CN117724410A - LSTM and particle swarm optimization-based coal-fired boiler flue gas prediction and feedback control system - Google Patents
LSTM and particle swarm optimization-based coal-fired boiler flue gas prediction and feedback control system Download PDFInfo
- Publication number
- CN117724410A CN117724410A CN202311563289.3A CN202311563289A CN117724410A CN 117724410 A CN117724410 A CN 117724410A CN 202311563289 A CN202311563289 A CN 202311563289A CN 117724410 A CN117724410 A CN 117724410A
- Authority
- CN
- China
- Prior art keywords
- cost
- model
- flue gas
- prediction
- lstm
- 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
Links
- UGFAIRIUMAVXCW-UHFFFAOYSA-N Carbon monoxide Chemical compound [O+]#[C-] UGFAIRIUMAVXCW-UHFFFAOYSA-N 0.000 title claims abstract description 55
- 239000003546 flue gas Substances 0.000 title claims abstract description 55
- 239000002245 particle Substances 0.000 title claims abstract description 34
- 238000005457 optimization Methods 0.000 title claims abstract description 26
- 238000000034 method Methods 0.000 claims abstract description 35
- 229910052602 gypsum Inorganic materials 0.000 claims abstract description 28
- 239000010440 gypsum Substances 0.000 claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 23
- 239000000428 dust Substances 0.000 claims abstract description 20
- 239000003344 environmental pollutant Substances 0.000 claims abstract description 18
- 231100000719 pollutant Toxicity 0.000 claims abstract description 18
- 238000010276 construction Methods 0.000 claims abstract description 11
- 238000002360 preparation method Methods 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims abstract description 10
- QGZKDVFQNNGYKY-UHFFFAOYSA-N Ammonia Chemical compound N QGZKDVFQNNGYKY-UHFFFAOYSA-N 0.000 claims description 44
- 239000002002 slurry Substances 0.000 claims description 27
- 238000004364 calculation method Methods 0.000 claims description 23
- 239000004071 soot Substances 0.000 claims description 22
- 238000006477 desulfuration reaction Methods 0.000 claims description 21
- 230000023556 desulfurization Effects 0.000 claims description 21
- 239000003054 catalyst Substances 0.000 claims description 16
- 230000005611 electricity Effects 0.000 claims description 15
- VTYYLEPIZMXCLO-UHFFFAOYSA-L Calcium carbonate Chemical compound [Ca+2].[O-]C([O-])=O VTYYLEPIZMXCLO-UHFFFAOYSA-L 0.000 claims description 12
- 235000019738 Limestone Nutrition 0.000 claims description 12
- 230000005684 electric field Effects 0.000 claims description 12
- 239000006028 limestone Substances 0.000 claims description 12
- 229910000069 nitrogen hydride Inorganic materials 0.000 claims description 12
- 239000000779 smoke Substances 0.000 claims description 10
- VHUUQVKOLVNVRT-UHFFFAOYSA-N Ammonium hydroxide Chemical compound [NH4+].[OH-] VHUUQVKOLVNVRT-UHFFFAOYSA-N 0.000 claims description 9
- 235000011114 ammonium hydroxide Nutrition 0.000 claims description 9
- 229910052925 anhydrite Inorganic materials 0.000 claims description 9
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 238000007664 blowing Methods 0.000 claims description 9
- OSGAYBCDTDRGGQ-UHFFFAOYSA-L calcium sulfate Chemical compound [Ca+2].[O-]S([O-])(=O)=O OSGAYBCDTDRGGQ-UHFFFAOYSA-L 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 9
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 9
- 230000008901 benefit Effects 0.000 claims description 8
- 239000000463 material Substances 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 7
- 229910021529 ammonia Inorganic materials 0.000 claims description 6
- 229910000019 calcium carbonate Inorganic materials 0.000 claims description 6
- 235000010216 calcium carbonate Nutrition 0.000 claims description 6
- 230000003009 desulfurizing effect Effects 0.000 claims description 6
- 239000012895 dilution Substances 0.000 claims description 6
- 238000010790 dilution Methods 0.000 claims description 6
- 239000012717 electrostatic precipitator Substances 0.000 claims description 6
- 230000003647 oxidation Effects 0.000 claims description 6
- 238000007254 oxidation reaction Methods 0.000 claims description 6
- 239000000243 solution Substances 0.000 claims description 5
- NINIDFKCEFEMDL-UHFFFAOYSA-N Sulfur Chemical compound [S] NINIDFKCEFEMDL-UHFFFAOYSA-N 0.000 claims description 4
- 238000005265 energy consumption Methods 0.000 claims description 4
- 229910052717 sulfur Inorganic materials 0.000 claims description 4
- 239000011593 sulfur Substances 0.000 claims description 4
- OYPRJOBELJOOCE-UHFFFAOYSA-N Calcium Chemical compound [Ca] OYPRJOBELJOOCE-UHFFFAOYSA-N 0.000 claims description 3
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 230000002745 absorbent Effects 0.000 claims description 3
- 239000002250 absorbent Substances 0.000 claims description 3
- XKMRRTOUMJRJIA-UHFFFAOYSA-N ammonia nh3 Chemical compound N.N XKMRRTOUMJRJIA-UHFFFAOYSA-N 0.000 claims description 3
- 239000011575 calcium Substances 0.000 claims description 3
- 229910052791 calcium Inorganic materials 0.000 claims description 3
- 239000003245 coal Substances 0.000 claims description 3
- 238000007865 diluting Methods 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000002347 injection Methods 0.000 claims description 3
- 239000007924 injection Substances 0.000 claims description 3
- 238000005507 spraying Methods 0.000 claims description 3
- RAHZWNYVWXNFOC-UHFFFAOYSA-N Sulphur dioxide Chemical compound O=S=O RAHZWNYVWXNFOC-UHFFFAOYSA-N 0.000 claims 6
- 230000000694 effects Effects 0.000 description 4
- IJGRMHOSHXDMSA-UHFFFAOYSA-N Atomic nitrogen Chemical compound N#N IJGRMHOSHXDMSA-UHFFFAOYSA-N 0.000 description 2
- 238000004140 cleaning Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000002411 adverse Effects 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 229910052757 nitrogen Inorganic materials 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000002195 synergetic effect Effects 0.000 description 1
