step 4, the PID controller receives the valve position deviation e at the current time_nCoefficient of proportionality K_PIntegral coefficient K_IDifferential coefficient K_DAfter the signal is received, the initial control output u at the current moment is obtained through calculation_nAnd transmitted to the second comparator;

initial control output u at present_nIs calculated as follows:

u_n＝u_n-1+Δu_n

Δu_n＝K_P(e_n-e_n-1)+K_Ie_n+K_D(e_n-2e_n-1+e_n-2)

wherein, Δ u_nOutput of an increment, u, for the initial control of the current moment_n-1Outputting for initial control at the previous moment; e.g. of the type_n-2The valve position deviation at the first two moments;

step 5, the second comparator outputs u the initial control at the current moment_nProportional to the valve back pressure I at the present moment_pnPerforming comparative calculation to obtain the valve back pressure deviation e at the current moment_1nAnd transmits it to the pressure controller;

current time post-valve pressure deviation e_1nIs calculated as follows:

e_1n＝u_n-I_pn

step 6, the pressure controller receives the pressure deviation e after the valve at the current time_1nSignal to generate a current time control output u_1nAnd transmitted to the torque motor;

current time control output u_1nIs calculated as follows:

wherein, K_P1Is a proportional gain;

step 7, the torque motor controls the output u according to the received current time_1nThe corresponding action is carried out, so that the pressure behind the proportional valve is changed, the input air pressure of the pneumatic actuator is changed after the pressure behind the proportional valve is changed, the pneumatic actuator is displaced up and down, and the valve position of the regulating valve is regulated;

and 8, assigning n +1 to n at the moment of n +1, and returning to the step 1 to control the valve position at the next moment.

2. The method according to claim 1, wherein the fuzzy neural network model is a 2-input and 3-output 5-layer network structure model, and the 5-layer network structures are respectively: the system comprises an input layer, a fuzzy layer, a rule layer, a normalization layer and an output layer.

3. The method according to claim 2, wherein the parameter to be trained and learned of the network model of the fuzzy neural network is a connection weight ω between the output layer and the normalization layer_klC center of the Gaussian membership function_ijAnd width σ_ijWherein k is 1,2,3, which represents the output number of the network; 1, 2., 49, which represents the fuzzy rule number of the network; i is 1 and 2, and represents the input number of the network; j 1, 2., 7, which represents the number of linguistic variables of the fuzzy subset; the learning algorithm of the network model adopts a BP algorithm, and training samples of the network are input and output data acquired on site.

4. The valve position cascade control method based on the fuzzy neural network PID controller according to claim 3, characterized in that the Gaussian membership function is a membership function of a fuzzy subset obtained by fuzzifying an input variable by a fuzzification layer, and the expression formula is as follows:

wherein the content of the first and second substances,

Technical Field

The invention relates to the field of industrial automatic control, in particular to a valve position cascade control method based on a fuzzy neural network PID controller.

Background

Regulating valves, also known as flow valves, are devices in process control systems that are powered to change the flow of a medium. The regulating quality of process control is influenced to a certain extent, the regulation of environmental factors such as temperature, pressure, liquid level, concentration and the like closely related to the product quality in an industrial production system is required to be regulated through the opening control of a regulating valve, and whether the position of the valve meets the production requirement is directly related to whether the whole production system can run efficiently and safely.

With the development of society and economy, the accuracy and rapidity of the valve position control of the industrial control system are required to be higher and higher. At present, most of domestic valve position control systems still adopt the traditional PID control algorithm, and the PID control algorithm has the characteristics of simple principle and easy realization, and has excellent control effect on the production process with clear mathematical models of controlled objects. Due to different working environments of the regulating valve, the valve position control process has strong nonlinearity, large inertia and time-varying property and cannot establish an accurate mathematical model under the influence of the viscosity of the medium flow flowing through the valve, the internal friction force, the unbalanced force and the like, so that the parameter setting of the traditional PID algorithm is extremely difficult, the ideal control effect is often difficult to achieve, and the safety problem and the huge economic loss are brought to the industrial production process.

In addition, at present, the research and application of the valve position control system at home and abroad mainly aim at a single-loop control system, the single-loop control system is simple in structure and easy to operate, but the anti-interference performance of the single-loop control system is poor, and the regulating valve is often interfered by a plurality of external factors in the industrial production process to cause the valve position control process to swing, so that a series of safety problems are caused. Therefore, the use of single-loop control in the valve position control system is difficult to ensure that the industrial production process is carried out stably and safely.

