CN114397814A - Thermal power generating unit optimal operation parameter searching method based on BP neural network - Google Patents

Thermal power generating unit optimal operation parameter searching method based on BP neural network Download PDF

Info

Publication number
CN114397814A
CN114397814A CN202111516066.2A CN202111516066A CN114397814A CN 114397814 A CN114397814 A CN 114397814A CN 202111516066 A CN202111516066 A CN 202111516066A CN 114397814 A CN114397814 A CN 114397814A
Authority
CN
China
Prior art keywords
concentration
neural network
input variables
thermal power
power generating
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.)
Pending
Application number
CN202111516066.2A
Other languages
Chinese (zh)
Inventor
李璐
聂宗鹏
李伟
王宏亮
黄莹
张明浩
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.)
PowerChina Guizhou Electric Power Engineering Co Ltd
Original Assignee
PowerChina Guizhou Electric Power Engineering 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 PowerChina Guizhou Electric Power Engineering Co Ltd filed Critical PowerChina Guizhou Electric Power Engineering Co Ltd
Priority to CN202111516066.2A priority Critical patent/CN114397814A/en
Publication of CN114397814A publication Critical patent/CN114397814A/en
Pending legal-status Critical Current

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)
  • Treating Waste Gases (AREA)

Abstract

The invention discloses a thermal power generating unit optimal operation parameter searching method based on a BP neural network, which comprises the following steps: s01, determining input variables; s02, dividing the input variables into a training set and a testing set; s03, screening input variables; s04, inputting the input variable into a BP neural network training model; and S05, outputting the SCR inlet NOx concentration and the FGD inlet SOx concentration as prediction variables, taking the lowest total cost value of removing the two pollutants as an optimization target, and finding out the optimal operation parameters through a genetic algorithm. Aims to solve the problems that the prior art can only realize the prediction of the real-time optimal control quantity of NOx and can not predict NOx and SOXSimultaneously predicting the real-time optimal control quantity; the model prediction results in the optimal control amount based on the lowest NOx emission amount, but the running cost is not necessarily the lowest problem.

Description

Thermal power generating unit optimal operation parameter searching method based on BP neural network
Technical Field
The invention relates to a thermal power generating unit optimal operation parameter searching method based on a BP neural network, and belongs to the technical field of thermal power.
Background
The total energy consumption of China is less than 10% of the total energy consumption of the world, but the emission of nitrogen oxides exceeds 10% of the total emission of the world. Pollutants released in the combustion process of coal have great influence on the ecological environment, not only influence the social and economic development of China, but also bring problems in the aspects of health and agriculture. Nitrogen oxides are one of the main causes of photochemical smog formation, causing acid rain. In addition to this, nitrous oxide (N)2O) can destroy stratospheric ozone, causing ozone voiding and even global warming problems.
SO generated by coal combustion in China2Is about SO286% of the total emissions, with the emissions from coal fired power plants being about 40%. Sulfur dioxide is an acidic gas, is chemically very reactive, and can react with various oxides in the atmosphere to generate sulfur trioxide. The moisture, dust, etc. are then combined with sulfur trioxide to form sulfuric acid and various acid corrosive aerosols, acid rain, acid snow, acid mist, etc. The acidic precipitation can not only pollute various water sources, but also change the pH value of soil, thereby causing the reduction of yield of crops and the death of livestock and poultry. Meanwhile, the health of human beings is directly threatened by acid rain. The economic loss of acid rain to human beings is not negligible, and the loss to the ecological environment is beyond the estimation.
NOx and SOXIs harmful to the environment and is used as NOxWith SOXCoal-fired power plant of main emission source how to reduce NO to the maximumxWith SOXDischarge becomes a big problem. How to reduce NOxWith SOXIn the prior art, for example, chinese patent CN202110020405.1 proposes an intelligent combustion control method for a thermal power plant based on cloud data and cloud computing, which, although proposed, has the following problems:
1) can only realize NOxPrediction of real-time optimal control quantity of NOxWith SOXSimultaneously predicting the real-time optimal control quantity;
2) the model prediction result is based on NOxThe emissions are the lowest to get the optimum control amount, but the running cost is not necessarily the lowest.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: the method for searching the optimal operation parameters of the thermal power generating unit based on the BP neural network is provided, so that the defects of the prior art are overcome.
The technical scheme of the invention is as follows: a thermal power generating unit optimal operation parameter searching method based on a BP neural network comprises the following steps:
s01, determining input variables;
s02, dividing the input variables into a training set and a testing set;
s03, screening input variables;
s04, inputting the input variable into a BP neural network training model;
and S05, outputting the SCR inlet NOx concentration and the FGD inlet SOx concentration as prediction variables, taking the lowest total cost value of removing the two pollutants as an optimization target, and finding out the optimal operation parameters through a genetic algorithm.
Specifically, the step S03 is to filter the input variables as follows:
s03-1, carrying out normalization processing on the data;
s03-2, training the BP model by using an unscreened sample, and predicting a training sample;
s03-3, respectively adding and subtracting 10% to the characteristic value of a certain input variable in the training sample P on the basis of the original value to obtain two new training samples P1And P2Then P is added1And P2Respectively using the built models to carry out prediction to obtain two corresponding prediction results A1And A2Obtaining A1And A2The difference value is an influence change value of changing the input variable on the output, and the input variables are changed in sequence until all the input variables are changed;
s03-4, calculating the average influence value of each input variable on the output variable, wherein the average influence value is obtained by averaging the influence change values according to the number of observation cases;
s03-5, sorting the input variables according to the absolute values of the average influence values of the input variables on the output variables, wherein the larger the absolute value is, the larger the influence of the independent variable on the output variables of the model is;
and S03-6, extracting input variables which have large influence on the output quantity of the model from the input variables, thereby realizing variable screening.
Specifically, the method for evaluating the input variables having a large influence on the model output quantity in step S03-6 is as follows:
screening out SCR Inlet NO by step S03XMean influence of concentration and FGD inlet SOXAnd the absolute value of the concentration average influence value is larger than 1.
Specifically, the formula of the step S03-1 for normalizing the data is as follows:
Xk＝2*(Xk-Xmin)/(Xmax-Xmin)+(-1)；
in the formula, XmaxIs the maximum number in the data sequence, XminX to the right of the equal sign as the smallest number in the data sequencekTo normalize previous data values, X to the left of equal signkTo normalize the subsequent data values.
Further, the genetic algorithm of step S05 takes the total cost of removing pollutants in unit exhaust smoke as the fitness function f (x).
Specifically, the fitness function f (x) is:
total cost of NOXUnit drop out cost X (SCR inlet NO)xConcentration of-NOxNational allowed emission Standard of concentration) + SO2Unit drop out cost x (FGD inlet SO)xconcentration-SOxConcentration country permitted discharge standard).
Further, let the running parameter vector be x ═ x1,x2,x3,x4,x5,x6,x7,x8,x9,x10]In the formula, x1～x6Respectively 6 secondary air compensation quantities with the unit of km3/h；x7Is the air quantity of over-fire air with the unit of km3/h；x8Is a primary wind compensation quantity with the unit of km3/h；x9Is the post-economizer oxygen content in%, x10Is total air volume, and has unit of km3The constraint condition of the fitness function f (x) is as follows:
Figure BDA0003394730990000031
preferably, the iteration number of the genetic algorithm is 1000, the cross probability is 0.4, and the variation probability is 0.2, and the population size is 30.
Further, after the input variable is screened in step S03, the SCR inlet NO will be at the previous timeXConcentration and previous time FGD inlet SO2Concentration adds an input variable.
The invention has the beneficial effects that: compared with the prior art, the invention utilizes the BP neural network intelligent algorithm to carry out on the FGD inlet SOxConcentration and SCR inlet NOXThe concentration is subjected to soft measurement, in order to enable the measurement result to be more accurate, the input variables are screened, the screening result is ideal, the model constructed by the screened variables is higher in prediction precision and higher in speed, meanwhile, the lowest total cost value is taken as an optimization target, the optimal operation parameters are found out through a genetic algorithm, the pollutant removal cost is effectively reduced, the pollutant emission can be reduced, and the optimal operation of a power plant is facilitated. Realizing the win-win of economic benefit and environmental benefit to a certain extent.
Drawings
FIG. 1 is a flow chart of an embodiment of the present invention;
FIG. 2 shows the respective models SO2Fitting a curve between the predicted value and the true value;
FIG. 3 shows the respective models NOXFitting a curve between the predicted value and the true value;
FIG. 4 shows the SO of each model2Relative error;
FIG. 5 shows each model NOXRelative error.
Detailed Description
In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, and are only for convenience of description and simplicity of description, but do not indicate or imply that the device or element being referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus, should not be construed as limiting the present invention; the terms "first," "second," and "third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance, and furthermore, unless otherwise explicitly stated or limited, the terms "mounted," "connected," and "connected" are to be construed broadly and may be, for example, fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.
In order to better understand the technical solutions, the technical solutions will be described in detail below with reference to the drawings and the detailed description.
Example 1 was carried out:
referring to fig. 1, a thermal power generating unit optimal operation parameter searching method based on a BP neural network includes:
s01, determining input variables;
s02, dividing the input variables into a training set and a testing set;
s03, screening input variables;
s04, inputting the input variable into a BP neural network training model;
and S05, outputting the SCR inlet NOx concentration and the FGD inlet SOx concentration as prediction variables, taking the lowest total cost value of removing the two pollutants as an optimization target, and finding out the optimal operation parameters through a genetic algorithm.
FGD entry SO using BP neural network intelligent algorithmxConcentration and SCR inlet NOXSoft measurement of concentration, and providing the above measures for higher accuracyThe method has the advantages that the screening is carried out by entering variables, the screening result is ideal, the model constructed by using the screened variables is higher in prediction accuracy and higher in speed, meanwhile, the lowest total cost value is taken as an optimization target, the optimal operation parameters are found out through a genetic algorithm, the pollutant removal cost is effectively reduced, the pollutant emission can be reduced, and the method is favorable for the optimized operation of a power plant. Realizing the win-win of economic benefit and environmental benefit to a certain extent.
Example 2 was carried out:
in this embodiment, in order to reduce the running time of the model in implementation example 1, increase the prediction accuracy of the model, and reduce the model training redundancy possibly caused by the possible correlation between the variables, the input variables may be filtered by the following method, and in this embodiment, except for the content of implementation example 1, the step of filtering the input variables in step S03 in this embodiment is specifically as follows:
s03-1, carrying out normalization processing on the data;
s03-2, training the BP model by using an unscreened sample, and predicting a training sample;
s03-3, respectively adding and subtracting 10% to the characteristic value of a certain input variable in the training sample P on the basis of the original value to obtain two new training samples P1And P2Then P is added1And P2Respectively using the built models to carry out prediction to obtain two corresponding prediction results A1And A2Obtaining A1And A2The difference value is an influence change value of changing the input variable on the output, and the input variables are changed in sequence until all the input variables are changed;
s03-4, calculating the average influence value of each input variable on the output variable, wherein the average influence value is obtained by averaging the influence change values according to the number of observation cases;
s03-5, sorting the input variables according to the absolute values of the average influence values of the input variables on the output variables, wherein the larger the absolute value is, the larger the influence of the independent variable on the output variables of the model is;
and S03-6, extracting input variables which have large influence on the output quantity of the model from the input variables, thereby realizing variable screening.
Calculating the average influence value of each parameter from X1To X27The average influence values of (a) are shown in table 1:
Figure BDA0003394730990000051
Figure BDA0003394730990000061
TABLE 1
The method for evaluating the input variables having a large influence on the model output quantity in step S03-6 is as follows: screening out SCR inlet NOXMean influence of concentration and FGD inlet SOXAnd the absolute value of the concentration average influence value is larger than 1. From Table 1, the combustion section X in the boiler can be obtained1-X3，X5，X6，X16，X18，X24，X26，X27All values exceed 1, so the 10 parameters are applied to SCR inlet NOxConcentration and FGD inlet SOxThe influence of concentration is large; the accuracy and comprehensiveness of indexes are comprehensively considered, and SCR inlet NO needs to be establishedxConcentration and FGD inlet SOxThe prediction model of concentration, the selected input variables are boiler load, total fuel quantity, total air quantity, primary air quantity, secondary air quantity, excess air coefficient, fly ash carbon content, exhaust gas temperature, main steam temperature, SCR inlet NO at the previous momentxConcentration, FGD inlet SO of previous momentxThe concentration was 10 parameters.
Example 3 of implementation:
in the embodiment, except for the content of the implementation example 2, the formula of the normalization processing is specifically as follows:
Xk＝2*(Xk-Xmin)/(Xmax-Xmin)+(-1)；
in the formula, XmaxIs the maximum number in the data sequence, XminX to the right of the equal sign as the smallest number in the data sequencekTo normalize previous data values, X to the left of equal signkTo normalize the subsequent data values.
Example 4 of implementation:
in order to find the lowest cost under the operating parameters, in this embodiment, in addition to the content of the implementation example 1, the step S05 genetic algorithm further takes the total cost of removing the pollutants in the unit exhaust smoke as a fitness function f (x).
Example 5 was carried out:
in order to optimize the combustion of a large-scale power plant boiler, reduce the emission of pollutants and remove the total cost, the win-win of economic benefit and environmental benefit is realized.
In this embodiment, except for the content of embodiment 4, specifically, the fitness function f (x) is:
total cost of NOXUnit drop out cost X (SCR inlet NO)xConcentration of-NOxNational allowed emission Standard of concentration) + SO2Unit drop out cost x (FGD inlet SO)xconcentration-SOxConcentration country permitted discharge standard).
Example 6 of implementation:
in order to eliminate the model operation interference, in addition to the content of embodiment example 5, in this embodiment, the operation parameter vector is further set to x ═ x1,x2,x3,x4,x5,x6,x7,x8,x9,x10]In the formula, x1～x6Respectively 6 secondary air compensation quantities with the unit of km3/h；x7Is the air quantity of over-fire air with the unit of km3/h；x8Is a primary wind compensation quantity with the unit of km3/h；x9Is the post-economizer oxygen content in%, x10Is total air volume, and has unit of km3The constraint condition of the fitness function f (x) is as follows:
Figure BDA0003394730990000071
example 7 was carried out:
in order to avoid obtaining a local optimum, the convergence speed is faster.
In addition to the content of the embodiment 6, in this embodiment, it is preferable that the number of iterations of the genetic algorithm is 1000, the cross probability is 0.4, and the variation probability is 0.2, and the population size is 30.
The following table is the model optimization results:
Figure BDA0003394730990000081
TABLE 2
In table 2, for the parameter optimization results of the optimization model, the total cost of removing NOx and SOx in unit volume of the exhaust smoke after optimization under the working conditions 11 and 12 is reduced by 3.67% and 2.43%, respectively. From the comparison of the data before and after optimization, when secondary air and over-fire air are increased, primary air quantity is reduced, oxygen content is also reduced, combustion is in reducing atmosphere, which is exactly consistent with the air distribution mode of low-nitrogen low-sulfur combustion, and the total cost required for removing the secondary air and the over-fire air is the lowest value under the air distribution mode of pollutant emission characteristics. Theoretically, it is likely that incomplete combustion of pulverized coal will occur when the oxygen content is reduced, but it is shown from data obtained from a certain power plant from which the data herein are derived that, when the oxygen content is 3.02%, 3.14%, 2.76%, respectively, the carbon content of fly ash is 1.4%, 1.46%, 1.33%, respectively, and thus it can be seen that the reduction of oxygen content in this range does not have a great influence on the carbon content of fly ash (loss of incomplete combustion of solids). Therefore, the optimization of the air distribution mode can reduce the pollutant removal cost of the power plant.
Therefore, for a given operation condition, the NO in the discharged smoke with unit volume can be reduced and removed by adjusting corresponding operation parametersxAnd SOxThe total cost target. I.e. the optimal combination of parameters can be found so that the singleton is removedAccumulation of NO in exhaust gasxAnd SOxThe total cost of the process is the lowest. The rapid convergence of the genetic algorithm and the accurate prediction of the BP neural network are very suitable for online optimization, the combination of the genetic algorithm and the BP neural network provides an effective way for reducing the cost of removing pollutants from a power plant, and the win-win of environmental benefits and economic benefits can be realized to a certain extent.
Example 8 was carried out:
in addition to the content of the embodiment 1, the embodiment further includes the specific selection of the input variable, and in the embodiment, further, the SCR inlet NO at the previous time is set in step S01XConcentration and previous time FGD inlet SO2Concentration adds an input variable. Aiming at a 1000MW coal-fired unit of a certain power plant, relevant monitoring is carried out on equipment such as an SCR system, an FGD system, a boiler, a primary fan, a secondary fan, an economizer, an air preheater and the like, and the SO at an inlet of the FGD is preliminarily determinedxConcentration and SCR inlet NOxThe parameters related to concentration are: boiler load, total fuel quantity, total air quantity, burnout air quantity, total furnace secondary air quantity, total furnace primary air quantity, 1 furnace mill A inlet primary air quantity, 1 furnace mill B inlet primary air quantity, 1 furnace mill C inlet primary air quantity, 1 furnace mill D inlet primary air quantity, 1 furnace mill E inlet primary air quantity, 1 furnace mill F inlet primary air quantity, fly ash carbon content, boiler efficiency, boiler negative pressure, excess air coefficient, economizer outlet oxygen content, main steam temperature, main steam pressure, high-temperature superheater temperature, solid incomplete combustion loss, ash heat loss, smoke exhaust temperature, smoke exhaust loss, SCR inlet NOX concentration at previous moment and FGD inlet SO at previous moment2And (4) concentration. The negative pressure of the boiler is the average value of all measuring points.
Figure BDA0003394730990000091
Figure BDA0003394730990000101
TABLE 3
The consideration of the SCR inlet NOx concentration at the previous time and the FGD inlet SO2 concentration at the previous time is based on the fact that:
synergy between nitrogen and sulfur contaminants: during the combustion of the coal dust, SO2The evolution of gases presents a very pronounced "bimodal" structure, since the various forms of sulfur in the coal fines are evolved at different stages in the combustion process; as the particle size of the pulverized coal increases, SO can be generated2A gradual merging of the two peaks on the precipitation curve of (1), SO2The total precipitation amount of the gas is reduced, and the time for sulfur precipitation is prolonged; when the excess air factor increases, sulfur precipitated from the pulverized coal is converted into SO2The amount of gas also increases, SO SO2The concentration of the gas is higher, and H2The S concentration decreases. If the proportion of the primary air and the secondary air is increased, SO2The concentration will also increase. With the addition of secondary air, H2The S concentration will drop sharply, and conversely SO2The concentration is significantly increased. If the proportion of the primary air and the secondary air is smaller at the moment, H2The larger the decrease of the concentration of S gas is, the more SO is generated2The higher the concentration of (c).
In the process of pulverized coal combustion, the precipitation of nitrogen pollutants comprises two stages of early-stage rapid generation and later-stage slow release, and the nitrogen pollutants are similar to the precipitation of sulfur pollutants and also have a 'double-peak' structure. The coal dust ignition process is the major stage of NOx production. The particle size and the humidity of the pulverized coal have obvious influence on the generation amount of NOx. The finer the coal fines, the lower the NO conversion, so that the fine coal fines enable low NOx combustion, but at the same time the coal mill output is increased. Generally, the discharge amount of NO of the pulverized coal with high humidity is less than that of the dried pulverized coal; the excess air factor has a large influence on the NO conversion. When the excess air ratio is increased, the NOx generation amount is increased; when the total air quantity is fixed, the generation quantity of NO is closely related to the primary air distribution ratio and the secondary air distribution ratio. When the proportion of the primary air and the secondary air is reduced, the concentration of NOx is rapidly reduced. At lower temperatures, the nitrogen precipitation curve forms a monomodal structure, while at higher temperatures, the nitrogen precipitation curve forms a bimodal structure, and the bimodal structure of the precipitation curve becomes more pronounced at higher temperatures, with the amount of nitrogen precipitated being greater.
According to the research, the interaction and influence between nitrogen and sulfur components exist in the combustion process, and the law of the synergistic action between nitrogen and sulfur pollutants in the pulverized coal combustion process is summarized as follows. In which the influence of different forms of sulphur on nitrogen evolution, and SO in different atmospheres are addressed2The influence of gas on NO generation and the influence rule of nitrogen on sulfur precipitation in the pulverized coal combustion process are briefly analyzed. The test results of the relevant researchers show that: the different forms of sulfur in coal have a large impact on the evolution of nitrogen. The most remarkable is the influence of the pyrite sulfur on nitrogen precipitation, and the pyrite sulfur can inhibit the reduction of NO under the oxygen-deficient working condition and is beneficial to the generation of NO gas. The influence of elemental sulfur on nitrogen precipitation is related to coal types, and the elemental sulfur mainly influences the precipitation peak value and the peak width of NO; when the combustion atmosphere is different, SO2The effect of the gas on the precipitation of nitrogen is obviously different, and the gas is separated from the nitrogen along with SO2Increase of gas concentration, NO in reducing atmosphereXThe amount of the produced NO is obviously increased on the basis of the original amount, but the NO is in an oxidizing atmosphereXThe production amount is obviously reduced on the basis of the original production amount. At the same time, NOXAlso has an effect on the precipitation behavior of sulfur in coal. NOXThe generation amount is increased, the precipitation and release of sulfur are advanced, and therefore SO2The initial product concentration of (a) also increases. The combustion temperature has a great influence on the synergistic effect in the precipitation of nitrogen and sulfur pollutants. At lower temperatures, the nitrogen precipitation curve forms a monomodal structure, while at higher temperatures, the nitrogen precipitation curve forms a bimodal structure, and the bimodal structure of the precipitation curve becomes more pronounced at higher temperatures, with the amount of nitrogen precipitated being greater. The temperature rises, the separation peak of the sulfur is advanced, the final separation rate is increased, and the separation time is shortened. The precipitation curves of nitrogen and sulfur change with the temperature, and the corresponding peaks shift, so the synergistic effect of nitrogen and sulfur under different combustion conditions changes.
At establishment of SCR inlet NOxConcentration and FGD inlet SOxIn the case of a concentration prediction model, if the 25 parameters are simultaneously used as the input of the modelThe variables inevitably make the model very complex, which not only increases the model running time, but also reduces the model prediction accuracy; in addition, there may be a certain correlation between variables, which may cause redundancy in model training, so that it is necessary to screen input variables, retain arguments that have a significant influence on the output quantity of the model, and reject arguments that have an insignificant influence.
Table 4 shows the relative error comparison of the prediction results of each model:
Figure BDA0003394730990000111
Figure BDA0003394730990000121
TABLE 4
It can be seen from table 4 that after the variables are extracted from the average influence values, the output result of the model is significantly better than that of the model without the extracted variables. Of the two models for screening input variables, the 10-input model is slightly more accurate than the 8-input MIV-BP model because it takes full account of the synergy between the pollutants.
The BP neural network model is trained by using 180 groups of data, and then the last 30 groups of data are predicted, and the prediction results are shown in the following figures 1-4. The three models for predicting SO are shown in FIGS. 1 and 22And NOXAnd the fitting condition of the real value shows that the fitting of the two processes in the three models is ideal, the difference between the original value and the predicted value is not large, and the BP network structure of the model is reasonable. Fig. 3 and 4 show the relative error between the actual value and the measured value of the test data of the FGD inlet SOx concentration and the SCR inlet NOx concentration, and it is obvious from the comparison curve of the relative error of 30 sets of test data that the prediction effect of the 10-input model is better and the accuracy is very high with the relative error less than 5%. Secondly, the relative error of the 8 input models is less than 8%, and the accuracy is high. Therefore, the model prediction effect after variable screening is improved, and the SCR inlet NOx concentration and the FGD inlet SOx concentration at the moment before the introduction of synergistic effect are taken as input parametersThe model prediction effect of the numbers is improved again.
The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, several simple deductions or substitutions can be made without departing from the spirit of the invention, and all shall be considered as belonging to the protection scope of the invention.

Claims (9)

1. A thermal power generating unit optimal operation parameter searching method based on a BP neural network is characterized by comprising the following steps:
s01, determining input variables;
s02, dividing the input variables into a training set and a testing set;
s03, screening input variables;
s04, inputting the input variable into a BP neural network training model;
and S05, outputting the SCR inlet NOx concentration and the FGD inlet SOx concentration as prediction variables, taking the lowest total cost value of removing the two pollutants as an optimization target, and finding out the optimal operation parameters through a genetic algorithm.
2. The thermal power generating unit optimal operation parameter searching method based on the BP neural network as claimed in claim 1, wherein the step S03 of screening the input variables comprises the following steps:
s03-1, carrying out normalization processing on the data;
s03-2, training the BP model by using an unscreened sample, and predicting a training sample;
s03-3, respectively adding and subtracting 10% to the characteristic value of a certain input variable in the training sample P on the basis of the original value to obtain two new training samples P1And P2Then P is added1And P2Respectively using the built models to carry out prediction to obtain two corresponding prediction results A1And A2Obtaining A1And A2The difference is the influence of the input variable on the outputChanging values, and sequentially changing input variables until all the input variables are changed;
s03-4, calculating the average influence value of each input variable on the output variable, wherein the average influence value is obtained by averaging the influence change values according to the number of observation cases;
s03-5, sorting the input variables according to the absolute values of the average influence values of the input variables on the output variables, wherein the larger the absolute value is, the larger the influence of the independent variable on the output variables of the model is;
and S03-6, extracting input variables which have large influence on the output quantity of the model from the input variables, thereby realizing variable screening.
3. The method for searching the optimal operating parameters of the thermal power generating unit based on the BP neural network as claimed in claim 2, wherein the method for evaluating the input variables having large influence on the model output quantity in the step S03-6 is as follows:
screening out SCR inlet NOXMean influence of concentration and FGD inlet SOXAnd the absolute value of the concentration average influence value is larger than 1.
4. The thermal power generating unit optimal operation parameter searching method based on the BP neural network as claimed in claim 3, wherein the formula of the step S03-1 for carrying out normalization processing on the data is as follows:
Xk＝2*(Xk-Xmin)/(Xmax-Xmin)+(-1)；
in the formula, XmaxIs the maximum number in the data sequence, XminX to the right of the equal sign as the smallest number in the data sequencekTo normalize previous data values, X to the left of equal signkTo normalize the subsequent data values.
5. The method for searching the optimal operating parameters of the thermal power generating unit based on the BP neural network as claimed in claim 1, wherein the genetic algorithm of the step S05 takes the total cost of removing pollutants in unit exhaust smoke as a fitness function f (x).
6. The thermal power generating unit optimal operation parameter searching method based on the BP neural network as claimed in claim 5, wherein the fitness function f (x) is:
total cost of NOXUnit drop out cost X (SCR inlet NO)xConcentration of-NOxNational allowed emission Standard of concentration) + SO2Unit drop out cost x (FGD inlet SO)xconcentration-SOxConcentration country permitted discharge standard).
7. The thermal power generating unit optimal operation parameter searching method based on the BP neural network as claimed in claim 6, wherein the operation parameter vector is set as x ═ x [ x ═ x1,x2,x3,x4,x5,x6,x7,x8,x9,x10]In the formula, x1～x6Respectively 6 secondary air compensation quantities with the unit of km3/h；x7Is the air quantity of over-fire air with the unit of km3/h；x8Is a primary wind compensation quantity with the unit of km3/h；x9Is the post-economizer oxygen content in%, x10Is total air volume, and has unit of km3The constraint condition of the fitness function f (x) is as follows:
Figure FDA0003394730980000021
8. the method for searching the optimal operation parameter of the thermal power generating unit based on the BP neural network as claimed in claim 7, wherein the iteration number of the genetic algorithm is 1000, the cross probability is 0.4, and the variation probability is 0.2, and the population size is 30.
9. The method for searching the optimal operating parameters of the thermal power generating unit based on the BP neural network as claimed in claim 1, wherein the SCR inlet NO at the previous moment is used in step S01XConcentration and previous time FGD inlet SO2Concentration adds an input variable.
CN202111516066.2A 2021-12-06 2021-12-06 Thermal power generating unit optimal operation parameter searching method based on BP neural network Pending CN114397814A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111516066.2A CN114397814A (en) 2021-12-06 2021-12-06 Thermal power generating unit optimal operation parameter searching method based on BP neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111516066.2A CN114397814A (en) 2021-12-06 2021-12-06 Thermal power generating unit optimal operation parameter searching method based on BP neural network

Publications (1)

Publication Number Publication Date
CN114397814A true CN114397814A (en) 2022-04-26

Family

ID=81226448

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111516066.2A Pending CN114397814A (en) 2021-12-06 2021-12-06 Thermal power generating unit optimal operation parameter searching method based on BP neural network

Country Status (1)

Country Link
CN (1) CN114397814A (en)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105954323A (en) * 2016-06-02 2016-09-21 中国石油大学(华东) Intelligent ultralow-concentration monitoring device for gaseous pollutants discharged by thermal power plant
CN109472406A (en) * 2018-11-02 2019-03-15 浙江大学 A kind of global optimization method of minimum discharge multi-pollutant cooperation-removal system
CN110032747A (en) * 2018-12-05 2019-07-19 太原理工大学 A kind of coal-burning boiler NOxDischarge On-line Measuring Method
CN110824586A (en) * 2019-10-23 2020-02-21 上海理工大学 Rainfall prediction method based on improved decision tree algorithm
CN112859780A (en) * 2021-01-07 2021-05-28 西安西热锅炉环保工程有限公司 Thermal power plant intelligent combustion control method based on cloud data and cloud computing
WO2021189739A1 (en) * 2020-03-23 2021-09-30 大连理工大学 Production plan-based iron and steel plant oxygen load prediction method

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105954323A (en) * 2016-06-02 2016-09-21 中国石油大学(华东) Intelligent ultralow-concentration monitoring device for gaseous pollutants discharged by thermal power plant
CN109472406A (en) * 2018-11-02 2019-03-15 浙江大学 A kind of global optimization method of minimum discharge multi-pollutant cooperation-removal system
CN110032747A (en) * 2018-12-05 2019-07-19 太原理工大学 A kind of coal-burning boiler NOxDischarge On-line Measuring Method
CN110824586A (en) * 2019-10-23 2020-02-21 上海理工大学 Rainfall prediction method based on improved decision tree algorithm
WO2021189739A1 (en) * 2020-03-23 2021-09-30 大连理工大学 Production plan-based iron and steel plant oxygen load prediction method
CN112859780A (en) * 2021-01-07 2021-05-28 西安西热锅炉环保工程有限公司 Thermal power plant intelligent combustion control method based on cloud data and cloud computing

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
廖永进;范军辉;杨维结;季鹏;张静;冯永新;: "基于RBF神经网络的SCR脱硝系统喷氨优化", 动力工程学报, no. 11 *
金晶亮;李晨宇;彭超华;冯俊丰;赵英全;: "风电并网背景下的低碳电力调度策略", 南京航空航天大学学报(社会科学版), no. 02 *

Similar Documents

Publication Publication Date Title
CN113050559B (en) Coal-fired power plant desulfurization system and electric precipitation system cooperative control method and system
CN112540158B (en) Method for testing utilization rate of limestone wet desulphurization forced oxidation air
CN105971738A (en) On-line monitoring method for flue gas flow of gas turbine unit
CN104331736A (en) RBF (Radial Basis Function) neural network based supercritical boiler nitric oxide discharging dynamic predication method
CN112710487B (en) Method for evaluating influence of desulfurization wastewater drying system on thermal efficiency of boiler
CN113440990A (en) EMD-LSTM based outlet SO2Concentration prediction method
CN114632624B (en) Electric dust removal operation optimization system and optimization method
CN107909531A (en) The definite method of coal-burned industrial boiler flue-gas denitration process
CN106055867A (en) Method for computing boiler efficiency with consideration of coal quality correction
CN111564184A (en) Limestone-gypsum wet desulphurization SO of coal-fired power plant3Collaborative removal efficiency prediction method
CN114397814A (en) Thermal power generating unit optimal operation parameter searching method based on BP neural network
CN117669858A (en) Quantitative analysis method for synergistic pollution and carbon reduction of coal-fired power plant
CN115113519A (en) Coal-gas co-combustion boiler denitration system outlet NO x Concentration early warning method
CN116805065A (en) Intelligent management method for monitoring data of coal powder heating furnace burner
CN112598167A (en) Power station boiler NO based on dragonfly algorithm and fast learning netxEmission amount prediction method
CN105808924A (en) Boiler combustion adjustment working conditions economy comparison method
Rashid et al. Evaluation of particulate emission from a palm oil mill boiler
CN116906910A (en) Efficient combustion control method and system based on deep convolutional neural network
CN106269250B (en) A kind of low temperature electric precipitation safe and highly efficient operation method based on coal property
CN115372285A (en) Tail gas analysis device and method
Azid et al. Predicting combined-cycle natural gas power plant emissions by using artificial neural networks
CN113639980A (en) Method and device for evaluating water lifting performance of flue gas water lifting device
Rabaçal et al. Particulate emissions from the combustion of biomass pellets
CN110647560B (en) Mechanism and brain-like intelligence combined online soft measurement method for coal quality of fired coal of power plant
CN108459628B (en) Method for controlling pollutant emission of coal power plant through mixed coal blending combustion

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