Method and device for predicting flue gas parameters of power plant

Abstract

The invention provides a power plant flue gas parameter prediction method and device, and belongs to the technical field of power plant prediction.The method comprises the steps that historical power plant operation parameters corresponding to multiple moments are imported, and an original data set is constructed through the historical power plant operation parameters; preprocessing the original data set to obtain a preprocessed data set; performing delay analysis on the preprocessed data set to obtain a data set after delay processing; and carrying out feature analysis and prediction modeling on the data set after delay processing to obtain a prediction value of the target. According to the method, accurate and rapid dynamic prediction of the flue gas parameters of the power plant is achieved, the phenomena of large inertia, large delay and large lag of the air pollutant parameters of the power plant are effectively solved, optimization guidance suggestions are provided for air pollutant control operation personnel of the power plant, meanwhile, the effects of pre-control and lag elimination are achieved, and the control efficiency of the power plant is improved. And the method is of great significance to air pollutant optimization control, energy conservation and emission reduction of a power plant.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of power plant prediction, and in particular to a method and device for predicting flue gas parameters of a power plant. Background Art

[0002] At present, my country's energy structure is dominated by coal, which determines that coal-fired power generation is still the main power generation method. In the process of coal combustion, atmospheric pollutants such as NOx and SO2 will be produced. Reasonable control of NOx and SO2 emissions has always been an important task facing coal-fired power plants. SCR flue gas denitrification technology and wet absorption tower flue gas desulfurization technology are the mainstream technologies for controlling NOx and SO2 emissions in power plants at this stage. Among them, SCR flue gas denitrification technology needs to determine the amount of reducing substance NH3 required according to the content of NOx in the flue gas, and use the reduction reaction of NH3 and NOx to reduce NOx to H2O and N2, thereby removing NOx from the flue gas. Similarly, wet absorption tower flue gas desulfurization technology needs to determine the required slurry absorbent content according to the content of SO2 in the flue gas, and use the neutralization and oxidation reaction of slurry absorbent and SO2 to convert SO2 into sulfate, thereby removing SO2 from the flue gas. Therefore, accurate measurement of the content of NOx and SO2 in flue gas is crucial to the control of atmospheric pollutant emissions in power plants.

[0003] At present, the flue gas parameters of power plants such as NOx, SO2, flue gas volume, etc. are monitored in real time through the continuous online flue gas monitoring system (CEMS) and related thermal control instruments. However, due to the complex operation of power plant boilers, strong load fluctuations, long flue gas movement process, and the physical consumption time required for the CEMS system itself, the monitoring of flue gas pollutants in large power plants shows large inertia, large delay, and large lag, making it difficult to achieve true "real-time" monitoring, which leads to a general control lag in pollutant emissions in power plants. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method and device for predicting flue gas parameters of a power plant in view of the deficiencies in the prior art.

[0005] The technical solution of the present invention to solve the above technical problem is as follows: A method for predicting flue gas parameters of a power plant comprises the following steps:

[0006] Import the historical power plant operation parameters corresponding to multiple time periods, and construct multiple original data sets through the historical power plant operation parameters corresponding to multiple time periods;

[0007] Preprocessing the multiple original data sets respectively to obtain corresponding multiple preprocessed data sets;

[0008] Performing time-delay analysis on the plurality of pre-processed data sets respectively to obtain a plurality of corresponding time-delay processed data sets;

[0009] Performing feature analysis on the plurality of delayed processed data sets respectively to obtain a plurality of corresponding feature data sets;

[0010] The plurality of feature data sets are trained respectively by using an LSTM long short-term memory network model to obtain corresponding plurality of prediction models;

[0011] The dynamically updated real-time power plant operating parameters are respectively input into the plurality of prediction models, thereby obtaining real-time prediction results of the flue gas parameters of the plurality of power plants.

[0012] Another technical solution of the present invention to solve the above technical problem is as follows: A device for predicting flue gas parameters of a power plant, comprising:

[0013] Import module, used to import historical power plant operating parameters corresponding to multiple moments;

[0014] A data set construction module is used to construct multiple original data sets through historical power plant operation parameters corresponding to multiple moments;

[0015] A preprocessing module, used to preprocess the multiple original data sets to obtain corresponding multiple preprocessed data sets;

[0016] A delay processing module, used for performing delay analysis on the plurality of pre-processed data sets to obtain a plurality of corresponding delay processed data sets;

[0017] A feature analysis module, used to perform feature analysis on the plurality of delayed processed data sets to obtain a plurality of corresponding feature data sets;

[0018] A model analysis module is used to train the plurality of feature data sets through an LSTM long short-term memory network model to obtain a plurality of corresponding prediction models;

[0019] The prediction result acquisition module is used to input the dynamically updated real-time power plant operation parameters into the multiple prediction models respectively, so as to obtain the real-time prediction results of the flue gas parameters of the multiple power plants.