Li yanqi in "implementation of an intelligent electrical valve positioner based on fuzzy PID control [ J ] (electronic technology and software engineering, 2017(17):94-94), proposes a valve position control method based on fuzzy PID control, which adopts a fuzzy algorithm to set PID controller control parameters, and uses a fuzzy rule to make the PID controller have a certain adaptability to time-varying and non-linear of a controlled object, but the performance of the fuzzy algorithm itself for eliminating system errors is poor, and it is difficult to achieve higher control accuracy, and fuzzy rules in the system are generally summarized and sorted by the opinions of domain experts, and knowledge in many fields is difficult to describe.

In pneumatic transmission system modeling and MATLAB simulation analysis [ J ] (machine manufacturing, 2016,54(01):22-25) of valve positioner, Yangmai and Dermav propose a cascade control system, which takes the pressure behind the regulating valve and the position of the valve as a secondary control object and a main control object respectively, so as to achieve the anti-interference effect to a certain extent.

To sum up, at present, the valve position control technology has the following technical problems:

1. because the valve position control process has strong nonlinearity, large inertia and time-varying property, an accurate mathematical model cannot be established, the control parameter setting of the conventional PID control algorithm is extremely difficult, and the effective control is difficult to realize;

2. the fuzzy algorithm is adopted to set the parameters of the controller, so that the controller has certain adaptive capacity to the time-varying property and nonlinearity of a controlled object, but as the fuzzy rule is generally summarized and sorted out by the opinion of a domain expert, and the knowledge in many fields is difficult to describe, the control is difficult to achieve higher precision;

3. the anti-interference performance of a single-loop control system is poor, the oscillation of a valve position control process is easily caused, the safe operation of industrial production is difficult to maintain, and the valve position cascade control system provided at present has the problems of limited anti-interference performance, simple control algorithm and general control effect.

Disclosure of Invention

Aiming at the defects of the prior art, the invention provides a valve position cascade control method based on a fuzzy neural network PID controller, and aims to solve the problems that the traditional PID control effect is poor and the external disturbance is difficult to eliminate in single-loop control due to the fact that the valve position control process is complex and changeable and an accurate mathematical model is difficult to establish.

The object of the invention is thus achieved. The invention provides a valve position cascade control method based on a fuzzy neural network PID controller, wherein a valve position control system related to the valve position cascade control method comprises a first comparator, a valve position controller, a second comparator, a pressure controller, a torque motor, a proportional valve, a pneumatic actuator and a regulating valve which are sequentially arranged, wherein the valve position controller comprises a fuzzy neural network and the PID controller, and the fuzzy neural network is in one-way electric connection with the PID controller; the first comparator is respectively in one-way electric connection with the fuzzy neural network and the PID controller, the PID controller is in one-way electric connection with the second comparator, the second comparator is in one-way electric connection with the pressure controller, the pressure controller is in one-way electric connection with the torque motor, the torque motor is in one-way pneumatic connection with the proportional valve, the proportional valve is in one-way pneumatic connection with the pneumatic actuator, and the pneumatic actuator is in one-way pneumatic connection with the regulating valve to regulate the valve position of the regulating valve; the position transmitter is arranged on the regulating valve, detects the actual valve position of the regulating valve in real time and transmits data to the first comparator through electric connection; a pressure transmitter is arranged between the proportional valve and the pneumatic actuator, and the pressure transmitter detects the pressure behind the proportional valve in real time and transmits data to the second comparator through electric connection;

the valve position cascade control method comprises the following specific steps:

step 1, setting the current time as n, and giving a valve position set value y at the current time_dnThe position transmitter detects and obtains the actual valve position y of the regulating valve at the current moment_nAnd the pressure I is transmitted to a first comparator, and the pressure transmitter detects the pressure I behind the proportional valve at the current moment_pnAnd transmitted to the second comparator;

step 2, the first comparator adjusts the actual valve position y of the valve at the current moment_nAnd the set value y of the valve position at the current moment_dnComparing and calculating to obtain the valve position deviation e at the current moment_nAnd the rate of change Δ e of the valve position deviation at the present time_nAnd the valve position at the current moment is deviated by e_nAnd the rate of change Δ e of the valve position deviation at the present time_nTransmitting to fuzzy neural network, and calculating the valve position deviation e at the current moment_nTransmitting to PID controller;

