CN105759607A - Design method for PAC controller based on intelligent control algorithms - Google Patents

Abstract

The invention discloses a design method for a PAC controller based on intelligent control algorithms. On the basis of a conventional PAC controller, the invention can improve the control accuracy and control response, and meet higher control requirements by using intelligent control algorithms such as an improved PID control algorithm, a variable universe fuzzy PID control algorithm, a PID control algorithm based on a neural network. In an aspect of network communication, the Ethernet, CAN bus and other means of communication are added on the basis of conventional serial ports 232 and 485, and Modbus-RTU, Modbus-TCP, CANopen, custom protocols and so forth are used for achieving network interconnection. An integral windup prevention algorithm, an algorithm for differentiation of control variables, a differential forward algorithm and the like are added on the basis of a conventional PID, the PID effect can be significantly improved, the response speed is increased, and the control overshoot is reduced.

Description

The method for designing of PAC controller based on intelligent control algorithm

Technical field

The invention belongs to field of industrial automatic control, relate to the design of embedded PAC controller, high accuracy temperature control, Intelligent control algorithm etc..

Background technology

In industrial manufacturing industry production process, more and more higher to the demand controlled.PAC controller is on control performance, letter Breath processes above and has some more significant advantages on Internet communication capacity.PAC controller combines intrinsic reliable of PLC Property, robustness and distribution character, control with PC simultaneously compared with, PAC uses real time operating system, has real-time in process performance The advantage that the PCs such as property, certainty are incomparable.But traditional PAC controller is the most single in control algolithm, is relating to During some complicated algorithms, through frequently with method be to be combined control with the mode such as PC, also expose some defects, example As: the weak effect of control, the retardance of communication, the increase of cost.

Summary of the invention

According to the problem proposed in above-mentioned prior art, it is an object of the invention to: improve the control performance of control system, adopt With advanced PID control algorithm, variable universe Fuzzy PID, pid control algorithm based on neutral net, it is possible to increase Control accuracy, improves and controls response.Additionally improve the extensibility gathering signal, analog input signal can be voltage letter Number or the signal such as current signal and resistance.In terms of network communication, increase on the basis of traditional serial ports 232,485 The communication modes such as Ethernet, CAN, uses Modbus-RTU, Modbus-TCP, CANopen, custom protocol etc., comes Realize interconnecting of network.

The present invention realizes based on a kind of intelligent controlling device, and described intelligent controlling device mainly includes following Part:

1, senser sampling circuit, this circuit includes the sampled signal modulate circuit of voltage, electric current, resistance signal, to Realize the signal high precision collection of sensor, such as pressure transmitter, PT100 temperature sensor, angular transducer, 4～20mA Current sensor etc..

2, control output circuit, containing isolated location, use solid-state relay to control big voltage high-current.Use PWM defeated Go out signal, control output accuracy high.

3, for participating in the microcontroller of computing, and Ethernet communication circuit, 232,485, CAN communicating circuit, Reset circuit, watchdog circuit.

4, the power module circuitry of power good is guaranteed.

Sensor is interacted with microcontroller by input signal conditioning circuit.

The present invention is to propose some intelligent control algorithms and realization, mainly includes following components:

Step one: devise PID Self-tuning System algorithm, common Industry Control object has non-linear, time variation and not The factors such as certainty, cause pid parameter to use the method manually adjusted to compare the consuming time, and the effect adjusted is the most poor, The present invention adds the algorithm of PID Self-tuning System, it is possible to automatically determine the pid parameter that equipment runs, save the time, improve Control effect.

Step 2: additionally on the basis of traditional PID controls, add and prevent the saturated algorithm of integration, control variables is entered Row differential, differential forward scheduling algorithm, it is possible to significantly improve the effect of PID, improve the response time, reduces and controls overshoot.

Step 3: due to factors such as non-linear, the time variation of control object and uncertainties, simply use PID control, Control effectiveness comparison poor.The method, on the basis of improving regulatory PID control method, adds FUZZY ALGORITHMS FOR CONTROL, passes through Set up the fuzzy reasoning table imitating human knowledge language, and membership function carries out fuzzy control operation, uses Fuzzy Control The mode that combines with regulatory PID control of system, it is possible to increase controller has non-linear, time variation and the most true in control object The control effect of qualitative factor.

Obtaining mostly from the experience of people, control effect of step 4: in fuzzy control, fuzzy rule and membership function The quality of fruit is also closely bound up with the selection of fuzzy rule and membership function.But neutral net has the strongest Nonlinear Quasi Conjunction ability, can learn and the advantage such as adaptive ability, controls to combine with PID by neutral net, it is possible to utilize neutral net Self-learning capability, calculates the optimized parameter that PID controls, reaches optimum control, and effect is notable.

The realization of the PID Self-tuning System that described step one is mentioned is specifically divided into following steps:

1.1, use Z-N (Ziegler-Nichols rule) relay feedback formula setting method to realize, compare Z-N method There is good advantage, i.e. use relay method to produce oscillation element, the wherein descriptive equation of relay characteristics:In formula, M is relay characteristics amplitude, and X tries to achieve for measuring output peak difference calculating.When meeting argG (j ω)=-π,In formula, A is the maximum exported by measurement and minimum of a value is tried to achieve, K_uFor threshold oscillation ratio Gain, d is the amplitude of the symmetrical relay characteristics divided.

1.2, by the oscillating curve produced, it is thus achieved that above K_uAnd critical period of the oscillation T_u, by Ziegler-Nichols certainly The computing formula of setting method, as shown in table 1, performance requirement as required, calculate the pid parameter adjusted, complete relay The PID Self-tuning System process of feedback.

Table 1 Z-N selftuning PID parameters (quick performance)

The realization improving Traditional PID that described step 2 is mentioned is specifically divided into following steps:

2.1, if normal PID lgorithm uses without modification, easy overshoot, system stability cycle time can be exposed Long, to have oscillatory occurences when stable shortcoming.

The common type of 2.2, PID is:

$u\left(t\right)=Kp\[e\left(t\right)+\frac{1}{Ti}{\∫}_0^{t}e\left(t\right)dt+Td\frac{de\left(t\right)}{d\left(t\right)}\]=k_{p}e\left(t\right)+k_i{\∫}_0^{t}e\left(t\right)dt+k_d\frac{de\left(t\right)}{d\left(t\right)},$

Use the saturated algorithm of anti-windup to limit integration and cause over control, can quickly improve the rise time again.Wherein, Umin≤u (t)≤Umax, whenTime, make u₁(t)=Umax, Qi ZhonglingLimit integration saturated phenomenon.Additionally improve the variable quantity of differential, due toWherein e (t)=r (t)-y (t), wherein r (t) is for arranging value, and y (t) is sensor sample value, but works as r The when of t () adjusts, necessarily cause the instantaneous variation of differential term, the unstability of system can be increased, therefore make

The realization of the Varied scope fuzzy control PID that described step 3 is mentioned is specifically divided into following steps:

3.1, theory-region changed adaptive-fuzzy control is using error e and error rate ec as the input of system, in difference Pid parameter is automatically adjusted by the moment according to the difference of e and ec, and Kp, Ki, Kd of pid parameter is carried out by fuzzy rule online Amendment.Fuzzy selftuning PID is on the basis of pid algorithm, by calculating current system error e and error rate ec, and profit Carrying out fuzzy reasoning with fuzzy rule, inquiry fuzzy reasoning table carries out parameter adjustment.

3.2, the foundation of fuzzy control table is not unique, for different control system differences, but for common system System, uses following fuzzy control table all can meet control and requires:

Δ KP fuzzy reasoning table:

Δ Ki fuzzy reasoning table:

Δ Kd fuzzy reasoning table:

The selection of 3.3 domains and the determination of membership function are determined by artificial experience, it is ensured that the most accurately.Additionally lead to Cross the change of monitoring e and ec, continue to optimize and reduce the scope of domain, the problem reaching variable universe Self-tuning System.Described step 4 The realization of the Neural network PID mentioned is specifically divided into following steps:

4.1 this method use MN Neural Network PID, it is possible to avoid as using BP neutral net, something lost The problem that propagation algorithm, particle cluster algorithm are computationally intensive, it is also possible to ensure to control in real time, the most constantly revise.

4.2 as shown in Figure 4, and Neuron PID structural model is the structure of 3 single outputs of input, and wherein 3 inputs are e (k),Neuron is output as u (k), and wherein the weights of neuron are the ratio of PID, integration, micro- Divide three coefficients i.e. Kp, Ki, Kd.

4.3 neurons are output as $u\left(k\right)=K_{p}e\left(k\right)+K_i{T}_s\underset{m=0}{\overset{k}{\mathrm{\Σ}}}e\left(m\right)+K_d\frac{e\left(k\right)-e(k-1)}{Ts},$ When wherein Ts is for sampling Between, error function is e (k)=r_in(k)-y_outK (), takes performance indicationsUse steepest descent method, neuron Weighed value adjusting as follows:

${\mathrm{\ΔK}}_p(k+1)=-\mathrm{\μ}\frac{\∂J}{\∂K_p}=\mathrm{\μ}\frac{\∂J}{\∂e(k+1)}\frac{\∂e(k+1)}{\∂y(k+1)}\frac{\∂y(k+1)}{\∂u\left(k\right)}\frac{\∂u\left(k\right)}{\∂K_p}=\mathrm{\μ}e(k+1)T_s\underset{m=0}{\overset{k}{\Σ}}e\left(m\right)\frac{\∂y(k+1)}{\∂u\left(k\right)}$

In like manner can obtain,

${\mathrm{\ΔK}}_i(k+1)=-\mathrm{\μ}\frac{\∂J}{\∂Ki}=\mathrm{\μ}e(k+1)T_s\underset{m=0}{\overset{k}{\Σ}}e\left(m\right)\frac{\∂y(k+1)}{\∂u\left(k\right)},$

${\mathrm{\ΔK}}_d(k+1)=-\mathrm{\μ}\frac{\∂J}{\∂K_d}=\mathrm{\μ}e(k+1)T_s\frac{e\left(k\right)-e(k-1)}{Ts}\frac{\∂y(k+1)}{\∂u\left(k\right)},$

Wherein μ is learning rate, 0 < μ < 1；

4.4 order $\frac{\&part;y(k+1)}{\&part;u\left(k\right)}=sign\left\{\frac{y\left(k\right)-y(k-1)}{u(k-1)-u(k-2)}\right\},$ Then can be tried to achieve successively by step 3:

Kp (k+1)=Kp (k)+Δ Kp (k+1),

Ki (k+1)=Ki (k)+Δ Ki (k+1),

Kd (k+1)=Kd (k)+Δ Kd (k+1).

The value that the PID automatic setting method that the initial value of 4.5 neural network weights uses step one to relate to calculates is as just Initial value, it is possible to accelerate the speed of Neural network PID Self-tuning System.

The present invention, in conventional PAC controller design, hardware improves acquisition precision, improves network interconnection intercommunication, Use CANopen and Modbus-TCP, Modbus-RTU as conventional communications protocol, software have employed improvement routine PID control method, variable universe fuzzy self-adaption method, Neural network PID algorithm, the best realizes industry control scene The control of equipment, it is possible to be preferably useful in such as petroleum pipeline electric tracer heating system, cotton textile industry Temperature and Humidity Control, boiler temperature The occasions such as control.

Accompanying drawing explanation

Fig. 1 system hardware structure block diagram；

Fig. 2 PID relay autotuner block diagram；

Fig. 3 Varied scope fuzzy control PID structured flowchart；

Fig. 4 Neural network PID structured flowchart；

Detailed description of the invention

According to the detailed description being embodied as the present invention below in conjunction with accompanying drawing, those skilled in the art will become more apparent from The above-mentioned advantage of the present invention and feature.

First it is hardware configuration, as shown in Figure 1.

Microcontroller, selects the STM32F207VET6 of ST Microelectronics, has jumbo memory space and high property The arithmetic speed of energy, it is ensured that run some control algolithms.

House dog and reset circuit, select SP706, for special watchdog reset chip, can improve stablizing of system Property and anti-interference.

Power module, uses Switching Power Supply 220v to turn 24v, uses LM2576 and SPX1117-in system equipment outside equipment 3.3 carry out voltage conversion, and to meet system module voltage requirements, additionally ADC sampling module needs reference voltage accurately, select TL431 source as the reference voltage.

Sensor and input signal conditioning circuit, for ensureing 4～20mA current modes, voltage-type, resistance change type sensing In the access systems such as device, devise and can emit selection switch according to selecting sensor difference to arrange corresponding jumping at same port. Such as temperature sensor uses PT100, and signal conditioning circuit selects electric bridge mode, and it is right that the resistance value of PT100 change is converted into The voltage form answered send the ADC of controller to sample.Obtain actual temperature value through conversion, the temperature of collection is delivered to micro-place Reason device processes.Other sensor circuit also similar process.

Output signal conditioning circuit and actuator, the digital control amount that controller computing obtains, can be by way of output root It is converted into switching value, PWM output etc. according to demand.Inside, containing electrical isolation module, prevents from being disturbed by the external world, in addition to Powerful output controls, in-built solid state relay.

Data memory module, selects FM24CL04, can deposit a small amount of data and power down preserves the function do not lost. Can be used to deposit some critical parameter information.

Communication module, use MAX3485, MAX3232, PCA82C250, DP83848 etc. respectively as RS485, RS232, CAN communication, the data processing chip of ethernet communication and physical layer transceiver circuit.The selection of multiple communication modes can be from different Equipment between to carry out having more the information of compatibility mutual.

PID relay autotuner method is as shown in Figure 2.

The oscillating curve produced by relay link, it is thus achieved that K_uAnd critical period of the oscillation T_u(seeing above described), by The computing formula of Ziegler-Nichols automatic setting method, as shown in table 2, can performance requirement as required, calculate whole Fixed pid parameter, completes the PID Self-tuning System process of relay feedback.

Table 2 relay feedback formula Z-N selftuning PID parameters (quick performance)

Pid parameter can be obtained through calculating, can be as fuzzy control and the initial value of ANN Control.

Varied scope fuzzy control method is as shown in Figure 3.Mainly it is made up of fuzzy control and PID controller two parts.

The input of Digital PID Controller is collection value r (t) obtained by sensor sample and the artificial desired value set Y (t), obtains numeral output controlled quentity controlled variable u (t) by advanced PID control algorithm. and controlled quentity controlled variable u (t) changes or logical through D/A Cross numeral output PWM ripple to carry out exporting control.And then control the operation of equipment.Reach to make output control the function in desired value.

Varied scope fuzzy control pid algorithm is as shown in Figure 3.

1. determine domain and membership function.Deviation e and deviation variation rate ec input range according to input signal determine Domain, the determination of domain is typically based on artificial experience, the fuzzy set of domain be divided into negative big, and negative in, negative little, zero, the least, just In, honest }.Concrete domain scope needs to be used in combination scene and determines.Membership function i.e. determines the signal value institute of input At the proportion of interval range, in embedded device, the general curve selecting rectilinear(-al) is as membership function model, in order to Facilitate flush bonding processor to calculate, prepare for following fuzzy rule and ambiguity solution.

2. determine fuzzy reasoning table.The establishment of fuzzy reasoning table should guarantee accurately as far as possible because the quality of control algolithm with The accuracy of fuzzy rule is closely bound up.Typically the scope of e and ec is divided into seven sections, it may be assumed that and negative big, in bearing, negative little, zero, just Little, center, honest.Therefore 49 complete fuzzy rule set can be set up.As a example by temperature rise curve, when the temperature rising stage Between time, it should allowing Kp take bigger value to guarantee quickly to respond as far as possible, reduce the rise time, Ki takes relatively small Value, prevents integration saturated, the generation of integration over control, and Kd takes relatively large value, in order to prevent overshoot, reduces the defeated of system Go out, control climbing speed.

Therefore rule 1: when e is " PB " and ec is " Z0 ", Δ Kp is " PM ", and Δ Ki is " PS ", and Δ Kd is " PM ".

