CN115527630A - Training method and device of prediction model and related equipment - Google Patents
Training method and device of prediction model and related equipment Download PDFInfo
- Publication number
- CN115527630A CN115527630A CN202211323854.4A CN202211323854A CN115527630A CN 115527630 A CN115527630 A CN 115527630A CN 202211323854 A CN202211323854 A CN 202211323854A CN 115527630 A CN115527630 A CN 115527630A
- Authority
- CN
- China
- Prior art keywords
- training data
- data set
- training
- model
- denitration
- 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
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
- G16C20/70—Machine learning, data mining or chemometrics
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
- G16C20/10—Analysis or design of chemical reactions, syntheses or processes
Landscapes
- Chemical & Material Sciences (AREA)
- Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Crystallography & Structural Chemistry (AREA)
- Theoretical Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Databases & Information Systems (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Analytical Chemistry (AREA)
- Chemical Kinetics & Catalysis (AREA)
- Exhaust Gas Treatment By Means Of Catalyst (AREA)
Abstract
本申请提供一种预测模型的训练方法、装置及相关设备，其中，所述方法包括：获取第一训练数据集；基于扰动数据对所述第一训练数据集中的训练数据进行更新，得到第二训练数据集，其中，所述扰动数据用于表征脱硝机组在运行过程中遭遇的干扰因素；根据所述第二训练数据集对初始模型进行训练，获得目标模型，其中，所述目标模型用于预测所述脱硝机组的氨逃逸浓度和所述脱硝机组的脱硝效率中的至少一者。基于扰动数据对第一训练数据集中的训练数据进行更新，以在模型训练过程中引入对抗攻击环节，使得用于训练模型的训练数据集所模拟环境趋近于脱硝机组的实际运行环境，使训练得到的目标模型具备较强的鲁棒性。
The present application provides a prediction model training method, device and related equipment, wherein the method includes: obtaining a first training data set; updating the training data in the first training data set based on disturbance data to obtain the second A training data set, wherein the disturbance data is used to characterize the disturbance factors encountered by the denitrification unit during operation; the initial model is trained according to the second training data set to obtain a target model, wherein the target model is used for At least one of the ammonia slip concentration of the denitration unit and the denitration efficiency of the denitration unit is predicted. Based on the disturbance data, the training data in the first training data set is updated to introduce an adversarial attack link in the model training process, so that the simulated environment of the training data set used to train the model is close to the actual operating environment of the denitrification unit, so that the training The obtained target model has strong robustness.
Description
Technical Field
The application relates to the technical field of fire coal denitration, in particular to a training method and device of a prediction model and related equipment.
Background
The selective catalytic reduction method is a widely used denitration process using ammonia (NH) 3 ) The ammonia gas is used as a reducing agent, and is fully mixed with the coal-fired flue gas after entering the flue, and the nitrogen oxide in the flue gas can be reduced into N under the action of a multilayer catalyst 2 And H 2 And O. The denitration efficiency and the ammonia escape concentration are two indexes which need to be focused in the monitoring process of the denitration unit.
Because the difficulty of directly measuring the ammonia escape concentration is high, the related technology can utilize a machine learning mode to indirectly measure the ammonia escape concentration of the denitration unit, namely, the trained prediction model is used for predicting the ammonia escape concentration of the denitration unit, and various interferences can be found in the application process of the denitration unit, so that the trained prediction model can not accurately predict the actual ammonia escape concentration of the denitration unit. That is, the predictive model trained by applying the correlation technique is weak in robustness.
Disclosure of Invention
The embodiment of the application aims to provide a method, a device and related equipment for training a prediction model, which are used for solving the problem that the robustness of the prediction model obtained by the training of the related technology is weak.
In a first aspect, an embodiment of the present application provides a method for training a prediction model, including:
acquiring a first training data set, wherein training data in the first training data set are acquired according to first smoke attribute information of a denitration unit in a historical time period;
updating the training data in the first training data set based on disturbance data to obtain a second training data set, wherein the disturbance data is used for representing interference factors encountered by the denitration unit in the operation process;
training an initial model according to the second training data set to obtain a target model, wherein the target model is used for predicting at least one of the ammonia escape concentration of the denitration unit and the denitration efficiency of the denitration unit.
Optionally, the updating the training data in the first training data set based on the disturbance data to obtain a second training data set includes:
performing iteration processing on the first training data set for n times according to a first countermeasure algorithm to obtain n iterative data sets, wherein the first iterative data set is an iterative data set generated based on the first training data set and the first countermeasure algorithm, the (i + 1) th iterative data set is an iterative data set generated based on the (i) th iterative data set and the first countermeasure algorithm, n is an integer larger than 2, and i is a positive integer smaller than or equal to n-1;
obtaining n iterative loss values according to the initial model and the n iterative data sets, wherein the n iterative loss values correspond to the n iterative data sets one to one, the iterative loss values are determined according to the difference between first prediction information and actual denitration information corresponding to the first training data set, and the first prediction information is information output after the initial model processes the corresponding iterative data sets;
and determining the iteration data set corresponding to the iteration loss value with the maximum value in the n iteration loss values as the second training data set.
Optionally, the training the initial model by the second training data set to obtain the target model includes:
performing iterative training on the initial model according to the second training data set to obtain a plurality of alternative models;
calculating a loss function value of each candidate model in the plurality of candidate models, wherein the loss function value is determined according to a difference between second prediction information and actual denitration information corresponding to the first training data set, and the second prediction information is information output after the corresponding candidate model processes the second training data set;
and determining the candidate model corresponding to the loss function value with the minimum value in the plurality of loss function values as the target model.
Optionally, the first flue gas attribute information includes at least one of a flue gas amount, a total ammonia injection amount, a partitioned manual ammonia injection amount under a fault condition, a dynamic ammonia nitrogen molar ratio, a gas flow rate difference at different positions on the same cross section of the flue, a reactor inlet nitrogen oxide concentration, a reactor outlet nitrogen oxide concentration, and an oxygen concentration of the denitration unit in a historical operation period.
Optionally, the updating the training data in the first training data set based on the disturbance data to obtain a second training data set includes:
acquiring first disturbance data corresponding to target training data and a correction coefficient corresponding to the first disturbance data, wherein the target training data is any training data in the first training data set, the first disturbance data is used for representing an interference factor of the target training data, and the update coefficient is used for representing the influence degree of the interference factor on the target training data;
updating the first disturbance data according to the updating coefficient to obtain second disturbance data;
and adding the second perturbation data in the target training data to obtain perturbation training data, wherein the second training data set comprises the perturbation training data.
Optionally, the interference factor includes at least one of scene migration interference, equipment fault interference, external cause interference, and noise interference corresponding to the denitration unit.
Optionally, after the initial model is trained according to the second training data set and a target model is obtained, the method further includes:
acquiring second flue gas attribute information of the denitration unit in a target time period;
inputting the second flue gas attribute information into the target model for prediction to obtain the predicted ammonia escape concentration and the predicted denitration efficiency of the denitration unit output by the target model.
In a second aspect, an embodiment of the present application provides a training apparatus for a prediction model, including:
the denitration unit comprises a sample acquisition module, a data acquisition module and a data processing module, wherein the sample acquisition module is used for acquiring a first training data set, and training data in the first training data set are acquired according to first smoke attribute information of the denitration unit in a historical period;
the sample updating module is used for updating the training data in the first training data set based on disturbance data to obtain a second training data set, wherein the disturbance data is used for representing interference factors encountered by the denitration unit in the operation process;
and the model training module is used for training an initial model according to the second training data set to obtain a target model, wherein the target model is used for predicting at least one of the ammonia escape concentration of the denitration unit and the denitration efficiency of the denitration unit.
In a third aspect, an embodiment of the present application further provides an electronic device, which includes a processor, a memory, and a computer program stored on the memory and executable on the processor, and when the computer program is executed by the processor, the steps of the training method for the prediction model described above are implemented.
In a fourth aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps of the training method for the prediction model described above.
In the embodiment of the application, the training data in the first training data set are updated based on the disturbance data, so that an anti-attack link is introduced in the model training process, the simulated environment of the training data set for training the model approaches to the actual operation environment of the denitration unit, and the trained target model has strong robustness.
Drawings
FIG. 1 is a flowchart of a training method of a prediction model according to an embodiment of the present disclosure;
FIG. 2 is a schematic structural diagram of a training apparatus for a predictive model according to an embodiment of the present disclosure;
fig. 3 is a schematic structural diagram of an electronic device according to an embodiment of the present application
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some, but not all, embodiments of the present application. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present application without making any creative effort belong to the protection scope of the present application.
An embodiment of the present application provides a training method of a prediction model, referring to fig. 1, where fig. 1 is a schematic flow diagram of the training method of the prediction model provided in the embodiment of the present application, and as shown in fig. 1, the method includes the following steps:
And training data in the first training data set are obtained according to the first smoke attribute information of the denitration unit in the historical time period.
The denitration unit can be understood as a unit formed by related equipment applying a Selective Catalytic Reduction (SCR) process, and the denitration process can be as follows: ammonia gas is sprayed into a flue through an ammonia spraying grid, is fully mixed with flue gas, then enters an SCR (selective catalytic reduction) reactor, and nitrogen oxides in the flue gas are reduced into nitrogen (N) under the action of a multilayer catalyst 2 ) And water (H) 2 O)。
The historical time period may be understood as any time period before the current time, and the time length of the single time period may be adaptively selected according to the requirement, for example, three days, one week, half month, one month, and the like, which is not limited in the embodiment of the present application.
In a preferred embodiment, after the time length of the single time period is determined, the historical time period is preferably set to be a time period which is before the current time and has the shortest time difference with the current time, so as to ensure the timeliness of the first smoke attribute information collected in the historical time period; for example, if the time length of a single time interval is set to be one month and the current time is april, the time interval corresponding to the whole 3 months can be determined as the historical time interval within the year of the current time.
The first flue gas attribute information can be understood as flue gas attribute data acquired by a sensor arranged inside/outside the denitration unit in the operation process of the denitration unit, for example: the flue gas volume input into the denitration unit in unit time (such as 1 second), the total ammonia injection volume of the denitration unit in unit time, the concentration of nitrogen oxides at the inlet of the SCR reactor, the concentration of nitrogen oxides at the outlet of the SCR reactor, the concentration of oxygen at the outlet of the SCR reactor and the like.
It should be noted that, when training data in the first training data set is formed according to the first smoke attribute information in the historical time period, smoke attribute data that does not meet the preset condition needs to be removed to ensure the reliability of the first training data set, where the meeting of the preset condition may mean that the corresponding smoke attribute data exceeds an allowable error range, for example, a numerical range of a certain smoke attribute data under a normal condition is (0, 25), and a numerical range when the certain smoke attribute data is interfered may be changed into (0, 40), where a range indicated by (0, 40) may mean the preset condition, and when a numerical value of the collected smoke attribute data exceeds 40, the smoke attribute data may be removed.
And 102, updating the training data in the first training data set based on the disturbance data to obtain a second training data set.
The disturbance data are used for representing interference factors encountered by the denitration unit in the operation process.
And updating the training data in the first training data set by applying a mode of resisting disturbance, so that the simulated environment of the training data set for training the model approaches to the actual operating environment of the denitration unit.
The interference factors can be at least one of interference caused by operation faults of the denitration unit (such as aging/blockage of a nitrogen oxide detector), interference caused by change of unit operation environments (such as different scales of power plants for supplying power to the unit under different unit operation environments, so that the generated energy is different, and the operation of various devices included in the denitration unit is influenced), interference caused by external force (such as interference or damage caused by environmental mutation, natural disasters or external force impact on various devices of the denitration unit), interference caused by malicious attack in a networking environment (such as deviation between data of an expected input model and data of an actual input model caused by network attack), interference caused by reduction of activity of a reaction catalyst in the denitration unit, and interference caused by random noise.
It should be noted that the disturbance data may be obtained in a computer simulation calculation manner, or in an experimental manner, or in at least one manner of collecting experimental operation data of the denitration unit. This is not limited by the examples of the present application.
And 103, training the initial model according to the second training data set to obtain a target model.
Wherein the target model is used for predicting at least one of the ammonia escape concentration of the denitration unit and the denitration efficiency of the denitration unit.
In the embodiment of the application, the training data in the first training data set are updated based on the disturbance data, so that an anti-attack link is introduced in the model training process, the simulated environment of the training data set for training the model approaches to the actual operating environment of the denitration unit, the trained target model has stronger robustness, and the target model can be guaranteed to be capable of accurately predicting the ammonia escape concentration and/or the denitration efficiency under the complex working condition.
Optionally, the updating the training data in the first training data set based on the disturbance data to obtain a second training data set includes:
performing iteration processing on the first training data set for n times according to a first countermeasure algorithm to obtain n iterative data sets, wherein the first iterative data set is an iterative data set generated based on the first training data set and the first countermeasure algorithm, the (i + 1) th iterative data set is an iterative data set generated based on the (i) th iterative data set and the first countermeasure algorithm, n is an integer larger than 2, and i is a positive integer smaller than or equal to n-1;
obtaining n iterative loss values according to the initial model and the n iterative data sets, wherein the n iterative loss values correspond to the n iterative data sets one by one, the iterative loss values are determined according to the difference between first prediction information and actual denitration information corresponding to the first training data set, and the first prediction information is information output after the initial model processes the corresponding iterative data set;
and determining the iteration data set corresponding to the iteration loss value with the maximum value in the n iteration loss values as the second training data set.
As described above, the first training data set is iteratively processed by applying the first countermeasure algorithm to obtain n iterative data sets, where n is a preset number of iterations, such as 100, 200, and so on.
For example, in the i-1 st iteration, an ith iteration data set may be obtained, the ith iteration data set is input into the initial model to obtain first prediction information output by the initial model and corresponding to the i-1 st iteration, and after actual denitration information corresponding to the first training data set is known, an iteration loss value corresponding to the i-1 st iteration, that is, an iteration loss value corresponding to the ith iteration is determined according to a difference between the first prediction information corresponding to the i-1 st iteration and the actual denitration information, so that an iteration processing process is completed.
It should be noted that, in the case where the target model is used only for predicting the ammonia slip concentration of the denitration unit, the iterative loss value may be understood as a numerical difference between the model-predicted ammonia slip concentration indicated by the first prediction information and the actual ammonia slip concentration indicated by the actual denitration information.
In the case where the target model is used only for predicting the denitration efficiency of the denitration unit, the iteration loss value may be understood as a numerical difference between the model-predicted denitration efficiency indicated by the first prediction information and the denitration efficiency indicated by the actual denitration information.
In the case where the target model is used to predict the denitration efficiency and the ammonia slip concentration of the denitration unit, the iteration loss value may be understood as a sum of a numerical difference between the model-predicted ammonia slip concentration indicated by the first prediction information and the actual ammonia slip concentration indicated by the actual denitration information, and a numerical difference between the model-predicted denitration efficiency indicated by the first prediction information and the denitration efficiency indicated by the actual denitration information, in which case, if the numerical difference between the model-predicted ammonia slip concentration indicated by the first prediction information and the actual ammonia slip concentration indicated by the actual denitration information is B1, and the numerical difference between the model-predicted denitration efficiency indicated by the first prediction information and the denitration efficiency indicated by the actual denitration information is B2, the iteration loss value B = B1+ B2, and in this application, different coefficients may be set for B1 and B2, the coefficients being used to represent the degree of influence of the ammonia slip concentration/denitration efficiency on the iteration loss value B.
After n iterative loss values are obtained, the iterative loss value with the largest value is selected from the n iterative loss values, and the iterative data set corresponding to the iterative loss value is determined as a second training data set, so that the determined second training data set can fully represent the complex working conditions of the actual operating environment of the denitration unit, and the robustness of the target model trained by the second training data set is improved.
It should be noted that the iterative data set determined as the second training data set may be any iterative data set of the n iterative data sets.
For example, the first countermeasure algorithm may be a Projection Gradient (PGD) algorithm.
Optionally, training the initial model according to the second training data set to obtain the target model, including:
performing iterative training on the initial model according to the second training data set to obtain a plurality of alternative models;
calculating a loss function value of each of the multiple candidate models, wherein the loss function value is determined according to a difference between second prediction information and actual denitration information corresponding to the first training data set, and the second prediction information is information output after the corresponding candidate model processes the second training data set;
and determining the candidate model corresponding to the loss function value with the minimum value in the plurality of loss function values as the target model.
As described above, after the second training data set is determined, the initial model is iteratively trained by applying a machine learning algorithm (e.g., a deep neural network) and the second training data set to obtain a plurality of candidate models, where the first candidate model is an candidate model obtained after the initial model is trained by applying the second training data set, the t +1 th candidate model is an candidate model obtained after the t-th candidate model is trained by applying the second training data set, and t is a positive integer.
In the case where the target model is used only for predicting the ammonia escape concentration of the denitration unit, the loss function value may be understood as: and constructing a loss function according to the difference between the ammonia escape concentration predicted by the model indicated by the second prediction information and the actual ammonia escape concentration indicated by the actual denitration information, and calculating to obtain a loss function value.
In the case where the target model is used only for predicting the denitration efficiency of the denitration unit, the loss function value may be understood as: and constructing a loss function according to the difference between the denitration efficiency predicted by the model indicated by the second prediction information and the denitration efficiency indicated by the actual denitration information, and calculating to obtain a loss function value.
In the case where the target model is used to predict the denitration efficiency and the ammonia slip concentration of the denitration unit, the loss function value may be understood as a sum of a numerical difference between the model-predicted ammonia slip concentration indicated by the second prediction information and the actual ammonia slip concentration indicated by the actual denitration information, and a numerical difference between the model-predicted denitration efficiency indicated by the second prediction information and the denitration efficiency indicated by the actual denitration information.
Optionally, the first flue gas attribute information includes at least one of a flue gas amount, a total ammonia injection amount, a partition manual ammonia injection amount under a fault condition, a dynamic ammonia nitrogen molar ratio, a gas flow rate difference at different positions on the same section of the flue, a reactor inlet nitrogen oxide concentration, a reactor outlet nitrogen oxide concentration, and an oxygen concentration of the denitration unit during a historical operation period.
In some embodiments, the target model and the initial model are both multi-input and multi-output prediction models, and specifically, in the model training stage, the input data α of the model i May include flue gas flow rate alpha 1 Total ammonia injection amount alpha 2 The amount of the injected ammonia in each zone is alpha 2j (j =1, \8230;, n 1), the manual ammonia spraying amount α of each partition in case of failure 3j (j =1, \8230;, n 1), NO at the reactor inlet x Concentration alpha 4 NO at the outlet of the reactor x Concentration alpha 5 Dynamic ammonia nitrogen molar ratio alpha 6 Gas flow velocity difference alpha of different positions on the same section of the flue 7 Composition, n1 is a positive integer greater than 1.
Wherein the flue gas flow rate alpha 1 Can be understood as the flow of smoke in unit time and the total ammonia injection quantity alpha 2 Can be understood as the total amount of ammonia gas passing through the smoke-ejecting grille in unit time, and the zoned ammonia ejecting amount alpha 2j Can be understood as the ammonia amount passing through different subareas of the denitration unit in unit time and the manual ammonia spraying amount alpha of each subarea under the fault condition 3j It can be understood that the amount of ammonia gas passing through different partitions of the denitration unit in unit time under the condition of fault, alpha 4 And alpha 5 The concentration of nitrogen oxide and the molar ratio alpha of dynamic ammonia nitrogen at the inlet and the outlet of a reactor included in a denitration unit respectively 6 Which is understood to be the ratio of the amount of substance injected with ammonia per unit time to the amount of nitrogen oxide substance at the inlet of the reactor.
In application, as permitted by gas flow velocity distribution in SCR reactionThe maximum deviation is 15 percent, therefore, a flow field adjusting device is arranged between the ammonia spraying equipment and the reactor of the denitration unit to adjust the flow of different areas in the flue gas treatment flue, further reduce the gas flow speed difference on the inlet cross section at the top of the reactor and achieve the purpose of improving the reaction effect, therefore, the gas flow speed difference alpha at different positions on the same cross section of the flue 7 At least the difference in gas flow rates over the inlet cross section at the top of the reactor.
In the case where the partition device does not have a failure, α of the corresponding partition is set to 4 Can ignore, or will correspond to, a of the partition 4 Set to a default value.
Seven indices (α) mentioned above 1 ，α 2 ，α 2j ，α 3j ，α 4 ，α 5 ，α 6 ，α 7 ) The method has a correlation relation with the ammonia escape concentration and the denitration efficiency, and therefore, the method can be used as input data of a model; in practical application, under the condition that a correlation exists between a new index and the ammonia escape concentration and the denitration efficiency, the new index can be added into the input data of the model, so that the training effect and the prediction accuracy of the model are improved.
Before adding a new index to the input data of the model, it is necessary to analyze the correlation between the new index and the prediction target (ammonia slip concentration and denitration efficiency), and when the correlation between the new index and the prediction target is greater than a correlation threshold, the new index is added to the input data of the model. By setting a correlation threshold, partial indexes with low correlation are prevented from being introduced into a model training process and a model prediction process, so that the model can obtain a better training effect and a better prediction effect with less system overhead in a training stage and a prediction stage, and the data processing efficiency and the data processing effect of the model are balanced.
Optionally, the updating the training data in the first training data set based on the disturbance data to obtain a second training data set includes:
acquiring first disturbance data corresponding to target training data and a correction coefficient corresponding to the first disturbance data, wherein the target training data is any training data in the first training data set, the first disturbance data is used for representing an interference factor of the target training data, and the update coefficient is used for representing the influence degree of the interference factor on the target training data;
updating the first disturbance data according to the updating coefficient to obtain second disturbance data;
and adding the second perturbation data in the target training data to obtain perturbation training data, wherein the second training data set comprises the perturbation training data.
Illustratively, the training data in the first training data set is α i Updating the training data in the first training data set based on the disturbance data, wherein the updated training data is alpha i ‘＝α i + k μ, where k is understood as the update coefficient, μ is understood as the first perturbation data, and k μ is understood as the second perturbation data.
Through the setting, the first disturbance data are prevented from being directly accumulated into the training data through the adjustment of the updating coefficient, so that the obtained disturbance training data can accurately reflect the data deviation under the complex working condition, and the target model obtained through training is guaranteed to have higher robustness.
Optionally, the interference factor includes at least one of scene migration interference, equipment fault interference, external cause interference, and noise interference corresponding to the denitration unit.
Wherein, the scene migration interference can be understood as the interference caused by the change of the working scene of the denitration unit, for example: interference caused by power supply source variation.
The disturbance caused by equipment failure can be understood as disturbance caused by aging and failure of relevant equipment of the denitration unit, such as: interference caused by aging of the flue gas sensor, interference caused by blockage of a flue, and interference caused by reduction of activity of a catalyst.
Exogenous disturbances are understood to be disturbances caused by environmental mutations, natural disasters and network attacks, for example: interference caused by sudden changes of the ambient temperature in extreme weather conditions, and interference caused by malicious network attacks.
Noise interference is understood to be interference due to random noise.
As above, the actual operation condition of the denitration device is fully simulated by synthesizing the multiple interference conditions, so that the target model obtained by training has higher robustness.
Optionally, after the initial model is trained according to the second training data set and a target model is obtained, the method further includes:
acquiring second flue gas attribute information of the denitration unit in a target time period;
inputting the second flue gas attribute information into the target model for prediction to obtain the predicted ammonia escape concentration and the predicted denitration efficiency of the denitration unit output by the target model.
The target time interval is understood to be a time interval including the current time, and the current time is any time in the target time interval.
The current ammonia escape concentration and denitration efficiency of the denitration unit are predicted by acquiring the current second flue gas attribute information of the denitration unit and applying a target model to process the second flue gas attribute information.
For ease of understanding, examples are illustrated below:
in the model training phase: obtaining a plurality of data of the unit in operation as model input alpha i And taking the corresponding data of the ammonia escape and denitration efficiency of the unit as target real data beta 1 ，β 2 (ii) a Wherein the input data alpha of the model i May include flue gas flow rate alpha 1 Total ammonia injection amount alpha 2 The amount of partitioned sprayed ammonia alpha 2j (j =1, \ 8230;, n 1), the amount of ammonia sprayed manually by each partition in the event of a fault, α 3j (j =1, \ 8230;, n 1), NO at the reactor inlet x Concentration alpha 4 NO at the outlet of the reactor x Concentration alpha 5 Dynamic ammonia nitrogen molar ratio alpha 6 Gas flow velocity difference alpha of different positions on the same section of the flue 7 Composition, n1 is a positive integer greater than 1.
The mathematical model for predicting ammonia slip can be expressed as:
R 1 ，R 2 ＝G(α i )
wherein G can be understood as a machine learning mapping function (e.g., a nonlinear deep convolutional neural network), R 1 Ammonia slip concentration, R, as model output 2 Denitration efficiency, beta, for model output 1 For true ammonia slip concentration, beta 2 The denitration efficiency is real.
The loss function is set as follows:
l(R，β)
the mathematical expression for the training target may be:
argmin∑l(R，β)＝argmin∑l[G(α 1 ，α 2 ，α 2j ，α 3j ，α 4 ，α 5 ，α 6 ，α 7 )，β]
in the model application phase: importing the trained robust model G into a machine learning network, and obtaining real-time input data of the unit
Wherein,
comprises the following steps: real-time flue gas flow
Real-time total ammonia injection
Real-time zoned ammonia injection
Manual ammonia injection amount of each subarea under fault condition
Real-time concentration of NOx at reactor inlet
Real time concentration of NOx at reactor outlet
Real-time dynamic ammonia nitrogen molar ratio
Real-time gas flow velocity difference of different positions on same section of flue
And (4) forming.
Using the model G to input data in real time
Mapping is performed
The result is real-time prediction output data: concentration of ammonia slip
And denitration efficiency
Wherein,
referring to fig. 2, fig. 2 is a block diagram of a training apparatus 200 for a prediction model according to an embodiment of the present disclosure. As shown in fig. 2, the training apparatus 200 for the prediction model includes:
the denitration unit comprises a sample acquisition module 201, a denitration unit and a denitration unit, wherein the sample acquisition module is used for acquiring a first training data set, and training data in the first training data set are acquired according to first flue gas attribute information of the denitration unit in a historical period;
the sample updating module 202 is configured to update the training data in the first training data set based on disturbance data to obtain a second training data set, where the disturbance data is used to represent interference factors encountered by the denitration unit in an operation process;
and the model training module 203 is configured to train an initial model according to the second training data set to obtain a target model, where the target model is used to predict at least one of the ammonia escape concentration of the denitration unit and the denitration efficiency of the denitration unit.
Optionally, the sample updating module 202 includes:
the iterative countermeasure submodule is used for carrying out iterative processing on the first training data set for n times according to a first countermeasure algorithm to obtain n iterative data sets, wherein the first iterative data set is an iterative data set generated based on the first training data set and the first countermeasure algorithm, the (i + 1) th iterative data set is an iterative data set generated based on the (i) th iterative data set and the first countermeasure algorithm, n is an integer larger than 2, and i is a positive integer smaller than or equal to n-1;
an iteration loss calculation submodule, configured to obtain n iteration loss values according to the initial model and the n iteration data sets, where the n iteration loss values correspond to the n iteration data sets one to one, the iteration loss value is determined according to a difference between first prediction information and actual denitration information corresponding to the first training data set, and the first prediction information is information output after the initial model processes the corresponding iteration data set;
and the target determining submodule is used for determining the iteration data set corresponding to the iteration loss value with the maximum value in the n iteration loss values as the second training data set.
Optionally, the model training module 203 includes:
the iterative training submodule is used for performing iterative training on the initial model according to the second training data set to obtain a plurality of alternative models;
a loss calculation sub-module, configured to calculate a loss function value of each candidate model in the multiple candidate models, where the loss function value is determined according to a difference between second prediction information and actual denitration information corresponding to the first training data set, and the second prediction information is information that is output after the corresponding candidate model processes the second training data set;
and the model determining submodule is used for determining the candidate model corresponding to the loss function value with the minimum value in the plurality of loss function values as the target model.
Optionally, the first flue gas attribute information includes at least one of a flue gas amount, a total ammonia injection amount, a partitioned manual ammonia injection amount under a fault condition, a dynamic ammonia nitrogen molar ratio, a gas flow rate difference at different positions on the same cross section of the flue, a reactor inlet nitrogen oxide concentration, a reactor outlet nitrogen oxide concentration, and an oxygen concentration of the denitration unit in a historical operation period.
Optionally, the sample updating module 202 includes:
a disturbance obtaining unit, configured to obtain first disturbance data corresponding to target training data and a correction coefficient corresponding to the first disturbance data, where the target training data is any training data in the first training data set, the first disturbance data is used to represent an interference factor of the target training data, and the update coefficient is used to represent an influence degree of the interference factor on the target training data;
the disturbance updating unit is used for updating the first disturbance data according to the updating coefficient to obtain second disturbance data;
and the data updating unit is used for adding the second perturbation data in the target training data to obtain perturbation training data, and the second training data set comprises the perturbation training data.
Optionally, the interference factor includes at least one of scene migration interference, equipment fault interference, external cause interference, and noise interference corresponding to the denitration unit.
The apparatus 200 further comprises:
the attribute acquisition module is used for acquiring second flue gas attribute information of the denitration unit in a target time period;
and the prediction module is used for inputting the second flue gas attribute information into the target model for prediction to obtain the predicted ammonia escape concentration and the predicted denitration efficiency of the denitration unit output by the target model.
The training device 200 for the prediction model provided in the embodiment of the present application can implement each process in the above method embodiments, and is not described here again to avoid repetition.
Referring to fig. 3, fig. 3 is a schematic structural diagram of an electronic device according to an embodiment of the present application, and as shown in fig. 3, the electronic device includes: may include a processor 301, a memory 302, and a program 3021 stored on the memory 302 and executable on the processor 301.
When executed by the processor 301, the program 3021 may implement any of the steps of the method embodiment shown in fig. 1 and achieve the same advantages, and thus, the description thereof is omitted here.
Those skilled in the art will appreciate that all or part of the steps of the method according to the above embodiments may be implemented by hardware related to program instructions, and the program may be stored in a readable medium.
An embodiment of the present application further provides a readable storage medium, where a computer program is stored, and when the computer program is executed by a processor, the computer program may implement any step in the method embodiment corresponding to fig. 1, and may achieve the same technical effect, and in order to avoid repetition, details are not repeated here.
The computer-readable storage media of the embodiments of the present application may take any combination of one or more computer-readable media. The computer readable medium may be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium may be, for example, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
A computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.
Program code embodied on a storage medium may be transmitted over any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
Computer program code for carrying out operations for aspects of the present application may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, smalltalk, C + +, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or terminal. In the latter scenario, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider).
While the foregoing is directed to the preferred embodiment of the present application, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the principles of the disclosure, and it is intended that such changes and modifications be considered as within the scope of the disclosure.
Claims (10)
1. A method of training a predictive model, the method comprising:
acquiring a first training data set, wherein training data in the first training data set are acquired according to first smoke attribute information of a denitration unit in a historical time period;
updating the training data in the first training data set based on disturbance data to obtain a second training data set, wherein the disturbance data is used for representing interference factors encountered by the denitration unit in the operation process;
training an initial model according to the second training data set to obtain a target model, wherein the target model is used for predicting at least one of the ammonia escape concentration of the denitration unit and the denitration efficiency of the denitration unit.
2. The method of claim 1, wherein the updating the training data in the first training data set based on the perturbation data to obtain a second training data set comprises:
performing iteration processing on the first training data set for n times according to a first countermeasure algorithm to obtain n iterative data sets, wherein the first iterative data set is an iterative data set generated based on the first training data set and the first countermeasure algorithm, the (i + 1) th iterative data set is an iterative data set generated based on the (i) th iterative data set and the first countermeasure algorithm, n is an integer larger than 2, and i is a positive integer smaller than or equal to n-1;
obtaining n iterative loss values according to the initial model and the n iterative data sets, wherein the n iterative loss values correspond to the n iterative data sets one by one, the iterative loss values are determined according to the difference between first prediction information and actual denitration information corresponding to the first training data set, and the first prediction information is information output after the initial model processes the corresponding iterative data set;
and determining the iteration data set corresponding to the iteration loss value with the maximum value in the n iteration loss values as the second training data set.
3. The method of claim 1, wherein training an initial model from the second training data set to obtain a target model comprises:
performing iterative training on the initial model according to the second training data set to obtain a plurality of alternative models;
calculating a loss function value of each candidate model in the plurality of candidate models, wherein the loss function value is determined according to a difference between second prediction information and actual denitration information corresponding to the first training data set, and the second prediction information is information output after the corresponding candidate model processes the second training data set;
and determining the candidate model corresponding to the loss function value with the minimum value in the plurality of loss function values as the target model.
4. The method according to claim 1, wherein the first flue gas attribute information includes at least one of flue gas volume, total ammonia injection volume, partitioned manual ammonia injection volume under fault condition, dynamic ammonia nitrogen molar ratio, gas flow velocity difference at different positions on the same section of a flue, reactor inlet nitrogen oxide concentration, reactor outlet nitrogen oxide concentration and oxygen concentration of the denitration unit in a historical operation period.
5. The method of claim 1, wherein the updating the training data in the first training data set based on the perturbation data to obtain a second training data set comprises:
acquiring first disturbance data corresponding to target training data and a correction coefficient corresponding to the first disturbance data, wherein the target training data is any training data in the first training data set, the first disturbance data is used for representing interference factors of the target training data, and the update coefficient is used for representing the influence degree of the interference factors on the target training data;
updating the first disturbance data according to the updating coefficient to obtain second disturbance data;
and adding the second perturbation data in the target training data to obtain perturbation training data, wherein the second training data set comprises the perturbation training data.
6. The method according to claim 1, wherein the interference factors include at least one of scene migration interference, equipment failure interference, exogenous interference and noise interference corresponding to the denitration unit.
7. The method of claim 1, wherein after training an initial model according to the second training data set and obtaining a target model, the method further comprises:
acquiring second flue gas attribute information of the denitration unit in a target time period;
inputting the second flue gas attribute information into the target model for prediction to obtain the predicted ammonia escape concentration and the predicted denitration efficiency of the denitration unit output by the target model.
8. An apparatus for training a predictive model, comprising:
the denitration unit comprises a sample acquisition module, a data acquisition module and a data processing module, wherein the sample acquisition module is used for acquiring a first training data set, and training data in the first training data set are acquired according to first smoke attribute information of the denitration unit in a historical period;
the sample updating module is used for updating the training data in the first training data set based on disturbance data to obtain a second training data set, wherein the disturbance data is used for representing interference factors encountered by the denitration unit in the operation process;
and the model training module is used for training an initial model according to the second training data set to obtain a target model, wherein the target model is used for predicting at least one of the ammonia escape concentration of the denitration unit and the denitration efficiency of the denitration unit.
9. An electronic device comprising a processor, a memory and a computer program stored on the memory and executable on the processor, the computer program, when executed by the processor, implementing the steps of the training method of a predictive model according to any one of claims 1 to 7.
10. A readable storage medium, characterized in that the readable storage medium has stored thereon a computer program which, when being executed by a processor, carries out the steps of the method of training a prediction model according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211323854.4A CN115527630A (en) | 2022-10-27 | 2022-10-27 | Training method and device of prediction model and related equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211323854.4A CN115527630A (en) | 2022-10-27 | 2022-10-27 | Training method and device of prediction model and related equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115527630A true CN115527630A (en) | 2022-12-27 |
Family
ID=84704571
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211323854.4A Pending CN115527630A (en) | 2022-10-27 | 2022-10-27 | Training method and device of prediction model and related equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115527630A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117382967A (en) * | 2023-10-19 | 2024-01-12 | 河北白沙烟草有限责任公司 | Methods, devices and equipment for determining the operating status of cigarette packaging machines |
| CN118173206A (en) * | 2024-03-22 | 2024-06-11 | 武汉理工大学 | Polyurea composite structure antiknock performance prediction model training method and prediction method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100281855A1 (en) * | 2009-05-06 | 2010-11-11 | Min Sun | Model based method for selective catalyst reducer urea dosing strategy |
| CN113488111A (en) * | 2021-07-05 | 2021-10-08 | 东南大学 | Ammonia injection amount optimization modeling method for SCR denitration system |
| CN114491448A (en) * | 2022-01-24 | 2022-05-13 | 支付宝(杭州)信息技术有限公司 | Automatic confrontation training method and device |
-
2022
- 2022-10-27 CN CN202211323854.4A patent/CN115527630A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100281855A1 (en) * | 2009-05-06 | 2010-11-11 | Min Sun | Model based method for selective catalyst reducer urea dosing strategy |
| CN113488111A (en) * | 2021-07-05 | 2021-10-08 | 东南大学 | Ammonia injection amount optimization modeling method for SCR denitration system |
| CN114491448A (en) * | 2022-01-24 | 2022-05-13 | 支付宝(杭州)信息技术有限公司 | Automatic confrontation training method and device |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117382967A (en) * | 2023-10-19 | 2024-01-12 | 河北白沙烟草有限责任公司 | Methods, devices and equipment for determining the operating status of cigarette packaging machines |
| CN118173206A (en) * | 2024-03-22 | 2024-06-11 | 武汉理工大学 | Polyurea composite structure antiknock performance prediction model training method and prediction method |
| CN118173206B (en) * | 2024-03-22 | 2024-10-22 | 武汉理工大学 | Polyurea composite structure antiknock performance prediction model training method and prediction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Young et al. | Simplicity out of complexity in environmental modelling: Occam's razor revisited | |
| KR102024953B1 (en) | Dynamic outlier bias reduction system and method | |
| KR102052217B1 (en) | Dynamic outlier bias reduction system and method | |
| CN115527630A (en) | Training method and device of prediction model and related equipment | |
| CN112364562B (en) | Flue gas environment-friendly island cooperative control method and system | |
| CN114742260A (en) | Method, system and medium for predicting large-volume concrete temperature monitoring data | |
| CN107786369A (en) | Based on the perception of IRT step analyses and LSTM powerline network security postures and Forecasting Methodology | |
| CN112016248B (en) | Elman neural network-based bad data identification method for SCR denitration system | |
| CN110716512A (en) | Environmental protection equipment performance prediction method based on coal-fired power plant operation data | |
| Anders et al. | Cost related reliability measures for power system equipment | |
| CN112128950B (en) | Machine room temperature and humidity prediction method and system based on multiple model comparisons | |
| CN118673317B (en) | Instrument equipment fault diagnosis method and device, electronic equipment and readable storage medium | |
| CN111489605B (en) | Ammonia spraying optimization control simulation system based on Simulink and WinCC | |
| Song et al. | Data-driven differentiable model for dynamic prediction and control in wastewater treatment | |
| CN116189791A (en) | An optimization method, device and storage medium for ammonia-nitrogen dual-control ammonia injection in an SCR system | |
| Radu et al. | Mixed finite element discretization and Newton iteration for a reactive contaminant transport model with nonequilibrium sorption: convergence analysis and error estimates | |
| Liu et al. | A multi‐perspective framework for seismic retrofit optimization of urban infrastructure systems | |
| CN115759373A (en) | Gas daily load prediction method, device and equipment | |
| CN119327271B (en) | Dynamic replacement method of catalyst in denitration device | |
| Cruz et al. | Machine learning-based indoor air quality baseline study of the offices and laboratories of the northwest and southwest building of Mapúa University–Manila | |
| CN120197525A (en) | A method and system for predicting power grid frequency based on time series prediction algorithm | |
| CN117252300A (en) | A coal consumption prediction device and method based on machine learning | |
| CN111400927A (en) | Method and device for predicting corrosion growth in pipeline based on generalized additive model | |
| CN114444295A (en) | NOx exhaust gas concentration prediction method based on combined time sequence prediction | |
| JP7561529B2 (en) | Prediction device, plant, prediction method, program, and configuration program |
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 | ||
| CB02 | Change of applicant information |
Country or region after: China Address after: 250100 floor 36, building 4, Lushang Guoao City, No. 9777, Jingshi Road, Lixia District, Jinan City, Shandong Province Applicant after: Shandong Chuangyu Energy Technology Co.,Ltd. Address before: 250100 floor 36, building 4, Lushang Guoao City, No. 9777, Jingshi Road, Lixia District, Jinan City, Shandong Province Applicant before: Shandong Chuangyu Environmental Protection Technology Co.,Ltd. Country or region before: China |
|
| CB02 | Change of applicant information |