valve position deviation e at the present time_nAnd the valve position at the current momentDifferential rate of change Δ e_nIs calculated as follows:

e_n＝y_dn-y_n

Δe_n＝e_n-e_n-1

wherein e is_n-1Indicating the valve position deviation at the previous moment;

step 3, the fuzzy neural network receives the valve position deviation e at the current moment sent by the first comparator_nAnd the rate of change Δ e of the valve position deviation at the present time_nAfter the signal is received, the valve position deviation e at the current moment is calculated_nAnd the rate of change Δ e of the valve position deviation at the present time_nAs network input, real-time online learning is carried out through a trained fuzzy neural network model, and the following network output signals are obtained: coefficient of proportionality K_PIntegral coefficient K_IAnd a differential coefficient K_DThe fuzzy neural network transmits the network output signal to the PID controller;

step 4, the PID controller receives the valve position deviation e at the current time_nCoefficient of proportionality K_PIntegral coefficient K_IDifferential coefficient K_DAfter the signal is received, the initial control output u at the current moment is obtained through calculation_nAnd transmitted to the second comparator;

initial control output u at present_nIs calculated as follows:

u_n＝u_n-1+Δu_n

Δu_n＝K_P(e_n-e_n-1)+K_Ie_n+K_D(e_n-2e_n-1+e_n-2)

wherein, Δ u_nOutput of an increment, u, for the initial control of the current moment_n-1Outputting for initial control at the previous moment; e.g. of the type_n-2The valve position deviation at the first two moments;

step 5, the second comparator outputs u the initial control at the current moment_nProportional to the valve back pressure I at the present moment_pnPerforming comparative calculation to obtain the valve back pressure deviation e at the current moment_1nAnd transmits it to the pressure controller;

at the present timePressure deviation e after carving valve_1nIs calculated as follows:

e_1n＝u_n-I_pn

step 6, the pressure controller receives the pressure deviation e after the valve at the current time_1nSignal to generate a current time control output u_1nAnd transmitted to the torque motor;

current time control output u_1nIs calculated as follows:

u_1n＝K_P1×e_1n

wherein, K_P1Is a proportional gain;

step 7, the torque motor controls the output u according to the received current time_1nThe corresponding action is carried out, so that the pressure behind the proportional valve is changed, the input air pressure of the pneumatic actuator is changed after the pressure behind the proportional valve is changed, the pneumatic actuator is displaced up and down, and the valve position of the regulating valve is regulated;

and 8, assigning n +1 to n at the moment of n +1, and returning to the step 1 to control the valve position at the next moment.

Preferably, the network model of the fuzzy neural network is a 2-input and 3-output 5-layer network structure model, and the 5-layer network structures are respectively: the system comprises an input layer, a fuzzy layer, a rule layer, a normalization layer and an output layer.

Preferably, the parameter of the network model of the fuzzy neural network, which needs to be trained and learned, is a connection weight ω between the output layer and the normalization layer_klC center of the Gaussian membership function_ijAnd width σ_ijWherein k is 1,2,3, which represents the output number of the network; 1, 2., 49, which represents the fuzzy rule number of the network; i is 1 and 2, and represents the input number of the network; j 1, 2., 7, which represents the number of linguistic variables of the fuzzy subset; the learning algorithm of the network model adopts a BP algorithm, and training samples of the network are input and output data acquired on site.

Preferably, the gaussian membership function is a membership function of a fuzzy subset obtained by fuzzifying the input variable by the fuzzification layer, and the expression formula is as follows:

wherein the content of the first and second substances,representing an input variable x_iIs the membership function of the jth linguistic variable.

Compared with the prior valve position control technology, the invention has the beneficial effects that:

1. according to the technical scheme, the cascade control method takes the valve position of the regulating valve as a control target, and the valve position cascade control system related to the method actually forms two loops, namely a valve position control loop of the regulating valve and a pressure control loop behind a proportional valve. The control loop of the valve position of the regulating valve is a main loop, and the valve position of the regulating valve is a main loop control object. The pressure control loop behind the proportional valve is an auxiliary ring, and the pressure behind the proportional valve is used as an auxiliary ring control object. In the valve position control process, the pressure control loop behind the proportional valve serving as the secondary ring can rapidly overcome the main disturbance which causes severe and frequent change and large amplitude of the air pressure of the system in the controlled process, so that the influence of the secondary disturbance on the control loop of the primary ring is very little, and then the interference which is not completely overcome by the valve position control loop of the regulating valve serving as the primary ring on the secondary ring is further eliminated, the valve position change amplitude in the whole control process is small and stable, and the response speed, the control precision and the stability of the system are greatly improved.

2. The valve position controller adopts the fuzzy neural network PID controller, combines the advantages of fuzzy control and neural network control, and realizes PID controller control parameter self-tuning by the fuzzy neural network, so that the system has strong self-learning and self-adaptive capabilities, the problems of large disturbance and system oscillation in the valve position control process are solved, and the control precision and robustness are greatly improved.

Drawings

Fig. 1 is a schematic system structure diagram of the valve position cascade control method of the present invention.

Fig. 2 is a flow chart of the valve position cascade control method of the present invention.

Fig. 3 is a schematic structural diagram of a fuzzy neural network in the valve level controller according to the embodiment of the present invention.

Wherein, 1 is a first comparator; 2, a valve level controller; 3 a second comparator; 4, a pressure controller; 5 a torque motor; 6, a proportional valve; 7, a pneumatic actuator; 8, adjusting a valve; 9 a position transducer; 10 pressure transmitter.

Detailed Description

The embodiments of the present invention will be further described with reference to the accompanying drawings.

Fig. 1 is a schematic system structure diagram of the valve position cascade control method of the present invention. As can be seen from FIG. 1, the valve position control system comprises a first comparator 1, a valve position controller 2, a second comparator 3, a pressure controller 4, a torque motor 5, a proportional valve 6, a pneumatic actuator 7 and a regulating valve 8 which are arranged in sequence, wherein the valve position controller 2 comprises a fuzzy neural network 21 and a PID controller 22, and the fuzzy neural network 21 and the PID controller 22 are in one-way electrical connection.

The first comparator 1 is respectively in one-way electric connection with the fuzzy neural network 21 and the PID controller 22, the PID controller 22 is in one-way electric connection with the second comparator 3, the second comparator 3 is in one-way electric connection with the pressure controller 4, the pressure controller 4 is in one-way electric connection with the torque motor 5, the torque motor 5 is in one-way pneumatic connection with the proportional valve 6, the proportional valve 6 is in one-way pneumatic connection with the pneumatic actuator 7, the pneumatic actuator 7 is in one-way pneumatic connection with the regulating valve 8, and the valve position of the regulating valve 8 is regulated.

A position transmitter 9 is arranged on the regulating valve, and the position transmitter 9 detects the actual valve position of the regulating valve 8 in real time and transmits data to the first comparator 1 through electric connection; a pressure transmitter 10 is arranged between the proportional valve 6 and the pneumatic actuator 7, which pressure transmitter 10 detects the post-valve pressure of the proportional valve 6 in real time and transmits the data to the second comparator 3 via an electrical connection.

In the present embodiment, the proportional valve 6 is also in communication with an external air supply provided by an air compressor.

Fig. 2 is a flow chart of the valve position cascade control method of the present invention. As can be seen from fig. 2, the valve position cascade control method for valve position control specifically includes the following steps:

step 1, setting the current time as n, and giving a valve position set value y at the current time_dnAnd the position transmitter 9 detects and obtains the actual valve position y of the regulating valve at the current moment_nAnd transmitted to the first comparator 1, and the pressure transmitter 10 detects the pressure I behind the proportional valve at the current moment_pnAnd is passed to the second comparator 3.

Step 2, the first comparator 1 adjusts the actual valve position y of the valve at the current moment_nAnd the set value y of the valve position at the current moment_dnComparing and calculating to obtain the valve position deviation e at the current moment_nAnd the rate of change Δ e of the valve position deviation at the present time_nAnd the valve position at the current moment is deviated by e_nAnd the rate of change Δ e of the valve position deviation at the present time_nTransmitting to fuzzy neural network 21, and transmitting the valve position deviation e at the current moment_nTo the PID controller 22.

Valve position deviation e at the present time_nAnd the rate of change Δ e of the valve position deviation at the present time_nIs calculated as follows:

e_n＝y_dn-y_n

Δe_n＝e_n-e_n-1

wherein e is_n-1Indicating the valve position deviation at the previous time.

Step 3, the fuzzy neural network 21 receives the current time valve position deviation e sent by the first comparator 1_nAnd the rate of change Δ e of the valve position deviation at the present time_nAfter the signal is received, the valve position deviation e at the current moment is calculated_nAnd the rate of change Δ e of the valve position deviation at the present time_nAs network input, real-time online learning is performed through the trained fuzzy neural network model 21, and the following network output signals are obtained: coefficient of proportionality K_PIntegral coefficient K_IAnd a differential coefficient K_DThe fuzzy neural network 21 transmits the network output signal to the PID controller 22.

Step 4, the PID controller 22 receives the valve position deviation e at the current time_nCoefficient of proportionality K_PIntegral coefficient K_IDifferential coefficient K_DAfter the signal is received, the initial control output u at the current moment is obtained through calculation_nAnd transmitted to the second comparator 3.

Initial control output u at present_nIs calculated as follows:

u_n＝u_n-1+Δu_n

Δu_n＝K_P(e_n-e_n-1)+K_Ie_n+K_D(e_n-2e_n-1+e_n-2)

wherein, Δ u_nOutput of an increment, u, for the initial control of the current moment_n-1Outputting for initial control at the previous moment; e.g. of the type_n-2The valve position deviation at the first two moments.

As can be seen from fig. 1, the valve level controller 2 includes two parts, a fuzzy neural network 21 and a PID controller 22. In the valve position cascade control method of the invention, the valve position controller 2 is a main controller, and a fuzzy neural network PID control algorithm is adopted, namely, the fuzzy neural network 21 carries out real-time online adjustment on the control parameters of the PID controller 22.

Step 5, the second comparator 3 outputs u the initial control at the current moment_nProportional to the valve back pressure I at the present moment_pnPerforming comparative calculation to obtain the valve back pressure deviation e at the current moment_1nAnd transmits it to the pressure controller 4.

Current time post-valve pressure deviation e_1nIs calculated as follows:

e_1n＝u_n-I_pn

step 6, the pressure controller 4 receives the pressure deviation e after the valve at the current time_1nSignal to generate a current time control output u_1nAnd transmitted to the torque motor 5.

Current time control output u_1nIs calculated as follows:

u_1n＝K_P1×e_1n

wherein, K_P1Is a proportional gain.

In the present invention, the pressure controller 4 is a sub-controller, and is required to compare the post-valve pressure of the proportional valveThe input of the control loop is quickly responded, the deviation of the secondary loop is quickly eliminated, the functions of quick adjustment and coarse adjustment are achieved, and the requirement on the accuracy is not high, so that the pressure controller 4 adopts a proportional control algorithm and a proportional gain K_P1Given a constant, in this embodiment, K_P1＝4.5。

Step 7, the torque motor 5 controls the output u according to the received current time_1nAnd (3) corresponding action is carried out, so that the pressure behind the proportional valve 6 is changed, after the pressure behind the proportional valve 6 is changed, the input air pressure of the pneumatic actuator 7 is changed, the pneumatic actuator 7 is displaced up and down, and the valve position of the regulating valve 8 is regulated.

And 8, assigning n +1 to n at the moment of n +1, and returning to the step 1 to control the valve position at the next moment.

Fig. 3 is a schematic structural diagram of the fuzzy neural network 22 in the valve level controller 2 according to the embodiment of the present invention. As can be seen from fig. 3, the network model of the fuzzy neural network 22 is a 2-input 3-output 5-layer network structure model. The 5-layer network structures are respectively as follows: the system comprises an input layer, a fuzzy layer, a rule layer, a normalization layer and an output layer.

The parameters of the network model of the fuzzy neural network 22 to be trained and learned are the connection weight ω between the output layer and the normalization layer_klC center of the Gaussian membership function_ijAnd width σ_ijWherein k is 1,2,3, which represents the output number of the network; 1, 2., 49, which represents the fuzzy rule number of the network; i is 1 and 2, and represents the input number of the network; j 1, 2., 7, which represents the number of linguistic variables of the fuzzy subset; the learning algorithm of the network model adopts a BP algorithm, and training samples of the network are input and output data acquired on site.

The Gaussian membership function is a membership function of a fuzzy subset obtained by fuzzifying an input variable by a fuzzification layer, and the expression formula of the Gaussian membership function is as follows:

wherein the content of the first and second substances,representing an input variable x_iIs the membership function of the jth linguistic variable.

