CN110673482B - Power station coal-fired boiler intelligent control method and system based on neural network prediction - Google Patents

Power station coal-fired boiler intelligent control method and system based on neural network prediction Download PDF

Info

Publication number
CN110673482B
CN110673482B CN201910954643.2A CN201910954643A CN110673482B CN 110673482 B CN110673482 B CN 110673482B CN 201910954643 A CN201910954643 A CN 201910954643A CN 110673482 B CN110673482 B CN 110673482B
Authority
CN
China
Prior art keywords
prediction
neural network
input
fitting
optimal
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.)
Active
Application number
CN201910954643.2A
Other languages
Chinese (zh)
Other versions
CN110673482A (en
Inventor
谷薇
王继明
戴维
翟海龙
张国瑞
冯贾华
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.)
Guoneng Xinkong Internet Technology Co Ltd
Original Assignee
Beijing Huadian Tianren Power Controlling Technology Co Ltd
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 Beijing Huadian Tianren Power Controlling Technology Co Ltd filed Critical Beijing Huadian Tianren Power Controlling Technology Co Ltd
Priority to CN201910954643.2A priority Critical patent/CN110673482B/en
Publication of CN110673482A publication Critical patent/CN110673482A/en
Application granted granted Critical
Publication of CN110673482B publication Critical patent/CN110673482B/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)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Feedback Control In General (AREA)

Abstract

The application discloses a power station coal-fired boiler intelligent control method and system based on neural network prediction, wherein the method comprises the following steps: establishing a prediction control model; establishing a neural network, training a fitting prediction control model on line, and iteratively updating the weight; predicting parameters on the basis of the weight values obtained by each iterative calculation; screening an optimal predicted value and calculating an optimal prediction deviation, and calculating an optimal control instruction based on the optimal prediction deviation; and acting the optimized control instruction on the field production system through DCS, and performing advanced adjustment on the actuating mechanism in advance, thereby realizing closed-loop control of the system. The integral control level of the unit is not lower than the national standard or the line standard requirement; the intelligent control strategy has strong adaptability to the working condition, and the parameters can be corrected in an online self-adaptive manner; the method and the device have little dependence on experienced human resources, greatly reduce the on-site debugging workload, and realize maintenance-free system.

Description

Power station coal-fired boiler intelligent control method and system based on neural network prediction
Technical Field
The invention belongs to the technical field of intelligent control of power station coal-fired boilers, and relates to a method and a system for intelligently controlling a power station coal-fired boiler based on neural network prediction.
Background
Along with the development of modern control theory and big data analysis and other technologies in recent years, few units try to adopt a new control strategy to carry out optimization and transformation in a small range, but the controller in the aspect is basically in the stages of theoretical research and trial in a small range, a few of long-term stable operation can be realized in practical application, and the optimization effect is very limited by the factors of variable working conditions of load coal quality and the like, and the two existing control modes have no universality.
(1) The large thermal power generating unit has the characteristics of large delay, non-linearity in the production process and the like in control characteristics, equipment characteristics generally change along with the increase of the operation time of the unit, the stability of partial working conditions of the traditional PID controller can be realized only by the time delay time-varying characteristics, and PID parameters and feedforward action quantity need to be continuously updated and adjusted manually along with the time lapse and the working condition change, so that the DCS maintenance quantity is increased, and the safety risk of the operation process is increased. Meanwhile, the PID control parameter setting has higher requirements on the experience background of manpower resources, and the traditional PID controller is greatly limited to exert the optimal control effect on the site.
(2) Although the theoretical basis of a new control strategy based on the technologies such as modern control theory, big data analysis and the like is advanced, the modeling process is too complex, accurate model parameters are needed, configuration parameters are numerous, many parameters need to be tested and set, and online self-correction cannot be achieved. Some models need to rely on huge historical data, training time is long, convergence is difficult, and the models can only be trained offline and cannot be updated online. This set of drawbacks results in many theoretically advanced models not being directly applied in engineering or being forced to exit due to model mismatch after a period of operation.
Disclosure of Invention
In order to solve the defects in the prior art, the method and the system for intelligently controlling the coal-fired boiler of the power station based on the neural network prediction are provided, the neural network prediction algorithm is used as a core, the rolling historical data analysis is utilized to predict process parameters and correct an online model, the advanced control is realized based on the prediction result, the hidden control rule of the historical data is fully excavated, the method and the system are essentially different from the traditional PID control strategy, the controllable unit state measurable system is ensured, and the unit operation level is greatly improved.
In order to achieve the above object, the first invention of the present application adopts the following technical solutions:
a power station coal-fired boiler intelligent control method based on neural network prediction comprises the following steps:
step 1: analyzing a process mechanism, establishing a predictive control model, and determining input and output of the predictive control model and key model parameters;
step 2: establishing a neural network, training a fitting prediction control model on line, and iteratively updating the weight until a fitting error meets a required range;
and step 3: predicting parameters on the basis of the weight values obtained by each iterative calculation;
and 4, step 4: screening an optimal predicted value based on the parameter prediction curve, calculating an optimal prediction deviation, and calculating an optimal control instruction based on the optimal prediction deviation;
and 5: and acting the optimized control instruction on the field production system through DCS, and performing advanced adjustment on the actuating mechanism in advance, thereby realizing closed-loop control of the system.
The invention further comprises the following preferred embodiments:
preferably, the analyzing the process mechanism in step 1, establishing a predictive control model, and determining input and output of the predictive control model and key model parameters includes the following steps:
step 101: analyzing each influence factor having causal or correlation relation with the regulated quantity according to the constraint of the time delay lag characteristic of the boiler combustion process, and screening the most critical parameter according to the importance degree as the model input, wherein the model input comprises the regulated quantity u;
step 102: and determining the rolling time length, the difference interval and the positive and negative acting directions of each input parameter one by one.
Preferably, in step 102, the scroll time is longer than the delay time of the corresponding model input to the output, and the upper limit is not more than 15 minutes;
the difference interval is taken for 3-5 seconds;
and the positive and negative acting directions are determined according to the actual technological process and are used for restraining the weight change direction in the training process of the predictive control model.
Preferably, in step 2, the neural network comprises an input layer, a hidden layer and an output layer;
the input of the neural network is a differential historical sequence delta X in the rolling time length of q relevant process parameters of a prediction control model acquired from DCSk；
The influence quantity of the input differential pair modulated quantity at each historical moment
Figure BDA0002226886560000021
The number of hidden layer nodes is consistent with the number of input nodes for the weight of each hidden layer node;
and the output of the neural network is a fitting value delta yfit and a predicted value delta ypre of the adjusted quantity difference calculated according to the fitting equation and the prediction equation.
Preferably, the step 2 of establishing a neural network and training a fitting prediction control model on line, and iteratively updating the weight until the fitting error meets the required range includes the following steps:
step 201: difference value delta y of current time by modulated quantityt0To fit the target value, a sequence of differential histories over a scrolling period of x with q inputs
Figure BDA0002226886560000031
Building a BP neural network for input;
step 202: let input xkThe weight sequence corresponding to each hidden layer node is as follows
Figure BDA0002226886560000032
The fitting equation is established as follows:
Figure BDA0002226886560000033
step 203: error of fitting
Figure BDA0002226886560000034
For each weight sequence corresponding to q input x
Figure BDA0002226886560000035
The numerical calculation formula of the sequence is as follows:
Figure BDA0002226886560000036
wherein, λ is a configured learning rate, and is used for adjusting the iteration step length of the weight value, so that the fitting output value quickly approaches the actual value change and is determined by target combined online debugging;
step 204: and repeating the iteration weight and the fitting value based on the formulas of the step 202 and the step 203 until the delta err is less than the epsilon, wherein the epsilon is a fitting error threshold value.
Preferably, in step 2, if the weight iteration result is not consistent with the input positive and negative acting directions, the original value is retained and not updated.
Preferably, the step 3 of predicting the parameters based on the weights obtained by each iterative computation includes the following steps:
step 301: let input xkThe weight sequence obtained by the current iteration is
Figure BDA0002226886560000037
And input sequence DeltaXkPerforming convolution operation to establish a prediction equation, and calculating to obtain a prediction sequence
Figure BDA0002226886560000041
Wherein the prediction equation is:
Figure BDA0002226886560000042
the pre-sequencing column is as follows:
Figure BDA0002226886560000043
step 302: based on predicted sequences
Figure BDA0002226886560000044
Linear conversion is carried out to obtain the final prediction sequence
Figure BDA0002226886560000045
Wherein, the conversion formula is:
Figure BDA0002226886560000046
the final predicted sequence is:
Figure BDA0002226886560000047
preferably, the step 4 of screening an optimal predicted value based on the parameter prediction curve and calculating an optimal control command based on the optimal prediction deviation includes the following steps:
step 401: observing the lead time of the prediction curve and the deviation from the true value, and selecting one prediction curve as the optimal prediction curve for lead control;
step 402: calculating the optimal prediction deviation delta errpre；
Step 403: replacing the actual deviation in the conventional PID control with the optimal predicted deviation to obtain an optimal control instruction:
Figure BDA0002226886560000048
preferably, in step 401, the last strip is selected as the optimal predicted value
Figure BDA0002226886560000049
Preferably, in step 402, the actual value of the current adjusted quantity is set as
Figure BDA00022268865600000410
Set value of
Figure BDA00022268865600000411
For a certain regulating variable, the deviation is predicted if the regulating variable is positive for the regulated variable
Figure BDA0002226886560000051
If it is a negative effect, then
Figure BDA0002226886560000052
The application also discloses another invention, namely a power station coal-fired boiler intelligent control system based on neural network prediction, which comprises an establishing unit, a training unit, a prediction unit, a calculation unit and an adjusting unit;
the establishing unit is used for analyzing a process mechanism, establishing a predictive control model and determining input and output of the predictive control model and key model parameters;
the training unit is used for establishing a neural network, training a fitting prediction control model on line, and iteratively updating the weight until the fitting error meets the required range;
the prediction unit is used for predicting parameters on the basis of the weight values obtained by each iterative calculation;
the calculation unit is used for screening an optimal predicted value based on the parameter prediction curve, calculating an optimal prediction deviation and calculating an optimal control instruction based on the optimal prediction deviation;
and the adjusting unit is used for acting the optimized control instruction on the field production system through the DCS, and performing advanced adjustment by the advanced action executing mechanism, so that closed-loop control of the system is realized.
The beneficial effect that this application reached:
1. the integral control level of the unit is not lower than the national standard or the line standard requirement;
2. the intelligent control strategy has strong adaptability to the working condition, and the parameters can be corrected in an online self-adaptive manner;
3. the method and the device have little dependence on experienced human resources, greatly reduce the on-site debugging workload, and realize maintenance-free system.
Drawings
FIG. 1 is a flow chart of a method for intelligently controlling a coal-fired boiler of a power plant based on neural network prediction according to the application;
FIG. 2 is a flow chart of an embodiment of a neural network prediction-based intelligent control method for a coal-fired boiler of a power plant according to the present application;
FIG. 3 is a schematic diagram of the BP neural network structure of the present application;
FIG. 4 is a block diagram of a power plant coal-fired boiler intelligent control system based on neural network prediction according to the present application.
Detailed Description
The present application is further described below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present application is not limited thereby.
As shown in fig. 1 and fig. 2, the method for intelligently controlling a coal-fired boiler of a power plant based on neural network prediction of the present application includes the following steps:
step 1: analyzing a process mechanism, establishing a predictive control model, and determining input and output of the predictive control model and a small amount of key model parameters; the method comprises the following steps:
step 101: analyzing each influence factor having causal or correlation relation with the regulated quantity according to the constraint of the time delay lag characteristic of the boiler combustion process, and screening 3-5 most critical parameters according to the importance degree as model input, wherein the model input comprises the regulated quantity u;
step 102: and determining the rolling time length, the difference interval and the positive and negative acting directions of each input parameter one by one.
In an embodiment, the scroll duration is slightly longer than the delay time of the input to the output of the corresponding model, and the upper limit is not more than 15 minutes;
the difference interval is taken for 3-5 seconds;
and the positive and negative action directions are determined according to the actual process flow and are used for restricting the weight change direction in the process of training the predictive control model.
Step 2: establishing a neural network, training a fitting prediction control model on line, and iteratively updating the weight until a fitting error meets a required range;
as shown in fig. 3, the neural network includes an input layer, a hidden layer, and an output layer;
the input of the neural network is a differential historical sequence delta X in the rolling time length of q relevant process parameters of a prediction control model acquired from DCSk；
The influence quantity of the input differential pair modulated quantity at each historical moment
Figure BDA0002226886560000061
The number of hidden layer nodes is consistent with the number of input nodes for the weight of each hidden layer node;
and the output of the neural network is a fitting value delta yfit and a predicted value delta ypre of the adjusted quantity difference calculated according to the fitting equation and the prediction equation.
In an embodiment, the establishing a neural network and training a fitting prediction control model on line in step 2, and iteratively updating a weight until a fitting error meets a required range includes the following steps:
step 201: difference value of current time by modulated quantity
Figure BDA0002226886560000062
To fit the target value, a sequence of differential histories over a scrolling period of x with q inputs
Figure BDA0002226886560000063
Establishing a BP neural network for input;
step 202: let input xkThe weight sequence corresponding to each hidden layer node is as follows
Figure BDA0002226886560000064
The fitting equation is established as follows:
Figure BDA0002226886560000065
step 203: error of fitting
Figure BDA0002226886560000071
For each weight sequence corresponding to q input x
Figure BDA0002226886560000072
The numerical calculation formula of the sequence is as follows:
Figure BDA0002226886560000073
wherein, λ is a configured learning rate, and is used for adjusting the iteration step length of the weight value, so that the fitting output value quickly approaches the actual value change and is determined by target combined online debugging;
step 204: and repeating the iteration weight and the fitting value based on the formulas of the step 202 and the step 203 until the delta err is less than the epsilon, wherein the epsilon is a fitting error threshold value.
In the embodiment, if the weight iteration result is inconsistent with the input positive and negative action directions, the original value is reserved and is not updated.
And step 3: predicting parameters on the basis of the weight values obtained by each iterative calculation; the method comprises the following steps:
step 301: let input xkThe weight sequence obtained by the current iteration is
Figure BDA0002226886560000074
And input sequence DeltaXkPerforming convolution operation to establish a prediction equation, and calculating to obtain a prediction sequence
Figure BDA0002226886560000075
Wherein the prediction equation is:
Figure BDA0002226886560000076
the pre-sequencing column is as follows:
Figure BDA0002226886560000077
step 302: based on predicted sequences
Figure BDA0002226886560000078
Linear conversion is carried out to obtain the final prediction sequence
Figure BDA0002226886560000079
Wherein, the conversion formula is:
Figure BDA00022268865600000710
the final predicted sequence is:
Figure BDA0002226886560000081
and 4, step 4: screening an optimal predicted value based on the parameter prediction curve, calculating an optimal prediction deviation, and calculating an optimal control command based on the optimal prediction deviation; the method comprises the following steps:
step 401: observing the lead time of the prediction curve and the deviation from the true value, selecting one as the optimal prediction curve for lead control, preferably selecting the last as the optimal predicted value
Figure BDA0002226886560000082
Step 402: calculating the optimal prediction deviation delta errpre：
Setting the current regulated quantity actual value as
Figure BDA0002226886560000083
Set value is
Figure BDA0002226886560000084
For a certain regulating variable, the deviation is predicted if the regulating variable is positive for the regulated variable
Figure BDA0002226886560000085
If it is a negative effect, then
Figure BDA0002226886560000086
Step 403: replacing the actual deviation in the conventional PID control with the optimal predicted deviation to obtain an optimal control instruction:
Figure BDA0002226886560000087
and 5: and the optimized control instruction acts on the field production system through DCS, and the advance action executing mechanism performs advance regulation, so that closed-loop control of the system is realized.
As shown in fig. 4, a power plant coal-fired boiler intelligent control system based on neural network prediction comprises an establishing unit, a training unit, a prediction unit, a calculation unit and an adjusting unit;
the establishing unit is used for analyzing a process mechanism, establishing a predictive control model and determining input and output of the predictive control model and a small amount of key model parameters;
the training unit is used for establishing a neural network, training a fitting prediction control model on line, and iteratively updating the weight until the fitting error meets the required range;
the prediction unit is used for predicting parameters on the basis of the weight obtained by each iterative calculation;
the calculation unit is used for screening an optimal predicted value based on the parameter prediction curve, calculating an optimal prediction deviation and calculating an optimal control instruction based on the optimal prediction deviation;
and the adjusting unit is used for acting the optimized control instruction on the field production system through the DCS, and performing advanced adjustment by the advanced action executing mechanism, so that closed-loop control of the system is realized.
In summary, according to the method, historical data of relevant process parameters are collected from the DCS for caching according to mechanism modeling analysis in a period of time, a BP neural network is established on the basis of the historical data for self-adaptive calculation, and the purposes of sensing and learning system changes in real time are achieved by recording stored data and training neural network weights on line.
According to the method and the device, the prediction based on the neural network model ensures the accuracy of process data prediction, the minimum action quantity under the condition of ensuring the optimal control quality is calculated based on the optimal instruction of the prediction result, the controllable allowance of a system is fully ensured, and the optimal energy consumption is achieved.
Finally, the optimized control instruction acts on the field system through the DCS executing mechanism to carry out closed-loop control, and meanwhile, the control output participates in subsequent prediction adaptive correction and optimal instruction calculation in a circulating mode, so that the system can quickly follow the change of a set value, and the optimal control quality is guaranteed to be realized under various working conditions.
The present applicant has described and illustrated embodiments of the present invention in detail with reference to the accompanying drawings, but it should be understood by those skilled in the art that the above embodiments are merely preferred embodiments of the present invention, and the detailed description is only for the purpose of helping the reader to better understand the spirit of the present invention, and not for limiting the scope of the present invention, and on the contrary, any improvement or modification made based on the spirit of the present invention should fall within the scope of the present invention.

Claims (9)

1. A power station coal-fired boiler intelligent control method based on neural network prediction is characterized in that:
the method comprises the following steps:
step 1: analyzing a process mechanism, establishing a predictive control model, and determining input and output of the predictive control model and key model parameters;
step 2: establishing a neural network, training a fitting prediction control model on line, and iteratively updating the weight until the fitting error meets the required range;
and step 3: predicting parameters on the basis of the weight values obtained by each iterative calculation;
and 4, step 4: screening an optimal predicted value based on the parameter prediction curve, calculating an optimal prediction deviation, and calculating an optimal control instruction based on the optimal prediction deviation;
and 5: the optimized control instruction acts on the field production system through DCS, and the advance action executing mechanism performs advance regulation, so that closed-loop control of the system is realized;
in step 2, the neural network comprises an input layer, a hidden layer and an output layer;
the input of the neural network is a differential historical sequence delta X in the rolling duration of q relevant process parameters of a prediction control model acquired from DCSk；
The influence quantity of the input differential pair modulated quantity at each historical moment
Figure FDA0003599669470000011
The number of hidden layer nodes is consistent with the number of input nodes for the weight of each hidden layer node;
the output of the neural network is a fitting value delta yfit and a predicted value delta ypre of the adjusted quantity difference calculated according to the fitting equation and the prediction equation;
step 2, establishing a neural network and training a fitting prediction control model on line, iteratively updating the weight until the fitting error meets the required range, and the method comprises the following steps:
step 201: difference value delta y of current time by modulated quantityt0To fit the target value, a sequence of differential histories over a scrolling period of x with q inputs
Figure FDA0003599669470000012
Building a BP neural network for input;
step 202: let input xkThe weight sequence corresponding to each hidden layer node is as follows
Figure FDA0003599669470000013
The fitting equation is established as follows:
Figure FDA0003599669470000014
step 203: error of fitting
Figure FDA0003599669470000021
For each weight sequence corresponding to q input x
Figure FDA0003599669470000022
The numerical calculation formula of the sequence is as follows:
Figure FDA0003599669470000023
wherein, λ is a configured learning rate, and is used for adjusting the iteration step length of the weight value, so that the fitting output value quickly approaches the actual value change and is determined by target combined online debugging;
step 204: and repeating the iteration weight and the fitting value based on the formulas of the step 202 and the step 203 until the delta err is less than the epsilon, wherein the epsilon is a fitting error threshold.
2. The intelligent control method for the coal-fired boiler of the power plant based on the neural network prediction as claimed in claim 1, characterized in that:
the process mechanism analysis in the step 1, the establishment of a predictive control model and the determination of input and output of the predictive control model and key model parameters comprise the following steps:
step 101: analyzing each influence factor having causal or correlation relation with the regulated quantity according to the constraint of the time delay lag characteristic of the boiler combustion process, and screening the most critical parameter according to the importance degree as the model input, wherein the model input comprises the regulated quantity u;
step 102: and determining the rolling time length, the difference interval and the positive and negative acting directions of each input parameter one by one.
3. The intelligent control method for the coal-fired boiler of the power plant based on the neural network prediction as claimed in claim 2, characterized in that:
in step 102, the rolling duration is greater than the delay time of the input to the output of the corresponding model, and the upper limit is not more than 15 minutes;
the difference interval is taken for 3-5 seconds;
and the positive and negative acting directions are determined according to the actual technological process and are used for restraining the weight change direction in the training process of the predictive control model.
4. The intelligent control method for the coal-fired boiler of the power plant based on the neural network prediction as claimed in claim 1, characterized in that:
in step 2, if the weight iteration result is inconsistent with the input positive and negative action directions, the original value is reserved and is not updated.
5. The intelligent control method for the coal-fired boiler of the power plant based on the neural network prediction as claimed in claim 1, characterized in that:
step 3, predicting parameters on the basis of the weights obtained by each iterative computation, comprising the following steps:
step 301: let input xkThe weight sequence obtained by the current iteration is
Figure FDA0003599669470000031
And input sequence DeltaXkPerforming convolution operation to establish a prediction equation, and calculating to obtain a prediction sequence
Figure FDA0003599669470000032
Wherein the prediction equation is:
Figure FDA0003599669470000033
the pre-sequencing column is as follows:
Figure FDA0003599669470000034
step 302: based on predicted sequences
Figure FDA0003599669470000035
Linear conversion is carried out to obtain the final prediction sequence
Figure FDA0003599669470000036
Wherein, the conversion formula is:
Figure FDA0003599669470000037
the final predicted sequence is:
Figure FDA0003599669470000038
6. the intelligent control method for the coal-fired boiler of the power plant based on the neural network prediction as claimed in claim 1, characterized in that:
and 4, screening an optimal predicted value based on the parameter prediction curve, and calculating an optimal control instruction based on the optimal prediction deviation, wherein the method comprises the following steps:
step 401: observing the lead time of the prediction curve and the deviation from the true value, and selecting one prediction curve as the optimal prediction curve for lead control;
step 402: calculating an optimal prediction deviation Δ errpre；
Step 403: replacing the actual deviation in the conventional PID control with the optimal predicted deviation to obtain an optimal control instruction:
Figure FDA0003599669470000041
7. the intelligent control method for the coal-fired boiler of the power plant based on the neural network prediction as claimed in claim 6, characterized in that:
in step 401, the last strip is selected as the optimal predicted value
Figure FDA0003599669470000042
8. The intelligent control method for the coal-fired boiler of the power plant based on the neural network prediction as claimed in claim 6, characterized in that:
in step 402, the actual value of the current adjusted quantity is set as
Figure FDA0003599669470000043
Set value is
Figure FDA0003599669470000044
For a certain adjustment quantity, the deviation is predicted if the adjustment quantity is positive for the adjusted quantity
Figure FDA0003599669470000045
If the effect is negative, then
Figure FDA0003599669470000046
Wherein the content of the first and second substances,
Figure FDA0003599669470000047
and the optimal predicted value is obtained.
9. The utility model provides a power plant coal fired boiler intelligence control system based on neural network prediction which characterized in that:
the system comprises an establishing unit, a training unit, a prediction unit, a calculation unit and an adjusting unit;
the establishing unit is used for analyzing a process mechanism, establishing a predictive control model and determining input and output of the predictive control model and key model parameters;
the training unit is used for establishing a neural network, training a fitting prediction control model on line, and iteratively updating the weight until a fitting error meets a required range;
the prediction unit is used for predicting parameters on the basis of the weight values obtained by each iterative calculation;
the calculation unit is used for screening an optimal predicted value based on the parameter prediction curve, calculating an optimal prediction deviation and calculating an optimal control instruction based on the optimal prediction deviation;
the adjusting unit is used for acting the optimized control instruction on the field production system through the DCS, and performing advanced adjustment by the advanced action executing mechanism so as to realize closed-loop control of the system;
in the training unit, the neural network comprises an input layer, a hidden layer and an output layer;
the input of the neural network is a differential historical sequence delta X in the rolling time length of q relevant process parameters of a prediction control model acquired from DCSk；
The influence quantity of the input differential pair modulated quantity at each historical moment
Figure FDA0003599669470000051
The number of hidden layer nodes is consistent with the number of input nodes for the weight of each hidden layer node;
the output of the neural network is a fitting value delta yfit and a predicted value delta ypre of the adjusted quantity difference calculated according to the fitting equation and the prediction equation;
the method for establishing the neural network and training the fitting prediction control model on line and iteratively updating the weight until the fitting error meets the required range comprises the following steps of:
1): difference value of current time by modulated quantity
Figure FDA0003599669470000052
To fit the target value, a sequence of differential histories over a scrolling period of x with q inputs
Figure FDA0003599669470000053
Establishing a BP neural network for input;
2): let input xkThe weight sequence corresponding to each hidden layer node is as follows
Figure FDA0003599669470000054
The fitting equation is established as follows:
Figure FDA0003599669470000055
3): error of fitting
Figure FDA0003599669470000056
For each weight sequence corresponding to q input x
Figure FDA0003599669470000057
The calculation formula of each numerical value of the sequence is as follows:
Figure FDA0003599669470000058
wherein, λ is a configured learning rate, and is used for adjusting the iteration step length of the weight value, so that the fitting output value quickly approaches the actual value change and is determined by target combined online debugging;
4): and (3) repeating iteration weight and fitting value based on the formulas of 2) and 3) until the delta err is less than the epsilon, wherein the epsilon is a fitting error threshold value.
CN201910954643.2A 2019-10-09 2019-10-09 Power station coal-fired boiler intelligent control method and system based on neural network prediction Active CN110673482B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910954643.2A CN110673482B (en) 2019-10-09 2019-10-09 Power station coal-fired boiler intelligent control method and system based on neural network prediction

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910954643.2A CN110673482B (en) 2019-10-09 2019-10-09 Power station coal-fired boiler intelligent control method and system based on neural network prediction

Publications (2)

Publication Number Publication Date
CN110673482A CN110673482A (en) 2020-01-10
CN110673482B true CN110673482B (en) 2022-07-05

Family

ID=69081197

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910954643.2A Active CN110673482B (en) 2019-10-09 2019-10-09 Power station coal-fired boiler intelligent control method and system based on neural network prediction

Country Status (1)

Country Link
CN (1) CN110673482B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112016754A (en) * 2020-08-31 2020-12-01 哈电发电设备国家工程研究中心有限公司 Power station boiler exhaust gas temperature advanced prediction system and method based on neural network
CN113091088B (en) * 2021-04-14 2022-10-25 南京邮电大学 Boiler combustion generalized predictive control method based on two-stage neural network
CN113239565B (en) * 2021-05-27 2022-07-12 中南大学 Soft measurement method and device for product quality of fluidized bed roaster

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101201591A (en) * 2006-12-11 2008-06-18 株式会社日立制作所 Control device for boiler equipment and gas concentration concluding apparatus
CN101498459A (en) * 2009-03-02 2009-08-05 杭州电子科技大学 Modeling method for boiler combustion optimization
CN101581454A (en) * 2009-06-09 2009-11-18 南通大学 Industrial boiler combustion controlling system adopting predictive functional controller based on characteristic model

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150260398A1 (en) * 2013-11-12 2015-09-17 SYNCRUDE CANADA LTD. in trust for the owners of the Syncrude Project, as such owners exist now and Furnace combustion cross limit control with real-time diagnostic features

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101201591A (en) * 2006-12-11 2008-06-18 株式会社日立制作所 Control device for boiler equipment and gas concentration concluding apparatus
CN101498459A (en) * 2009-03-02 2009-08-05 杭州电子科技大学 Modeling method for boiler combustion optimization
CN101581454A (en) * 2009-06-09 2009-11-18 南通大学 Industrial boiler combustion controlling system adopting predictive functional controller based on characteristic model

Also Published As

Publication number Publication date
CN110673482A (en) 2020-01-10

Similar Documents

Publication Publication Date Title
CN110673482B (en) Power station coal-fired boiler intelligent control method and system based on neural network prediction
CN110285403B (en) Main steam temperature control method based on controlled parameter estimation
CN104181895A (en) Strategy for optimizing short-term and ultra-short-term coordination rolling schedules adapting to access of new energy resources
CN113623719B (en) Heat exchange station prediction control method based on effective room temperature detection
CN112286057A (en) Coal amount optimizing and predicting control method based on AGC optimization of thermal power plant
CN106773682A (en) Based on the glass furnace bottom of pond temperature intelligent forecast Control Algorithm that time lag is dynamically determined
CN112330012B (en) Building energy consumption prediction method and equipment based on transfer learning
WO2019085446A1 (en) Self-order-reducing multi-loop centralized prediction control system for reheat steam temperature of double reheat unit
CN101709867B (en) Hybrid control method for drum water level system of coal-fired boiler
CN104456513B (en) A kind of thermal power plant Stream temperature estimation optimizing control method
CN112016754A (en) Power station boiler exhaust gas temperature advanced prediction system and method based on neural network
CN101709863A (en) Hybrid control method for furnace pressure system of coal-fired boiler
CN105652666B (en) Large-scale drop press upper beam prediction of speed control method based on BP neural network
CN114721263A (en) Intelligent regulation and control method for cement decomposing furnace based on machine learning and intelligent optimization algorithm
CN103605284B (en) The cracking waste plastics stove hearth pressure control method that dynamic matrix control is optimized
CN114928049A (en) Feedforward predictive control method and system for thermal power generating unit
CN105511270B (en) A kind of PID controller parameter optimization method and system based on coevolution
CN116880191A (en) Intelligent control method of process industrial production system based on time sequence prediction
CN112947606A (en) Boiler liquid level control system and method based on BP neural network PID predictive control
CN111998383A (en) Over-fire air control method based on boiler load and flame central point coordinate quantization
CN216772212U (en) Two-furnace one-chimney desulfurization intelligent control system
CN111663032B (en) Active disturbance rejection temperature control method for amorphous iron core annealing furnace
CN114326395B (en) Intelligent generator set control model online updating method based on working condition discrimination
Schlag et al. Neural networks for process control in steel manufacturing
Zhang et al. Research on delay prediction compensation of teleoperation robot network control system

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
CP03 Change of name, title or address

Address after: 102299 rooms 618 and 619, 6 / F, building 301, Guodian New Energy Institute, No. 9, Yingcai North 2nd Street, future science and Technology City, Changping District, Beijing

Patentee after: Guoneng xinkong Internet Technology Co.,Ltd.

Address before: Building 1, yard 16, West Fourth Ring Middle Road, Haidian District, Beijing 100039

Patentee before: BEIJING HUADIAN TIANREN ELECTRIC POWER CONTROL TECHNOLOGY Co.,Ltd.

CP03 Change of name, title or address