CN113779858B - Combustion optimization method and system, storage medium and electronic equipment - Google Patents
Combustion optimization method and system, storage medium and electronic equipment Download PDFInfo
- Publication number
- CN113779858B CN113779858B CN202010522836.3A CN202010522836A CN113779858B CN 113779858 B CN113779858 B CN 113779858B CN 202010522836 A CN202010522836 A CN 202010522836A CN 113779858 B CN113779858 B CN 113779858B
- Authority
- CN
- China
- Prior art keywords
- combustion
- neural network
- layer
- connection weight
- network model
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/04—Constraint-based CAD
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses a combustion optimization method, a system for optimizing a combustion industrial process, a storage medium and an electronic device. The combustion optimization method comprises the following steps: s1, acquiring original combustion data and constructing a combustion neural network model; s2, initializing a neural network model connection weight; s3, updating the connection weight of each layer based on the backward propagation process; s4, obtaining adjusted input and output parameters based on a forward propagation process; s5, judging the adjusted input and output parameters based on the combustion mechanism, and if the adjusted input and output parameters accord with the combustion mechanism, taking the adjusted corresponding weight as a third connection weight and executing the step S6; if the combustion mechanism is not met, replacing the second connection weight in the step S3 with the adjusted corresponding connection weight, and continuing to execute the step S4 and the step S5; s6, carrying out iterative computation until the root-mean-square error and the decision coefficient meet a specified threshold or the iteration meets specified times, and obtaining an optimized neural network model; and S7, optimizing the combustion.
Description
Technical Field
The invention relates to the technical field of industrial information and control, in particular to a method, a system, a storage medium and electronic equipment for optimizing a combustion industrial process by adopting a neural network model.
Background
The combustion of the coal-fired unit is a complex physical and chemical process, part of the process can be explained through a mechanism model, the complex process reflected by modeling of original operation data through a neural network is fuzzy, and the trend of model sensitivity analysis and excavation is not completely consistent with mechanism knowledge inference under the condition that the data cannot cover the whole process.
The mechanism model of the industrial process usually can not accurately reflect the complex industrial process, the combustion of the coal-fired unit involves the processes of chemical reaction, heat transfer and the like, and meanwhile, different air doors and coal amount can affect the combustion effect to different degrees. With the development of machine learning technology, the reflecting of complex industrial processes through neural network modeling becomes a trend, but the structure of a 'black box' model of a neural network is influenced by super parameters such as a model structure, initialization weight, iteration times and the like, and uncertain factors exist in a fitting or classification result.
Although the mechanism model cannot accurately reflect the whole complex process, the mechanism knowledge can reflect the trend of partial processes. Taking the industrial combustion process of a coal-fired unit as an example, no accurate mechanism model exists at present to reflect the combustion process, but the rules of the primary air door, the secondary air door and the coal quantity on the combustion efficiency and the pollutant generation are clear.
In the prior art, a decision method suitable for a text sequence is provided, but the text sequence is different from structural data of industrial operation, the context in a text statement has strong correlation, and a neural network optimization method aiming at the characteristic cannot be extended to the structural data, so that the method is not suitable for the construction of the neural network of the industrial data. The invention also provides a neural network optimization mechanism, which limits the NN topology to have one or more macro layers; adjusting the one or more macro layers to fit input and output components of the NN; and train the NN based on the one or more macro layers. However, the invention aims to optimize and improve the operation efficiency of the neural network, and does not relate to the accuracy of the neural network model and the conformity degree with the actual operation process.
In the current combustion industrial process of the coal-fired unit, the condition that the output parameter trend discovered by the sensitivity analysis of the existing neural network model is not completely consistent with the inference of the combustion mechanism knowledge exists, the iterative process of the neural network needs to be constrained by the law of the combustion mechanism knowledge to obtain the neural network model which accords with the actual combustion industrial process, so that the technology or the operator is guided to carry out the matching adjustment of the production parameters in the combustion industrial process, the combustion industrial process of the coal-fired unit is optimized to obtain better output, the combustion efficiency is improved, and the environmental pollution is reduced.
Disclosure of Invention
By combining the rules of mechanism knowledge, the invention adds the updating condition of the constraint parameters in the neural network iteration process, and solves the technical problem that the goodness of fit of the neural network model and the combustion process is not matched in the industrial operation process.
The invention provides a combustion optimization method, which comprises the following steps:
s1, acquiring original combustion data, and constructing a combustion neural network model based on the combustion data, wherein the neural network model comprises an input layer, a hidden layer and an output layer, and the original combustion data comprises original input parameter data and original output parameter data;
s2, initializing the neural network model weight to obtain the initial connection weight of each layer of the model;
s3, updating the connection weight of each layer in the model based on a back propagation process to be used as a second connection weight;
s4, carrying out forward propagation calculation on the neural network model based on a forward propagation process to obtain an input parameter and an output parameter after adjustment of the model;
s5, judging the adjusted input parameters and the adjusted output parameters based on the combustion mechanism, and if the adjusted input parameters and the adjusted output parameters accord with the combustion mechanism, taking weights corresponding to the adjusted input parameters and the adjusted output parameters as third connection weights, and executing a step S6;
if the adjusted input parameters and the adjusted output parameters do not accord with the combustion mechanism, replacing the second connection weight in the step S3 with the connection weight corresponding to the adjusted input parameters and the adjusted output parameters, and continuing to execute the step S4 and the step S5;
s6, carrying out iterative computation on the third connection weight until the root mean square error and the decision coefficient of the verification data set meet a specified threshold or the iteration meets a specified number of times, and then carrying out iterative computation to obtain an optimized neural network model reflecting a real combustion mechanism process;
and S7, optimizing combustion based on the optimized neural network model.
In an embodiment of the present invention in which,
updating the connection weight of each layer in the model as a second connection weight based on a back propagation process, comprising the steps of:
obtaining a predicted value of the output layer output parameter based on initializing each connection weight of the neural network model;
obtaining a loss function error based on the output layer output parameter prediction value and the original output parameter of the output layer;
and calculating the partial derivative of the error to each connection weight of the initialized hidden layer and the initialized input layer based on the loss function error, and obtaining the second connection weight based on the partial derivative of each connection weight.
In an embodiment of the present invention in which,
the method for calculating the forward propagation of the neural network model based on the forward propagation process to obtain the adjusted input parameters and the adjusted output parameters of the model comprises the following steps:
after the nodes of the neural network input layer and each node of the first layer of hidden layer are weighted, the weighted nodes are transmitted to the second layer of hidden layer after being calculated through an activation function;
after weighting the node of the N-th hidden layer of the neural network and each node of the (N + 1) -th hidden layer, transmitting the weighted node to an output layer after calculation of an activation function, wherein N is more than or equal to 2;
and the neural network output layer obtains the nitrogen oxide concentrations corresponding to different ranges of each input parameter by adopting a normalized exponential function.
In an embodiment of the present invention, it is,
the judgment based on the combustion mechanism comprises the following steps:
construction of the Combustion mechanism function y j ＝＜k·x i >. Wherein, x i The method comprises the following steps of (1) loading, primary air volume, secondary air volume, SOFA air door opening degree of each layer, OFA air door opening degree and total coal quantity parameters; y is j As the concentration of nitrogen oxides formed;
setting a calculation formula of a combustion mechanism judgment rule, wherein the calculation formula comprises:
and eliminating the connection weight of the calculation result of the forward process in the data set which is not in accordance with the combustion mechanism judgment rule in batches, and keeping the consistent connection weight to obtain the third connection weight.
In an embodiment of the present invention in which,
the time for acquiring the original data is 1 minute and is acquired once;
the number of the hidden layers is at least 8;
the specified times when the iteration meets the specified times are more than or equal to 50;
the calculation formula of the activation function Sigmoid is:
in an embodiment of the present invention in which,
the original combustion data comprises original input data and original output data, wherein the original input data comprises load, primary air volume, total coal volume, SOFA air door opening degree of each layer and OFA air door opening degree, and the original output data comprises nitrogen oxide concentration.
In an embodiment of the present invention, it is,
the root mean square error is calculated as:
the calculation formula of the decision coefficient is:
The present invention also provides a system for optimizing a combustion industrial process, comprising:
the data acquisition module is used for acquiring original combustion data required in the neural network model;
the neural network module is used for establishing a neural network model based on the original combustion data, initializing the neural network model and obtaining initial connection weight of each layer of the model, and comprises an input layer, a hidden layer and an output layer;
the backward propagation module is used for updating the connection weight of each layer in the model as a second connection weight based on a backward propagation process;
a forward propagation module: the neural network model is used for carrying out forward propagation calculation on the neural network model based on a forward propagation process to obtain the adjusted input parameters and the adjusted output parameters of the model;
a combustion mechanism module: judging the adjusted input parameters and the adjusted output parameters based on the combustion mechanism, and if the adjusted input parameters and the adjusted output parameters accord with the combustion mechanism, taking the weights corresponding to the adjusted input parameters and the adjusted output parameters as third connection weights, and transferring the third connection weights to an iterative computation module;
if the adjusted input parameters and output parameters do not accord with the combustion mechanism, replacing the second connection weight in the backward propagation module with the connection weight corresponding to the adjusted input parameters and output parameters, and then transferring to the forward propagation module and the combustion mechanism module;
an iterative computation module: and the third connection weight is used for carrying out iterative computation until the root mean square error and the decision coefficient of the verification data set meet a specified threshold or the iteration meets a specified number of times, then carrying out iterative computation to obtain an optimized neural network model reflecting the real combustion mechanism process, and optimizing the combustion based on the optimized neural network model.
The present invention also provides a storage medium having a computer program stored thereon, wherein the computer program is used for implementing the steps of the neural network model building method according to any one of the above contents when being executed by a processor.
The present invention also provides an electronic device, comprising:
a memory having a computer program stored thereon; and
a processor for executing the computer program in the memory to implement the steps of the neural network model construction method of any one of the above.
One or more embodiments of the present invention may have the following advantages over the prior art:
the invention establishes the neural network model by combining the combustion mechanism knowledge and the BP algorithm, and adds the mechanism constraint condition of the combustion industrial process in the weight calculation BP algorithm iteration process, so that the connection weight gradually conforms to the rules of the actual process in the iteration process, the adaptability of the BP iteration algorithm to the training data is ensured, the stability of the model in an excessive state is improved, the generalization capability of the model in an excessive working condition is improved, the technology or an operator is also guided to carry out the matching adjustment of the production parameters of the combustion industrial process, the combustion industrial process of the coal-fired unit is optimized, the better output is obtained, the combustion efficiency is improved, and the environmental pollution is reduced.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by the practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention. In the drawings:
FIG. 1 is a schematic diagram of a neural network model construction and combustion optimization process of the present invention;
FIG. 2 is a schematic diagram of a neural network model structure based on the knowledge of combustion mechanism;
FIG. 3 is a schematic diagram of a neural network optimization flow based on knowledge of combustion mechanisms according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a SOFA damper opening sensitivity analysis of layer 2 without adding combustion mechanism determination rules in accordance with an embodiment of the present invention;
FIG. 5 is a schematic diagram of a layer 2 SOFA damper opening sensitivity analysis incorporating rules for determining a combustion mechanism in accordance with embodiments of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the following detailed description of the present invention is made with reference to the accompanying drawings, so that how to apply technical means to solve the technical problems and achieve the technical effects can be fully understood and implemented. It should be noted that, as long as there is no conflict, the embodiments and the features of the embodiments of the present invention may be combined with each other, and the technical solutions formed are within the scope of the present invention.
First embodiment
FIG. 1 is a schematic diagram of a neural network model construction and combustion optimization process of the present invention;
FIG. 2 is a schematic diagram of a neural network model structure based on the knowledge of combustion mechanism;
FIG. 3 is a schematic diagram of a neural network optimization flow based on knowledge of combustion mechanisms according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a SOFA damper opening sensitivity analysis of layer 2 without adding combustion mechanism determination rules in accordance with an embodiment of the present invention;
FIG. 5 is a schematic diagram of a SOFA damper opening sensitivity analysis of layer 2 incorporating a combustion mechanism determination rule according to an embodiment of the present invention.
The invention provides a neural network model construction method based on a combustion mechanism, which comprises the following steps:
s1, acquiring original combustion data, and constructing a combustion neural network model based on the combustion data, wherein the neural network model comprises an input layer, a hidden layer and an output layer, and the original combustion data comprises original input parameter data and original output parameter data;
s2, initializing the neural network model weight to obtain the initial connection weight of each layer of the model;
s3, updating the connection weight of each layer in the model based on a back propagation process to be used as a second connection weight;
s4, performing forward propagation calculation on the neural network model based on a forward propagation process to obtain an input parameter and an output parameter after adjustment of the model;
s5, judging the adjusted input parameters and the adjusted output parameters based on the combustion mechanism, and if the adjusted input parameters and the adjusted output parameters accord with the combustion mechanism, taking the weights corresponding to the adjusted input parameters and the adjusted output parameters as third connection weights, and executing the step S6;
if the adjusted input parameters and the adjusted output parameters do not accord with the combustion mechanism, replacing the second connection weight in the step S3 with the connection weight corresponding to the adjusted input parameters and the adjusted output parameters, and continuing to execute the step S4 and the step S5;
s6, carrying out iterative computation on the third connection weight until the root mean square error and the decision coefficient of the verification data set meet a specified threshold or the iteration meets a specified number of times, and then carrying out iterative computation to obtain an optimized neural network model reflecting a real combustion mechanism process;
and S7, optimizing combustion based on the optimized neural network model.
Description of the nouns: OFA is over-fire air, SOFA is air-out over-fire air, loss is a Loss function, and SCR is a denitration system.
Specifically, a combustion neural network model based on a machine learning algorithm is established by adopting 3-month running data of a certain 300MW coal-fired unit, the combustion industrial process of the combustion unit is optimized, and the combustion parameter adjustment of the coal-fired unit is guided.
The method comprises the steps of S1, obtaining original combustion data including original input parameter data and original output parameter data, and constructing a combustion neural network model based on the combustion data, wherein the neural network model comprises an input layer, a hidden layer and an output layer.
The method comprises the steps that a coal-fired unit distributed control system DCS and a plant-level monitoring information system SIS in a data acquisition system of a power plant acquire original data set data of about 43000 groups of parameters in a neural network model, 100 groups of data are randomly selected from the original data set data to serve as test verification set data, and the rest data serve as training set data to be used for machine learning training.
Each parameter raw data comprises a model input data set x ij I ∈ {1.. N }, j ∈ {1.. M } and output data y j J is in the range of {1.. M }. n is a data set which is used for collecting input data, the input parameter data comprises load, total air volume, total secondary air volume, coal feed quantity of each layer of coal mills and air door opening degrees of primary air, secondary air, SOFA air and OFA air of each layer, model output variable parameters are NOx concentration at a side inlet B of the SCR system, and the collection time interval is 1 minute and is m minutes in total.
Building neural networks from combustion characteristics
The neural network structure includes an input layer
Hidden layer H 1...t :{h ks,s },ks>10,s e {1.. T } and output layer O: y j A t-layer hidden layer structure is adopted. Wherein h is ks,s Parameters corresponding to the s-th hidden layer; ks represents the number of hidden layer elements in the s-th layer, ks > 108230t, t > 8, and t is preferably set to 15 in this embodiment.
And S2, initializing the weight of the neural network model to obtain the initial connection weight of each layer of the model.
Setting a neural network according to the data content, and initializing connection weight omega of each layer of the neural network ij ,β jk ,......，
Wherein, ω is ij For the connection weight of the ith element of the input layer and the jth element of the first layer hidden layer,
β jk weights are connected for the jth element of the second layer hidden layer and the kth element of the third layer hidden layer.
The number of the connection weights depends on the selection of the hidden layer number t, and so on, the embodiment is preferably set to 15 hidden layers, and after the data are sequentially substituted into the neural network structure for initialization calculation, a predicted value of the NOx concentration after the neural network model is initialized can be obtained.
And S3, updating the connection weight of each layer in the model as a second connection weight based on the back propagation process.
Firstly, a predicted value of an output layer output parameter is obtained based on initializing each connection weight of the neural network model, and a predicted value of NOx concentration output by the neural network model is obtained.
Then obtaining a loss function error based on the output parameter predicted value of the output layer and the original output parameter of the output layer;
according to the actual output NOx concentration data in the training set and the NOx concentration predicted value of the neural network structure, calculating a back propagation process, and according to a Loss function Loss calculation formula Loss = Σ (y-y') 2 And calculating errors, wherein y is the real output NOx concentration data in the training set, and y' is the predicted value of the NOx concentration of the neural network structure.
And then, acquiring second connection weight based on the partial derivative of the Loss error to the connection weight of the hidden layer and the input layer.
And solving partial derivatives delta omega and delta beta of the Loss error to the weights of the hidden layer and the input layer, and updating the connection weights omega '= omega + delta omega and beta' = beta + delta beta of each layer to obtain a second connection weight.
And S4, carrying out forward propagation calculation on the neural network model based on a forward propagation process to obtain the adjusted input parameters and the adjusted output parameters of the model.
After updating the connection weight of each layer, ensuring that other input parameter data in the data are stable and unchanged, adjusting the opening degree value of the SOFA door of the layer 2 from 0 to 100 percent, performing a forward propagation process by a neural network, and continuously calculating the nitrogen oxide concentration of the SOFA door of the layer 2 under the new connection weight within different ranges, wherein the method comprises the following steps:
after the nodes of the neural network input layer and each node of the first hidden layer are weighted, the weighted nodes are transmitted to the second hidden layer after being calculated through an activation function s igmo id, and the calculation formula of the Sigmoid function is as follows:
after weighting the node of the N-th hidden layer of the neural network and each node of the N + 1-th hidden layer, transmitting the weighted node to an output layer after calculation of an activation function, wherein N is more than or equal to 2; specifically, after the nodes of the second hidden layer and each node of the third hidden layer of the neural network are weighted, the subsequent hidden layers are sequentially weighted through sigmoid activation function calculation and then transmitted to the output layer;
the neural network output layer obtains the nitrogen oxide concentrations of the 2 nd layer SOFA air door opening parameter in different ranges through a normalization index function softmax, and the calculation formula of the softmax function is
Wherein z is i And outputting the ith element of the second top layer, and obtaining a curve of the predicted value of the concentration of the NOx.
If the input parameter data of the main operation at a certain moment comprises unit load 296MW, total secondary air quantity 881.91kNm3/h, oxygen quantity 3.12% at the previous moment, total air quantity 881.91kNm3/h, total coal quantity 122.2t/h, opening degree of each layer of secondary air door: 49.25%,29.79%,31.63%,29.75%,29.4% by weight, 11.96%, each primary damper opening: 73.56%,75.24%,72.2%,74.26%,0%,0%.
The calculation results from this time data are shown in FIG. 4, which is a 2-layer SOFA wind sensitivity analysis, and the blue curve is the influence of the 2 nd SOFA damper opening of 0-100% opening on NOx production and the NOx production amount corresponding to the 2 nd SOFA damper opening under a certain condition. The red points in the graph are actual operation data, and the SOFA air door opening of the layer 2 in the historical data is only operated at an opening of more than 95% under the similar working condition, so that the predicted trend of the curve within 90% is deviated from the actual empirical judgment, and the curve is obviously inconsistent with the combustion mechanism knowledge.
S5, judging the adjusted input parameters and the adjusted output parameters based on the combustion mechanism, and if the adjusted input parameters and the adjusted output parameters accord with the combustion mechanism, taking weights corresponding to the adjusted input parameters and the adjusted output parameters as third connection weights, and executing a step S6;
if the adjusted input parameters and output parameters do not accord with the combustion mechanism, replacing the second connection weight in the step S3 with the connection weight corresponding to the adjusted input parameters and output parameters, and continuing to execute the step S4 and the step S5;
specifically, the introduced combustion mechanism judgment rule comprises the following steps:
construction of the Combustion mechanism function y j ＝＜k·x i >. Wherein, x i The method comprises the following steps of (1) loading, primary air volume, secondary air volume, SOFA1-4 layer air door opening degree, OFA air door opening degree and total coal quantity parameters; y is j As the concentration of nitrogen oxides formed;
setting a combustion mechanism judgment rule calculation formula, comprising:
a)
wherein x is i The method comprises a primary air volume and a secondary air volume, wherein y is the concentration of nitrogen oxides;
b)
wherein x is i The method comprises the steps of loading, coal quantity, SOFA1-4 layer air door opening degree and OFA air door opening degree, wherein y is the concentration of nitrogen oxide;
and removing the connection weight which is not in accordance with the combustion mechanism judgment rule in batches, namely removing the result of the forward process calculation in the data set which is in conflict with the judgment rule, and keeping the consistent connection weight to obtain a third connection weight.
From FIG. 4, it can be seen that after the model is iterated this time, the operation result violates the actual combustion mechanism knowledge rule, i.e. the operation result does not conform to the SOFA throttle opening of layer 2 and is in the range of 0-100%
And removing the weight updating result of the data, keeping the weight unchanged, obtaining a third connection weight, and adding the group of operation data into the next iteration.
Step S6, carrying out iterative computation on the third connection weight until the root mean square error and the decision coefficient of the verification data set meet a specified threshold or the iteration meets a specified number of times, and then carrying out iterative computation to obtain an optimized neural network model reflecting the real combustion mechanism process;
after multiple combustion mechanism knowledge rule adjustment and iteration, the preset maximum iteration numerical value is reached for 50 times, or when the decision coefficient R2_ score is more than or equal to 95% and the root mean square error RMSE is less than or equal to 5mg/Nm < 3 >, the iteration updating of the test set data is stopped.
Wherein the root mean square error is calculated as:
wherein y is the concentration of nitrogen oxides output by the original combustion data, and y' is the predicted value of the concentration of nitrogen oxides output by the neural network model;
the calculation formula for the decision coefficient is:
The calculation result is shown in fig. 5, and the result is not only consistent with the knowledge of the combustion mechanism, but also close to the actual situation of the power plant after the verification of the test set. Through the optimization and adjustment of the algorithm, the weight which is contrary to the knowledge trend of the combustion mechanism in the iterative process is removed, and a neural network model representing all working conditions can be obtained.
And S7, optimizing the combustion based on the optimized neural network model.
In each iteration period, each group of data is sequentially input, the weight is updated while the weight of each layer before updating is reserved, forward calculation is carried out and mechanism knowledge rules a) and b) are compared, if the operation result accords with the rules, calculation is carried out continuously, if the operation result does not accord with the rules, the weight before updating is reserved, and the output predicted value in the parameter potential adjustable range can still be accurately deduced under the condition that the historical data of the model after each iteration updating is incomplete under similar working conditions, so that the technology or operating personnel are guided to carry out matching adjustment on the production parameters in the combustion industrial process, the combustion industrial process of the coal-fired unit is optimized to obtain better output, the combustion efficiency is improved, the environmental pollution is reduced, and production accidents are prevented.
In conclusion, the neural network model is established by combining combustion mechanism knowledge and a BP algorithm, mechanism constraint conditions of the combustion industrial process are added in the BP algorithm iteration process of weight calculation, and the weights of random walk are compared and calculated in the forward direction, so that the connection weights gradually accord with the rules of the actual process in the iteration process, the adaptability of the BP iteration algorithm to training data is ensured, the final neural network model accords with the mechanism knowledge rules while being suitable for operation data, the stability of the model in an excessive state is improved, the generalization capability of the model in an excessive working condition is improved, the technology or operators are guided to carry out combustion industrial process production parameter matching adjustment, the combustion industrial process of a coal-fired unit is optimized, better output is obtained, the combustion efficiency is improved, and the environmental pollution is reduced.
Second embodiment
The present embodiments provide a system for optimizing a combustion industrial process, comprising:
the data acquisition module is used for acquiring original combustion data required in the neural network model;
the neural network module is used for establishing a neural network model based on original combustion data, initializing the neural network model to obtain initial connection weights of each layer of the model, and comprises an input layer, a hidden layer and an output layer;
the backward propagation module is used for updating the connection weight of each layer in the model based on the backward propagation process to be used as a second connection weight;
a forward propagation module: the device is used for carrying out forward propagation calculation on the neural network model based on a forward propagation process to obtain an input parameter after model adjustment and an output parameter after adjustment;
a combustion mechanism module: judging the adjusted input parameters and the adjusted output parameters based on the combustion mechanism, and if the adjusted input parameters and the adjusted output parameters accord with the combustion mechanism, taking the weights corresponding to the adjusted input parameters and the adjusted output parameters as third connection weights, and transferring the third connection weights to an iterative computation module;
if the adjusted input parameters and output parameters do not accord with the combustion mechanism, replacing second connection weights in the backward propagation module with connection weights corresponding to the adjusted input parameters and output parameters, and then transferring to the forward propagation module and the combustion mechanism module;
an iterative calculation module: and the third connection weight is used for carrying out iterative computation until the root mean square error and the decision coefficient of the verification data set meet a specified threshold or the iteration meets a specified number of times, then carrying out iterative computation to obtain an optimized neural network model reflecting the real combustion mechanism process, and optimizing the combustion based on the optimized neural network model.
In conclusion, the system for reflecting the combustion industrial process of the neural network model is established by combining the combustion mechanism knowledge and the BP algorithm, the mechanism constraint condition of the combustion industrial process is added in the BP algorithm iteration process of weight calculation, so that the connection weight gradually conforms to the rules of the actual process in the iteration process, the adaptability of the BP iteration algorithm to training data is ensured, the stability of the system in an excessive state is improved, the generalization capability of the system in an excessive working condition is improved, the technology or an operator is guided to perform the production parameter matching adjustment of the combustion industrial process, the combustion industrial process of a coal-fired unit is optimized, better output is obtained, the combustion efficiency is improved, and the environmental pollution is reduced.
Third embodiment
FIG. 1 is a schematic diagram of a neural network model construction and combustion optimization process of the present invention;
the present embodiment provides a storage medium on which a computer program is stored, which when executed by a processor implements the steps of the neural network model construction method described in any one of the above.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
Fourth embodiment
FIG. 1 is a schematic diagram of a neural network model construction and combustion optimization process of the present invention;
the embodiment provides an electronic device, including:
a memory having a computer program stored thereon; and a processor for executing the computer program in the memory to implement the steps of the neural network model construction method of any one of the above.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Although the embodiments of the present invention have been described above, the above description is only for the convenience of understanding the present invention, and is not intended to limit the present invention. It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as disclosed, and that the scope of the invention is not to be limited to the particular embodiments disclosed herein but is to be accorded the full scope of the claims.
Claims (10)
1. A method of combustion optimization, comprising the steps of:
s1, acquiring original combustion data, and constructing a combustion neural network model based on the combustion data, wherein the neural network model comprises an input layer, a hidden layer and an output layer, and the original combustion data comprises original input parameter data and original output parameter data;
s2, initializing the weight of the neural network model to obtain the initial connection weight of each layer of the model;
s3, updating the connection weight of each layer in the model based on a back propagation process to be used as a second connection weight;
s4, carrying out forward propagation calculation on the neural network model based on a forward propagation process to obtain an input parameter and an output parameter after adjustment of the model;
s5, judging the adjusted input parameters and the adjusted output parameters based on the combustion mechanism, and if the adjusted input parameters and the adjusted output parameters accord with the combustion mechanism, taking weights corresponding to the adjusted input parameters and the adjusted output parameters as third connection weights, and executing a step S6;
if the adjusted input parameters and the adjusted output parameters do not accord with the combustion mechanism, replacing the second connection weight in the step S3 with the connection weight corresponding to the adjusted input parameters and the adjusted output parameters, and continuing to execute the step S4 and the step S5;
s6, carrying out iterative computation on the third connection weight until the root mean square error and the decision coefficient of the verification data set meet a specified threshold or the iteration meets a specified number of times, and then carrying out iterative computation to obtain an optimized neural network model reflecting a real combustion mechanism process;
and S7, optimizing combustion based on the optimized neural network model.
2. The combustion optimization method of claim 1,
updating the connection weight of each layer in the model as a second connection weight based on a back propagation process, comprising the steps of:
obtaining a predicted value of the output layer output parameter based on initializing each connection weight of the neural network model;
obtaining a loss function error based on the output layer output parameter prediction value and the original output parameter of the output layer;
and calculating the partial derivative of the error to each connection weight of the initialized hidden layer and the initialized input layer based on the loss function error, and obtaining the second connection weight based on the partial derivative of each connection weight.
3. The combustion optimization method of claim 1, wherein the forward propagation calculation of the neural network model based on the forward propagation process to obtain the adjusted input parameters and the adjusted output parameters of the model comprises the following steps:
after the nodes of the neural network input layer and each node of the first layer of hidden layer are weighted, the weighted nodes are calculated through an activation function and then are transmitted to the second layer of hidden layer;
after the node of the N-th hidden layer of the neural network and each node of the (N + 1) -th hidden layer are weighted, the weighted nodes are transmitted to an output layer after being calculated through an activation function, and N is more than or equal to 2;
and the neural network output layer obtains the nitrogen oxide concentrations corresponding to different ranges of each input parameter by adopting a normalized exponential function.
4. The combustion optimization method of claim 3,
the judgment based on the combustion mechanism comprises the following steps:
construction of the Combustion mechanism function y j ＝＜k·x i > (ii) wherein x i The method comprises the following steps of (1) loading, primary air volume, secondary air volume, SOFA air door opening degree of each layer, OFA air door opening degree and total coal quantity parameters; y is j As the concentration of nitrogen oxides formed;
setting a calculation formula of a combustion mechanism judgment rule, wherein the calculation formula comprises:
and eliminating the connection weight of the calculation result of the forward process in the data set which is not in accordance with the combustion mechanism judgment rule in batches, and keeping the consistent connection weight to obtain the third connection weight.
5. The combustion optimization method of claim 4,
the time for acquiring the original data is 1 minute and is acquired once;
the number of the hidden layers is at least 8;
the specified times when the iteration meets the specified times are more than or equal to 50;
the calculation formula of the activation function Sigmoid is:
6. the combustion optimization method of claim 1,
the original combustion data comprises original input data and original output data, wherein the original input data comprises load, primary air volume, total coal volume, SOFA air door opening degree of each layer and OFA air door opening degree, and the original output data comprises nitrogen oxide concentration.
7. The combustion optimization method of claim 6,
the root mean square error is calculated as:
the calculation formula of the decision coefficient is:
8. A system for optimizing a combustion industrial process, comprising:
the data acquisition module is used for acquiring original combustion data required in the neural network model;
the neural network module is used for establishing a neural network model based on the original combustion data, initializing the neural network model and obtaining initial connection weight of each layer of the model, and comprises an input layer, a hidden layer and an output layer;
the backward propagation module is used for updating the connection weight of each layer in the model based on the backward propagation process to be used as a second connection weight;
a forward propagation module: the neural network model is used for carrying out forward propagation calculation on the neural network model based on a forward propagation process to obtain the adjusted input parameters and the adjusted output parameters of the model;
a combustion mechanism module: judging the adjusted input parameters and the adjusted output parameters based on the combustion mechanism, and if the adjusted input parameters and the adjusted output parameters accord with the combustion mechanism, taking the weights corresponding to the adjusted input parameters and the adjusted output parameters as third connection weights, and switching to an iterative computation module;
if the adjusted input parameters and output parameters do not accord with the combustion mechanism, replacing the second connection weight in the backward propagation module with the connection weight corresponding to the adjusted input parameters and output parameters, and switching to the forward propagation module and the combustion mechanism module;
an iterative calculation module: and the third connection weight is used for carrying out iterative computation until the root mean square error and the decision coefficient of the verification data set meet a specified threshold or the iteration meets a specified number of times, then carrying out iterative computation to obtain an optimized neural network model reflecting the real combustion mechanism process, and optimizing the combustion based on the optimized neural network model.
9. A storage medium on which a computer program is stored, which program, when being executed by a processor, carries out the steps of the neural network model building method of any one of claims 1 to 7.
10. An electronic device, comprising:
a memory having a computer program stored thereon; and
a processor for executing the computer program in the memory to implement the steps of the neural network model building method of any one of claims 1 through 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010522836.3A CN113779858B (en) | 2020-06-10 | 2020-06-10 | Combustion optimization method and system, storage medium and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010522836.3A CN113779858B (en) | 2020-06-10 | 2020-06-10 | Combustion optimization method and system, storage medium and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113779858A CN113779858A (en) | 2021-12-10 |
| CN113779858B true CN113779858B (en) | 2022-12-16 |
Family
ID=78834758
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010522836.3A Active CN113779858B (en) | 2020-06-10 | 2020-06-10 | Combustion optimization method and system, storage medium and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113779858B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117871771A (en) * | 2024-03-13 | 2024-04-12 | 山东国研自动化有限公司 | Big data-based gas energy monitoring method |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107977539A (en) * | 2017-12-29 | 2018-05-01 | 华能国际电力股份有限公司玉环电厂 | Improvement neutral net boiler combustion system modeling method based on object combustion mechanism |
| CN109034388A (en) * | 2018-07-27 | 2018-12-18 | 湖北中烟工业有限责任公司 | A kind of prediction model of cigarette material and mainstream smoke constituents based on Genetic Algorithm Optimized Neural Network |
| CN109063359A (en) * | 2018-08-16 | 2018-12-21 | 燕山大学 | A kind of dynamic modelling method of Process of Circulating Fluidized Bed Boiler |
-
2020
- 2020-06-10 CN CN202010522836.3A patent/CN113779858B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN113779858A (en) | 2021-12-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106874581B (en) | Building air conditioner energy consumption prediction method based on BP neural network model | |
| US10466659B2 (en) | Method for designing a non-linear controller for non-linear processes | |
| CN109828089B (en) | DBN-BP-based water quality parameter nitrous acid nitrogen online prediction method | |
| CN109887284B (en) | Smart city traffic signal control recommendation method, system and device | |
| CN106453293A (en) | Network security situation prediction method based on improved BPNN (back propagation neural network) | |
| CN112884056A (en) | Optimized LSTM neural network-based sewage quality prediction method | |
| CN110598929B (en) | Wind power nonparametric probability interval ultrashort term prediction method | |
| CN103559537B (en) | Based on the template matching method of error back propagation in a kind of out of order data stream | |
| CN113705922B (en) | Improved ultra-short-term wind power prediction algorithm and model building method | |
| Baklacioglu et al. | Metaheuristic approach for an artificial neural network: exergetic sustainability and environmental effect of a business aircraft | |
| CN111461466A (en) | Heating household valve adjusting method, system and equipment based on L STM time sequence | |
| CN113779858B (en) | Combustion optimization method and system, storage medium and electronic equipment | |
| CN112270442A (en) | IVMD-ACMPSO-CSLSTM-based combined power load prediction method | |
| CN116992759A (en) | Freeze-thawing concrete strength evaluation method and device based on interpretable neural network | |
| CN114819316A (en) | Complex optimization method for multi-agent task planning | |
| CN113051828A (en) | Online prediction method for natural gas water dew point driven by technological parameters | |
| CN112598050A (en) | Ecological environment data quality control method | |
| Li et al. | Neural networks and genetic algorithms can support human supervisory control to reduce fossil fuel power plant emissions | |
| CN108319146B (en) | Discrete particle swarm training-based method for radial basis function neural network | |
| CN115392125A (en) | Temperature prediction method for rotary cement kiln | |
| CN113759709A (en) | Method and device for training strategy model, electronic equipment and storage medium | |
| CN113112072A (en) | NOx emission content prediction method based on deep bidirectional LSTM | |
| CN117190173B (en) | Optimal control method and control system for flue gas recirculation and boiler coupling system | |
| CN113379455B (en) | Order quantity prediction method and equipment | |
| Song | Evaluation and analysis of an industrial cluster based on the BP neural network and LM algorithm |
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 |