CN112947094A - Temperature control PID parameter self-adjusting method for rotary cement kiln - Google Patents

Temperature control PID parameter self-adjusting method for rotary cement kiln Download PDF

Info

Publication number
CN112947094A
CN112947094A CN202110373367.8A CN202110373367A CN112947094A CN 112947094 A CN112947094 A CN 112947094A CN 202110373367 A CN202110373367 A CN 202110373367A CN 112947094 A CN112947094 A CN 112947094A
Authority
CN
China
Prior art keywords
cement kiln
temperature
rotary cement
network
control
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110373367.8A
Other languages
Chinese (zh)
Other versions
CN112947094B (en
Inventor
刘世
陈特欢
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ningbo University
Original Assignee
Ningbo University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ningbo University filed Critical Ningbo University
Priority to CN202110373367.8A priority Critical patent/CN112947094B/en
Publication of CN112947094A publication Critical patent/CN112947094A/en
Application granted granted Critical
Publication of CN112947094B publication Critical patent/CN112947094B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
    • G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance

Landscapes

  • Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Feedback Control In General (AREA)
  • Muffle Furnaces And Rotary Kilns (AREA)
  • Curing Cements, Concrete, And Artificial Stone (AREA)

Abstract

The invention discloses a temperature control PID parameter self-adjusting method of a rotary cement kiln, which is characterized by comprising the following steps: (1) obtaining and setting initial parameters of the incremental PID control module through reinforcement learning; (2) acquiring the ideal temperature of the rotary cement kiln, the deviation between the actual temperature and the ideal temperature and the actual temperature in real time as input values of a BP neural network module; (3) inputting the data acquired in real time in the step (2) into a BP neural network module, carrying out online network operation, outputting PID (proportion integration differentiation) parameters by a network to obtain control parameters, and controlling the opening of a coal feeding valve of the rotary cement kiln by the rotary cement kiln model module in real time by using the control parameters to control the internal working temperature of the rotary cement kiln; and (4) repeating the step (2) and the step (3), and continuously updating the network weight and the threshold of the BP neural network module until the internal temperature of the rotary cement kiln reaches a set condition.

Description

Temperature control PID parameter self-adjusting method for rotary cement kiln
Technical Field
The invention relates to the technical field of PID controller parameter adjustment, in particular to a method for self-adjusting a temperature control PID parameter of a rotary cement kiln.
Background
The cement rotary kiln is the most important part for producing cement, and is mainly used for completing a thermal processing link, and calcium carbonate and mineral substances containing silicon dioxide and the like are subjected to thermal reaction in the rotary kiln to form a calcium silicate mixture. The working quality of a cement rotary kiln usually determines the quality of produced cement, wherein the key factor is the working temperature control of the rotary kiln, the temperature control in the rotary kiln relates to the coal feeding amount, the rotary kiln has the characteristics of hysteresis, nonlinearity and the like, an accurate mathematical model is not easy to deduce, at present, cement production plants mostly select operators with abundant experience to carry out manual parameter adjustment or semi-automatic parameter adjustment, the cement production efficiency and the quality are based on the experience of the operators, and the acceptable performance is obtained with lower production efficiency. Therefore, those skilled in the art are devoted to developing a method for self-adjusting the temperature control PID parameters of the rotary cement kiln.
Disclosure of Invention
In view of the above defects in the prior art, the technical problem to be solved by the present invention is that the PID controller parameters of the rotary cement kiln cannot be quickly and accurately adjusted by human operators to meet the temperature requirements in the actual production process.
In order to achieve the aim, the invention provides a temperature control PID parameter self-adjusting method of a rotary cement kiln, which is characterized by comprising the following steps:
(1) obtaining and setting initial parameters of the incremental PID control module through reinforcement learning;
(2) acquiring the ideal temperature of the rotary cement kiln, the deviation between the actual temperature and the ideal temperature and the actual temperature in real time as input values of a BP neural network module;
(3) inputting the data acquired in real time in the step (2) into a BP neural network module, carrying out online network operation, outputting PID (proportion integration differentiation) parameters by a network to obtain control parameters, and controlling the opening of a coal feeding valve of the rotary cement kiln by the rotary cement kiln model module in real time by using the control parameters to control the internal working temperature of the rotary cement kiln; and (4) repeating the step (2) and the step (3), and continuously updating the network weight and the threshold of the BP neural network module until the internal temperature of the rotary cement kiln reaches a set condition.
Further, the initial parameters in the step (1) are obtained by using array data of the working process of the rotary cement kiln as preparation data and by using reinforcement learning, and the method comprises the following substeps:
(101) determining decision network pi and desired reward network VπInitializing the weight, the threshold and the learning rate of the two;
(102) sampling an ideal flame temperature r (t-1) and an actual flame temperature state st-1And calculating the error r (t-1) -st-1Updating the decision network pi input valueCalculating control parameters, and adjusting the opening of the coal feeding valve to adjust the flame temperature to a state stIf r (t) -st|<|r(t-1)-st-1If yes, the reward is given as r (t) 1, otherwise, the reward is given as r (t) 0;
(103) get stAnd r (t) as network VπAnd calculating a process expected reward Vπ(st) While repeating step (102) to obtain the actual reward of the process
Figure BDA0003010232270000021
(104) Training network VπThe mean square error between the process expected reward and the process actual reward is made as small as possible;
(105) based on trained network VπAnd (5) updating the decision network pi, and repeating the steps (102), (103) and (104) until the control parameters obtained by the decision network pi meet the ideal temperature control requirement of the rotary kiln.
Further, step (3) comprises the sub-steps of:
(301) determining the structure of a BP neural network module, and determining an initial learning rate and a momentum coefficient according to the initial parameters of the incremental PID control module determined in the step (1);
(302) sampling an ideal flame temperature r (t) and an actual output temperature y (t), calculating an error e (t), r (t) -y (t), and updating an input value of a BP neural network module;
(303) calculating the control parameters of the incremental PID control module by a BP neural network forward propagation algorithm, and further calculating a control signal u (t);
(304) the rotary cement kiln model module acquires a control signal u (t), controls the opening of a coal feeding valve of the rotary cement kiln in real time, and updates the weight and the threshold of the network according to the mean square error and an improved BP neural network back propagation algorithm;
(305) returning to the step (302), repeating the step (304) to reduce the mean square error until the internal temperature of the rotary cement kiln reaches the set condition.
Further, the rotary cement kiln temperature control PID parameter self-adjusting method as claimed in claim 3, wherein,
u(t)＝u(t-1)+△Kp·[e(t)-e(t-1)]+△Ki·e(t)
+△Kd·[e(t)-2·e(t-1)+e(t-2)]；
in the formula, Delta Kp、△Ki、△KdThe proportional, integral and differential coefficients of the incremental PID control module are provided.
Further, the reinforced learning decision network pi parameter thetaπThe update rule is defined as follows:
Figure BDA0003010232270000022
in the formula (I), the compound is shown in the specification,