Specifically, the layers of the fuzzy neural network are specifically divided and function as follows:

the first layer is an input layer, and has 2 nodes which are respectively deviated from the current valve position of the input variable at the moment_nAnd the valve position deviation change rate delta e of the input variable at the current moment_nConnected to input variable e_nAnd input variable Δ e_nAnd transmitting to the blurring layer.

The second layer is a fuzzification layer which inputs a variable e_nAnd Δ e_nFuzzification is carried out to obtain fuzzy subsets of the fuzzy subsets, the fuzzy subsets are divided into 7 linguistic variables respectively, and input variables e are obtained_nAnd input variable Δ e_nAnd the membership functions belong to fuzzy sets of language variable values respectively, so that the fuzzy layer has 14 nodes in total, and the membership functions of the fuzzy subsets adopt Gaussian functions.

The third layer is a rule layer, fuzzy rules are matched through connection with the fuzzification layer to obtain the logic relation between input and output, fuzzy reasoning is carried out, and the applicability alpha of each rule is calculated through multiplication operation_lWhere l denotes the number of fuzzy rules of the network, and l is 1, 2., 49, there are 49 nodes, and each node represents a control rule.

The fourth layer is a normalization layer for realizing the normalization calculation of the rule applicability, which is beneficial to accelerating the convergence speed of the fuzzy neural network, and the number of the nodes is 49 as that of the rule layer.

The fifth layer is an output layer which realizes defuzzification operation, has 3 nodes in total and outputs a control parameter proportionality coefficient K of a PID controller_PIntegral coefficient K_IDifferential coefficient K_DAnd K is_P、K_I、K_DThe fuzzy linguistic variables of (1) are also 7, and the membership function is also a gaussian function.

In the present embodiment, the parameters of the network model of the fuzzy neural network 22 that need to be trained and learned are the outputConnection weight omega between layer and normalization layer_klC center of the Gaussian membership function_ijAnd width σ_ijThe learning algorithm of the network model adopts a BP algorithm, and training samples of the network are input and output data acquired on site.

The specific learning process is as follows:

defining the target cost function E as follows:

the connection weight omega of the normalization layer and the output layer_klThe learning algorithm of (1) is as follows:

wherein eta is₁Is a connection weight omega of the normalization layer and the output layer_klThe learning rate of (c); alpha is a momentum factor; (n +1) is the next moment; (n-1) is the previous moment;

center c of the Gaussian membership function_ijThe learning algorithm of (1) is as follows:

wherein eta is₂Is the center c of a Gaussian membership function_ijThe learning rate of (c);

width σ of the Gaussian membership function_ijThe learning algorithm of (1) is as follows:

wherein eta is₃Is the width σ of the Gaussian membership function_ijThe learning rate of (2).

The valve position cascade control method adopts a cascade control strategy and establishes a valve position cascade control model comprising a regulating valve position control loop and a proportional valve back pressure control loop. Wherein, governing valve position control circuit includes: the device comprises a first comparator 1, a valve position controller 2, a second comparator 3, a pressure controller 4, a torque motor 5, a proportional valve 6, a pneumatic actuator 7, a regulating valve 8 and a position transmitter 9. The proportional valve post-valve pressure control loop includes: a second comparator 3, a pressure controller 4, a torque motor 5, a proportional valve 6, and a pressure transmitter 10. In the cascade control method, a control loop of the valve position of the regulating valve is a main loop, and the valve position of the regulating valve is taken as a main loop control object. The pressure control loop behind the proportional valve is an auxiliary ring, and the pressure behind the proportional valve is used as an auxiliary ring control object. Specifically, the valve level controller 2 is a main controller, and a fuzzy neural network PID control algorithm is adopted, that is, the fuzzy neural network 21 performs real-time online adjustment on the control parameters of the PID controller 22. The pressure controller 4 is a sub-controller and adopts a proportional control algorithm.

The cascade control method for the valve position effectively inhibits external disturbance in the valve position control process by adopting a cascade control strategy, realizes stable and rapid control on the valve position of the regulating valve 8, simultaneously adopts a fuzzy neural network PID control algorithm in a main loop, combines the advantages of the fuzzy neural network and the traditional PID control, realizes real-time online adjustment on PID control parameters, solves the problem that the traditional PID control is difficult to effectively control the complicated and changeable valve position control process, and is beneficial to the stable and continuous work of the regulating valve 8.