Rule 2: when e is " PB " and ec is " NS ", Δ Kp is " PM ", and Δ Ki is " PS ", and Δ Kd is " PM ".

Rule 3: when e is " PB " and ec is " NM ", Δ Kp is " PS ", and Δ Ki is " Z0 ", and Δ Kd is " PM ".

Pushing away with this, specific rules list is:

Δ KP fuzzy reasoning table:

Δ Ki fuzzy reasoning table:

Δ Kd fuzzy reasoning table:

3., in order to obtain real PID controller parameter Kp and Ki and Kd, change as follows:

Kp=Kp0+ Δ Kp；Ki=Ki0+ Δ Ki；(wherein Kp0, Ki0, Kd0 represent PID coefficient to Kd=Kd0+ Δ Kd respectively Initial value)；

Neuron PID structural model as shown in Figure 4,

1. structure is the structure of triple input single output, and wherein three inputs are e (k),Neuron Being output as u (k), wherein the weights of neuron are the ratio of PID, integration, three coefficients (Kp, Ki, Kd) of differential.

2, neuron is output as $u\left(k\right)=K_{p}e\left(k\right)+K_i{T}_s\underset{m=0}{\overset{k}{\&Sigma;}}e\left(m\right)+K_d\frac{e\left(k\right)-e(k-1)}{Ts},$ Error function is e (k) =r_in(k)-y_outK (), takes performance indicationsUsing steepest descent method, the weighed value adjusting of neuron is as follows:

${\mathrm{\&Delta;K}}_p(k+1)=-\mathrm{\&mu;}\frac{\&part;J}{\&part;K_p}=\mathrm{\&mu;}\frac{\&part;J}{\&part;e(k+1)}\frac{\&part;e(k+1)}{\&part;y(k+1)}\frac{\&part;y(k+1)}{\&part;u\left(k\right)}\frac{\&part;u\left(k\right)}{\&part;K_p}=\mathrm{\&mu;}e(k+1)T_s\underset{m=0}{\overset{k}{\&Sigma;}}e\left(m\right)\frac{\&part;y(k+1)}{\&part;u\left(k\right)}$

(wherein μ is learning rate, 0 < μ < 1)

In like manner can obtain,

${\mathrm{\&Delta;K}}_i(k+1)=-\mathrm{\&mu;}\frac{\&part;J}{\&part;Ki}=\mathrm{\&mu;}e(k+1)T_s\underset{m=0}{\overset{k}{\&Sigma;}}e\left(m\right)\frac{\&part;y(k+1)}{\&part;u\left(k\right)},$

${\mathrm{\&Delta;K}}_d(k+1)=-\mathrm{\&mu;}\frac{\&part;J}{\&part;K_d}=\mathrm{\&mu;}e(k+1)T_s\frac{e\left(k\right)-e(k-1)}{Ts}\frac{\&part;y(k+1)}{\&part;u\left(k\right)},$

3, order $\frac{\&part;y(k+1)}{\&part;u\left(k\right)}=sign\left\{\frac{y\left(k\right)-y(k-1)}{u(k-1)-u(k-2)}\right\},$ Then can be tried to achieve successively by above formula: Kp (k+1)=Kp (k) + Δ Kp (k+1), Ki (k+1)=Ki (k)+Δ Ki (k+1), Kd (k+1)=Kd (k)+Δ Kd (k+1).

4, the PID automatic setting method that the initial value of neural network weight can use this method step one to mention calculates Value is as initial value, it is possible to accelerate the speed of Neural network PID Self-tuning System.

Claims (1)

1. the method for designing of PAC controller based on intelligent control algorithm, the method realizes based on a kind of intelligent controlling device, Described intelligent controlling device mainly includes following components:

1) senser sampling circuit, this circuit includes the sampled signal modulate circuit of voltage, electric current, resistance signal, to realizing The signal high precision collection of sensor, such as pressure transmitter, PT100 temperature sensor, angular transducer, 4～the electric current of 20mA Sensor etc.；

2) control output circuit, containing isolated location, use solid-state relay to control big voltage high-current；Use PWM output letter Number, control output accuracy high；

3) for participating in the microcontroller of computing, and Ethernet communication circuit, 232,485, CAN communicating circuit, reset Circuit, watchdog circuit；

4) power module circuitry of power good is guaranteed；

Sensor is interacted with microcontroller by input signal conditioning circuit；

It is characterized in that: the method comprises the following steps:

Step one: design PID Self-tuning System algorithm, common Industry Control object has non-linear, time variation and uncertainty Etc. factor, causing pid parameter to use the method manually adjusted to compare the consuming time, the effect adjusted is the most poor, in this method In add the algorithm of PID Self-tuning System, it is possible to automatically determine the pid parameter that equipment runs, save the time, improve control effect Really；

Step 2: additionally on the basis of traditional PID controls, add and prevent the saturated algorithm of integration, control variables is carried out micro- Point, differential forward scheduling algorithm, it is possible to significantly improve the effect of PID, improve the response time, reduce and control overshoot；

Step 3: due to factors such as non-linear, the time variation of control object and uncertainties, simply use PID control, controls Effectiveness comparison is poor；The method, on the basis of improving regulatory PID control method, adds FUZZY ALGORITHMS FOR CONTROL, by setting up Imitate the fuzzy reasoning table of human knowledge language, and membership function carry out fuzzy control operation, use fuzzy control with The mode that regulatory PID control combines, it is possible to increase controller has non-linear, time variation and uncertainty in control object The control effect of factor；

Obtaining mostly from the experience of people of step 4: in fuzzy control, fuzzy rule and membership function, controls effect Quality is also closely bound up with the selection of fuzzy rule and membership function；But neutral net has the strongest nonlinear fitting energy Power, can learn and the advantage such as adaptive ability, controls to combine with PID by neutral net, it is possible to utilize the self-study of neutral net Habit ability, calculates the optimized parameter that PID controls, reaches optimum control, and effect is notable；

The realization of the PID Self-tuning System that described step one is mentioned is specifically divided into following steps:

1.1, use Z-N (Ziegler-Nichols rule) relay feedback formula setting method to realize, comparing Z-N method has very Good advantage, i.e. uses relay method to produce oscillation element, the wherein descriptive equation of relay characteristics: In formula, M is relay characteristics amplitude, and X tries to achieve for measuring output peak difference calculating；When meeting argG (j ω)=-π,In formula, A is the maximum exported by measurement and minimum of a value is tried to achieve, K_uFor threshold oscillation proportional gain, d is The amplitude of the symmetrical relay characteristics divided；

1.2, by the oscillating curve produced, it is thus achieved that above K_uAnd critical period of the oscillation T_u, by Ziegler-Nichols Self-tuning System The computing formula of method, as shown in table 1, performance requirement as required, calculate the pid parameter adjusted, complete relay feedback PID Self-tuning System process；

Table 1Z-N selftuning PID parameters (quick performance)

The realization improving Traditional PID that described step 2 is mentioned is specifically divided into following steps:

2.1, if normal PID lgorithm uses without modification, can expose easy overshoot, system stability length cycle time, The shortcoming having oscillatory occurences when stable；

The common type of 2.2, PID is:

$u\left(t\right)=Kp\&lsqb;e\left(t\right)+\frac{1}{Ti}{\&Integral;}_0^{t}e\left(t\right)dt+Td\frac{de\left(t\right)}{d\left(t\right)}\&rsqb;=k_{p}e\left(t\right)+k_i{\&Integral;}_0^{t}e\left(t\right)dt+k_d\frac{de\left(t\right)}{d\left(t\right)},$

Use the saturated algorithm of anti-windup to limit integration and cause over control, can quickly improve the rise time again；Wherein, Umin≤ U (t)≤Umax, whenTime, make u₁(t)=Umax, Qi ZhonglingLimit integration saturated phenomenon；Additionally improve the variable quantity of differential, due toWherein e (t)=r (t)-y (t), wherein r (t) is for arranging value, and y (t) is sensor sample value, but works as r The when of t () adjusts, necessarily cause the instantaneous variation of differential term, the unstability of system can be increased, therefore make

The realization of the Varied scope fuzzy control PID that described step 3 is mentioned is specifically divided into following steps:

3.1, theory-region changed adaptive-fuzzy control is using error e and error rate ec as the input of system, the most in the same time Pid parameter is automatically adjusted by the difference according to e and ec, and Kp, Ki, Kd of pid parameter is modified by fuzzy rule online； Fuzzy selftuning PID is on the basis of pid algorithm, by calculating current system error e and error rate ec, utilizes fuzzy Rule carries out fuzzy reasoning, and inquiry fuzzy reasoning table carries out parameter adjustment；

3.2, the foundation of fuzzy control table is not unique, for different control system differences, but for ubiquitous system, makes All can meet control to require in order to lower fuzzy control table:

Δ KP fuzzy reasoning table:

Δ Ki fuzzy reasoning table:

Δ Kd fuzzy reasoning table:

The selection of 3.3 domains and the determination of membership function are determined by artificial experience, it is ensured that the most accurately；Additionally by prison Survey the change of e and ec, continue to optimize and reduce the scope of domain, the problem reaching variable universe Self-tuning System；

The realization of the Neural network PID that described step 4 is mentioned is specifically divided into following steps:

4.1 this method use MN Neural Network PID, it is possible to avoid as using BP neutral net, heredity to calculate The problem that method, particle cluster algorithm are computationally intensive, it is also possible to ensure to control in real time, the most constantly revise；

4.2 Neuron PID structural models are the structure of 3 single outputs of input, and wherein 3 inputs areNeuron is output as u (k), and wherein the weights of neuron are the ratio of PID, integration, micro- Divide three coefficients i.e. Kp, Ki, Kd；

4.3 neurons are output as $u\left(k\right)=K_{p}e\left(k\right)+K_i{T}_s\underset{m=0}{\overset{k}{\&Sigma;}}e\left(m\right)+K_d\frac{e\left(k\right)-e(k-1)}{Ts},$ Wherein Ts is the sampling time, error Function is e (k)=r_in(k)-y_outK (), takes performance indicationsUsing steepest descent method, the weights of neuron are adjusted Whole as follows:

${\mathrm{\&Delta;K}}_p(k+1)=-\mathrm{\&mu;}\frac{\&part;J}{\&part;K_p}=\mathrm{\&mu;}\frac{\&part;J}{\&part;e(k+1)}\frac{\&part;e(k+1)}{\&part;y(k+1)}\frac{\&part;y(k+1)}{\&part;u\left(k\right)}\frac{\&part;u\left(k\right)}{\&part;K_p}=\mathrm{\&mu;}e(k+1)T_s\underset{m=0}{\overset{k}{\&Sigma;}}e\left(m\right)\frac{\&part;y(k+1)}{\&part;u\left(k\right)}$

In like manner can obtain,

${\mathrm{\&Delta;K}}_i(k+1)=-\mathrm{\&mu;}\frac{\&part;J}{\&part;Ki}=\mathrm{\&mu;}e(k+1)T_s\underset{m=0}{\overset{k}{\&Sigma;}}e\left(m\right)\frac{\&part;y(k+1)}{\&part;u\left(k\right)},$

${\mathrm{\&Delta;K}}_d(k+1)=-\mathrm{\&mu;}\frac{\&part;J}{\&part;K_d}=\mathrm{\&mu;}e(k+1)T_s\frac{e\left(k\right)-e(k-1)}{Ts}\frac{\&part;y(k+1)}{\&part;u\left(k\right)},$

Wherein μ is learning rate, 0 < μ < 1；

4.4 order $\frac{\&part;y(k+1)}{\&part;u\left(k\right)}=sign\left\{\frac{y\left(k\right)-y(k-1)}{u(k-1)-u(k-2)}\right\},$ Then can be tried to achieve successively by step 3:

Kp (k+1)=Kp (k)+Δ Kp (k+1),

Ki (k+1)=Ki (k)+Δ Ki (k+1),

Kd (k+1)=Kd (k)+Δ Kd (k+1)；

The value that the PID automatic setting method that the initial value of 4.5 neural network weights uses step one to relate to calculates as initial value, The speed of Neural network PID Self-tuning System can be accelerated.