Figure BDA0003010232270000023
is the learning rate;
Figure BDA0003010232270000024
defined according to the following formula:
Figure BDA0003010232270000025
wherein, N is the total number of times of off-line simulation operation, a represents the control parameter obtained by the decision network pi, R (t): if r (t) -st|<|r(t-1)-st-1If yes, the reward is given as r (t) 1, otherwise, the reward is given as r (t) 0;
Figure BDA0003010232270000026
representing the temperature of the flame of the rotary cement kiln in a state stLower sampling control parameter atAnd is transferred to the state st+1The probability of (c).
Further, reinforcement learning expectation reward network VπParameter sigmaπThe update rule is defined as follows:
Figure BDA0003010232270000031
where ρ is a learning rate; e is
Figure BDA0003010232270000032
And Vπ(st) Mean square error between.
Further, the input layer node input of the BP neural network module is:
Figure BDA0003010232270000033
corresponding to a system expected value r (t), an actual output y (t) of the rotary cement kiln and an error e (t) between the actual output and the expected output, wherein the error e (t) is r (t) -y (t);
the input and output of the hidden layer node of the BP neural network module are respectively:
Figure BDA0003010232270000034
Figure BDA0003010232270000035
Figure BDA0003010232270000036
the input and output of the output layer node of the BP neural network module are respectively:
Figure BDA0003010232270000037
Figure BDA0003010232270000038
Figure BDA0003010232270000039
wherein, the superscripts (1), (2) and (3) represent a neural network input layer, a hidden layer and an output layer respectively; the functions f (x) and g (x) are activation functions of a hidden layer and an output layer of the neural network respectively; output of a network
Figure BDA00030102322700000310
Control parameters delta K corresponding to incremental PID control modules respectivelyp、△Ki、△Kd。
Further, the cement rotary kiln model comprises:
Figure BDA00030102322700000311
further, the performance error function is:
Figure BDA00030102322700000312
the weight variation quantity is iteratively adjusted according to the following rules:
Figure BDA00030102322700000313
wherein η (t) and α (t) are adjusted according to the following modification rules:
η(t)＝η(t-1)·(1+0.1·cosθ)；
Figure BDA00030102322700000314
；
where θ is the angle between the current weight update direction and the last weight update direction.
Further, when the included angle θ is smaller than 90 ° and tends to zero, the learning rate η (t-1) of the last iteration is increased by (1+0.1 · cos θ) times (cos θ >0) to accelerate the optimization speed in the iteration direction; and when the included angle theta is larger than 90 DEG and tends to 180 DEG, the learning rate is reduced to be (1+0.1 · cos theta) times (cos theta <0) before, so that the algorithm searches for a minimum point along the negative gradient.
The invention provides a rotary cement kiln temperature control PID parameter self-adjusting method based on an improved BP neural network for updating a learning rate and a momentum coefficient simultaneously. PID parameters can be dynamically self-tuned on line by monitoring the parameters in real time in the rotary kiln; the initial PID control parameter obtained by reinforcement learning can effectively overcome the defect that the error between the ideal temperature and the actual output temperature is large due to the lack of key parameters at the initial operation stage of the equipment; the defect that the algorithm only searches for the steepest descent path in an unchanged search range and only can search for a local optimal value due to the fact that the learning rate and the momentum coefficient are unchanged in the traditional BP neural network propagation process can be overcome; the problem of slow algorithm convergence in the optimization process can be solved; the BP neural network can fully know the structure, parameters, lag and nonlinear relation of the rotary kiln system through self continuous updating and learning; the problem that the PID parameters cannot be optimized quickly due to insufficient experience of operators of the controller can be effectively solved; has strong self-adaptive capacity and excellent stability.
The invention adopts a method of combining an improved BP neural network and a PID based on updating the learning rate and the momentum coefficient at the same time, can accurately and rapidly control key equipment parameters of the opening and closing degree of the coal feeding valve on the basis of real-time monitoring of production process parameters such as ideal temperature, deviation between actual temperature and ideal temperature, actual temperature and the like, further can realize real-time adjustment of the temperature of the cement rotary kiln, improve the resource utilization rate and the quality of produced cement, reduce the labor capacity of operators, optimize the allocation of human resources, and play a positive role in promoting the improvement of the production efficiency of enterprises and the cement production process.
The conception, the specific structure and the technical effects of the present invention will be further described with reference to the accompanying drawings to fully understand the objects, the features and the effects of the present invention.
Drawings
FIG. 1 is a diagram of an improved BP neural network PID control system architecture in accordance with a preferred embodiment of the present invention;
FIG. 2 is a flow chart of reinforcement learning calculation of initial control parameters in a preferred embodiment of the present invention;
FIG. 3 is an online self-tuning process for PID controller parameters in a preferred embodiment of the invention;
FIG. 4 is a diagram of the improved BP neural network and the reinforcement learning decision network pi structure in accordance with a preferred embodiment of the present invention;