[0020] The beneficial effects of the present invention are as follows: an original data set is constructed through historical power plant operating parameters, the original data set is preprocessed to obtain a preprocessed data set, a delay analysis is performed on the preprocessed data set to obtain a delayed processed data set, a feature analysis is performed on the delayed processed data set to obtain a feature data set, a prediction model is obtained by training the feature data set through an LSTM long short-term memory network model, and dynamically updated real-time power plant operating parameters are input into the prediction model to obtain real-time prediction results of power plant flue gas parameters, thereby achieving accurate and rapid dynamic prediction of power plant flue gas parameters, effectively solving the large inertia, large delay and large lag phenomena of power plant atmospheric pollutant parameters, providing optimization guidance and suggestions for power plant atmospheric pollutant control operators, and at the same time, achieving the effects of pre-control and lag elimination, which is of great significance to the optimization control of power plant atmospheric pollutants and energy conservation and emission reduction. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A schematic diagram of a flow chart of a method for predicting flue gas parameters of a power plant provided by an embodiment of the present invention;

[0022] Figure 2 A module block diagram of a power plant flue gas parameter prediction device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0024] Figure 1 A schematic flow chart of a method for predicting flue gas parameters in a power plant provided in an embodiment of the present invention.

[0025] like Figure 1 As shown, a method for predicting flue gas parameters of a power plant includes the following steps:

[0026] Import the historical power plant operation parameters corresponding to multiple time periods, and construct multiple original data sets through the historical power plant operation parameters corresponding to multiple time periods;

[0027] Preprocessing the multiple original data sets respectively to obtain corresponding multiple preprocessed data sets;

[0028] Performing time-delay analysis on the plurality of pre-processed data sets respectively to obtain a plurality of corresponding time-delay processed data sets;

[0029] Performing feature analysis on the plurality of delayed processed data sets respectively to obtain a plurality of corresponding feature data sets;

[0030] The plurality of feature data sets are trained respectively by using an LSTM long short-term memory network model to obtain corresponding plurality of prediction models;

[0031] The dynamically updated real-time power plant operating parameters are respectively input into the plurality of prediction models, thereby obtaining real-time prediction results of the flue gas parameters of the plurality of power plants.

[0032] It should be understood that according to the NOx and SO2 generation and desulfurization and denitrification reaction mechanism, a total of 154 relevant auxiliary variables (i.e. historical power plant operating parameters) on the boiler side, denitrification side and desulfurization side are selected, among which:

[0033] (1) Boiler side variables include:

[0034] Host load, total air volume of boiler, coal feed of each coal mill, main steam parameters, secondary air door opening of each boiler, primary air parameters of boiler, secondary air parameters of boiler, etc., a total of 105 variables

[0035] (2) Denitrification side variables include:

[0036] A total of 18 variables including NO2 and O2 concentrations at the inlet and outlet of the denitrification reactor, flue gas temperature and pressure, and flue gas flow at the inlet of the denitrification reactor

[0037] (3) Desulfurization side variables include:

[0038] There are a total of 31 variables, including the inlet and outlet flue gas flow rate of the absorption tower, the inlet and outlet flue gas temperature and pressure of the absorption tower, the SO2 and O2 concentration of the inlet and outlet flue gas of the absorption tower, the absorption tower liquid level, the pH value of the absorption tower slurry, the gypsum slurry density, the current of each slurry circulation pump, and the outlet pressure of each slurry circulation pump.

[0039] It should be understood that more than 100,000 sets of historical data of auxiliary variables (ie, historical power plant operating parameters) are extracted from power plant DCS, PLC, CEMS and other systems at intervals of 10 seconds to construct the original data set.

[0040] Specifically, the auxiliary variables obtained in real time are used as test set data (i.e., real-time power plant operating parameters), which are dynamically updated and input into the optimization prediction model (i.e., prediction model) to quickly obtain dynamic prediction values ​​of target parameters such as flue gas NOX, SO2, and flue gas volume, providing a basis for subsequent pollutant control.

[0041] In the above embodiment, an original data set is constructed by historical power plant operating parameters, the original data set is preprocessed to obtain a preprocessed data set, a delay analysis is performed on the preprocessed data set to obtain a delayed processed data set, a feature analysis is performed on the delayed processed data set to obtain a feature data set, and the feature data set is trained by an LSTM long short-term memory network model to obtain a prediction model, and the dynamically updated real-time power plant operating parameters are input into the prediction model to obtain real-time prediction results of the power plant flue gas parameters, thereby achieving accurate and rapid dynamic prediction of the power plant flue gas parameters, effectively solving the large inertia, large delay and large lag phenomena of the power plant's atmospheric pollutant parameters, and providing optimization guidance and suggestions for the power plant's atmospheric pollutant control operators. At the same time, the effects of pre-control and elimination of lag are achieved, which is of great significance to the optimization control of atmospheric pollutants in power plants and energy conservation and emission reduction.

[0042] Optionally, as an embodiment of the present invention, the historical power plant operating parameters and the real-time power plant operating parameters include power plant boiler operating parameters, denitrification reactor operating parameters and desulfurization tower operating parameters.

[0043] In the above embodiment, accurate and rapid dynamic prediction of power plant flue gas parameters is achieved through historical power plant operating parameters and real-time power plant operating parameters, so that the large inertia, large delay and large lag of power plant atmospheric pollutant parameters are effectively solved, and optimization guidance and suggestions are provided for power plant atmospheric pollutant control operators.

[0044] Optionally, as an embodiment of the present invention, the multiple original data sets include three types, namely, an original data set for predicting the NOx concentration of the flue gas at the inlet of the denitrification reactor, an original data set for predicting the SO2 concentration of the flue gas at the inlet of the desulfurization tower, and an original data set for predicting the flue gas flow rate at the inlet of the desulfurization tower, wherein:

[0045] The independent variables of the original data set for predicting the NOx concentration of the flue gas at the inlet of the denitrification reactor are composed of the operating parameters of the power plant boiler at multiple moments, and the dependent variable is composed of the NOx concentration of the flue gas at the inlet of the denitrification reactor at the corresponding moment.

[0046] The independent variables of the original data set for predicting the SO2 concentration of the flue gas at the desulfurization tower inlet are composed of the power plant boiler operating parameters and the denitrification reactor operating parameters at multiple times, and the dependent variable is composed of the SO2 concentration of the flue gas at the desulfurization tower inlet at the corresponding time.

[0047] The independent variables of the original data set used to predict the flue gas flow rate at the desulfurization tower inlet are composed of the power plant boiler operating parameters, denitrification reactor operating parameters, and desulfurization tower operating parameters at multiple times, and the dependent variable is composed of the flue gas flow rate at the desulfurization tower inlet at the corresponding time.

[0048] It should be understood that according to the prediction target, the variables in the original data set are divided into independent variables and dependent variables, where the dependent variables are the NOx concentration at the SCR inlet, the SO2 concentration in the raw flue gas of the absorber, and the flue gas flow rate of the raw flue gas of the absorber, and the remaining variables are independent variables.

[0049] Specifically, the original data set for predicting the NOx concentration of the flue gas at the inlet of the denitrification reactor includes independent variables consisting of the power plant boiler operating parameters at multiple times and the dependent variable consisting of the NOx concentration of the flue gas at the inlet of the denitrification reactor at the corresponding times; the original data set for predicting the SO2 concentration of the flue gas at the inlet of the desulfurization tower includes independent variables consisting of the power plant boiler operating parameters and the denitrification reactor operating parameters at multiple times and the dependent variable consisting of the SO2 concentration of the flue gas at the inlet of the desulfurization tower at the corresponding times; the original data set for predicting the flue gas flow rate at the inlet of the desulfurization tower includes independent variables consisting of the power plant boiler operating parameters, the denitrification reactor operating parameters, and the desulfurization tower operating parameters at multiple times and the dependent variable consisting of the flue gas flow rate at the inlet of the desulfurization tower at the corresponding times.

[0050] In the above embodiment, accurate and rapid dynamic prediction of power plant flue gas parameters is achieved through the original data set, so that the large inertia, large delay and large lag phenomenon of power plant atmospheric pollutant parameters are effectively solved.

[0051] Optionally, as an embodiment of the present invention, the power plant boiler operating parameters include unit load, total boiler air volume, boiler pressure, main steam pressure, main steam flow, main steam temperature, reheat steam pressure, reheat steam flow, reheat steam temperature, superheater outlet temperature, superheater outlet oxygen content, cold primary air flow, cold primary air pressure, cold primary air temperature, hot primary air flow, hot primary air pressure, hot primary air temperature, secondary air flow, secondary air pressure, secondary air temperature, coal feed rate of coal mill, secondary air door opening and SOFA air door opening;

[0052] The denitration reactor operation parameters include the denitration reactor inlet flue gas NOx concentration, the denitration reactor inlet flue gas O2 concentration, the denitration reactor inlet flue gas temperature, the denitration reactor inlet flue gas pressure, the denitration reactor inlet flue gas flow rate and the denitration reactor outlet flue gas NOx concentration;

[0053] The desulfurization tower operating parameters include the flue gas temperature at the desulfurization tower inlet, the flue gas pressure at the desulfurization tower inlet, the SO2 concentration at the desulfurization tower inlet, the O2 concentration at the desulfurization tower inlet, the flue gas temperature at the desulfurization tower outlet, the flue gas pressure at the desulfurization tower outlet, the flue gas flow rate at the desulfurization tower outlet, the desulfurization tower liquid level, the desulfurization tower slurry pH, the gypsum slurry density, the circulating slurry pump current and the circulating slurry pump outlet pressure.

[0054] In the above embodiment, accurate and rapid dynamic prediction of the flue gas parameters of the power plant is achieved through the power plant boiler operating parameters, the denitrification reactor operating parameters and the desulfurization tower operating parameters, so that the large inertia, large delay and large lag phenomena of the power plant atmospheric pollutant parameters are effectively solved, and optimization guidance and suggestions are provided for the power plant atmospheric pollutant control operators. At the same time, the effect of pre-control and elimination of lag is achieved, which is of great significance to the optimization control of atmospheric pollutants in power plants and energy conservation and emission reduction.

[0055] Optionally, as an embodiment of the present invention, the process of preprocessing the plurality of original data sets respectively to obtain the corresponding plurality of preprocessed data sets includes:

[0056] The independent variables of the three original data sets were integrated, replaced and deleted with the same type of parameters to obtain three simplified data sets;

[0057] The three simplified data sets are respectively subjected to missing value processing, outlier processing and purge point processing to obtain three preprocessed data sets;

[0058] The missing value processing adopts the deletion method;

[0059] The outlier processing includes identification and deletion of illogical parameters;

[0060] The purge point processing includes purge point identification and deletion, specifically:

[0061] When the absolute value of the difference between the dependent variables of two adjacent moments is greater than a set value, the previous moment is identified as the starting moment of the purge point, and the simplified data set within the set time period before and after the starting moment of the purge point is deleted.

[0062] It should be understood that the independent variables belonging to the power plant boiler operating parameters, the denitrification reactor operating parameters and the desulfurization tower operating parameters are respectively integrated, replaced and deleted with the same type of parameters to obtain three simplified data sets, which include a simplified data set of power plant boiler operating parameters, a simplified data set of denitrification reactor operating parameters and a simplified data set of desulfurization tower operating parameters.

[0063] Specifically, the purge point treatment is as follows:

[0064] Determine whether the absolute value of the difference between the simplified data after outlier processing at the current moment and the simplified data after outlier processing at the previous moment or the simplified data after outlier processing at the next moment is greater than the set value. If so, delete all simplified data after outlier processing within the difference between the current moment and the set time period to the sum of the current moment and the set time period to obtain the preprocessed data set.

[0065] Specifically, the purge point is handled as follows:

[0066] (1) When the absolute value of the difference between the NOx, SO2, and flue gas volume data at the previous and next time points is greater than a certain set value, it is determined that the time point is in the purge process;

[0067] (2) During a period of time before and after the purge process, the flue gas NOx, SO2, and flue gas volume data displayed by the instrument are distorted, so the data points within a certain period of time before and after the purge point are also treated as abnormal values.

[0068] In the above embodiment, multiple original data sets are preprocessed respectively to obtain corresponding multiple preprocessed data sets, and the timestamps of the respective variables and the dependent variables can be adjusted so that the time of the respective variables and the dependent variables are synchronized, achieving the effect of pre-control and eliminating lag, which is of great significance to the optimization control of atmospheric pollutants in power plants and energy conservation and emission reduction.

[0069] Optionally, as an embodiment of the present invention, the process of performing delay analysis on the plurality of preprocessed data sets to obtain the corresponding plurality of delay processed data sets includes:

[0070] The mutual information between each independent variable and the dependent variable at each moment in the set time period is calculated for the three preprocessed data sets, and the moment with the largest mutual information with the dependent variable is selected as the delay compensation value to unify the timestamps of each independent variable and the dependent variable, thus obtaining three delayed processed data sets.

[0071] It should be understood that mutual information (MI) is a metric in information theory that is used to quantify the dependency between two random variables. It measures the amount of information one random variable contains about another random variable.

[0072] Specifically, the mutual information method is used to calculate the delay value of each variable and the dependent variable in the data set, and the delay value is used to adjust the timestamps of each variable and the dependent variable so that the time of each variable and the dependent variable is synchronized. The calculation formula of the mutual information method is:

[0073]

[0074] In the formula, x is the independent variable, y is the dependent variable, X and Y are the sets of independent variables x and dependent variables y respectively, and I(X,Y) is the mutual information value between x and y. Using the above formula, the mutual information between the independent variable and the dependent variable at different time points can be calculated, and then the time delay with the largest mutual information can be found to obtain the optimal delay value between the independent variable and the dependent variable.

[0075] It should be understood that the independent variables corresponding to all delay compensation values ​​are collected as a delay-processed data set.

[0076] In the above embodiment, the preprocessed data sets are subjected to delay analysis to obtain corresponding multiple delayed processed data sets, and the timestamps of the respective variables and the dependent variables can be adjusted to synchronize the time of the respective variables and the dependent variables, thereby achieving the effect of pre-control and elimination of lag, which is of great significance to the optimization control of atmospheric pollutants in power plants and energy conservation and emission reduction.

[0077] Optionally, as an embodiment of the present invention, the process of performing feature analysis on the plurality of delayed processed data sets respectively to obtain the corresponding plurality of feature data sets includes:

[0078] S1: The three delayed processed data sets were evaluated using four algorithm models, including the Random Forest algorithm model, the GBDT gradient boosting decision tree, the XGBoost extreme gradient boosting decision tree, and the LightBGM lightweight gradient boosting machine;

[0079] S2: Select the algorithm model with the highest model evaluation score corresponding to the three delayed processed data sets respectively, and use the selected algorithm model to rank the feature importance of the corresponding delayed processed data sets, and select several features with the highest feature importance ranking from the three delayed processed data sets respectively, to obtain three data sets after simplified features;

[0080] S3: Use the KL divergence method to perform correlation analysis on the three simplified feature data sets, and retain several features with high KL divergence values ​​to form three feature data sets.

[0081] Specifically, the ensemble learning algorithm adopts four algorithms, namely Random Forest, Gradient Boosting Decision Tree (GBDT), Extreme Gradient Boosting Decision Tree (XGBoost), and Lightweight Gradient Boosting Machine (LightBGM). The optimal model is used to evaluate the importance of each variable, and then combined with actual process experience, features are selected according to their importance.

[0082] It should be understood that each delayed processed data set is evaluated using four algorithm models, and the algorithm model with the highest model evaluation score is selected from the four algorithm models corresponding to each delayed processed data set. The selected algorithm model is used to rank the feature importance of the corresponding delayed processed data set, and several features with high feature importance rankings are selected from the three delayed processed data sets to obtain three data sets with simplified features.

[0083] Specifically, after feature selection, the correlation between each feature is calculated. For several highly correlated features, only one feature is retained, thereby further reducing the data dimension and obtaining a feature data set. The correlation between each feature is calculated using KL divergence, and the formula is as follows:

[0084]

[0085] Where P(x) and Q(x) are the probability distributions of the two variables respectively, and KL(P||Q) is the divergence value of the two distributions. The closer the KL divergence between the two probability distributions P and Q is to 0, the higher the correlation between P and Q.

[0086] In the above embodiment, feature analysis is performed on multiple delayed processed data sets to obtain corresponding multiple feature data sets, which realizes accurate and rapid dynamic prediction of power plant flue gas parameters, effectively solves the large inertia, large delay and large lag phenomenon of power plant atmospheric pollutant parameters, and provides optimization guidance and suggestions for power plant atmospheric pollutant control operators. At the same time, it also achieves the effect of pre-control and elimination of lag, which is of great significance to the optimization control of atmospheric pollutants in power plants and energy conservation and emission reduction.

[0087] Optionally, as an embodiment of the present invention, the process of training the plurality of feature data sets through the LSTM long short-term memory network model to obtain the corresponding plurality of prediction models includes:

[0088] The three feature data sets are respectively split into three corresponding training sets and a validation set and a test set corresponding to each of the training sets;

[0089] The LSTM long short-term memory network model is used to train three training sets respectively, and the random search algorithm is used to tune the model hyperparameters in the validation sets corresponding to the training sets. Then, the generalization ability of the corresponding models is evaluated using the test sets corresponding to the training sets, thereby obtaining three prediction models.

[0090] It should be understood that the characteristic data set is split into a training set, a validation set, and a test set, the minimum and maximum normalization method is used for data normalization, and the LSTM (long short-term memory neural network model) is used for regression between independent variables and dependent variables.

[0091] It should be understood that each feature dataset is split into a training set, a validation set, and a test set.

[0092] It should be understood that the normalization adopts the minimum and maximum normalization method, and the calculation formula is as follows:

[0093]

[0094] In the formula, x' is the normalized value, min(x) and max(x) are the minimum and maximum values ​​of each variable sample, respectively.

[0095] Specifically, the hyperparameters of the LSTM model are tuned to maximize the prediction accuracy of the model, and finally an optimized prediction model (i.e., prediction model) for NOx, SO2, and flue gas volume of power plant flue gas is obtained, wherein the hyperparameters include time step, number of hidden layers, number of neurons, weight initialization, activation function, optimization algorithm, learning rate, number of traversals, batch size, and regularization. The hyperparameter tuning adopts the random search algorithm.

[0096] It should be understood that the random search algorithm is an optimization method based on random numbers, which gradually approaches the optimal solution by continuously randomly generating random points within the allowed variation range of the variables and calculating the objective function values ​​of these points. This method is particularly suitable for complex problems that are difficult to solve by traditional optimization methods, such as nonlinear, non-convex, non-differentiable or high-dimensional problems.

[0097] In the above embodiment, the feature data set is trained through the LSTM long short-term memory network model to obtain corresponding multiple prediction models, which realizes accurate and rapid dynamic prediction of the flue gas parameters of the power plant, so that the large inertia, large delay and large lag phenomenon of the atmospheric pollutant parameters of the power plant are effectively solved, and optimization guidance and suggestions are provided for the atmospheric pollutant control operators of the power plant. At the same time, the effect of pre-control and elimination of lag is achieved, which is of great significance to the optimization control of atmospheric pollutants in power plants and energy conservation and emission reduction.

[0098] Optionally, as an embodiment of the present invention, the process of inputting the dynamically updated real-time power plant operation parameters into the multiple prediction models respectively to obtain the real-time prediction results of the flue gas parameters of the multiple power plants includes:

[0099] The dynamically updated real-time power plant operating parameters are respectively input into the three prediction models to obtain real-time prediction results of the NOx concentration of the flue gas at the denitrification reactor inlet, the SO2 concentration of the flue gas at the desulfurization tower inlet, and the flue gas flow rate at the desulfurization tower inlet.

[0100] It should be understood that the dynamically updated real-time power plant operating parameters are respectively input into the three prediction models to obtain real-time prediction results of the NOx concentration in the flue gas at the denitrification reactor inlet, the real-time prediction results of the SO2 concentration in the flue gas at the desulfurization tower inlet, and the real-time prediction results of the flue gas flow rate at the desulfurization tower inlet.

[0101] In the above embodiment, the dynamically updated real-time power plant operating parameters are respectively input into multiple prediction models to obtain real-time prediction results of multiple power plant flue gas parameters, thereby achieving accurate and rapid dynamic prediction of power plant flue gas parameters, and effectively solving the large inertia, large delay and large lag phenomena of power plant atmospheric pollutant parameters, providing optimization guidance and suggestions for power plant atmospheric pollutant control operators.

[0102] Optionally, as another embodiment of the present invention, the present invention is based on the existing thermal control detection equipment to quickly obtain easily measurable auxiliary variables, and then use these auxiliary variables to accurately estimate or predict key variables such as NOx, SO2, flue gas volume, etc. that are difficult to obtain in real time through machine learning modeling or mechanism modeling. This has obvious advantages in the real-time, accuracy and cost of power plant flue gas pollutant control. Therefore, it is of great significance to establish a flexible and efficient modeling method for predicting NOx, SO2, and flue gas volume.

[0103] Optionally, as another embodiment of the present invention, the steps of the present invention are as follows:

[0104] 1. Data selection

[0105] 1.1. According to the generation of NOx and SO2 and the reaction mechanism of desulfurization and denitrification, 154 related auxiliary variables were selected from the boiler side, denitrification side and desulfurization side.

[0106] 1.2. Extract historical data of auxiliary variables from the power plant DCS, PLC, CEMS and other systems at equal time intervals to construct the original data set.

[0107] 1.3. According to the prediction target, the variables in the original data set are divided into independent variables and dependent variables, where the dependent variables are the NOx concentration at the SCR inlet, the SO2 concentration in the original flue gas of the absorption tower, and the flue gas flow rate of the original flue gas of the absorption tower, and the remaining variables are independent variables.

[0108] 2. Data preprocessing

[0109] 2.1. Integrate and calculate the independent variables of the same type in the original data set, reduce the dimension of the original data set, and construct a new data set.

[0110] 2.2. The new data sets are preprocessed by missing value processing, outlier processing, purge point processing, etc.

[0111] 3. Data delay calculation

[0112] The mutual information method is used to calculate the delay value between each variable and the dependent variable in the data set, and the delay value is used to adjust the timestamps of each variable and the dependent variable so that the time of each variable and the dependent variable is synchronized.

[0113] 4. Data feature engineering

[0114] 4.1. Split the data set into training set and test set, and perform data normalization. Use the integrated learning algorithm model to regress the independent variables and dependent variables, and perform model evaluation to select the optimal model.

[0115] 4.2. Use the optimal model to evaluate the importance of each variable, and then combine it with actual process experience to select features based on their importance.

[0116] 4.3. After feature selection, the correlation between the features is calculated. For several highly correlated features, only one of them is retained, thereby further reducing the data dimension and obtaining a feature data set.

[0117] 5. Model prediction

[0118] 5.1. Split the feature data set into a training set and a validation set, use the minimum and maximum normalization method to normalize the data, and use LSTM (long short-term memory neural network model) to perform regression between independent variables and dependent variables.

[0119] 5.2. Perform hyperparameter tuning of the LSTM model to maximize the prediction accuracy of the model and ultimately obtain an optimized prediction model for power plant flue gas NOx, SO2, and flue gas volume.

[0120] 5.3. The auxiliary variables obtained in real time are used as test set data, dynamically updated and input into the above-mentioned optimization prediction model to quickly obtain dynamic prediction values ​​of target parameters such as flue gas NOX, SO2, flue gas volume, etc., providing a basis for subsequent pollutant control.

[0121] Optionally, as another embodiment of the present invention, the present invention can use the existing easy-to-measure auxiliary variable real-time data to realize accurate and rapid dynamic prediction of key parameters such as NOx, SO2, flue gas volume, etc. of power plant flue gas, so that the large inertia, large delay, and large lag phenomenon of power plant atmospheric pollutant parameters are effectively solved, and optimization guidance suggestions are provided for power plant atmospheric pollutant control operators. The actual application results of the present invention in power plants show that the NOx, SO2, and flue gas volume prediction method based on machine learning has high measurement accuracy and good generalization performance. The prediction accuracy of NOx, SO2, and flue gas volume is 99.1%, 98.7%, and 99.2%, respectively, and the prediction root mean square error is 0.0050, 0.0187, and 0.0153, respectively, which meets the measurement requirements of NOx, SO2, and flue gas volume. At the same time, the acquisition time of the predicted values ​​of NOx, SO2, and flue gas volume is 10 to 40 seconds earlier than the traditional instrument monitoring value, achieving the effect of pre-control and eliminating lag, which is of great significance to the optimization control of atmospheric pollutants in power plants and energy conservation and emission reduction.

[0122] Figure 2A module block diagram of a power plant flue gas parameter prediction device provided in an embodiment of the present invention.

[0123] Optionally, as another embodiment of the present invention, Figure 2 As shown, a device for predicting flue gas parameters of a power plant includes:

[0124] Import module, used to import historical power plant operating parameters corresponding to multiple moments;

[0125] A data set construction module is used to construct multiple original data sets through historical power plant operation parameters corresponding to multiple moments;

[0126] A preprocessing module, used to preprocess the multiple original data sets to obtain corresponding multiple preprocessed data sets;

[0127] A delay processing module, used for performing delay analysis on the plurality of pre-processed data sets to obtain a plurality of corresponding delay processed data sets;

[0128] A feature analysis module, used to perform feature analysis on the plurality of delayed processed data sets to obtain a plurality of corresponding feature data sets;

[0129] A model analysis module is used to train the plurality of feature data sets through an LSTM long short-term memory network model to obtain a plurality of corresponding prediction models;

[0130] The prediction result acquisition module is used to input the dynamically updated real-time power plant operation parameters into the multiple prediction models respectively, so as to obtain the real-time prediction results of the flue gas parameters of the multiple power plants.

[0131] Optionally, another embodiment of the present invention provides a power plant flue gas parameter prediction system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the power plant flue gas parameter prediction method as described above is implemented. The system may be a computer or other system.

[0132] Optionally, another embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the power plant flue gas parameter prediction method as described above is implemented.

[0133] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0134] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0135] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0136] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of the present invention.

[0137] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0138] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0139] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting flue gas parameters in a power plant, characterized in that: The steps include: Import the historical power plant operation parameters corresponding to multiple time periods, and construct multiple original data sets through the historical power plant operation parameters corresponding to multiple time periods; Preprocessing the multiple original data sets respectively to obtain corresponding multiple preprocessed data sets; Performing time-delay analysis on the plurality of pre-processed data sets respectively to obtain a plurality of corresponding time-delay processed data sets; Performing feature analysis on the plurality of delayed processed data sets respectively to obtain a plurality of corresponding feature data sets; The plurality of feature data sets are trained respectively by using an LSTM long short-term memory network model to obtain corresponding plurality of prediction models; The dynamically updated real-time power plant operating parameters are respectively input into the plurality of prediction models, thereby obtaining real-time prediction results of the flue gas parameters of the plurality of power plants.

2. The power plant flue gas parameter prediction method according to claim 1, characterized in that: The historical power plant operating parameters and the real-time power plant operating parameters include power plant boiler operating parameters, denitrification reactor operating parameters and desulfurization tower operating parameters.

3. The power plant flue gas parameter prediction method according to claim 1, characterized in that: There are three types of the multiple original data sets, namely, an original data set for predicting the NOx concentration of the flue gas at the inlet of the denitrification reactor, an original data set for predicting the SO2 concentration of the flue gas at the inlet of the desulfurization tower, and an original data set for predicting the flue gas flow rate at the inlet of the desulfurization tower, wherein: The independent variables of the original data set for predicting the NOx concentration of the flue gas at the inlet of the denitrification reactor are composed of the operating parameters of the power plant boiler at multiple moments, and the dependent variable is composed of the NOx concentration of the flue gas at the inlet of the denitrification reactor at the corresponding moment. The independent variables of the original data set for predicting the SO2 concentration of the flue gas at the desulfurization tower inlet are composed of the power plant boiler operating parameters and the denitrification reactor operating parameters at multiple times, and the dependent variable is composed of the SO2 concentration of the flue gas at the desulfurization tower inlet at the corresponding time. The independent variables of the original data set used to predict the flue gas flow rate at the desulfurization tower inlet are composed of the power plant boiler operating parameters, denitrification reactor operating parameters, and desulfurization tower operating parameters at multiple times, and the dependent variable is composed of the flue gas flow rate at the desulfurization tower inlet at the corresponding time.

4. The power plant flue gas parameter prediction method according to claim 3, characterized in that: The power plant boiler operating parameters include unit load, total boiler air volume, boiler pressure, main steam pressure, main steam flow, main steam temperature, reheat steam pressure, reheat steam flow, reheat steam temperature, superheater outlet temperature, superheater outlet oxygen content, cold primary air flow, cold primary air pressure, cold primary air temperature, hot primary air flow, hot primary air pressure, hot primary air temperature, secondary air flow, secondary air pressure, secondary air temperature, coal feed rate of coal mill, secondary air door opening and SOFA air door opening; The denitration reactor operation parameters include the denitration reactor inlet flue gas NOx concentration, the denitration reactor inlet flue gas O2 concentration, the denitration reactor inlet flue gas temperature, the denitration reactor inlet flue gas pressure, the denitration reactor inlet flue gas flow rate and the denitration reactor outlet flue gas NOx concentration; The desulfurization tower operating parameters include the flue gas temperature at the desulfurization tower inlet, the flue gas pressure at the desulfurization tower inlet, the SO2 concentration at the desulfurization tower inlet, the O2 concentration at the desulfurization tower inlet, the flue gas temperature at the desulfurization tower outlet, the flue gas pressure at the desulfurization tower outlet, the flue gas flow rate at the desulfurization tower outlet, the desulfurization tower liquid level, the desulfurization tower slurry pH, the gypsum slurry density, the circulating slurry pump current and the circulating slurry pump outlet pressure.

5. The power plant flue gas parameter prediction method according to claim 4, characterized in that: The process of preprocessing the plurality of original data sets respectively to obtain the corresponding plurality of preprocessed data sets comprises: The independent variables of the three original data sets were integrated, replaced and deleted with the same type of parameters to obtain three simplified data sets; The three simplified data sets are respectively subjected to missing value processing, outlier processing and purge point processing to obtain three preprocessed data sets; The missing value processing adopts the deletion method; The outlier processing includes identification and deletion of illogical parameters; The purge point processing includes purge point identification and deletion, specifically: When the absolute value of the difference between the dependent variables of two adjacent moments is greater than a set value, the previous moment is identified as the starting moment of the purge point, and the simplified data set within the set time period before and after the starting moment of the purge point is deleted.

6. The power plant flue gas parameter prediction method according to claim 5, characterized in that: The process of performing delay analysis on the plurality of preprocessed data sets respectively to obtain the corresponding plurality of delay processed data sets comprises: The mutual information between each independent variable and the dependent variable at each moment in the set time period is calculated for the three preprocessed data sets, and the moment with the largest mutual information with the dependent variable is selected as the delay compensation value to unify the timestamps of each independent variable and the dependent variable, thus obtaining three delayed processed data sets.

7. The power plant flue gas parameter prediction method according to claim 6, characterized in that: The process of performing feature analysis on the plurality of delayed processed data sets to obtain the corresponding plurality of feature data sets comprises: S1: The three delayed processed data sets were evaluated using four algorithm models, including the Random Forest algorithm model, the GBDT gradient boosting decision tree, the XGBoost extreme gradient boosting decision tree, and the LightBGM lightweight gradient boosting machine; S2: Select the algorithm model with the highest model evaluation score corresponding to the three delayed processed data sets respectively, and use the selected algorithm model to rank the feature importance of the corresponding delayed processed data sets, and select several features with the highest feature importance ranking from the three delayed processed data sets respectively, to obtain three data sets after simplified features; S3: Use the KL divergence method to perform correlation analysis on the three simplified feature data sets, and retain several features with high KL divergence values ​​to form three feature data sets.

8. The power plant flue gas parameter prediction method according to claim 7, characterized in that: The process of training the plurality of feature data sets through the LSTM long short-term memory network model to obtain the corresponding plurality of prediction models includes: The three feature data sets are respectively split into three corresponding training sets and a validation set and a test set corresponding to each of the training sets; The LSTM long short-term memory network model is used to train three training sets respectively, and the random search algorithm is used to tune the model hyperparameters in the validation sets corresponding to the training sets. Then, the generalization ability of the corresponding models is evaluated using the test sets corresponding to the training sets, thereby obtaining three prediction models.

9. The power plant flue gas parameter prediction method according to claim 8, characterized in that: The process of inputting the dynamically updated real-time power plant operating parameters into the plurality of prediction models respectively, thereby obtaining the real-time prediction results of the flue gas parameters of the plurality of power plants comprises: The dynamically updated real-time power plant operating parameters are respectively input into the three prediction models to obtain real-time prediction results of the NOx concentration of the flue gas at the denitrification reactor inlet, the SO2 concentration of the flue gas at the desulfurization tower inlet, and the flue gas flow rate at the desulfurization tower inlet.

10. A device for predicting flue gas parameters in a power plant, characterized in that: include: Import module, used to import historical power plant operating parameters corresponding to multiple moments; A data set construction module is used to construct multiple original data sets through historical power plant operation parameters corresponding to multiple moments; A preprocessing module, used to preprocess the multiple original data sets to obtain corresponding multiple preprocessed data sets; A delay processing module, used for performing delay analysis on the plurality of pre-processed data sets to obtain a plurality of corresponding delay processed data sets; A feature analysis module, used to perform feature analysis on the plurality of delayed processed data sets to obtain a plurality of corresponding feature data sets; A model analysis module is used to train the plurality of feature data sets through an LSTM long short-term memory network model to obtain a plurality of corresponding prediction models; The prediction result acquisition module is used to input the dynamically updated real-time power plant operation parameters into the multiple prediction models respectively, so as to obtain the real-time prediction results of the flue gas parameters of the multiple power plants.

Images