- 238000011282 treatment Methods 0.000 description 1
Landscapes
- Treating Waste Gases (AREA)
Abstract
The invention discloses a coal-fired boiler flue gas prediction and feedback control system based on an LSTM and particle swarm optimization algorithm, and relates to the technical field of coal-fired boiler operation control. The method comprises an LSTM prediction model, an ultralow emission system cost control model and a particle swarm model solution, wherein the construction of the LSTM prediction model comprises a data preparation stage, a data processing and selecting stage, a model construction stage and a model training and evaluating stage; the ultra-low emission system cost control model comprises an SCR system cost model, a dust removal system cost model and a limestone-gypsum wet desulphurization system cost model; the particle swarm model solving comprises the steps of establishing a cost optimization objective function, calculating an optimization control quantity by using a particle swarm algorithm, and performing feedback control; and after the system is built, optimizing, controlling and evaluating the historical data by using the system model. The invention realizes the pollutant prediction and collaborative optimization control.
Description
Technical Field
The invention relates to the technical field of operation control of coal-fired boilers, in particular to a coal-fired boiler flue gas prediction and feedback control system based on LSTM and particle swarm optimization.
Background
Coal-fired power plants are the most important power supply mode in China, but are one of the main sources causing atmospheric pollution, and with the promotion of the ultralow emission process of the coal-fired unit flue gas, the cooperative removal control of conventional pollutants such as sulfur, nitrogen, dust and the like in the coal-fired unit flue gas is also of great concern.
However, the current ultralow emission process of the coal-fired unit flue gas is usually independently operated, mainly meets the continuously enhanced pollutant emission standard in a step-by-step execution mode, and is constructed according to the progressive processes of dust removal, desulfurization and denitration; the flue gas pollutant control process has complex design and has the following problems:
1) Single pollutant treatments may adversely affect other pollutant control devices, making the overall process difficult to operate in an efficient manner.
2) The prediction of pollutant emission concentration is severely dependent on real-time detection data of a power plant flue gas pollutant control device.
Therefore, it is necessary to develop a conventional pollutant control system integrating predictive feedback and collaborative optimization control.
Disclosure of Invention
The invention aims to overcome the defects of the background technology and provide a coal-fired boiler flue gas prediction and feedback control system based on an LSTM and particle swarm algorithm.
In order to achieve the above purpose, the technical scheme of the invention is as follows:
coal-fired boiler flue gas prediction and feedback control system based on LSTM and particle swarm optimization, its characterized in that: the method comprises solving an LSTM prediction model, an ultralow emission system cost control model and a particle swarm model, wherein the LSTM prediction model utilizes data acquired from a DCS to train the prediction model, finally, the real-time pollutant emission concentration prediction is realized, and the boiler flue gas pollutants comprise SO 2 Emissions, NO X Emissions and smoke emissions;
the construction of the LSTM prediction model comprises a data preparation stage, a data processing and selecting stage, a model construction stage and a model training and evaluating stage;
in the data preparation stage, collecting and preparing a data set for training, wherein the data set is a plurality of sample points, and each sample point has a boiler load characteristic, a flue gas flow characteristic and a flue gas temperature characteristic;
in the data processing and selecting stage, the data are cleaned and screened, firstly, the abnormal value is replaced by using the 3 sigma principle, then the data are normalized or standardized by using the Min-max method, and the data are ensured to be in the same scale, and the corresponding formula is as follows:
where j represents the j-th feature and i is the i-th sample point;
in the model construction stage, a predictive model is constructed by using a pytorch platform, the constructed model structure is a 1-layer LSTM neural network and a 3-layer Dense neural network, and each super parameter of the model is as follows: the dimension of the LSTM layer hidden layer is 32,3, the dimension of the Dense layer neural network is 128, 32 and 3 respectively in sequence, the model learning rate is 0.006, and the training round is 1000;
the ultra-low emission system cost control model comprises an SCR system cost model, a dust removal system cost model and a limestone-gypsum wet desulphurization system cost model; SCR system cost model coupled NO X The prediction of emission is coupled with the cost model of the dust removal system, and the cost model of the limestone-gypsum wet desulfurization system is coupled with SO 2 Prediction of emissions;
the particle swarm model solving comprises the steps of establishing a cost optimization objective function, calculating an optimization control quantity by using a particle swarm algorithm to perform feedback control, wherein the control quantity in the particle swarm model solving comprises the following steps: ammonia injection amount, electric field voltage of an electrostatic precipitator, pH value of limestone slurry and running number of circulating pumps;
and after the system is built, optimizing, controlling and evaluating the historical data by using the system model.
In the technical scheme, in the data preparation stage, a plurality of sample points are collected, and in order to fully utilize the data set, the data set is divided into a training set and a testing set, namely 80% of data is randomly selected as the training set, and the other 20% is selected as the testing set; the training set is used to train the model and the test set is used to ultimately evaluate the model performance.
In the technical scheme, after the data processing and selecting stage finishes cleaning, the input features are screened by using a correlation coefficient method, namely, the pearson correlation coefficients between all variables and the prediction targets are calculated, and the features with the correlation coefficients larger than 0.3 are reserved as the input features.
In the above technical scheme, the cost model of the SCR system is generated by a soot blower, an induced draft fan and a dilution fan, and the cost formula of the cost model of the SCR system is as follows:
wherein COST is SCR-idf COST for induced draft fan sb COST of soot blower adf For diluting the cost of the fan, q is the real-time load of the boiler, P E For electricity price, U i 、I i Voltage and current of ith equipment, n idf 、n sb 、n adf The running numbers of the induced draft fan, the soot blower and the dilution blower are respectively, the cos psi is the power factor, 0.8 is taken, and P steam Is the empirical steam energy consumption, CV s Is the empirical reference catalyst amount, CV is the actual catalyst amount; alpha SCR Represents the resistance pressure drop P of the denitration reactor SCR Accounting for total resistance pressure drop P of flue gas system idf The calculating method comprises the following steps:
the soot blowing system has different soot blowing modes and different operation COST calculation formulas, wherein the material consumption of the SCR system mainly comprises liquid ammonia and a catalyst, and the COST of the liquid ammonia is COST NH3 The method comprises the following steps:
catalyst COST C The cost is as follows:
wherein C is NOx-in And C NOx-out Respectively representing the concentration of NOx contained in inlet flue gas and outlet flue gas of a denitration system, V is the flue gas flow, ae is the concentration of ammonia water, and V m Is the flow rate of ammonia water, M NH3 And M is as follows NO Respectively NH 3 Molecular weight with NO, delta 2 Taking 1 to 1.2 of ammonia nitrogen molar ratio and P NH3 Is the price of ammonia water, P C The price of the catalyst is that Q is the unit capacity, and h is the unit annual operation hours;
SCR system COST model COST SCR Expressed as:
in the above technical scheme, the cost model sources of the dust removal system are the power consumption of the induced draft fan and the power supply, and the calculation formulas are respectively as follows:
wherein COST is as follows ESP-idf COST is the electricity consumption COST of the induced draft fan ESP-e N is the electricity consumption cost of the power supply e Indicating the number of electric fields, alpha ESP Resistance pressure drop P for dust collector ESP Accounting for total resistance pressure drop P of flue gas system idf The calculation formula of the ratio is as follows:
then the COST model COST of the dust removal system ESP Expressed as:
COST ESP ＝COST ESP_idf +COST ESP_e (equation 12)
In the technical scheme, the cost model of the limestone-gypsum wet desulfurization system consists of electricity consumption and material consumption, wherein the electricity consumption source comprises a booster fan or an induced draft fan, an oxidation fan, a slurry circulating pump and a slurry stirrer, and the calculation formula is as follows:
wherein COST is as follows bf COST for booster fan or induced draft fan sa To oxidize the COST of the fan, COST SCP COST for slurry circulation pump oab N is the cost of the slurry mixer bf 、n sa 、n scp 、n oab Respectively representing the operation number of the booster fan, the oxidation fan, the slurry circulating pump and the slurry stirrer, alpha WFGD The calculation formula of the resistance of the desulfurizing tower accounting for the total resistance of the flue gas system is as follows:
wherein P is dt Is the pressure drop of the desulfurizing tower, P ESP Is the resistance pressure drop of the electric dust collector, P gd2 Is the resistance pressure drop of the flue part;
the absorbent of the desulfurization system is limestone slurry, and the cost calculation mode is as follows:
wherein C is SO2-in And C SO2-out Respectively represent SO contained in inlet flue gas and outlet flue gas of desulfurization system 2 Concentration, M CaCO3 And M is as follows SO2 CaCO respectively 3 With SO 2 Molecular weight, delta 1 The ratio of calcium to sulfur is set to be 1.02-1.05, lambda is the purity of limestone, P CaCO3 For the price of limestone, V represents the flue gas flow and is in direct proportion to the boiler load, and the calculation formula is as follows:
V＝m×q×V tc (equation 19)
Wherein V is tc The amount of smoke generated by unit coal;
the desulfurization system also consumes process water, COST of the process water WFGD-w The method comprises the following steps:
in addition, the limestone-gypsum wet desulfurization system also produces gypsum, which can be used as a benefit, the benefit R of the gypsum CaSO4 The method comprises the following steps:
wherein M is CaSO4 Is the molecular weight of gypsum, P W For the price of water, P CaSO4 Is the price of gypsum;
COST model COST of limestone-gypsum wet desulfurization system WFGD Expressed as:
in the above technical solution, the cost optimization objective function minCOST and constraint conditions in the particle swarm model solution are:
wherein m is NH3 Is the ammonia spraying flow; u (U) 1 -U 4 The voltages of the first electric field to the fourth electric field of the electrostatic precipitator respectively; n (N) PUMP The input quantity of the slurry circulating pump is calculated; c NOX Controlling a concentration value for NOx; c SO2 Controlling a concentration value for SO 2; c PM The concentration value is controlled for the soot.
In the technical scheme, after the system is built, the system model is applied to perform optimal control evaluation on the historical data SO 2 、NO x The smoke emission optimization results are listed in the following table:
table 1.
Compared with the prior art, the invention has the following advantages:
1) Aiming at the characteristics of multiple inputs, multiple outputs and synergistic effect of multiple devices of a coal-fired ultralow emission system, the invention takes the whole ultralow emission system as a black box model, key influence parameters of all subsystems are taken as input characteristics, key controllable parameters are taken as input variable characteristics, pollutant emission concentration is taken as output characteristics, real-time prediction of pollutant emission is realized based on intelligent algorithms such as LSTM (least squares) and the like, meanwhile, the sum of material consumption and energy consumption of the subsystems is taken as a global cost model, the safety limit of the key controllable parameters is taken as an optimization limit condition, and the particle swarm algorithm is utilized to perform optimizing calculation on the key controllable parameters so as to realize pollutant prediction and cooperative optimization control.
Drawings
Fig. 1 is a system block diagram of the present invention.
Fig. 2 is a flow chart of the present invention.
FIG. 3 is SO of LSTM predictive model 2 And (5) predicting effect comparison graphs.
FIG. 4 is a NO of the LSTM prediction model X And (5) predicting effect comparison graphs.
FIG. 5 is a graph comparing smoke prediction effects of LSTM prediction models.
Detailed Description
The following detailed description of the invention is, therefore, not to be taken in a limiting sense, but is made merely by way of example. While the advantages of the invention will become apparent and readily appreciated by reference to the following description.
As can be seen with reference to the accompanying drawings:
coal-fired boiler flue gas prediction and feedback control system based on LSTM and particle swarm optimization, its characterized in that: the method comprises an LSTM prediction model 1, an ultra-low emission system cost control model 2 and a particle swarm model solution 3, wherein the LSTM prediction model 1 utilizes data acquired from a DCS to train the prediction model, finally, the real-time pollutant emission concentration prediction is realized, and boiler pollutants comprise SO 2 Emissions, NO X Emissions and smoke emissions;
the construction of the LSTM prediction model 1 comprises a data preparation stage, a data processing and selecting stage, a model construction stage and a model training and evaluating stage;
in the data preparation stage, collecting and preparing a data set for training, wherein the data set is a plurality of sample points, and each sample point has a boiler load characteristic, a flue gas flow characteristic and a flue gas temperature characteristic;
in the data processing and selecting stage, cleaning and screening the data, firstly replacing an abnormal value by using a 3 sigma principle, namely replacing by using a mu+3 sigma value when the data value is larger than the mean mu plus 3 standard deviations sigma, and replacing by using a mu-3 sigma value when the data value is smaller than the mean mu minus 3 standard deviations sigma;
the data were then normalized or normalized using the Min-max method, ensuring that the data were on the same scale. The corresponding formula is as follows:
where j represents the j-th feature and i is the i-th sample point.
In the model construction stage, a pytorch platform is used for constructing a prediction model, the constructed model structure is a 1-layer LSTM neural network and a 3-layer Dense layer neural network, meanwhile, in order to prevent overfitting, a discarding method (dropout) is introduced between the LSTM and the Dense layers, the discarding proportion is 20%, and a Relu function is added between the Dense layers; the number of the super-parameter hidden layer units of the model, the learning rate, the iteration number and the like are all obtained by a manual adjustment method; meanwhile, in order to fully utilize the training data set, 5-fold cross validation is adopted in model training, and finally the training and adjustment are carried out, so that each super parameter of the model is obtained: the dimension of the LSTM layer hidden layer is 32,3, the dimension of the Dense layer neural network is 128, 32 and 3 respectively in sequence, the model learning rate is 0.006, and the training round is 1000; fig. 3, 4 and 5 are comparison diagrams of prediction effects of the prediction model after training.
The ultra-low emission system cost control model 2 comprises an SCR system cost model 21, a dust removal system cost model 22 and a limestone-gypsum wet desulphurization system cost model 23; SCR system cost model 21 is coupled with NO X Prediction of emissions, a dust removal system cost model 22 coupled to prediction of soot emissions, and a limestone-gypsum wet desulfurization system cost model 23 coupled to SO 2 Prediction of emissions;
the particle swarm model solving 3 comprises the steps of establishing a cost optimization objective function, calculating an optimization control quantity by using a particle swarm algorithm to perform feedback control, wherein the control quantity in the particle swarm model solving 3 comprises the following steps: ammonia injection amount, electric field voltage of an electrostatic precipitator, pH value of limestone slurry and running number of circulating pumps;
and after the system is built, optimizing, controlling and evaluating the historical data by using the system model.
In the data preparation stage, a large number of (8640) sample points are collected, and in order to fully utilize the data set, the data set is divided into a training set and a testing set, namely 80% of data is randomly selected as the training set, and the other 20% is selected as the testing set; the training set is used to train the model and the test set is used to ultimately evaluate the model performance.
In the data processing and selecting stage, after the data are cleaned, in order to reduce the data dimension and the feature space, thereby improving the performance and generalization capability of the model, reducing the calculation cost and reducing the risk of overfitting, the input features are screened by using a correlation coefficient method, namely, the pearson correlation coefficients between all variables and the prediction targets are calculated, and the features with the correlation coefficient larger than 0.3 are reserved as the input features.
The SCR system cost model 21 is generated by a soot blower, an induced draft fan and a dilution fan, and the cost formula of the SCR system cost model 21 is as follows:
wherein COST is SCR-idf COST for induced draft fan sb COST of soot blower adf For diluting the cost of the fan, q is the real-time of the boilerLoad, P E For electricity price, U i 、I i Voltage and current of ith equipment, n idf 、n sb 、n adf The running numbers of the induced draft fan, the soot blower and the dilution blower are respectively, the cos psi is the power factor, 0.8 is taken, and P steam Is the empirical steam energy consumption, CV s Is the empirical reference catalyst amount, CV is the actual catalyst amount; alpha SCR Represents the resistance pressure drop P of the denitration reactor SCR Accounting for total resistance pressure drop P of flue gas system idf The calculating method comprises the following steps:
the soot blowing system has different soot blowing modes and different operation COST calculation formulas, wherein the material consumption of the SCR system mainly comprises liquid ammonia and a catalyst, and the COST of the liquid ammonia is COST NH3 The method comprises the following steps: the material consumption of the SCR system mainly comprises liquid ammonia and a catalyst, and the cost of the liquid ammonia is as follows:
the catalyst cost is as follows:
wherein C is NOx-in And C NOx-out Respectively representing the concentration of NOx contained in inlet flue gas and outlet flue gas of a denitration system, V is the flue gas flow, ae is the concentration of ammonia water, and V m Is the flow rate of ammonia water, M NH3 And M is as follows NO Respectively NH 3 Molecular weight with NO, delta 2 Taking 1 to 1.2 of ammonia nitrogen molar ratio and P NH3 Is the price of ammonia water, P C The price of the catalyst is that Q is the unit capacity, and h is the unit annual operation hours;
then SCR System COST model (21) COST SCR Expressed as:
the dust removal system cost model 22 is derived from the power consumption of an induced draft fan and a power supply, and the calculation formulas are respectively as follows:
wherein COST is as follows ESP-idf COST is the electricity consumption COST of the induced draft fan ESP-e N is the electricity consumption cost of the power supply e Indicating the number of electric fields, alpha ESP Resistance pressure drop P for dust collector ESP Accounting for total resistance pressure drop P of flue gas system idf The calculation formula of the ratio is as follows:
then the dust removal system COST model 22COST ESP Expressed as:
COST ESP ＝COST ESP_idf +COST ESP_e (equation 12)
The limestone-gypsum wet desulfurization system cost model 23 is composed of electricity consumption and material consumption, wherein the electricity consumption sources comprise a booster fan, an oxidation fan, a slurry circulating pump and a slurry stirrer, and the calculation formula is as follows:
wherein COST is as follows bf COST for booster fan or induced draft fan sa To oxidize the COST of the fan, COST SCP COST for slurry circulation pump oab N is the cost of the slurry mixer bf 、n sa 、n scp 、n oab Respectively representing the operation number of a booster fan or an induced draft fan, an oxidation fan, a slurry circulating pump and a slurry stirrer, alpha WFGD The calculation formula of the resistance of the desulfurizing tower accounting for the total resistance of the flue gas system is as follows:
wherein p is dt Is the pressure drop of the desulfurizing tower, p ESP Is the resistance pressure drop of the electric dust collector, p gd2 Is the partial resistance pressure drop of the flue.
The absorbent of the desulfurization system is limestone slurry, and the cost calculation mode is as follows:
wherein C is SO2-in And C SO2-out Respectively represent SO contained in inlet flue gas and outlet flue gas of desulfurization system 2 Concentration, M CaCO3 And M is as follows SO2 CaCO respectively 3 With SO 2 Molecular weight, delta 1 The ratio of calcium to sulfur is set to be 1.02-1.05, lambda is the purity of limestone, P CaCO3 For the price of limestone, V represents the flue gas flow and is in direct proportion to the boiler load, and the calculation formula is as follows:
V＝m×q×V tc (equation 19)
Wherein V is tc The amount of smoke generated by unit coal;
the desulfurization system also consumes process water, COST of the process water WFGD-w The method comprises the following steps:
in addition, the limestone-gypsum wet desulfurization system also produces gypsum, which can be used as a benefit, the benefit R of the gypsum CaSO4 The method comprises the following steps:
wherein M is CaSO4 Is the molecular weight of gypsum, P W For the price of water, P CaSO4 Is the price of gypsum;
limestone-gypsum wet desulfurization system COST model (23) COST WFGD Expressed as:
the cost optimization objective function minCOST and constraint conditions in the particle swarm model solution 3 are as follows:
wherein m is NH3 Is the ammonia spraying flow; u (U) 1 -U 4 The voltages of the first electric field to the fourth electric field of the electrostatic precipitator respectively; n (N) PUMP The input quantity of the slurry circulating pump is calculated; c NOX Controlling a concentration value for NOx; c SO2 Controlling a concentration value for SO 2; c PM The concentration value is controlled for the soot.
After the system is built, the system model is applied to perform optimal control evaluation on the historical data SO 2 、NO x The smoke emission optimization results are listed in the following table:
table 1.
The real-time input data of the invention extracts DCS parameters by each point physical sensor; physical sensors at each point: and the system is used for acquiring the control parameters of each DCS and providing real-time data calculated by each module. The industrial personal computer: and a system algorithm is built in and used for controlling the boiler system through the DCS. A data line: the device is used for connecting the physical sensors at all the points with the industrial personal computer.
Other non-illustrated parts are known in the art.
Claims (8)
1. Coal-fired boiler flue gas prediction and feedback control system based on LSTM and particle swarm optimization, its characterized in that: the method comprises an LSTM prediction model (1), an ultralow emission system cost control model (2) and a particle swarm model solution (3), wherein the LSTM prediction model (1) trains the prediction model by utilizing data acquired from a DCS (distributed control system), and finally realizes the prediction of the emission concentration of pollutants in real time, and the pollutants in boiler flue gas comprise SO (sulfur dioxide) 2 Emissions, NO X Emissions and smoke emissions;
the construction of the LSTM prediction model (1) comprises a data preparation stage, a data processing and selecting stage, a model construction stage and a model training and evaluating stage;
in the data preparation stage, collecting and preparing a data set for training, wherein the data set is a plurality of sample points, and each sample point has a boiler load characteristic, a flue gas flow characteristic and a flue gas temperature characteristic;
in the data processing and selecting stage, the data are cleaned and screened, firstly, the abnormal value is replaced by using the 3 sigma principle, then the data are normalized or standardized by using the Min-max method, and the data are ensured to be in the same scale, and the corresponding formula is as follows:
where j represents the j-th feature and i is the i-th sample point;
in the model construction stage, a predictive model is constructed by using a pytorch platform, the constructed model structure is a 1-layer LSTM neural network and a 3-layer Dense neural network, and each super parameter of the model is as follows: the dimension of the LSTM layer hidden layer is 32,3, the dimension of the Dense layer neural network is 128, 32 and 3 respectively in sequence, the model learning rate is 0.006, and the training round is 1000;
the ultra-low emission system cost control model (2) comprises an SCR system cost model (21), a dust removal system cost model (22) and a limestone-gypsum wet desulphurization system cost model (23); SCR System cost model (21) coupled with NO X Prediction of emissions, a dust removal system cost model (22) coupled to prediction of soot emissions, a limestone-gypsum wet desulfurization system cost model (23) coupled to SO 2 Prediction of emissions;
the particle swarm model solving (3) comprises the steps of establishing a cost optimization objective function, calculating an optimization control quantity by using a particle swarm algorithm to perform feedback control, wherein the control quantity in the particle swarm model solving (3) comprises the following steps: ammonia injection amount, electric field voltage of an electrostatic precipitator, pH value of limestone slurry and running number of circulating pumps;
and after the system is built, optimizing, controlling and evaluating the historical data by using the system model.
2. The LSTM and particle swarm algorithm based coal-fired boiler flue gas prediction and feedback control system of claim 1, wherein: in the data preparation stage, a plurality of sample points are collected, and in order to fully utilize a data set, the data set is divided into a training set and a testing set, namely 80% of data is randomly selected as the training set, and the other 20% is selected as the testing set; the training set is used to train the model and the test set is used to ultimately evaluate the model performance.
3. The LSTM and particle swarm algorithm based coal-fired boiler flue gas prediction and feedback control system of claim 2, wherein: and in the data processing and selecting stage, after the data are cleaned, the input features are screened by using a correlation coefficient method, namely, the pearson correlation coefficients between all variables and the prediction targets are calculated, and the features with the correlation coefficients larger than 0.3 are reserved as the input features.
4. The LSTM and particle swarm algorithm based coal-fired boiler flue gas prediction and feedback control system according to claim 3, wherein: the SCR system cost model (21) is generated by a soot blower, an induced draft fan and a dilution fan, and the cost formula of the SCR system cost model (21) is as follows:
wherein COST is SCR-idf COST for induced draft fan sb COST of soot blower adf For diluting the cost of the fan, q is the real-time load of the boiler, P E For electricity price, U i 、I i Voltage and current of ith equipment, n idf 、n sb 、n adf The running numbers of the induced draft fan, the soot blowing fan and the dilution fan are respectively, P i Steam blowing means electric power of steam blower, P i The sound wave soot blowing means the electric power of the sound wave soot blower, cos psi is the power factor, 0.8, P steam Is the empirical steam energy consumption, CV s Is the empirical reference catalyst amount, CV is the actual catalyst amount; alpha SCR Represents the resistance pressure drop P of the denitration reactor SCR Accounting for total resistance pressure drop P of flue gas system idf The calculating method comprises the following steps:
the soot blowing system has different soot blowing modes and different operation COST calculation formulas, wherein the material consumption of the SCR system mainly comprises liquid ammonia and a catalyst, and the COST of the liquid ammonia is COST NH3 The method comprises the following steps:
catalyst COST C The cost is as follows:
wherein C is NOx-in And C NOx-out Respectively representing the concentration of NOx contained in inlet flue gas and outlet flue gas of a denitration system, V is the flue gas flow, ae is the concentration of ammonia water, and V m Is the flow rate of ammonia water, M NH3 And M is as follows NO Respectively NH 3 Molecular weight with NO, delta 2 Taking 1 to 1.2 of ammonia nitrogen molar ratio and P NH3 Is the price of ammonia water, P C The price of the catalyst is that Q is the unit capacity, and h is the unit annual operation hours;
then SCR System COST model (21) COST SCR Expressed as:
5. the LSTM and particle swarm algorithm based coal-fired boiler flue gas prediction and feedback control system of claim 4, wherein: the dust removal system cost model (22) is derived from the power consumption of an induced draft fan and a power supply, and the calculation formulas are respectively as follows:
wherein COST is as follows ESP-idf COST is the electricity consumption COST of the induced draft fan ESP-e N is the electricity consumption cost of the power supply e Indicating the number of electric fields, alpha ESP Resistance pressure drop P for dust collector ESP Accounting for total resistance pressure drop P of flue gas system idf The calculation formula of the ratio is as follows:
then the dedusting System COST model (22) COST ESP Expressed as:
COST ESP ＝COST ESP_idf +COST ESP_e (equation 12)
6. The LSTM and particle swarm algorithm based coal-fired boiler flue gas prediction and feedback control system of claim 5, wherein: the limestone-gypsum wet desulphurization system cost model (23) consists of electricity consumption and material consumption, wherein the electricity consumption sources comprise a booster fan or an induced draft fan, an oxidation fan, a slurry circulating pump and a slurry stirrer, and the calculation formula is as follows:
wherein COST is as follows bf COST for booster fan or induced draft fan sa To oxidize the COST of the fan, COST SCP COST for slurry circulation pump oab N is the cost of the slurry mixer bf 、n sa 、n scp 、n oab Respectively representing the operation number of the booster fan, the oxidation fan, the slurry circulating pump and the slurry stirrer, alpha WFGD The calculation formula of the resistance of the desulfurizing tower accounting for the total resistance of the flue gas system is as follows:
wherein P is dt Is the pressure drop of the desulfurizing tower, P ESP Is the resistance pressure drop of the electric dust collector, P gd2 Is the resistance pressure drop of the flue part;
the absorbent of the desulfurization system is limestone slurry, and the cost calculation mode is as follows:
wherein C is SO2-in And C SO2-out Respectively represent SO contained in inlet flue gas and outlet flue gas of desulfurization system 2 Concentration, M CaCO3 And M is as follows SO2 CaCO respectively 3 With SO 2 Molecular weight, delta 1 The ratio of calcium to sulfur is set to be 1.02-1.05, lambda is the purity of limestone, P CaCO3 For the price of limestone, V represents the flue gas flow and is in direct proportion to the boiler load, and the calculation formula is as follows:
V＝m×q×V tc (equation 19)
Wherein V is tc The amount of smoke generated by unit coal;
desulfurization systemAlso consume process water, COST of process water WFGD-w The method comprises the following steps:
in addition, the limestone-gypsum wet desulfurization system also produces gypsum, which can be used as a benefit, the benefit R of the gypsum CaSO4 The method comprises the following steps:
wherein M is CaSO4 Is the molecular weight of gypsum, P W For the price of water, P CaSO4 Is the price of gypsum;
limestone-gypsum wet desulfurization system COST model (23) COST WFGD Expressed as:
7. the LSTM and particle swarm algorithm based coal-fired boiler flue gas prediction and feedback control system of claim 6, wherein: the cost optimization objective function minCOST and constraint conditions in the particle swarm model solving (3) are as follows:
wherein m is NH3 Is the ammonia spraying flow; u (U) 1 -U 4 The voltages of the first electric field to the fourth electric field of the electrostatic precipitator respectively; n (N) PUMP The input quantity of the slurry circulating pump is calculated; c NOX Controlling a concentration value for NOx; c SO2 Controlling a concentration value for SO 2; c PM The concentration value is controlled for the soot.
8. According to the weightsThe LSTM and particle swarm optimization based coal-fired boiler flue gas prediction and feedback control system of claim 7, wherein: after the system is built, the system model is applied to perform optimal control evaluation on the historical data SO 2 、NO x The smoke emission optimization results are listed in the following table:
table 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311563289.3A CN117724410A (en) | 2024-01-15 | 2024-01-15 | LSTM and particle swarm optimization-based coal-fired boiler flue gas prediction and feedback control system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311563289.3A CN117724410A (en) | 2024-01-15 | 2024-01-15 | LSTM and particle swarm optimization-based coal-fired boiler flue gas prediction and feedback control system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117724410A true CN117724410A (en) | 2024-03-19 |
Family
ID=90207713
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311563289.3A Pending CN117724410A (en) | 2024-01-15 | 2024-01-15 | LSTM and particle swarm optimization-based coal-fired boiler flue gas prediction and feedback control system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117724410A (en) |
-
2024
- 2024-01-15 CN CN202311563289.3A patent/CN117724410A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113082954B (en) | Whole-process intelligent operation regulation and control system of wet desulphurization device | |
| CN109472406B (en) | Global optimization method of ultralow-emission multi-pollutant collaborative removal system | |
| CN114191953B (en) | Flue gas desulfurization and denitrification control method based on convolutional neural network and XGBoost | |
| CN109034457B (en) | Low-cost collaborative removal modeling and optimization method for pollutants of coal-fired power plant | |
| CN107748955B (en) | Energy efficiency assessment method for ultra-low emission environmental protection island of coal-fired power plant | |
| CN114225662B (en) | Hysteresis model-based flue gas desulfurization and denitrification optimal control method | |
| CN110263452B (en) | Flue gas time distribution characteristic analysis method and system in flue and denitration system | |
| US20060121616A1 (en) | Method and system for increasing efficiency of FGD operation in fossil fuel boilers | |
| CN111460726B (en) | Optimization method for ammonia escape of coal slime fluidized bed boiler denitration system | |
| CN112364562B (en) | Flue gas environment-friendly island cooperative control method and system | |
| CN112465239B (en) | Desulfurization system operation optimization method based on improved PSO-FCM algorithm | |
| CN116688754A (en) | Ship flue gas desulfurization automatic control system and method thereof | |
| CN110299188A (en) | SCR flue gas denitrification system GRNN modeling method based on GA variables choice | |
| CN112365065B (en) | WFGD self-adaptive online optimization scheduling method | |
| CN109766666A (en) | Boiler smoke based on low nitrogen burning and SNCR-SCR Collaborative Control discharges NOxConcentration prediction method | |
| EP3024561A1 (en) | Controlling aqcs parameters in a combustion process | |
| CN113175678A (en) | Method and device for monitoring garbage incineration | |
| CN111564184A (en) | Limestone-gypsum wet desulphurization SO of coal-fired power plant3Collaborative removal efficiency prediction method | |
| JP2023081485A (en) | Quantity-of-state prediction device and quantity-of-state prediction method | |
| CN117724410A (en) | LSTM and particle swarm optimization-based coal-fired boiler flue gas prediction and feedback control system | |
| CN109933884B (en) | Neural network inverse control method for SCR denitration system of coal-fired unit | |
| CN117180937A (en) | Method and system for eliminating white smoke from desulfurized flue gas | |
| CN114429236A (en) | Natural gas purification process control method | |
| CN116050643A (en) | Method for predicting emission concentration of process industrial pollutants based on integrated model | |
| CN115509189A (en) | Big data regulation and control method and system applied to wet desulphurization system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |