CN111804146A - Intelligent ammonia injection control method and intelligent ammonia injection control device - Google Patents

Intelligent ammonia injection control method and intelligent ammonia injection control device Download PDF

Info

Publication number
CN111804146A
CN111804146A CN202010606351.2A CN202010606351A CN111804146A CN 111804146 A CN111804146 A CN 111804146A CN 202010606351 A CN202010606351 A CN 202010606351A CN 111804146 A CN111804146 A CN 111804146A
Authority
CN
China
Prior art keywords
ammonia injection
model
nox emission
prediction model
emission concentration
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010606351.2A
Other languages
Chinese (zh)
Other versions
CN111804146B (en
Inventor
唐静
侯静静
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.)
Yuanguang Software Co Ltd
Original Assignee
Yuanguang Software 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 Yuanguang Software Co Ltd filed Critical Yuanguang Software Co Ltd
Priority to CN202010606351.2A priority Critical patent/CN111804146B/en
Publication of CN111804146A publication Critical patent/CN111804146A/en
Application granted granted Critical
Publication of CN111804146B publication Critical patent/CN111804146B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
    • B01D—SEPARATION
    • B01D53/00—Separation of gases or vapours; Recovering vapours of volatile solvents from gases; Chemical or biological purification of waste gases, e.g. engine exhaust gases, smoke, fumes, flue gases, aerosols
    • B01D53/34—Chemical or biological purification of waste gases
    • B01D53/74—General processes for purification of waste gases; Apparatus or devices specially adapted therefor
    • B01D53/86—Catalytic processes
    • B01D53/90—Injecting reactants
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
    • B01D—SEPARATION
    • B01D53/00—Separation of gases or vapours; Recovering vapours of volatile solvents from gases; Chemical or biological purification of waste gases, e.g. engine exhaust gases, smoke, fumes, flue gases, aerosols
    • B01D53/34—Chemical or biological purification of waste gases
    • B01D53/74—General processes for purification of waste gases; Apparatus or devices specially adapted therefor
    • B01D53/86—Catalytic processes
    • B01D53/8621—Removing nitrogen compounds
    • B01D53/8625—Nitrogen oxides
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F23—COMBUSTION APPARATUS; COMBUSTION PROCESSES
    • F23J—REMOVAL OR TREATMENT OF COMBUSTION PRODUCTS OR COMBUSTION RESIDUES; FLUES 
    • F23J15/00—Arrangements of devices for treating smoke or fumes
    • F23J15/003—Arrangements of devices for treating smoke or fumes for supplying chemicals to fumes, e.g. using injection devices
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield 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
    • G06N3/04—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
    • G06N3/12—Computing arrangements based on biological models using genetic models
    • G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
    • B01D—SEPARATION
    • B01D2251/00—Reactants
    • B01D2251/20—Reductants
    • B01D2251/206—Ammonium compounds
    • B01D2251/2062—Ammonia
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
    • B01D—SEPARATION
    • B01D2258/00—Sources of waste gases
    • B01D2258/02—Other waste gases
    • B01D2258/0283—Flue gases

Abstract

The invention discloses an intelligent ammonia injection control method and an intelligent ammonia injection control device, and relates to the technical field of denitration and automatic ammonia injection in the thermal power industry; the intelligent ammonia injection control method comprises the following steps: s1, determining an analysis measuring point; s2, collecting time T according to sampling interval SHHistorical data acquisition history of each analysis measuring point and NOx emission concentration measuring point in the deviceA data set; s3, carrying out optimal prediction model training of NOx emission concentration through a deep learning algorithm; s4, carrying out field deployment on the optimal prediction model; s5, collecting real-time data of each analysis measuring point, and predicting future NOx emission concentration through the optimal prediction model; and the desulfurization and denitrification control system controls the ammonia injection amount according to the prediction result so as to realize intelligent quantitative ammonia injection. The intelligent ammonia injection control method has the advantages that the prediction of the NOx emission concentration is accurate, the prediction result is used as the input of the desulfurization and denitrification control system, the accurate ammonia injection in the full-load range is realized, and the problems of excessive ammonia injection and insufficient ammonia injection are thoroughly solved.

Description

Intelligent ammonia injection control method and intelligent ammonia injection control device
Technical Field
The invention relates to the technical field of denitration and automatic ammonia spraying in the thermal power industry, in particular to an intelligent ammonia spraying control method and an intelligent ammonia spraying control device.
Background
Driven by economic benefits and increasingly stringent environmental requirements, power plants need to increase competitiveness by increasing unit efficiency and reducing pollutant emissions. NOx (nitrogen oxide) is used as one of the atmospheric pollutants discharged by the combustion of coal fired in a boiler of a thermal power plant, the discharge amount of the NOx is a key index for measuring the combustion quality of the boiler, and the NOx is a basis for judging whether the boiler operates in a green mode or not. The SCR denitration ammonia spraying system is an effective technology for reducing NOx emission through chemical reaction, is generally applied to a large-scale power plant boiler at present, and achieves good effect on the problem of NOx environmental pollution. However, the existing denitration ammonia injection system adopts fixed ammonia distribution, cannot adapt to the variability of denitration inlet flue gas NOx under the condition of unit flexibility peak shaving, and has the problems of excessive ammonia injection and insufficient ammonia injection amount. Under the condition of meeting the unit outlet emission, excessive ammonia injection can cause the increase of consumables and the blockage of a boiler air preheater and other problems, and the ammonia injection amount is too small to meet the NOx outlet emission standard.
At present, in the process of generating the combustion NOx emission concentration of a power plant boiler in real time, on one hand, instruments and meters are generally adopted to measure the concentration, the measurement and calculation process has certain time delay, and the monitored NOx concentration time does not correspond to the real emission time; on the other hand, the prediction generally adopts modes such as empirical formula, data mining and traditional machine learning, and can not adapt to the change of variable working condition operation parameters, the NOx emission concentration is difficult to predict accurately, the practical application effect is very poor, and the two technical defects result in that the NOx generated by boiler combustion can not be processed accurately.
In addition, when the time series prediction is performed by adopting the LSTM neural network in deep learning in the prior art, the LSTM neural network has higher prediction precision only by continuously adjusting and optimizing each parameter of the LSTM neural network, so that the calculation cost is high, the parameter adjusting time is long, and the optimal parameter combination of the LSTM neural network cannot be obtained.
Disclosure of Invention
The first purpose of the invention is to provide an intelligent ammonia injection control method, which can accurately predict the NOx emission concentration and implement intelligent ammonia injection, and solves the problem that the real-time ammonia injection amount is inconsistent with the NOx emission concentration; the first purpose of the invention is realized by the following technical scheme:
an intelligent ammonia injection control method is characterized by comprising the following steps:
s1, determining analysis measuring points, wherein the analysis measuring points comprise fixed measuring points and random measuring points, and the fixed measuring points comprise key operation measuring points and key monitoring measuring points of the coal-fired unit;
s2, collecting time T according to sampling interval SHHistorical data of each analysis measuring point and NOx emission concentration measuring point in the NOx emission concentration measuring device is obtained to obtain a historical data set;
s3, carrying out optimal prediction model training of NOx emission concentration through a deep learning algorithm based on the historical data set;
s4, carrying out field deployment on the optimal prediction model;
s5, acquiring real-time data of each analysis measuring point on site by using the optimal prediction model, and predicting future NOx emission concentration through the optimal prediction model; and the desulfurization and denitrification control system controls the ammonia injection amount in real time at a future time point according to the prediction result so as to realize intelligent quantitative ammonia injection.
Specifically, in step S2, the historical data set is divided into a training data set and a test data set;
step S3 specifically comprises constructing a rolling LSTM neural network prediction model, optimizing each hyper-parameter of the LSTM neural network prediction model by adopting a genetic algorithm, training the LSTM neural network prediction model by using the optimized hyper-parameters and the training data set, performing iterative optimization and evaluation, and extracting an optimal prediction model; and predicting the trained optimal prediction model by using the test data set, and evaluating the model error.
Specifically, step S2 further includes preprocessing the history data, and step S2 specifically includes:
step S21, collecting time T from the data storage module according to the sampling interval SHHistorical data of each analysis measuring point and NOx emission concentration measuring point in the NOx emission concentration measuring device are obtained to obtain a historical data set
Figure BDA0002559245230000021
Wherein m represents the number of samples; k represents the measurement point dimension, k is 1 to P, wherein k is 1 to (P-1) represents the analysis measurement point, k is P represents the NOx emission concentration measurement point, bfi kHistorical data representing a measuring point k;
step S22, setting delay time D, and performing windowing processing on the collected historical data; obtaining a model parameter matrix BF of each window t:
Figure BDA0002559245230000022
wherein k is 1 to P; t is the time window length in the LSTM neural network prediction model; d represents the number of sampling points of the delay time length, D is equal to D/S, and D + n represents the number of sampling points of the prediction time length;
step S23, standardizing each data of each window to obtain a standardized parameter matrix AF of each window t:
Figure BDA0002559245230000031
wherein k is 1 to P;
step S24, dividing the data of each window into input data and output data as the input and output of the training model;
taking the first T data of each analysis measuring point in the standardized matrix of each window as a training model input value X, and taking the last (d + n) value of the NOx emission concentration measuring point in the standardized matrix of each window as a training model output value Y;
namely, it is
Figure BDA0002559245230000032
Wherein k is 1 to (P-1);
Figure BDA0002559245230000033
wherein T is the time window length in the LSTM neural network prediction model and the number of input data of the LSTM neural network prediction model; the training of the prediction model is performed by dividing the historical data into a plurality of windows, inputting the first T data of each analysis measuring point of each window, outputting the last (d + n) data of the NOx emission concentration of each window, and training the model according to the input data and the output data. Model training is completed, and in practical application, (d + n) NOx emission concentration data are predicted according to T data of each analysis measuring point.
Specifically, in step S23, the formula for each data to be normalized is:
the formula I is as follows:
Figure BDA0002559245230000034
wherein k is 1 to P, q is 0 to T + d + n,
Figure BDA0002559245230000035
the first sample point data of the window t.
Preferably, the ratio of the number of data in the training data set to the number of data in the test data set is 9: 1.
preferably, the sampling interval S includes 2S, 5S, and 10S.
Preferably, the time window length T > d + n.
Specifically, step S3 specifically includes:
s31, constructing an initial model of the LSTM neural network prediction model;
s32, optimizing partial hyper-parameters of the LSTM prediction model by using a genetic algorithm, and performing parameter combination optimization in a parameter search space by using the minimum error as a fitness function;
step S33, updating the initial model by using the super-parameter optimization result, and performing model training and iterative optimization by using the training model input value X and the training model output value Y of each window of the normalized training data set; training a Loss function Loss by using a mean square error as a model, and performing gradient calculation by using an Adam optimization algorithm;
and S34, using the standardized test data set to perform rolling test on the LSTM neural network model trained in the step S33, performing anti-standardization on the output result of the model to obtain a predicted value of NOx emission concentration, comparing the predicted value and the actual value of the NOx emission concentration, evaluating the error of the model, and determining the optimal prediction model.
Specifically, the hyper-parameters optimized by the genetic algorithm in step S33 include: time window length T, training times epoch, batch size per training.
In particular, the model training loss function of step S33 is
Figure BDA0002559245230000041
Wherein: c is the number of data in the training data set, afi PFor normalized actual value of NOx emission concentration, afi P*Is a predicted value of the normalized NOx emission concentration.
Specifically, the formula for denormalization of the predicted value is as follows:
the formula II is as follows:
Figure BDA0002559245230000042
wherein, bfi P*A predicted value of NOx emission concentration is obtained;
thereby finally predicting the value
Figure BDA0002559245230000043
Specifically, step S34 employs accuracy Q and average absolute percentage error E when comparing the predicted and actual values of NOx emission concentrationMAPEAs an evaluation index. Wherein, the accuracy Q:
the formula III is as follows:
Figure BDA0002559245230000044
mean absolute valueFor percentage error EMAPE：
The formula four is as follows:
Figure BDA0002559245230000045
in formula three and formula four, c is the number of data in the training data set, bfi PActual value of NOx emission concentration, bfi P*Is a predicted value of the NOx emission concentration.
Specifically, step S5 specifically includes:
step S51, collecting the real-time data of each analysis measuring point at sampling interval S to obtain a real-time data set
Figure BDA0002559245230000046
The sampling number is the time window length T; standardizing each acquired real-time data, and taking standardized real-time parameters as input values of the optimal prediction model to obtain a prediction output value Ypre of the optimal prediction model; carrying out anti-standardization processing on the prediction output value Ypre to finally obtain d + n real-time prediction sequences of the NOx emission concentration:
Figure BDA0002559245230000051
step S52, predicting result Ypre *Uploading the data to a desulfurization and denitrification control system, and removing the delay data number D of the delay time D to obtain an actual pre-sequencing sequence:
Figure BDA0002559245230000052
step S53, actual prediction sequence Y of NOx emission concentrationac *And controlling the ammonia injection amount, and controlling the ammonia injection amount in real time by the desulfurization and denitrification control system at a future time point according to a prediction result so as to realize intelligent quantitative ammonia injection.
Particularly, the key operation measuring points comprise total air volume, total secondary air volume, flue gas O2 content before denitration, coal quantity and unit load; the key monitoring measuring points comprise inlet NOx concentration; and screening the random measuring points from the monitoring database through random logistic regression and a random forest algorithm.
Specifically, the optimal prediction model trained in step S3 includes a fixed station model and a random station model; in step S4, the optimal prediction model deployed on site is one of the fixed measuring point model and the random measuring point model.
The second purpose of the invention is to provide an intelligent ammonia injection control device, and the second purpose of the invention is realized by the following technical scheme:
the intelligent ammonia injection control device is characterized by comprising a NOx emission concentration prediction model generation module, a historical data storage module and a prediction model application module;
the NOx emission concentration prediction model generation module comprises an analysis measuring point acquisition module, a historical data acquisition and preprocessing module, a prediction model training module and a model test module;
the analysis measuring point acquisition module is used for acquiring analysis measuring points required by model training, the analysis measuring points comprise fixed measuring points and random measuring points, and the fixed measuring points comprise key operation measuring points and key monitoring measuring points of the coal-fired unit;
the historical data acquisition and preprocessing module comprises a historical data acquisition submodule and a historical data preprocessing submodule;
the historical data acquisition submodule acquires a historical data set from the historical data storage module according to the analysis measuring point and the NOx emission concentration measuring point;
the historical data preprocessing submodule is used for carrying out standardization processing on each historical data in the historical data set;
the prediction model training module performs model training according to the historical data after standardization processing to obtain an optimal prediction model of the NOx emission concentration;
the model testing module tests the optimal prediction model;
and the prediction model application module predicts the NOx emission concentration at a future time point according to the optimal prediction model and the real-time data of the analysis measuring point.
Particularly, the prediction model application module comprises a real-time data acquisition and pretreatment module, a desulfurization and denitrification control system and an ammonia spraying module;
the real-time data acquisition and preprocessing module acquires real-time data of each analysis measuring point in an industrial field and carries out standardized processing on the real-time data; the output end of the real-time data acquisition and preprocessing module is connected with the input end of the optimal prediction model;
the output end of the optimal prediction model is connected with an ammonia injection amount calculation module of the desulfurization and denitrification control system;
the ammonia injection amount calculation module of the desulfurization and denitrification control system is used for calculating the ammonia injection amount at a future time point; the desulfurization and denitrification control system sends an ammonia injection automatic regulation opening command to the ammonia injection module at a future time point according to the ammonia injection amount calculated by the ammonia injection amount calculation module;
the ammonia injection module performs automatic ammonia injection.
The invention has the beneficial technical effects that:
according to the intelligent ammonia injection control method and the intelligent ammonia injection control device, the NOx emission concentration in the flue gas is predicted by means of big data analysis and artificial intelligence according to the load change condition in the temperature range allowed by the catalyst, the prediction result is used as the input of a desulfurization and denitrification control system, accurate ammonia injection in the full load range is realized, and the problems of excessive ammonia injection and insufficient ammonia injection are thoroughly solved.
Drawings
FIG. 1 is a block diagram of an intelligent ammonia injection control device according to an embodiment of the present invention;
FIG. 2 is a flow chart of an intelligent ammonia injection control method provided by an embodiment of the invention;
FIG. 3 is a graph comparing predicted values to actual values of NOx on a test data set for two predictive models provided by an embodiment of the present invention;
FIG. 4 is a graph comparing the real-time predicted value and the actual value of the NOx emission concentration of the industrial site provided by the embodiment of the invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.
It is noted that, unless otherwise indicated, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
With reference to fig. 3, the present invention provides an intelligent ammonia injection control method, which includes the following steps:
and S1, determining analysis measuring points, wherein the analysis measuring points comprise fixed measuring points and random measuring points, and the fixed measuring points comprise key operation measuring points and key monitoring measuring points of the coal-fired unit.
Specifically, each selected analysis measuring point is determined by analyzing the mechanism of the NOx gas emission monitoring of the industrial unit and the relevant monitoring measuring point set. The fixed measuring points are obtained according to a business principle and data analysis and confirmed by unit expert business experience, and the key operation measuring points of the coal-fired unit comprise: total air volume, total secondary air volume, flue gas O2 content before denitration, coal quantity and unit load; key monitoring points include inlet NOx concentration; the random measuring points are screened in the full measuring points (namely a monitoring database) through random logistic regression and random forest algorithm; the full measuring points refer to all monitoring points needing to be monitored during the operation of the industrial unit, such as pressure, temperature data and the like of a discharge port.
In this step, after each analysis measuring point is determined, data acquisition is performed according to each selected analysis measuring point in each subsequent step.
S2, collecting time T according to sampling interval SHHistorical data of each analysis measuring point and NOx emission concentration in the NOx emission measuring device are obtained, and a historical data set is obtained; and pre-processing the collected historical data.
Specifically, the step S2 specifically includes the following steps S21 to S24:
step S21, collecting time T from the historical data storage module according to the sampling interval SHThe historical data of each analysis measuring point and NOx emission concentration in the device are obtained to obtain a historical data set
Figure BDA0002559245230000071
Wherein m represents the number of samples; k represents the measurement point dimension, k is 1 to P, wherein k is 1 to (P-1) represents the analysis measurement point, k is P represents the NOx emission concentration measurement point, bfi kRepresenting historical data for station k.
The historical data set comprises a training data set DFTrainAnd test data set DFTest(ii) a Training data set DF for this embodimentTrainAnd test data set DFTestThe data ratio of (a) 9: 1.
the sampling interval S is set according to actual prediction requirements, such as the requirements of measuring and calculating speed and measuring and calculating precision of a computer (server) for measuring and calculating; the smaller the sampling interval S is, the accurate measurement result is, but the measurement amount is larger, and the requirement on the measurement performance of the server is higher; the sampling time interval may be 2s, 3s, 10 s; the sampling time interval employed in this embodiment takes 5 s.
Step S22, setting delay time D, and performing windowing processing on the collected historical data; obtaining a model parameter matrix BF of each window t;
Figure BDA0002559245230000081
wherein k is 1 to P; t is the time window length in the LSTM neural network prediction model;
Figure BDA0002559245230000082
the data of the first sampling point of the window t; d represents the number of delay sampling points, D is equal to D/S, D + n represents the number of sampling points of the predicted time length, and n represents the number of sampling points of the target predicted time length.
During actual field operation, on one hand, when an external monitoring system measures the numerical values of all analysis measuring points, the measurement and calculation process has certain time delay; on the other hand, the transmission time is also required for uploading the prediction result to the desulfurization and denitrification control system. Therefore, considering the two unavoidable time delay situations in the field, the delay time D is taken into account in the prediction result, so that each prediction requires output data of a future (D + N) time length to be predicted, where the time length N is the target prediction time length and N is N/S.
More specifically, for the training data set, each data is divided into a plurality of windows, and the data of each window is not repeatedly used; while the data in the various windows of the test data set are used in a rolling fashion.
In this step, in order to fully mine the history information, the time window length T must satisfy T > (d + n), i.e., the history data length is greater than the predicted data length.
Step S23, standardizing each data of each window to obtain a standardized parameter matrix AF of each window t:
Figure BDA0002559245230000083
wherein k is 1 to P;
because the sampling magnitude of each analysis measuring point is different, for example, the unit of NOx emission is mg, the unit of temperature is Kpa, and the like; the sampled values of different magnitude attributes are not directly calculable, and therefore, a normalization process (or normalization) is required for each data of the window.
Specifically, the data is sampled at the first of each window
Figure BDA0002559245230000087
The normalized data is used as a normalized base number, each data of the window is normalized to obtain normalized data, and the normalized base number of the window is ensured to be a true value when real-time prediction is carried out, wherein a formula for data normalization refers to a formula I:
the formula I is as follows:
Figure BDA0002559245230000084
step S24, dividing the data of each window into input data and output data as the input and output of the training model;
taking the first T data of each analysis measuring point of the standardized matrix of each window as a training model input value X, and taking the last (d + n) data of the NOx emission concentration measuring point of the standardized matrix of each window as a training model output value Y;
namely, it is
Figure BDA0002559245230000085
Wherein k is 1 to (P-1);
Figure BDA0002559245230000086
since the output is only required to take into account the output NOx emission concentration, and prediction processing is not required for data at each measurement point, the output data of the model here is only required to be the NOx emission concentration.
And S3, carrying out optimal prediction model training of the NOx emission concentration through a deep learning algorithm.
In the embodiment, a rolling LSTM neural network prediction model is constructed, each hyper-parameter of the LSTM neural network prediction model is optimized by adopting a genetic algorithm, the optimized hyper-parameter and the training data set are used for training, iterative optimization and evaluation are carried out on the LSTM neural network prediction model, and an optimal prediction model is extracted; and predicting the trained optimal prediction model by using the test data set, and evaluating the model error.
Specifically, step S3 includes the following steps S31 to S34:
s31, constructing an initial model of the LSTM neural network prediction model;
s32, optimizing partial hyper-parameters of the LSTM prediction model by using a genetic algorithm, and performing parameter combination optimization in a parameter search space by using the minimum error as a fitness function; in this embodiment, the hyper-parameters optimized by the genetic algorithm include: time window length T, training times epoch, batch size per training.
Step S33, updating the initial model by using the super-parameter optimization result, and performing model training and iterative optimization by using the training model input value X and the training model output value Y of each window of the normalized training data set; and training a Loss function Loss by using a mean square error as a model, and performing gradient calculation by using an Adam optimization algorithm.
Specifically, the model training loss function is
Figure BDA0002559245230000091
Wherein: c is the number of data of the training data set, afi PFor normalized actual value of NOx emission concentration, afi P*Is a predicted value of the normalized NOx emission concentration.
And S34, using the standardized test data set to perform rolling test on the LSTM neural network model trained in the step S33, performing inverse standardization on the output result of the model to obtain a predicted value of NOx emission concentration, comparing the predicted value and the actual value of the NOx emission concentration, evaluating the error of the model, and determining the optimal prediction model.
The formula for denormalization of the predicted value refers to formula two:
the formula II is as follows:
Figure BDA0002559245230000092
wherein
Figure BDA0002559245230000093
For the first sample data, af, of each window that is not normalizedi P*For the predicted value of the normalized NOx emission concentration, bfi P*And predicting the NOx emission concentration of each sampling point in the window.
Thereby obtaining a predicted value sequence of the final NOx emission concentration
Figure BDA0002559245230000101
And the true value series of the NOx emission concentration at each time point (sampling point) corresponding to the predicted value is:
Figure BDA0002559245230000102
in the step, when the model error is evaluated by comparing the predicted value and the actual value of the NOx emission concentration, the accuracy Q and the average absolute percentage error EMAPE are used as evaluation indexes. Wherein, the formula of the accuracy Q refers to formula three:
the formula III is as follows:
Figure BDA0002559245230000103
the average absolute percentage error EMAPE refers to the formula four:
the formula four is as follows:
Figure BDA0002559245230000104
in the above formula three and formula four, c is the number of data in the training data set, bfi PActual value of NOx emission concentration, bfi P*Is a predicted value of denormalized NOx emission concentration.
Thus, an optimal predictive model for NOx emission concentration is obtained.
In this embodiment, the optimal prediction model trained in step S2 includes a random measuring point prediction model and a fixed measuring point prediction model; the random measuring point model is trained according to historical data collected by random measuring points in each analysis measuring point, and the fixed measuring point model is trained according to historical data collected by fixed measuring points in each analysis measuring point; in practical application of industrial fields, the model can be selected automatically according to use requirements.
S4, carrying out field deployment on the optimal prediction model;
s5, collecting real-time data of each analysis measuring point on site according to sampling intervals S, and predicting future (d + n) NOx emission concentration data through the optimal prediction model; and uploading the prediction result to a desulfurization and denitrification control system, and controlling the ammonia injection amount in real time at a future time point by the desulfurization and denitrification control system according to the prediction result so as to realize intelligent quantitative ammonia injection.
The step S5 specifically includes the following steps S51 to S53:
step S51, collecting the real-time data of each analysis measuring point on site at sampling interval S to obtain a real-time data set
Figure BDA0002559245230000105
The sampling number is the time window length T; standardizing each acquired real-time data, and taking the standardized real-time data as an input value of the optimal prediction model to obtain a prediction output value Ypre of the optimal prediction model; carrying out anti-standardization processing on the prediction output value Ypre to finally obtain d + n real-time prediction sequences of the NOx emission concentration:
Figure BDA0002559245230000111
step S52, predicting result Ypre *Uploading the data to a desulfurization and denitrification control system, and removing the delay data number D of the delay time D to obtain an actual pre-sequencing sequence:
Figure BDA0002559245230000112
step S53, actual prediction sequence Y of NOx emission concentrationac *The ammonia spraying amount is controlled, and intelligent quantitative ammonia spraying is realized.
The effectiveness of the present invention can be further illustrated by the following experiments in which the parameters do not affect the general applicability of the invention.
An experiment platform: the processor is Intel i5-6500, and the memory is 32.0 GB; the system is Windows10(64 bits); the program language version is Python 3.6.5; the integrated development environment is Pycharm 2018.2.1.
The experimental contents are as follows:
12 days (T) from a certain power plant in YangzhouH) The historical data of the boiler measuring points is used as a time sequence prediction sample, 5S is selected as a data acquisition time interval S according to the output data frequency of a DCS control system, and the prediction is carried out on three minutes in the future (namely D + N is 3min, D + N is 36). The method comprises the steps of firstly carrying out windowing processing on historical data in a historical data set, and then normalizing (namely standardizing) the data by adopting window standardization, wherein the first 90% of the historical data are training data sets, and the last 10% of the historical data are testing data sets.
Constructing an LSTM prediction model, training an initial model by adopting default parameters, verifying the correctness of the network and data, then optimizing LSTM model parameter values by adopting a genetic algorithm, setting the number of individuals in a population to be 50, the iteration number to be 100, the cross probability pc to be 0.6 and the variation probability pm to be 0.01.
Setting a hyper-parameter search space: the length T of the time window is 36-180, and the step length is 12; the iteration time epoch is 1-20, and the step length is 1; the batch size of each training is 10-100, and the step size is 10.
In genetic algorithms, a near-optimal solution can be quickly found in a search space. Before iterative optimization, initializing an optimal fitness and an optimal parameter value list. After each iteration is finished, searching the current population optimal fitness function value and the optimal parameter combination, comparing the current population optimal fitness function value and the current population optimal parameter combination with the previous optimal value, if the current iteration fitness function value is more optimal, replacing the optimal fitness and optimal parameter value list, and if not, keeping the current population optimal fitness function value and optimal parameter value list unchanged; then, the population is updated through selection, crossover and mutation, and the next iteration is carried out.
And (3) respectively carrying out modeling training on the fixed measuring points and the random measuring points of each analysis measuring point in the experiment to obtain a fixed measuring point model and a random measuring point model, testing the two models by using a test data set, and carrying out model stability test by adopting data after 2 months and 6 months.
Wherein FIG. 3 is NOx emission concentration data on a test data set for two models of the present invention; due to the limited representation form of the attached drawings, the specific predicted value curve and the actual value curve cannot be clearly represented in the graph shown in the figure 3, but the fitting degree of each curve can be high, and the fact that the predicted value and the actual value of the NOx emission concentration of the two models are highly fitted is illustrated.
Table 1 is a test data set, two model NOx emission concentration test set performance comparison data after 2 months and 6 months.
Figure BDA0002559245230000121
TABLE 1 comparison of two model NOx emission concentration Performance
The data in Table 1 show that the precision Q can reach more than 99 percentMean absolute percentage error EMAPEAlso below 0.01. The intelligent ammonia injection control device provided by the invention can obtain a lower prediction error and has good performance. Mean absolute percentage error E of the two modelsMAPEAll are less than 1%, and the prediction precision is high. On the data after 2 months and 6 months, the model still has low prediction error, and the method is proved to have high stability and good adaptability.
Then, the model is deployed on the site to predict the NOx emission concentration in real time, fig. 4 is a comparison graph of the real-time predicted value and the actual value of the industrial site NOx, and also because the expression form of the attached drawing is limited, fig. 4 can not clearly show a specific predicted value curve and an actual value curve, but can show that the fitting degree of each curve is very high, which indicates that the predicted value and the actual value of the NOx emission concentration of the two models are highly fitted; and table 2 shows the comparison of the prediction performances of the two models, wherein the field real-time data 1 and the field real-time data 2 are real-time prediction conditions under stable working conditions and variable working conditions, respectively. Because the current field data acquisition has the phenomenon of packet loss and poor data quality, the prediction effect is slightly poor compared with a test data set, but the average absolute percentage error E of the NOx emission concentration isMAPEStill control under 1%, prediction accuracy is higher. The results show that the method has certain reference value and practical economic benefit.
Figure BDA0002559245230000122
TABLE 2 comparison of NOx emission concentration test set Performance for two models
In other embodiments, the best prediction model may be obtained by other deep learning algorithms, such as, but not limited to, CNN (convolutional neural network), RNN (cyclic neural network), DNN (deep neural network), and so on.
In practical application, one of the two models is selected to be deployed to field application, and the selection of the model is selected according to use requirements.
With reference to fig. 1, the present invention further provides an intelligent ammonia injection control device, which includes a NOx emission concentration prediction model generation module, a historical data storage module, and a prediction model application module;
the NOx emission concentration prediction model generation module comprises an analysis measuring point acquisition module, a historical data acquisition and preprocessing module, a prediction model training module and a model test module;
the analysis measuring point acquisition module is used for acquiring analysis measuring points required by model training, the analysis measuring points comprise fixed measuring points and random measuring points, and the fixed measuring points comprise key operation measuring points and key monitoring measuring points of the coal-fired unit;
the historical data acquisition and preprocessing module comprises a historical data acquisition submodule and a historical data preprocessing submodule;
the historical data acquisition submodule acquires a historical data set from the historical data storage module according to the analysis measuring point and the NOx emission concentration measuring point;
the historical data preprocessing submodule is used for carrying out standardization processing on each historical data in the historical data set;
the prediction model training module performs model training according to the historical data after standardization processing to obtain an optimal prediction model of the NOx emission concentration;
the model testing module tests the optimal model;
the prediction model application module comprises a real-time data acquisition and pretreatment module, a desulfurization and denitrification control system and an ammonia spraying module;
the real-time data acquisition module acquires real-time data of each analysis measuring point in an industrial field; the output end of the real-time data acquisition submodule is connected with the input end of the optimal prediction model;
the output end of the optimal prediction model is connected with an ammonia injection amount calculation module of the desulfurization and denitrification control system;
the ammonia injection amount calculation module of the desulfurization and denitrification control system is used for calculating the ammonia injection amount at a future time point; the desulfurization and denitrification control system sends an ammonia injection automatic regulation opening command to the ammonia injection module at a future time point according to the ammonia injection amount calculated by the ammonia injection amount calculation module;
the ammonia injection module performs automatic ammonia injection.
According to the intelligent ammonia injection control method and the intelligent ammonia injection control device, the NOx emission concentration in the flue gas is predicted by utilizing big data analysis and artificial intelligence means according to the load change condition in the temperature range allowed by the catalyst, the prediction result is used as the input of an external controller (a PLC), the accurate ammonia injection in the full load range is realized, and the problems of excessive ammonia injection and insufficient ammonia injection are thoroughly solved.
Aiming at the problems, the method carries out accurate prediction on the NOx emission concentration in a future period of (D + N) time according to the real-time collected NOx emission concentration, uploads the prediction result to a desulfurization and denitrification control system, and controls the ammonia injection amount of an SCR of an ammonia injection and denitrification system. The series of processes are packaged into a set of control device, the problems of delay and accuracy of boiler combustion NOx emission concentration measurement are solved, intelligent quantitative ammonia injection is realized, equipment risks are reduced, NOx outlet emission standards are met, and environmental protection is enhanced.
The above description is only one specific embodiment of the present invention, but the present invention is not limited to the above specific embodiment, and the functional effects produced by the changes made according to the technical scheme of the present invention are also regarded as the disclosure of the present invention when the technical scheme of the present invention is not beyond the scope of the technical scheme of the present method.

Claims (15)

1. An intelligent ammonia injection control method is characterized by comprising the following steps:
s1, determining analysis measuring points, wherein the analysis measuring points comprise fixed measuring points and random measuring points, and the fixed measuring points comprise key operation measuring points and key monitoring measuring points of the coal-fired unit;
s2, collecting time T according to sampling interval SHHistorical data of each analysis measuring point and NOx emission concentration measuring point in the NOx emission concentration measuring device is obtained to obtain a historical data set;
s3, carrying out optimal prediction model training of NOx emission concentration through a deep learning algorithm based on the historical data set;
s4, carrying out field deployment on the optimal prediction model;
s5, acquiring real-time data of each analysis measuring point on site by using the optimal prediction model, and predicting future NOx emission concentration through the optimal prediction model; and the desulfurization and denitrification control system controls the ammonia injection amount in real time at a future time point according to the prediction result so as to realize intelligent quantitative ammonia injection.
2. The intelligent ammonia injection control method according to claim 1,
in step S2, the historical data set is divided into a training data set and a test data set;
step S3 specifically comprises constructing a rolling LSTM neural network prediction model, optimizing each hyper-parameter of the LSTM neural network prediction model by adopting a genetic algorithm, training the LSTM neural network prediction model by using the optimized hyper-parameters and the training data set, performing iterative optimization and evaluation, and extracting an optimal prediction model; and predicting the trained optimal prediction model by using the test data set, and evaluating the model error.
3. The intelligent ammonia injection control method according to claim 2, wherein the step S2 further comprises preprocessing historical data, and the step S2 specifically comprises:
step S21, collecting time T from the data storage module according to the sampling interval SHHistorical data of each analysis measuring point and NOx emission concentration measuring point in the NOx emission concentration measuring device are obtained to obtain a historical data set
Figure FDA0002559245220000011
Wherein m represents the number of samples; k represents the measurement point dimension, k is 1 to P, wherein k is 1 to (P-1) represents the analysis measurement point, k is P represents the NOx emission concentration measurement point, bfi kHistorical data representing a measuring point k;
step S22, setting delay time D, and performing windowing processing on the collected historical data; obtaining a model parameter matrix BF of each window t:
Figure FDA0002559245220000012
wherein k is 1 to P; t is the time window length in the LSTM neural network prediction model; d represents the number of sampling points of the delay time length, D is equal to D/S, and D + n represents the number of sampling points of the prediction time length;
step S23, standardizing each data of each window to obtain a standardized parameter matrix AF of each window t:
Figure FDA0002559245220000021
wherein k is 1 to P;
step S24, dividing the data of each window into input data and output data as the input and output of the training model;
taking the front T data of each analysis measuring point in the standardized matrix of each window as a training model input value X, and taking the rear (d + n) data of the NOx emission concentration measuring point in the standardized matrix of each window as a training model output value Y;
namely, it is
Figure FDA0002559245220000022
Wherein k is 1 to (P-1);
Figure FDA0002559245220000023
4. the intelligent ammonia injection control method according to claim 3,
in step S23, the formula for normalizing each data is:
the formula I is as follows:
Figure FDA0002559245220000024
wherein k is 1 to P, q is 0 to T + d + n,
Figure FDA0002559245220000025
the first sample point data of the window t.
5. The intelligent ammonia injection control method of claim 4, wherein the time window length T > d + n.
6. The intelligent ammonia injection control method according to claim 4, wherein the step S3 specifically comprises:
s31, constructing an initial model of the LSTM neural network prediction model;
s32, optimizing partial hyper-parameters of the LSTM prediction model by using a genetic algorithm, and performing parameter combination optimization in a parameter search space by using the minimum error as a fitness function;
step S33, updating the initial model by using the super-parameter optimization result, and performing model training and iterative optimization by using the training model input value X and the training model output value Y of each window of the normalized training data set; training a Loss function Loss by using a mean square error as a model, and performing gradient calculation by using an Adam optimization algorithm;
and S34, using the standardized test data set to perform rolling test on the LSTM neural network model trained in the step S33, performing anti-standardization on the output result of the model to obtain a predicted value of NOx emission concentration, comparing the predicted value and the actual value of the NOx emission concentration, evaluating the error of the model, and determining the optimal prediction model.
7. The intelligent ammonia injection control method according to claim 6, wherein the step S33 of optimizing the hyper-parameters by using a genetic algorithm comprises: time window length T, training times epoch, batch size per training.
8. The intelligent ammonia injection control method according to claim 6,
the model training loss function of step S33 is
Figure FDA0002559245220000031
Wherein: c is the number of data in the training data set, afi PFor normalized actual value of NOx emission concentration, afi P*Is a predicted value of the normalized NOx emission concentration.
9. The intelligent ammonia injection control method according to claim 8,
the formula for denormalization of the predicted value is as follows:
the formula II is as follows:
Figure FDA0002559245220000032
i ═ t +1 to (t +1+ d); wherein, bfi P*A predicted value of NOx emission concentration is obtained;
thereby finally predicting the value
Figure FDA0002559245220000033
10. The intelligent ammonia injection control method of claim 9, wherein the accuracy Q and the average absolute percentage error E are used when comparing the predicted value and the actual value of the NOx emission concentration in step S34MAPEAs an evaluation index; wherein, the accuracy Q:
the formula III is as follows:
Figure FDA0002559245220000034
mean absolute percentage error EMAPE：
The formula four is as follows:
Figure FDA0002559245220000035
in formula three and formula four, c is the number of data in the training data set, bfi PActual value of NOx emission concentration, bfi P*For NOx emissionsAnd (5) predicting the concentration.
11. The intelligent ammonia injection control method according to claim 9, wherein the step S5 specifically includes:
step S51, collecting the real-time data of each analysis measuring point at sampling interval S to obtain a real-time data set
Figure FDA0002559245220000041
The sampling number is the time window length T; standardizing each acquired real-time data, and taking standardized real-time parameters as input values of the optimal prediction model to obtain a prediction output value Ypre of the optimal prediction model; carrying out anti-standardization processing on the prediction output value Ypre to finally obtain d + n real-time prediction sequences of the NOx emission concentration:
Figure FDA0002559245220000042
step S52, predicting result Ypre *Uploading the data to a desulfurization and denitrification control system, and removing the delay data number D of the delay time D to obtain an actual pre-sequencing sequence:
Figure FDA0002559245220000043
step S53, actual prediction sequence Y of NOx emission concentrationac *And controlling the ammonia injection amount, and controlling the ammonia injection amount in real time by the desulfurization and denitrification control system at a future time point according to a prediction result so as to realize intelligent quantitative ammonia injection.
12. The intelligent ammonia injection control method according to any one of claims 1 to 11, wherein the key operation measurement points comprise total air volume, total secondary air volume, flue gas O2 content before denitration, coal quantity and unit load; the key monitoring measuring points comprise inlet NOx concentration; and screening the random measuring points from the monitoring database through random logistic regression and a random forest algorithm.
13. The intelligent ammonia injection control method according to any one of claims 1 to 11, wherein the optimal prediction model trained in step S3 includes a fixed point model and a random point model; in step S4, the optimal prediction model deployed on site is one of the fixed measuring point model and the random measuring point model.
14. The intelligent ammonia injection control device is characterized by comprising a NOx emission concentration prediction model generation module, a historical data storage module and a prediction model application module;
the NOx emission concentration prediction model generation module comprises an analysis measuring point acquisition module, a historical data acquisition and preprocessing module, a prediction model training module and a model test module;
the analysis measuring point acquisition module is used for acquiring analysis measuring points required by model training, the analysis measuring points comprise fixed measuring points and random measuring points, and the fixed measuring points comprise key operation measuring points and key monitoring measuring points of the coal-fired unit;
the historical data acquisition and preprocessing module comprises a historical data acquisition submodule and a historical data preprocessing submodule;
the historical data acquisition submodule acquires a historical data set from the historical data storage module according to the analysis measuring point and the NOx emission concentration measuring point;
the historical data preprocessing submodule is used for carrying out standardization processing on each historical data in the historical data set;
the prediction model training module performs model training according to the historical data after standardization processing to obtain an optimal prediction model of the NOx emission concentration;
the model testing module tests the optimal prediction model;
and the prediction model application module predicts the NOx emission concentration at a future time point according to the optimal prediction model and the real-time data of the analysis measuring point.
15. The intelligent ammonia injection control device of claim 14, wherein the predictive model application module comprises a real-time data acquisition and preprocessing module, a desulfurization and denitrification control system, and an ammonia injection module;
the real-time data acquisition and preprocessing module acquires real-time data of each analysis measuring point in an industrial field and carries out standardized processing on the real-time data; the output end of the real-time data acquisition and preprocessing module is connected with the input end of the optimal prediction model;
the output end of the optimal prediction model is connected with an ammonia injection amount calculation module of the desulfurization and denitrification control system;
the ammonia injection amount calculation module of the desulfurization and denitrification control system is used for calculating the ammonia injection amount at a future time point; the desulfurization and denitrification control system sends an ammonia injection automatic regulation opening command to the ammonia injection module at a future time point according to the ammonia injection amount calculated by the ammonia injection amount calculation module;
the ammonia injection module performs automatic ammonia injection.
CN202010606351.2A 2020-06-29 2020-06-29 Intelligent ammonia injection control method and intelligent ammonia injection control device Active CN111804146B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010606351.2A CN111804146B (en) 2020-06-29 2020-06-29 Intelligent ammonia injection control method and intelligent ammonia injection control device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010606351.2A CN111804146B (en) 2020-06-29 2020-06-29 Intelligent ammonia injection control method and intelligent ammonia injection control device

Publications (2)

Publication Number Publication Date
CN111804146A true CN111804146A (en) 2020-10-23
CN111804146B CN111804146B (en) 2022-07-01

Family

ID=72855588

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010606351.2A Active CN111804146B (en) 2020-06-29 2020-06-29 Intelligent ammonia injection control method and intelligent ammonia injection control device

Country Status (1)

Country Link
CN (1) CN111804146B (en)

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112379892A (en) * 2020-10-29 2021-02-19 远光软件股份有限公司 Ammonia spraying prediction code processing method and device, storage medium and terminal equipment
CN112508234A (en) * 2020-11-19 2021-03-16 中国大唐集团科学技术研究院有限公司火力发电技术研究院 Prediction method for ammonia injection amount of SCR denitration system based on neural network
CN112699600A (en) * 2020-12-23 2021-04-23 中国大唐集团科学技术研究院有限公司火力发电技术研究院 Thermal power operating parameter and NOxAnalysis method for bias between emission concentrations
CN112862180A (en) * 2021-02-04 2021-05-28 华电国际电力股份有限公司技术服务分公司 Denitration system inlet NOx concentration prediction method
CN112933913A (en) * 2021-02-08 2021-06-11 国家能源集团国源电力有限公司 Ammonia injection control method and device and coal combustion system
CN113813770A (en) * 2021-10-26 2021-12-21 深圳开云智能有限公司 Fusion-based artificial intelligence nitrogen oxide emission prediction and ammonia injection control system
CN114191953A (en) * 2021-12-07 2022-03-18 国网河北能源技术服务有限公司 Flue gas desulfurization and denitrification control method based on convolutional neural network and XGboost
CN114413247A (en) * 2022-01-14 2022-04-29 西安热工研究院有限公司 Boiler combustion heating surface overtemperature monitoring and active inhibition system
CN114664389A (en) * 2022-03-24 2022-06-24 南方电网电力科技股份有限公司 Method and device for predicting reaction conditions for urea hydrolysis ammonia production
CN115660211A (en) * 2022-11-11 2023-01-31 天瑞集团信息科技有限公司 Control method for reducing nitrogen oxides at tail of cement kiln based on big data and Internet of things
TWI800154B (en) * 2020-12-31 2023-04-21 南韓商Ｅｍｋｏ有限公司 Method for treating flue gas of thermal power plant by using artificial intelligence and device for treating flue gas of thermal power plant by using artificial intelligence
CN116046533A (en) * 2023-01-10 2023-05-02 中国人民解放军陆军工程大学 Crack tip stress intensity factor measuring method based on DIC and stress field reconstruction
CN116832614A (en) * 2023-05-24 2023-10-03 华能国际电力股份有限公司上海石洞口第二电厂 Ammonia spraying amount control method and system for SCR denitration system

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109298697A (en) * 2018-11-13 2019-02-01 远光软件股份有限公司 Thermal power plant's various parts working state evaluation method based on DBM Dynamic Baseline Model
CN110188383A (en) * 2019-04-23 2019-08-30 华中科技大学 A kind of power station SCR denitration modeling method based on selective ensemble model
CN110689171A (en) * 2019-09-05 2020-01-14 哈尔滨工程大学 Turbine health state prediction method based on E-LSTM
KR20200014048A (en) * 2018-07-31 2020-02-10 (주)휴엔릭스 Method for treating pollutant based on A.I data analysis and apparatus for treating pollutant based on A.I data analysis
CN110975597A (en) * 2019-10-15 2020-04-10 杭州电子科技大学 Neural network hybrid optimization method for cement denitration

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20200014048A (en) * 2018-07-31 2020-02-10 (주)휴엔릭스 Method for treating pollutant based on A.I data analysis and apparatus for treating pollutant based on A.I data analysis
CN109298697A (en) * 2018-11-13 2019-02-01 远光软件股份有限公司 Thermal power plant's various parts working state evaluation method based on DBM Dynamic Baseline Model
CN110188383A (en) * 2019-04-23 2019-08-30 华中科技大学 A kind of power station SCR denitration modeling method based on selective ensemble model
CN110689171A (en) * 2019-09-05 2020-01-14 哈尔滨工程大学 Turbine health state prediction method based on E-LSTM
CN110975597A (en) * 2019-10-15 2020-04-10 杭州电子科技大学 Neural network hybrid optimization method for cement denitration

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112379892A (en) * 2020-10-29 2021-02-19 远光软件股份有限公司 Ammonia spraying prediction code processing method and device, storage medium and terminal equipment
CN112508234A (en) * 2020-11-19 2021-03-16 中国大唐集团科学技术研究院有限公司火力发电技术研究院 Prediction method for ammonia injection amount of SCR denitration system based on neural network
CN112699600A (en) * 2020-12-23 2021-04-23 中国大唐集团科学技术研究院有限公司火力发电技术研究院 Thermal power operating parameter and NOxAnalysis method for bias between emission concentrations
TWI800154B (en) * 2020-12-31 2023-04-21 南韓商Ｅｍｋｏ有限公司 Method for treating flue gas of thermal power plant by using artificial intelligence and device for treating flue gas of thermal power plant by using artificial intelligence
CN112862180A (en) * 2021-02-04 2021-05-28 华电国际电力股份有限公司技术服务分公司 Denitration system inlet NOx concentration prediction method
CN112933913A (en) * 2021-02-08 2021-06-11 国家能源集团国源电力有限公司 Ammonia injection control method and device and coal combustion system
CN112933913B (en) * 2021-02-08 2022-08-30 国家能源集团国源电力有限公司 Ammonia injection control method and device and coal combustion system
CN113813770A (en) * 2021-10-26 2021-12-21 深圳开云智能有限公司 Fusion-based artificial intelligence nitrogen oxide emission prediction and ammonia injection control system
CN114191953A (en) * 2021-12-07 2022-03-18 国网河北能源技术服务有限公司 Flue gas desulfurization and denitrification control method based on convolutional neural network and XGboost
CN114191953B (en) * 2021-12-07 2024-02-20 国网河北能源技术服务有限公司 Flue gas desulfurization and denitrification control method based on convolutional neural network and XGBoost
CN114413247A (en) * 2022-01-14 2022-04-29 西安热工研究院有限公司 Boiler combustion heating surface overtemperature monitoring and active inhibition system
CN114664389A (en) * 2022-03-24 2022-06-24 南方电网电力科技股份有限公司 Method and device for predicting reaction conditions for urea hydrolysis ammonia production
CN114664389B (en) * 2022-03-24 2023-09-22 南方电网电力科技股份有限公司 Prediction method and device for reaction conditions of urea hydrolysis ammonia production
CN115660211A (en) * 2022-11-11 2023-01-31 天瑞集团信息科技有限公司 Control method for reducing nitrogen oxides at tail of cement kiln based on big data and Internet of things
CN116046533A (en) * 2023-01-10 2023-05-02 中国人民解放军陆军工程大学 Crack tip stress intensity factor measuring method based on DIC and stress field reconstruction
CN116046533B (en) * 2023-01-10 2023-09-22 中国人民解放军陆军工程大学 Crack tip stress intensity factor measuring method based on DIC and stress field reconstruction
CN116832614A (en) * 2023-05-24 2023-10-03 华能国际电力股份有限公司上海石洞口第二电厂 Ammonia spraying amount control method and system for SCR denitration system

Also Published As

Publication number Publication date
CN111804146B (en) 2022-07-01

Similar Documents

Publication Publication Date Title
CN111804146B (en) Intelligent ammonia injection control method and intelligent ammonia injection control device
EP1864193B1 (en) Predictive emissions monitoring system and method
CN112183709B (en) Method for predicting and early warning excessive dioxin in waste incineration gas
US8768664B2 (en) Predictive emissions monitoring using a statistical hybrid model
CN110716512A (en) Environmental protection equipment performance prediction method based on coal-fired power plant operation data
CN109670625A (en) NOx emission concentration prediction method based on Unscented kalman filtering least square method supporting vector machine
CN113175678B (en) Garbage incineration monitoring method and device
CN112488145A (en) NO based on intelligent methodxOnline prediction method and system
CN113450880A (en) Desulfurization system inlet SO2Intelligent concentration prediction method
CN114225662A (en) Flue gas desulfurization and denitrification optimization control method based on hysteresis model
CN112613237A (en) CFB unit NOx emission concentration prediction method based on LSTM
CN113192569A (en) Harmful gas monitoring method based on improved particle swarm and error feedback neural network
CN115755624A (en) Coal-fired boiler multi-objective optimization method based on evolutionary algorithm
CN116484675A (en) Crack propagation life prediction method and system for ship engine blade
CN113780639B (en) Urban solid waste incineration nitrogen oxide NOx emission prediction method based on multitask learning framework
CN110852496A (en) Natural gas load prediction method based on LSTM recurrent neural network
CN115113519A (en) Coal-gas co-combustion boiler denitration system outlet NO x Concentration early warning method
CN113592163A (en) Method and equipment for predicting concentration of nitrogen oxide at inlet of SCR (Selective catalytic reduction) denitration reactor
CN113869359A (en) Modular neural network-based prediction method for nitrogen oxides in urban solid waste incineration process
CN111651938B (en) Variable coal quality unit output prediction method based on thermodynamic calculation and big data
CN117744704B (en) Flue gas pollution source acquisition monitoring system, method and readable storage medium
CN112862180A (en) Denitration system inlet NOx concentration prediction method
CN116187187A (en) Method for predicting oxygen content of flue gas in urban solid waste incineration process in real time
Tang et al. A hybrid approach for the dynamic monitoring and forecasting of NOx emissions in power plants
CN115146833A (en) Method for predicting generation concentration of boiler nitrogen oxide

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