FIG. 5 is a reinforcement learning decision evaluation network in accordance with a preferred embodiment of the present invention;
FIG. 6 is a graph of the comparison of the ideal output and the actual output of the simulation system in a preferred embodiment of the present invention;
FIG. 7 is a comparison of the initial stages of operation of the simulation system in accordance with a preferred embodiment of the present invention;
FIG. 8 is a diagram of simulation system PID parameter tuning in a preferred embodiment of the invention.
Detailed Description
The technical contents of the preferred embodiments of the present invention will be more clearly and easily understood by referring to the drawings attached to the specification. The present invention may be embodied in many different forms of embodiments and the scope of the invention is not limited to the embodiments set forth herein.
The present invention generally comprises the following modules and steps:
(1) initial value training: using array data of the working process of the rotary cement kiln as preparation data and using Reinforcement Learning (RL) to obtain available PID control parameters as initial parameters of the whole PID control system;
(2) data acquisition: acquiring the ideal temperature of the rotary cement kiln, the deviation between the actual temperature and the ideal temperature and the actual temperature in real time according to the actual production condition as input values of a BP (back propagation) neural network;
(3) network establishment and operation: establishing an improved BP neural network, inputting data acquired in real time into the network, performing online network operation, outputting the network as a required PID parameter, further obtaining a control parameter, controlling the opening of a coal feeding valve in real time by using the control parameter, controlling the coal feeding amount by controlling and adjusting the key factors, further controlling the internal working temperature of the rotary kiln, updating the weight value and the threshold value of the network, and continuously repeating the step 2) and the step 3) until the internal temperature condition of the rotary kiln meets the actual production temperature requirement.
As shown in fig. 1, in an embodiment of the present invention, a system for self-adjusting temperature control PID parameters of a rotary cement kiln based on an improved BP neural network that simultaneously updates learning rate and momentum factor includes an improved BP neural network module (BPNN), an incremental PID controller module, and a rotary cement kiln approximate model module. In the figure, r (t) is an ideal temperature signal of the rotary kiln, y (t) is an actual output temperature signal of the rotary kiln of the controlled object, e (t) is an error between a reference output and an actual output, u (t) is a control signal output by PID, and delta Kp、△Ki、△KdIs a three-position control parameter signal of an incremental PID controller.
As shown in fig. 2, the specific steps of acquiring control initial parameters by reinforcement learning in an embodiment of the present invention are as follows:
(1) determining decision network pi and desired reward network VπInitializing the weight values, the threshold values and the learning rate of the two;
(2) sampling an ideal flame temperature r (t-1) and an actual flame temperature state st-1And calculating the error r (t-1) -st-1Updating the pi input value of decision network and calculating the control parameter delta Kp、△Ki、△KdAnd adjusting the opening degree of the coal feeding valve by using a PID controller so as to adjust the flame temperature to a state stIf r (t) -st|<|r(t-1)-st-1If yes, the reward is given as r (t) 1, otherwise, the reward is given as r (t) 0;
(3) get stAnd r (t) as network VπAnd calculating a process expected reward Vπ(st) And (3) repeating the step (2) at the same time to obtain the actual process reward
Figure BDA0003010232270000051
(4) Training network VπThe mean square error between the process expected reward and the process actual reward is made as small as possible;
(5) based on trained network VπAnd (5) updating the decision network pi, and repeating the steps (2), (3) and (4) until the control parameters obtained by the decision network pi meet the ideal temperature control requirement of the rotary kiln.
As shown in fig. 3, the concrete adjusting steps of the temperature control PID parameter self-adjusting method of the rotary cement kiln in one embodiment of the present invention are as follows:
(1) determining the structure of a BP neural network, using array data of the working process of the rotary cement kiln as preparation data and using reinforcement learning to obtain available PID control parameters as initial parameters of the whole PID control system, and determining an initial learning rate and a momentum coefficient;
(2) sampling an ideal flame temperature r (t) and an actual output temperature y (t), calculating an error e (t), r (t) -y (t), and updating an input value of a BP network;
(3) calculating three-position control parameter delta K of incremental PID by BP neural network forward propagation algorithmp、△Ki、△KdFurther, calculating a control signal u (t);
(4) the equipment model acquires a control signal, obtains a real-time output, and updates the weight and the threshold of the network according to the mean square error and an improved BP neural network back propagation algorithm;
(5) and (4) returning to the step (2), and repeating the reduction of the mean square error in the step (4) until the real-time output meets the ideal output requirement.
Wherein, the incremental PID controller algorithm:
u(t)＝u(t-1)+△Kp·[e(t)-e(t-1)]+△Ki·e(t)
+△Kd·[e(t)-2·e(t-1)+e(t-2)] (1)
in the formula, Delta Kp、△Ki、△KdProportional, integral and derivative coefficients of the PID controller, respectively.
Pi parameter theta of reinforcement learning decision network of the inventionπThe update rule is defined as follows:
Figure BDA0003010232270000061
in the formula (I), the compound is shown in the specification,
Figure BDA0003010232270000062
for the learning rate, set to 0.3 in this embodiment,
Figure BDA0003010232270000063
defined according to the following formula:
Figure BDA0003010232270000064
in the formula, N is the total number of times of off-line analog simulation operation, a represents a three-bit control parameter obtained by the decision network pi, and r (t) reward and punishment is expressed in the embodiment as follows: if r (t) -st|<|r(t-1)-st-1If yes, the reward is given as r (t) 1, otherwise, the reward is given as r (t) 0;
Figure BDA0003010232270000065
representing the temperature of the rotary kiln flame in a state stLower sampling control parameter atAnd is transferred to the state st+1The probability of (c).
Wherein the reinforcement learning expectation-reward network VπParameter sigmaπThe update rule is defined as follows:
Figure BDA0003010232270000066
where ρ is the learning rate, set to 0.2 in this embodiment, and E is
Figure BDA0003010232270000067
And Vπ(st) Mean square error between.
The input layer node input of the BP neural network is as follows:
Figure BDA0003010232270000068
the system is respectively corresponding to a system expected value r (t), an actual output y (t) of the controlled object and an error e (t) between the system actual output and the expected output, wherein the error e (t) is r (t) -y (t).
The inputs and outputs of the hidden layer nodes are respectively:
Figure BDA0003010232270000071
Figure BDA0003010232270000072
Figure BDA0003010232270000073
wherein, in this example, N1＝3,h＝1～5。
The inputs and outputs of the output layer nodes are respectively:
Figure BDA0003010232270000074
Figure BDA0003010232270000075
Figure BDA0003010232270000076
wherein, in this example, N2J is 5, 1-3; the superscripts (1), (2) and (3) represent a neural network input layer, a hidden layer and an output layer respectively; the functions f (x) and g (x) are the hidden layer of the neural network andoutputting a layer activation function; output of a network
Figure BDA0003010232270000077
Three-position control parameter delta K corresponding to PID controller respectivelyp、△Ki、△Kd。
For the rotary cement kiln model of the controlled equipment, the model can be simplified into the following complex nonlinear model according to the nonlinear and hysteresis characteristics:
Figure BDA0003010232270000078
the performance error function is as follows:
Figure BDA0003010232270000079
according to an improved method for updating the weight and the threshold based on the simultaneous updating of the learning rate and the momentum factor, the weight change quantity is iteratively adjusted according to the following rules:
Figure BDA00030102322700000710
wherein η (t) and α (t) are adjusted according to the following modification rules:
η(t)＝η(t-1)·(1+0.1·cosθ) (15)
Figure BDA00030102322700000711
when the included angle theta is smaller than 90 degrees and particularly tends to zero, the continuous two times of updating are carried out towards the minimum point of an error curve, at the moment, on the basis of the learning rate eta (t-1) of the last iteration, (1+ 0.1. cos theta) times (cos theta >0) are increased, and the optimizing speed in the iteration direction is accelerated; and when the included angle theta is larger than 90 degrees and particularly tends to 180 degrees, the learning rate is possibly overlarge, so that a minimum point is skipped in a certain iterative process of the algorithm, and the learning rate is reduced to be (1+0.1 · cos theta) times (cos theta <0) before, so that the algorithm searches for the minimum point along the negative gradient.
Fig. 4 is a structure diagram of an improved BP neural network and a reinforced learning decision network pi, which adopts a three-layer (3-5-3) network architecture, wherein the inputs respectively correspond to: in the modified BP neural network: x is the number of1＝r(t)、x2＝e(t)、x3Y (t), in the decision network pi: x is the number of1＝r(t)、x2＝r(t)-st、x3＝st(stTo reinforce the actual temperature state of the rotary cement kiln in the learning process); outputting three-position control parameters corresponding to the incremental PID controllers respectively: delta Kp、△Ki、△Kd。
FIG. 5 is a diagram of a reinforcement learning decision evaluation network with inputs of r (t) being the flame ideal temperature, stThe actual temperature state of the flame; outputting reward expectation value V corresponding to reinforcement learning processπ(st)。
FIG. 6 is a comparison graph of ideal output and actual output of the simulation system, the ideal output being a staircase signal; as shown in a comparison diagram of the initial operation stage of the simulation system in fig. 7, the initial output of the system under the reinforcement learning acquisition control initial parameter method is 0.7476, so that the initial error is greatly reduced compared with the traditional parameter adjusting method; the maximum peak error of the obtained actual output relative to the ideal output is only 0.0074, and the time for the actual output to completely reach the requirement is only 0.064 second.
FIG. 8 is a PID parameter tuning chart of the simulation system of the present invention, the control parameter corresponding to Δ K under the reinforcement learning acquisition control initial parameter method of the present inventionp＝0.9962、△Ki＝0.9674、△Kd0.9938; for each stage with different ideal outputs, the parameter searching method can quickly search the optimal value: in the first stage (0-1 s), the optimal value delta K is found within 0.019 secondsp＝0.9962、△Ki＝0.9650、△Kd0.9927; in the second stage (1-2 s), the optimal value delta K is found within 0.008 sp＝0.9988、△Ki＝0.9856、△Kd0.9975; the third stage (2-3 s) finds the optimal value delta K within 0.007 sp＝0.9991、△Ki＝0.9893、△Kd0.9982; the fourth stage (3-4 s) finds the optimal value delta K within 0.003 sp＝0.9992、△Ki＝0.9906、△Kd＝0.9984。
The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations could be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.

Claims (10)

1. A temperature control PID parameter self-adjusting method for a rotary cement kiln is characterized by comprising the following steps:
(1) obtaining and setting initial parameters of the incremental PID control module through reinforcement learning;
(2) acquiring the ideal temperature of the rotary cement kiln, the deviation between the actual temperature and the ideal temperature and the actual temperature in real time as input values of a BP neural network module;
(3) inputting the data acquired in real time in the step (2) into a BP neural network module, carrying out online network operation, outputting PID (proportion integration differentiation) parameters by a network to obtain control parameters, and controlling the opening of a coal feeding valve of the rotary cement kiln by the rotary cement kiln model module in real time by using the control parameters to control the internal working temperature of the rotary cement kiln; and (4) repeating the step (2) and the step (3), and continuously updating the network weight and the threshold of the BP neural network module until the internal temperature of the rotary cement kiln reaches a set condition.
2. A rotary cement kiln temperature control PID parameter self-adjusting method as claimed in claim 1, wherein the initial parameter in step (1) is obtained by using array data of rotary cement kiln working process as preliminary data and by using reinforcement learning, comprising the sub-steps of:
(101) determination blockPolicy network pi and expected reward network VπInitializing the weight, the threshold and the learning rate of the two;
(102) sampling an ideal flame temperature r (t-1) and an actual flame temperature state st-1And calculating the error r (t-1) -st-1Updating the pi input value of the decision network, calculating the control parameter, and adjusting the opening of the coal feeding valve so as to adjust the flame temperature to the state stIf r (t) -st|<|r(t-1)-st-1If yes, the reward is given as r (t) 1, otherwise, the reward is given as r (t) 0;
(103) get stAnd r (t) as network VπAnd calculating a process expected reward Vπ(st) While repeating step (102) to obtain the actual reward of the process
Figure FDA0003010232260000011
(104) Training network VπThe mean square error between the process expected reward and the process actual reward is made as small as possible;
(105) based on trained network VπAnd (5) updating the decision network pi, and repeating the steps (102), (103) and (104) until the control parameters obtained by the decision network pi meet the ideal temperature control requirement of the rotary kiln.
3. A rotary cement kiln temperature controlled PID parameter self-adjusting method as claimed in claim 2, wherein the step (3) comprises the sub-steps of:
(301) determining the structure of a BP neural network module, and determining an initial learning rate and a momentum coefficient according to the initial parameters of the incremental PID control module determined in the step (1);
(302) sampling an ideal flame temperature r (t) and an actual output temperature y (t), calculating an error e (t), r (t) -y (t), and updating an input value of a BP neural network module;
(303) calculating the control parameters of the incremental PID control module by a BP neural network forward propagation algorithm, and further calculating a control signal u (t);
(304) the rotary cement kiln model module acquires a control signal u (t), controls the opening of a coal feeding valve of the rotary cement kiln in real time, and updates the weight and the threshold of the network according to the mean square error and an improved BP neural network back propagation algorithm;
(305) returning to the step (302), repeating the step (304) to reduce the mean square error until the internal temperature of the rotary cement kiln reaches the set condition.
4. A rotary cement kiln temperature-controlled PID parameter self-adjusting method as claimed in claim 3, wherein,
u(t)＝u(t-1)+△Kp·[e(t)-e(t-1)]+△Ki·e(t)+△Kd·[e(t)-2·e(t-1)+e(t-2)]；
in the formula, Delta Kp、△Ki、△KdThe proportional, integral and differential coefficients of the incremental PID control module are provided.
5. A rotary cement kiln temperature-controlled PID parameter self-adjusting method as claimed in claim 4, wherein,
enhanced learning decision network pi parameter thetaπThe update rule is defined as follows:
Figure FDA0003010232260000021
in the formula (I), the compound is shown in the specification,
Figure FDA0003010232260000022
is the learning rate;
Figure FDA0003010232260000023
defined according to the following formula:
Figure FDA0003010232260000024
wherein N is the total number of off-line simulation operation times, and a represents the control parameter obtained by the decision network piAnd R (t): if r (t) -st|<|r(t-1)-st-1If yes, the reward is given as r (t) 1, otherwise, the reward is given as r (t) 0;
Figure FDA0003010232260000025
representing the temperature of the flame of the rotary cement kiln in a state stLower sampling control parameter atAnd is transferred to the state st+1The probability of (c).
6. A rotary cement kiln temperature-controlled PID parameter self-adjusting method as claimed in claim 5, wherein,
reinforcement learning expectation reward network VπParameter sigmaπThe update rule is defined as follows:
Figure FDA0003010232260000026
where ρ is a learning rate; e is
Figure FDA0003010232260000027
And Vπ(st) Mean square error between.
7. A rotary cement kiln temperature-controlled PID parameter self-adjusting method as claimed in claim 6, wherein,
the input layer node input of the BP neural network module is as follows:
Figure FDA0003010232260000028
corresponding to a system expected value r (t), an actual output y (t) of the rotary cement kiln and an error e (t) between the actual output and the expected output, wherein the error e (t) is r (t) -y (t);
the input and output of the hidden layer node of the BP neural network module are respectively:
Figure FDA0003010232260000029
Figure FDA00030102322600000210
Figure FDA00030102322600000211
the input and output of the output layer node of the BP neural network module are respectively:
Figure FDA0003010232260000031
Figure FDA0003010232260000032
Figure FDA0003010232260000033
wherein, the superscripts (1), (2) and (3) represent a neural network input layer, a hidden layer and an output layer respectively; the functions f (x) and g (x) are activation functions of a hidden layer and an output layer of the neural network respectively; output of a network
Figure FDA0003010232260000034
Control parameters delta K corresponding to incremental PID control modules respectivelyp、△Ki、△Kd。
8. The rotary cement kiln temperature-controlled PID parameter self-adjusting method as claimed in claim 7, wherein the rotary cement kiln model comprises:
Figure FDA0003010232260000035
9. a rotary cement kiln temperature controlled PID parameter self-adjusting method as claimed in claim 8, wherein the performance error function has:
Figure FDA0003010232260000036
the weight variation quantity is iteratively adjusted according to the following rules:
Figure FDA0003010232260000037
wherein η (t) and α (t) are adjusted according to the following modification rules:
η(t)＝η(t-1)·(1+0.1·cosθ)；
Figure FDA0003010232260000038
where θ is the angle between the current weight update direction and the last weight update direction.
10. The rotary cement kiln temperature-controlled PID parameter self-adjusting method as claimed in claim 9, wherein when the included angle θ is smaller than 90 ° and tends to zero, the learning rate η (t-1) of the last iteration is increased by (1+0.1 · cos θ) times (cos θ >0) to accelerate the optimizing speed in the iteration direction; and when the included angle theta is larger than 90 DEG and tends to 180 DEG, the learning rate is reduced to be (1+0.1 · cos theta) times (cos theta <0) before, so that the algorithm searches for a minimum point along the negative gradient.
CN202110373367.8A 2021-04-07 2021-04-07 Temperature control PID parameter self-adjusting method for rotary cement kiln Active CN112947094B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110373367.8A CN112947094B (en) 2021-04-07 2021-04-07 Temperature control PID parameter self-adjusting method for rotary cement kiln

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110373367.8A CN112947094B (en) 2021-04-07 2021-04-07 Temperature control PID parameter self-adjusting method for rotary cement kiln

Publications (2)

Publication Number Publication Date
CN112947094A true CN112947094A (en) 2021-06-11
CN112947094B CN112947094B (en) 2022-11-15

Family

ID=76230830

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110373367.8A Active CN112947094B (en) 2021-04-07 2021-04-07 Temperature control PID parameter self-adjusting method for rotary cement kiln

Country Status (1)

Country Link
CN (1) CN112947094B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113277761A (en) * 2021-06-23 2021-08-20 湖南师范大学 Cement formula limestone proportion adjusting method based on model prediction framework
CN114967790A (en) * 2022-05-05 2022-08-30 山西大学 Laser temperature control system and method based on machine learning
CN116151022A (en) * 2023-03-07 2023-05-23 浙江大学 Real-time cement rotary kiln temperature estimation method based on heat balance calculation

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101216715A (en) * 2008-01-11 2008-07-09 宁波大学 PID control temperature instrument using nerve cell network adjustment parameter and its control method
CN102096373A (en) * 2010-12-07 2011-06-15 昆明理工大学 Microwave drying PID (proportion integration differentiation) control method based on increment improved BP (back propagation) neural network
CN102629104A (en) * 2011-12-01 2012-08-08 燕山大学 Calcination predictive control system and method for rotary cement kiln
CN102644926A (en) * 2011-02-21 2012-08-22 上海电气自动化设计研究所有限公司 Loop control method of medical garbage combustion rotary kiln based on DCS (Distributed Control System)
CN106682735A (en) * 2017-01-06 2017-05-17 杭州创族科技有限公司 BP neural network algorithm based on PID adjustment
CN207663309U (en) * 2017-08-25 2018-07-27 北方奥钛纳米技术有限公司 A kind of kiln temperature control system
CN108809167A (en) * 2018-06-26 2018-11-13 长春工业大学 A kind of BP neural network PID speed regulating control algorithms based on fuzzy control
CN108919850A (en) * 2018-06-15 2018-11-30 光大环保技术研究院（南京）有限公司 Counter current rotary kiln Temperature Fuzzy Control method and system
CN111580384A (en) * 2020-06-08 2020-08-25 济南大学 Automatic adjusting method for parameters of PID control system for decomposing furnace temperature in cement production

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101216715A (en) * 2008-01-11 2008-07-09 宁波大学 PID control temperature instrument using nerve cell network adjustment parameter and its control method
CN102096373A (en) * 2010-12-07 2011-06-15 昆明理工大学 Microwave drying PID (proportion integration differentiation) control method based on increment improved BP (back propagation) neural network
CN102644926A (en) * 2011-02-21 2012-08-22 上海电气自动化设计研究所有限公司 Loop control method of medical garbage combustion rotary kiln based on DCS (Distributed Control System)
CN102629104A (en) * 2011-12-01 2012-08-08 燕山大学 Calcination predictive control system and method for rotary cement kiln
CN106682735A (en) * 2017-01-06 2017-05-17 杭州创族科技有限公司 BP neural network algorithm based on PID adjustment
CN207663309U (en) * 2017-08-25 2018-07-27 北方奥钛纳米技术有限公司 A kind of kiln temperature control system
CN108919850A (en) * 2018-06-15 2018-11-30 光大环保技术研究院（南京）有限公司 Counter current rotary kiln Temperature Fuzzy Control method and system
CN108809167A (en) * 2018-06-26 2018-11-13 长春工业大学 A kind of BP neural network PID speed regulating control algorithms based on fuzzy control
CN111580384A (en) * 2020-06-08 2020-08-25 济南大学 Automatic adjusting method for parameters of PID control system for decomposing furnace temperature in cement production

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
PENGKAI WANG 等: "PID Parameters Auto-Tuning Method for Industrial Temperature Adjustment", 《IEEE》 *
YAOHUA GUO 等: "The Neural Network PID Controller for Cement Rotary Kiln Temperature Based on FPGA", 《IEEE》 *
孟卓: "基于神经网络PID的水泥回转窑温度控制研究", 《中国优秀博硕士学位论文全文数据库(硕士) 工程科技Ⅰ辑》 *
胡小军 等: "基于神经网络——PID的水泥回转窑温度控制研究", 《河南建材》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113277761A (en) * 2021-06-23 2021-08-20 湖南师范大学 Cement formula limestone proportion adjusting method based on model prediction framework
CN114967790A (en) * 2022-05-05 2022-08-30 山西大学 Laser temperature control system and method based on machine learning
CN116151022A (en) * 2023-03-07 2023-05-23 浙江大学 Real-time cement rotary kiln temperature estimation method based on heat balance calculation
CN116151022B (en) * 2023-03-07 2024-04-23 浙江大学 Real-time cement rotary kiln temperature estimation method based on heat balance calculation

Also Published As

Publication number Publication date
CN112947094B (en) 2022-11-15

Similar Documents

Publication Publication Date Title
CN112947094B (en) Temperature control PID parameter self-adjusting method for rotary cement kiln
CN111045326B (en) Tobacco shred drying process moisture prediction control method and system based on recurrent neural network
CN104776446B (en) Combustion optimization control method for boiler
CN109901403A (en) A kind of face autonomous underwater robot neural network S control method
CN107272403A (en) A kind of PID controller parameter setting algorithm based on improvement particle cluster algorithm
CN101750964B (en) Multivariable predictive control method with firing system based on pulse response die
CN111812968B (en) Fuzzy neural network PID controller-based valve position cascade control method
JPH07502354A (en) Closed-loop neural network automatic tuner
CN108229721A (en) The Forecasting Methodology of pyrolysis of coal product based on Speed Controlling Based on Improving BP Neural Network
CN111413872A (en) Air cavity pressure rapid active disturbance rejection method based on extended state observer
CN113623719B (en) Heat exchange station prediction control method based on effective room temperature detection
CN112712189B (en) Heat supply demand load prediction method
CN110673482B (en) Power station coal-fired boiler intelligent control method and system based on neural network prediction
Dai et al. Multi-rate layered operational optimal control for large-scale industrial processes
CN114001562B (en) Cement pre-demodulation furnace temperature-regulating PID parameter self-setting method
Lakhani et al. Stability-preserving automatic tuning of PID control with reinforcement learning
CN116880191A (en) Intelligent control method of process industrial production system based on time sequence prediction
CN107490962A (en) A kind of servo-drive system method for optimally controlling of data-driven
CN108388115A (en) NCS method for compensating network delay based on generalized predictive control
CN111077771A (en) Self-tuning fuzzy PID control method
CN109101683B (en) Model updating method for pyrolysis kettle of coal quality-based utilization and clean pretreatment system
CN111538232A (en) Unmanned driving anti-shaking positioning method and system based on adaptive neural fuzzy control
CN110412873A (en) Time lag batch process 2D iterative learning forecast Control Algorithm based on end conswtraint
CN113126498A (en) Optimization control system and control method based on distributed reinforcement learning
CN115877811B (en) Flow process treatment method, device and equipment

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant